arXiv CSopen access

Adaptation Interfaces for In-Context Tabular Foundation Models in Time-to-Event Prediction

· arxiv_cs

arXiv CS · Papers · License: Open Access

machine learning, deep learning, neural networks

Adaptation Interfaces for In-Context Tabular Foundation Models in Time-to-Event Prediction Minh-Khoi Pham∗ [email protected] ADAPT Centre, School of Computing, Dublin City University DCU Glasnevin Campus, Dublin 9, D09 V209, Ireland

arXiv:2609.04901v1 [cs.LG] 4 Sep 2026

Luca Cotugno, Dan Cernei, Alina Sîrbu {luca.cotugno, dan.cernei}@studio.unibo.it, [email protected] Department of Computer Science and Engineering, University of Bologna, Italy

Stefano Masi, Giuseppe Prencipe {stefano.masi, giuseppe.prencipe}@unipi.it University of Pisa, Italy

Alessandro Pingitore, Patrizia Landi {alessandro.pingitore, patrizia.landi}@cnr.it Institute of Clinical Physiology, CNR, Pisa, Italy

Working Group on Uric Acid and Cardiovascular Risk of the Italian Society of Hypertension Tai Tan Mai, Martin Crane, Marija Bezbradica {tai.mai, martin.crane, marija.bezbradica}@dcu.ie School of Computing, Dublin City University, Ireland

Abstract Tabular foundation models (TabFMs) achieve strong performance on structured data, particularly for standard classification and regression problems. Yet, extending them to censored time-to-event prediction is challenging because it requires properly handling censoring and event-time dynamics. Building on our prior work, we further link TabFMs with CoxPH and DeepHit and revise the context-resampled training procedure. We evaluate temporal zero-shot reformulation, classification-based fine-tuning, and survival-head adaptation using frozen TabFM backbones on 74 single-risk data sets, and we additionally study 4 competingrisk data sets. Zero-shot inference is effective on smaller single-risk data sets, whereas supervised adaptation becomes increasingly advantageous as data sets scale. Cox provides the most reliably strong interface, especially for Integrated Brier Score (IBS) on larger data sets. DeepHit is relatively stronger for the time-dependent Concordance Index (Ctd ) than for IBS, while cause-specific MTLR ranks highest among the TabFM survival heads in the four-data-set competing-risk analysis. Classification fine-tuning becomes more competitive with zero-shot inference as data sets grow but remains weaker for probabilistic prediction. Overall, our results indicate that effective TabFM transfer depends on the data regime and on the statistical structure represented by the chosen adaptation interface. The implementation scripts used for this work are available at https://github.com/kaylode/survival-fm. Keywords: tabular foundation models, in-context learning, time-to-event prediction, survival analysis, transfer learning, competing risks ∗. Corresponding author

©2026 Minh-Khoi Pham, Luca Cotugno, Dan Cernei, Alina Sîrbu, Stefano Masi, Giuseppe Prencipe, Alessandro Pingitore, Patrizia Landi, Working Group on Uric Acid and Cardiovascular Risk of the Italian Society of Hypertension, Tai Tan Mai, Martin Crane, and Marija Bezbradica. License: CC-BY 4.0, see https://creativecommons.org/licenses/by/4.0/.

Minh-Khoi Pham et al.

1 Introduction Time-to-event prediction addresses not only if an event happens, but also the timing of that event, and is widely applied in domains such as healthcare, finance, and marketing, where making predictions from heterogeneous tabular data remains an ongoing challenge (Chi et al., 2026). Traditional survival-analysis methods, including Kaplan–Meier estimation and Cox proportional hazards (CoxPH), continue to serve as key baselines (Kaplan and Meier, 1958; Cox, 1972), but they provide limited flexibility for capturing complex relationships between covariates and time. As a result, a wide variety of machine-learning methods have been developed for censored time-to-event prediction. Tabular foundation models (TabFMs), including TabPFN (Hollmann et al., 2023), TabDPT (Ma et al., 2025), and TabICL (Qu et al., 2025), provide pretrained priors that transfer across heterogeneous tabular prediction tasks. Their standard interfaces, however, assume fully observed classification or regression targets. Time-to-event prediction violates this interface in several ways: outcomes may be censored, prediction targets depend on time, and competing-risk settings involve multiple mutually exclusive event types. The challenge is therefore not simply whether a TabFM can be used for time-to-event prediction, but how its pretrained prior should be adapted to a downstream task with fundamentally different statistical structure. This calls for learning objectives or prediction mechanisms that differ from standard classification and regression formulations (Kalbfleisch and Prentice, 2002). Existing work has approached this mismatch from two directions. One strategy reformulates survival prediction as censoring-aware classification over discretized time horizons, allowing pretrained tabular classifiers to be used directly without changing their underlying task interface (Kim et al., 2026). This provides a convenient route for generic classifiers, but introduces subject–time expansion, class imbalance, dependence on the choice of discretization horizons, and the native interface constraints of the underlying classifier. Another direction incorporates time-to-event structure during pretraining itself, producing prior-fitted foundation models specifically designed for survival inference (Seletkov et al., 2026; Qi et al., 2026). In our previous work (Pham et al., 2026a), we established multi-task logistic regression (MTLR)-head adaptation over frozen TabFM representations and evaluated it on standard public clinical survival benchmarks and two large ICU cohorts. That study was clinically focused, single-risk, and restricted to MTLR as the survival-specific TabFM head. The present work treats that MTLR interface as a prior baseline and asks how alternative adaptation interfaces behave across a substantially broader benchmark, additional survival objectives, probabilistic evaluation, and competing risks. In this work, we study time-to-event prediction with pretrained TabFMs as an adaptation interface problem under objective mismatch. We compare three interfaces for transferring a generic pretrained tabular prior to censored prediction: (i) the existing zero-shot horizon-wise reformulation; (ii) classification fine-tuning on temporally expanded survival examples; and (iii) survival-head adaptation, where CoxPH and DeepHit are compared directly with the previously established MTLR interface. In the supervised interfaces, the pretrained backbone remains frozen and adaptation is performed through the task-specific classification or survival head. Our central question is therefore: which adaptation interfaces best transfer generic pretrained TabFMs to time-to-event prediction?

TabFMs for Time-to-Event Prediction

Our main contributions are: • We formulate the transfer of generic TabFMs to time-to-event prediction as an adaptation-interface problem and systematically compare existing zero-shot reformulation, censoring-aware classification fine-tuning, and survival-head adaptation across a broad multi-domain benchmark. • We extend survival-head adaptation beyond the previously studied MTLR interface to CoxPH and DeepHit under context-resampled, context-conditioned training, and compare these heads in terms of discrimination and probabilistic prediction across data regimes while retaining MTLR as a reference baseline. • We extend the analysis to competing risks, comparing survival-head interfaces only, and distinguish cause-specific adaptation from jointly normalized cause–time modeling to test whether the conclusions from single-risk transfer persist when multiple event types are present.

2 Related Work 2.1 Modern Survival Modeling Survival analysis has progressed from classical statistical methods such as Kaplan–Meier estimation, CoxPH, and tree-based survival models toward increasingly flexible machine-learning approaches for censored time-to-event prediction (Cox, 1972; Ishwaran et al., 2008). Deep survival models including DeepSurv, MTLR, DeepHit, and related continuous- and discretetime extensions expand this design space by relaxing proportional-hazards assumptions, modeling individualized event-time distributions, and supporting competing risks through specialized objectives (Katzman et al., 2018; Yu et al., 2011; Kvamme et al., 2019; Lee et al., 2018; Nagpal et al., 2021). More recent architectures incorporate transformers, attention mechanisms, latent-variable formulations, and longitudinal encoders (Wang et al., 2022; Mesinovic et al., 2026). These methods show the importance of survival-specific statistical structure, but they are predominantly trained from scratch for individual tasks. 2.2 Foundation Models for Tabular Learning TabFMs shift structured-data prediction from fitting a separate model for each data set toward broad pretraining followed by in-context prediction. The present benchmark uses three representative pretrained backbones with deliberately different pretraining and inference designs. TabPFN v2 is a prior-fitted transformer pretrained on synthetic tabular tasks and performs prediction by conditioning on labeled examples supplied in context (Hollmann et al., 2025). TabDPT combines real-data pretraining with retrieval-based conditioning, explicitly using retrieved examples to support transfer to unseen tables (Ma et al., 2025). TabICL v1 uses a two-stage architecture that first builds fixed-dimensional row embeddings through column-then-row processing and then applies a transformer for scalable in-context learning (Qu et al., 2025). These are the exact model generations used in our experiments; for ease of reading, we refer to them hereafter without version numbers: TabPFN, TabDPT, and TabICL. 3

Minh-Khoi Pham et al.

The appeal of these models is that a reusable pretrained prior can reduce task-specific optimization and perform strongly in small- and medium-data regimes. At the same time, recent broader evaluations caution against treating this advantage as universal. BeyondArena finds that current TabFMs are particularly strong on tiny- to medium-sized IID problems, whereas conventional tree ensembles and supervised deep models can regain an advantage on non-IID, large, or high-dimensional data sets (Purucker et al., 2026). Adaptation and retrieval have therefore become increasingly important for extending pretrained tabular priors beyond their most favorable operating regime (Liu and Ye, 2025; Pham et al., 2026b). Native in-context interfaces can also impose architectural constraints; for example, TabFM’s native classifier is limited to a fixed number of output classes in its released interface (Kong and Das, 2026). Such constraints motivate distinguishing direct reuse of a pretrained prediction interface from adaptation strategies that retain the backbone while replacing the downstream head. Their standard objectives, however, still do not directly represent censoring or event-time distributions. 2.3 Foundation Models for Censored Time-to-Event Prediction Recent work addresses the mismatch between generic foundation model interfaces and censored time-to-event prediction through several distinct routes. Kim et al. (2026) reformulate survival analysis as a sequence of censoring-aware binary classification problems, allowing existing TabFMs to be used directly through their pretrained classification interface. Because this route leaves the native classifier unchanged, it also inherits backbone-specific interface restrictions; the horizon-wise formulation itself is binary, but the broader output interface remains fixed by the pretrained classifier. In contrast, Survival In-Context (SIC) (Seletkov et al., 2026) and SurvivalPFN (Qi et al., 2026) introduce survival structure during prior-fitted pretraining, amortizing survival inference directly rather than adapting a generic classificationpretrained model post hoc. These approaches demonstrate the value of incorporating event timing directly into the foundation model learning objective, but require survival- or time-toevent-aware pretraining rather than reusing an otherwise generic tabular foundation model unchanged. Our setting is complementary: rather than redesigning the pretraining distribution or requiring a survival-native foundation model, we study how generic pretrained TabFMs can be transferred through different downstream interfaces. Our earlier study (Pham et al., 2026a) introduced MTLR-based adaptation over frozen TabFM representations and evaluated it alongside zero-shot inference on public clinical survival benchmarks and two large ICU cohorts. In the present work, that MTLR interface is retained as a reference baseline rather than presented as a new contribution. We remove the ICU cohorts from the benchmark and broaden the study to 74 heterogeneous single-risk data sets, alternative CoxPH and DeepHit adaptation heads, censoring-aware classification fine-tuning, probabilistic evaluation with IBS, context-resampled training, data-regime analysis, and competing-risk formulations.

3 Methods We consider three transfer regimes: (i) zero-shot in-context reformulation, with no parameter updates; (ii) classification fine-tuning on temporally expanded examples; and (iii) survivalhead adaptation with CoxPH, DeepHit, or the previously established MTLR interface (Pham 4

TabFMs for Time-to-Event Prediction

et al., 2026a). The horizon-wise zero-shot mechanism follows prior work (Kim et al., 2026; Pham et al., 2026a). In both supervised regimes, the pretrained TabFM backbone remains frozen and only the task-specific classification or survival head is optimized. Temporal discretization and temporal expansion are distinct operations: classification adaptation expands each subject into multiple horizon-specific examples, whereas MTLR and DeepHit discretize the event-time distribution internally without horizon-wise data expansion. For discrete-time formulations, the time intervals are selected adaptively from the observed training event-time distribution. This allows finer temporal resolution when observed events are abundant while avoiding sparsely supervised bins on smaller data sets. 3.1 Problem Setup d We consider right-censored survival data D = {(xi , T̃i , δi )}N i=1 , where xi ∈ R denotes covariates, Ti is the event time, Ci is the censoring time, T̃i = min(Ti , Ci ) is the observed time, and δi = 1[Ti ≤ Ci ] indicates whether the event was observed. In competing-risk settings, we replace the binary event indicator with ∆i ∈ {0, 1, . . . , C}, where ∆i = 0 denotes censoring and ∆i = c denotes an observed event of cause c. The target is either the survival function S(t | x) = P (T > t | x) or, for cause c, the cumulative incidence function Fc (t | x) = P (T ≤ t, ∆ = c | x). To assess generalization while tuning model hyperparameters, we adopt a nested cross-validation scheme in which each outer fold is further split into training and validation subsets Dtrain and Dval , with the remaining outer fold reserved for testing Dtest . Let fθ⋆ denote a pretrained in-context learning TabFM. In zero-shot inference, θ⋆ remains fixed and predictions are obtained through the pretrained classification interface. For both classification fine-tuning and survival-head adaptation, the reported experiments also keep θ⋆ fixed but optimize a new task-specific head. The resulting representations remain context dependent because the pretrained TabFM is conditioned on sampled labeled support examples rather than used as a one-time deterministic feature extractor.

3.2 Context-Conditioned Training and Inference Context conditioning is shared by the classification and survival-head interfaces. Our earlier MTLR study conditioned each representation on the training cohort as a fixed support set (Pham et al., 2026a). Here, supervised training instead uses sampled support contexts and resamples them across optimization steps, making the downstream head robust to variation in the in-context support while preventing a query from appearing in its own context. Specifically, each query minibatch B ⊂ Dtrain is paired with train Dcontext ⊆ Dtrain \ B.

For survival-head adaptation, each sampled support subject is supplied to the pretrained backbone through its native labeled-context interface. Because the pretrained TabFMs expect classification-style labels rather than censored time-to-event targets, we encode the support label as the binary observed-event indicator ej = 1[δj > 0]. For a sampled support index set C, the query representation is therefore hi = fθ⋆ (xi ; {(xj , ej ) : j ∈ C}) . 5

Minh-Khoi Pham et al.

The observed time T̃j is not supplied as a context label to the pretrained backbone; event-time and censoring information instead enter through the downstream CoxPH, MTLR, or DeepHit objective used to optimize the survival head. In competing-risk settings, all observed causes are mapped to ej = 1, so cause identity is introduced only by the competing-risk survival head. Contexts are resampled across optimization steps, so even with frozen pretrained parameters the representation of the same query can vary under different labeled support samples. The classification-based adaptations use a different labeled context because they operate on temporally expanded subject–horizon examples. A generic classification backbone has no built-in representation of event time, so each classification query is augmented with fixed temporal features representing the requested horizon. We denote the resulting input by x̃ik . For single-risk data, support examples take the form (x̃jk , yjk ) ,

yjk = 1[δj = 1 ∧ T̃j ≤ tk ],

with censored subject–horizon pairs retained only while their status is known. Thus, the single-risk classification context label indicates whether an observed event has occurred by the queried horizon, whereas the survival-head interface uses only the subject-level observed-event indicator to condition the pretrained representation. The competing-risk extension uses multiclass cause labels as defined in Section 3.5. The detailed temporal expansion and validity mask are defined in Section 3.3. Validation queries come from Dval during fine-tuning and from Dtest during inference, while their support context is sampled only from the training set. Neither validation nor test observations are ever included in a support context, val Dcontext ⊆ Dtrain ,

test Dcontext ⊆ Dtrain .

The two classification-based interfaces use the temporally augmented input x̃ik but differ in whether a downstream head is learned. The underlying zero-shot mechanism is not new to this work: as in prior horizon-wise formulations (Kim et al., 2026; Pham et al., 2026a), expanded training examples act as labeled in-context demonstrations and no parameters are optimized. For a single-risk test query, p̂ik = gnative (fθ⋆ (x̃ik ; Dcontext )) ,

Ŝ(tk | xi ) = 1 − p̂ik ,

and the horizon-wise survival probabilities are interpolated to the evaluation grid. In classification fine-tuning, the pretrained backbone is frozen and a task-specific classification head is optimized with the censoring-aware objective defined in Section 3.3. Replacing the native head removes dependence on its fixed output dimensionality; the single-risk horizon-wise target is binary, while the competing-risk extension is multiclass. Survival-head adaptation uses the same train/validation context separation but operates on subject representations rather than subject–horizon classification queries. For a test subject x∗ ∈ Dtest , (r) h∗ = fθ⋆ x∗ ; {(xj , ej ) : j ∈ C (r) } , C (r) ⊆ Dtrain , (r)

and the fitted survival head produces Ŝ (r) (t | x∗ ), or F̂c (t | x∗ ) for competing risks. No model parameters are updated during validation or test inference. When context ensembling 6

TabFMs for Time-to-Event Prediction

is enabled, inference is repeated over separately sampled training contexts and the predictions are averaged, R 1 X (r) Ŝ (t | x∗ ), Ŝ(t | x∗ ) = R r=1

with analogous averaging applied independently to each cause-specific cumulative incidence function. The reported benchmarks use R = 5 context samples for both single-risk and competing-risk inference. The zero-shot interface averages over the same number of support samples, but through its native in-context mechanism: the pretrained classifier is fed with each of five stratified subsamples of the expanded context and the predicted probabilities are averaged. Zero-shot inference is also run per horizon, so a prediction requires one in-context fit per time bin (i.e., a query with 100 time bins requires 100 forward passes). For evaluations that require a scalar risk score, predicted trajectories are summarized after inference. In single-risk zero-shot inference, rZS (x) = 1 − Ŝ(tmax | x), whereas the supervised single-risk interfaces use the negative area under the predicted survival curve, Z rsup (x) = −

Ŝ(t | x) dt,

so shorter predicted survival corresponds to larger risk. For supervised competing-risk models, the analogous cause-specific score integrates each predicted cumulative incidence function over the evaluation grid and normalizes by the grid span. These scalar summaries are used only by evaluations that require a one-dimensional risk score; the reported Antolini Ctd is computed directly from the predicted survival trajectories. IBS is likewise computed from the survival or cumulative-incidence trajectories themselves. 3.3 Temporal Classification Reformulation The horizon-wise censoring-aware classification mechanism follows the existing reformulation literature (Kim et al., 2026; Pham et al., 2026a). For the present benchmark, we use an adaptive, event-balanced discretization rather than fixing a common temporal resolution across data sets. Let K = {10, 20, 30, 50, 100} denote candidate numbers of intervals. For each candidate K, quantile boundaries are computed from the observed (uncensored) event times and the number of events falling in each interval is counted. The procedure selects the largest feasible resolution (E) ⋆ K = max K : min nk ≥ m , K∈K

(E)

where nk is the number of observed events in interval k and m = 5 is the default minimum event support per interval. For the selected resolution K ⋆ , the initial quantile boundaries are j qj = QE , j = 0, . . . , K ⋆ , K⋆ 7

Minh-Khoi Pham et al.

where QE is the empirical quantile function of the observed event times. We additionally include t = 0 as the initial temporal coordinate and remove duplicate boundaries caused by tied event times. We therefore denote the effective ordered horizon set by ⋆ T = {t0 , . . . , tK−1 } = unique {0} ∪ {qj }K j=0 , where K = |T | is the effective number of temporal coordinates. This procedure provides finer temporal resolution when observed events are abundant and coarser resolution when event information is sparse, without tuning the number of bins separately for each data set. Because duplicate quantile boundaries are removed, the effective number of temporal coordinates K can be smaller than the number implied by the requested resolution K ⋆ . Each subject is then paired with each horizon to form time-augmented examples. For single-risk data, the binary target is yik = 1[δi = 1 ∧ T̃i ≤ tk ]. For an observed event, all horizons remain informative: horizons before the observed event time are negative and horizons at or after the event are positive. For a censored subject, horizons before censoring are known negatives, whereas later horizons have unknown event status and are excluded. The implementation therefore uses the validity mask mik = 1[δi = 1 ∨ T̃i > tk ]. The masked binary cross-entropy (BCE) objective is X LCE = mik BCE(p̂ik , yik ) . i,k

To control the size and class imbalance of the expanded classification data, all positive subject–horizon examples are retained and negative examples are randomly subsampled. With sampling ratio ρ, at most ρN+ negative examples are retained, where N+ denotes the number of positive expanded examples; the reported classification fine-tuning configuration uses ρ = 0.5. This subsampling is distinct from the event-balanced minibatch sampler used during optimization. This construction expands the data from N subjects to up to O(N K) subject–horizon examples before subsampling. Temporal discretization and temporal expansion are distinct: MTLR and DeepHit also use discretized event-time grids, but they optimize structured survival objectives without expanding subjects across horizon-specific classification examples, while CoxPH operates directly through continuous-time risk sets. 3.4 Survival Head Adaptation Survival-head adaptation freezes the pretrained TabFM parameters and optimizes a survivalspecific prediction head over context-conditioned representations. Our earlier work already established the MTLR variant (Pham et al., 2026a); we therefore retain MTLR as a reference baseline and focus the methodological comparison here on extending the same frozen-backbone interface to CoxPH and DeepHit. Gradients update only the survival-head parameters Φ 8

TabFMs for Time-to-Event Prediction

and are not propagated into θ⋆ . The head is a single hidden layer of 64 units with dropout, followed by the output parameterization required by the corresponding objective: a scalar logrisk for CoxPH, and a K ⋆ -dimensional logit vector for MTLR and DeepHit. CoxPH operates in continuous time. DeepHit is discrete-time and uses the adaptive event-balanced quantile discretization described above. For consistency in the benchmark, the MTLR baseline is evaluated under the same current data splits, preprocessing, context-resampling protocol, and adaptive discretization, but its standard single-risk formulation is not reintroduced here. 3.4.1 Cox Proportional Hazards (CoxPH) The survival head outputs a scalar log-risk ri = gΦ (hi ) and minimizes the negative partial log-likelihood   X X ri − log LCox = − exp(rj ) , R(Ti ) = {j : T̃j ≥ Ti }. i:δi =1

j∈R(Ti )

Censoring is handled through the event-indexed risk sets. 3.4.2 Single-Risk DeepHit For single-risk data, DeepHit predicts a discrete probability mass function pik = P (T ∈ Ik | xi ) over time intervals and uses the pycox DeepHitSingle objective implemented in our pipeline: LSR DH = αLNLL + (1 − α)Lrank . For an observed event in interval ki , the likelihood contribution is − log pi,ki ; for censoring in interval ki , it is the negative log of the remaining tail probability. The pairwise ranking term encourages subjects with earlier observed events to receive larger cumulative event probability at the corresponding horizon. Thus, the single-risk implementation combines discrete event-time likelihood and ranking supervision. 3.5 Competing-Risk Adaptation The competing-risk implementations use two distinct survival-head strategies that should not be conflated: cause-specific modeling for Cox and MTLR, and a joint cause–time distribution for DeepHit. As in the single-risk survival-head variants, the TabFM backbone remains frozen while the corresponding task-specific output layers are optimized. The horizon-wise classification interfaces admit a natural multiclass extension, described below for completeness, but we do not report competing-risk results for it; the reasons are given in Section 5.3. 3.5.1 Multiclass Horizon Reformulation For competing-risk classification, class 0 denotes that no observed event has occurred by horizon tk , while classes 1, . . . , C identify the observed cause. For a valid subject–horizon pair, ( ∆i , ∆i > 0 and T̃i ≤ tk , CR yik = 0, T̃i > tk . 9

Minh-Khoi Pham et al.

As in the single-risk construction, censored subject–horizon pairs after T̃i are excluded because their status is unknown. At inference, the probability assigned to class c at horizon tk provides the corresponding horizon-wise cause-c event probability. Unlike the joint DeepHit formulation below, this horizon-wise classification construction does not define a single normalized probability mass jointly over cause and event time, and the cause-specific probabilities it produces are not constrained to be mutually consistent. 3.5.2 Cause-Specific Cox and MTLR For Cox and MTLR, competing risks are handled through separate cause-specific models. For each cause c, we define (c) δi = 1[∆i = c], so that events from the other causes are treated as censoring for the cause-specific task. For Cox, the model produces one relative-risk score per cause and is optimized using the sum of the corresponding cause-specific partial likelihoods, where only events of cause c contribute as cases to the c-th objective. At prediction time, the cause-specific Breslow baseline cumulative hazards are combined with the predicted relative risks to obtain cause-specific cumulative hazards. Overall survival is then reconstructed from the sum of these hazards, and cumulative incidence functions are obtained from the resulting hazard increments. For MTLR, the same cause-specific recoding is used to define a binary single-risk task for each cause, and an independent MTLR head is fitted using the corresponding singlerisk discretization procedure. Because the cause-specific MTLR models are optimized independently, they do not impose a global normalization constraint across competing causes. Thus, Cox and MTLR both use cause-specific competing-risk formulations, whereas the DeepHit variant below models a single joint distribution over cause–time outcomes. 3.5.3 Joint Competing-Risk DeepHit The competing-risk DeepHit implementation predicts a joint cause–time probability mass function XX pc,k (x) = P (∆ = c, T ∈ Ik | x), pc,k (x) = 1, c

using a joint softmax. Unlike the single-risk DeepHit implementation, the competing-risk loss used here is a joint likelihood without the pairwise ranking term. For an observed event of cause ci in interval ki , Lobs,i = − log pci ,ki (xi ), whereas for censoring assigned to interval ki the implemented discrete-bin convention uses   XX Lcens,i = − log  pc,k (xi ) . c k≥ki

The inclusive boundary preserves censoring supervision for observations clipped P to the final discrete interval. The cumulative incidence for cause c follows as Fc (tm | x) = k≤m pc,k (x). This joint softmax is the formulation in our benchmark that most directly enforces probability conservation across competing causes and event times. The joint competing-risk time grid 10

TabFMs for Time-to-Event Prediction

Algorithm 1 Adaptation interfaces for transferring a pretrained TabFM to survival prediction Require: Pretrained TabFM fθ⋆ ; survival data D = {(xi , T̃i , zi )}N i=1 , where zi = δi for single risk and zi = ∆i for competing risks; adaptation strategy s ∈ {zero-shot, CE, survival-head} 1: if s ∈ {zero-shot, CE} then 2: {Classification interface; Section 3.3} 3: Select effective horizons t0 < · · · < tK−1 and construct temporally augmented inputs x̃ik For single risk, set zi = δi and define yik = 1[zi = 1 ∧ T̃i ≤ tk ] and mik = 1[zi = 4: 1 ∨ T̃i > tk ] 5: if s = zero-shot then 6: p̂ik ← gnative (fθ⋆ (x̃ik ; Dcontext )) {no parameter updates} 7: else 8: Freeze pretrained parameters θ⋆ 9: Optimize the task-specific classification head on valid expanded examples 10: end if 11: Recover horizon-wise survival probabilities, or cause probabilities for competing risks 12: else 13: {Survival head interface; Section 3.4} 14: Freeze pretrained parameters θ⋆ 15: Define native context labels ej = 1[zj > 0] 16: Sample C ⊂ Dtrain \ B for query minibatch B 17: Compute hi = fθ⋆ (xi ; {(xj , ej ) : j ∈ C}) 18: Attach gΦ ∈ {CoxPH, MTLR, DeepHit} and optimize Φ⋆ = arg min Lsurv gΦ (hi ), T̃i , zi Φ

Predict Ŝ(t | xi ), or F̂c (t | xi ) for competing risks, from gΦ⋆ (hi ) 20: end if 21: During supervised training, use event-balanced minibatches and resample support contexts across optimization steps 19:

uses equal-width intervals over the observed event-time range, in contrast to the adaptive quantile-based cuts used by the single-risk discrete heads. Algorithm 1 summarizes the three adaptation interfaces and the points at which they differ: target construction, whether a downstream head is optimized, and which objective supplies the event-time and censoring information.

4 Experimental Setup 4.1 Data Set Taxonomy We curate a broad benchmark of public tabular survival data sets spanning diverse domains, sample sizes, censoring rates, and event structures. Following Kim et al. (2026), the singlerisk benchmark includes data sets from SurvSet (Drysdale, 2022). For competing risks, the reported main results use SUPPORT2-CR, FRAMINGHAM, PBC2, and SYNTHETIC (Lee 11

Minh-Khoi Pham et al.

et al., 2018). Missing values are imputed using training-set medians; continuous covariates are clipped to the [0.1%, 99.9%] training quantile range and standardized using training-set statistics only. For the size-stratified analysis in Section 5.2 we group data sets by cohort size into small (N < 500), medium (500 ≤ N < 4,000), and large (N ≥ 4,000) regimes. Full data set characteristics are reported in Table 1 for the single-risk benchmark and Table 2 for the competing-risk benchmark. For TabFM interfaces whose native feature capacity is smaller than the input dimensionality, we apply principal component analysis (PCA) before backbone inference. Within each outer cross-validation fold, PCA is fitted only on the training partition and the learned projection is applied unchanged to validation and test samples. If d denotes the original number of covariates and dmax the feature capacity exposed by the corresponding backbone wrapper, PCA is applied only when d > dmax , with the number of retained components additionally bounded by the available number of training observations. Data sets already satisfying the native feature-capacity constraint are passed without PCA reduction. For classification-based interfaces, the capacity calculation also reserves dimensions for the appended temporal features. Backbones that natively support the observed feature dimensionality, such as the TabICL used here, do not invoke this PCA reduction. 4.2 Pretrained Backbones and Baselines We evaluate TabPFN, TabDPT, and TabICL under the adaptation framework above, together with non-pretrained neural and classical survival baselines. Supervised TabFM variants keep the pretrained backbone frozen; classification fine-tuning is denoted by -CE, while survivalhead variants are denoted by -Cox, -DH, and -MTLR. The -MTLR variants implement the adaptation interface established in our earlier work (Pham et al., 2026a) and are treated here as reference baselines under the current benchmark protocol. Other baselines include CoxPH, random survival forests (RSF), gradient-boosted survival models (GBSA), DeepSurv, SurvTRACE (Wang et al., 2022), and DySurv (Mesinovic et al., 2026). Two further non-pretrained neural baselines, denoted MLP-MTLR and MLP-DH, pair the same multilayer-perceptron trunk used by DeepSurv with the MTLR and DeepHit heads; they are the matched non-pretrained counterparts used for the backbone comparison in Section 5.1. The competing-risk benchmark adds SurvBoost, a gradient-boosted competing-risk baseline. All baselines are described in Appendix B. 4.3 Training Details All methods are evaluated with 5-fold cross-validation and fixed seed. We use batch size 128 for both single-risk and competing-risk experiments. Classification fine-tuning uses a learning rate of 10−5 and 5 fine-tuning epochs in the shared wrapper. Cox, DeepHit, and MTLR heads use learning rate 10−5 for up to 50 epochs in the benchmark configuration. Context-conditioned supervised forward passes use at most 512 labeled support examples; when fewer eligible examples are available, the full eligible support pool is used. During training, the support context is sampled from the training partition while excluding the current query minibatch. Observed events can be sparse, so supervised training uses eventbalanced minibatches with a minimum quota of observed-event examples; the remaining positions are sampled from the full training pool and the resulting minibatch is shuffled 12

TabFMs for Time-to-Event Prediction

before optimization. Hyperparameter search is applied to non-pretrained baselines with model-specific search spaces; each tuned baseline receives a 20-trial Optuna search within every training fold. No search is performed for any TabFM variant—zero-shot, classification fine-tuning, and survival-head adaptation all use the fixed settings above on every data set. Full implementation settings and baseline search spaces are summarized in Appendix C and Table 3. 4.4 Evaluation Protocol We report the Antolini time-dependent concordance index Ctd (Antolini et al., 2005) and Integrated Brier Score (IBS) (Graf et al., 1999) for discrimination and probabilistic prediction accuracy, respectively. IBS evaluates squared error of predicted survival probabilities over time under inverse-probability-of-censoring weighting; it is sensitive to both calibration and discrimination and should not be interpreted as a calibration-only metric. Mean dynamic area under the ROC curve (AUC), evaluated at the 25th, 50th, and 75th percentile time horizons, is reported only as a supplementary metric in the full-results appendix and is not used for the main comparative claims. IBS is integrated over 100 evenly spaced points spanning the observed follow-up, and the AUC horizons are percentiles of the observed event times in the test fold.

5 Results 5.1 Adaptation and Backbone Effects We first separate two sources of performance variation: the effect of changing the adaptation interface while keeping the pretrained TabFM backbone fixed, and the effect of replacing a non-pretrained neural backbone with a pretrained TabFM under the same survival objective. Figure 1 summarizes these as an adaptation effect, comparing survival-head adaptation with CE under the same frozen TabFM backbone, and a backbone effect, comparing a TabFM-based survival model with its matched non-pretrained neural counterpart. For Ctd , both effects are generally centered near or above zero but vary substantially across data sets and model families. The separation is clearer for IBS: Cox shows the strongest and most consistent improvement over CE, MTLR also benefits with greater variability, while DeepHit gains are more pronounced for Ctd than for IBS. These results indicate that the value of pretrained representations depends on the downstream survival head and evaluation metric, rather than yielding a universal advantage across objectives. 5.2 Performance Across Data Regimes We next examine how the relative performance of the full model interfaces changes with data-set size. Figure 2 shows a clear interaction between adaptation interface and data-set size; per-data-set results are reported in Table 5. On small data sets, RSF ranks best overall and zero-shot TabPFN remains competitive for Ctd . On medium data sets, supervised TabFM survival heads move toward the top of the rankings, with MTLR and DeepHit competitive for Ctd and Cox-based adaptations among the strongest for IBS. On large data sets, Cox-adapted TabFMs lead for both Ctd and IBS, while zero-shot and CE variants fall substantially behind. 13

Minh-Khoi Pham et al.

Objective effect: survival head vs. CE

Backbone effect: TabFM vs. MLP / DeepSurv

(a) Discrimination

(b) Calibration

TabPFN (Cox)

Foundation model and survival objective

TabDPT (Cox) TabICL (Cox) TabPFN (MTLR) TabDPT (MTLR) TabICL (MTLR) TabPFN (DeepHit) TabDPT (DeepHit) TabICL (DeepHit)

0.6

0.4

0.2 0.0 0.2 Improvement in Ctd (positive = better)

0.4

0.3

0.2

0.1

0.0 0.1 0.2 0.3 0.4 Improvement in IBS (positive = better)

0.5

Figure 1: Adaptation and backbone effects across 74 single-risk data sets. Blue compares survival-head adaptation with CE using the same frozen TabFM backbone; orange compares TabFM-based survival models with matched non-pretrained neural baselines. Panel (a) shows changes in Ctd and panel (b) changes in IBS, sign-reversed so that positive values indicate improvement. Across metrics, Cox is the most consistently strong interface for IBS and becomes especially competitive for Ctd on large data sets. DeepHit is relatively stronger for Ctd than for IBS, while MTLR remains competitive, particularly in the medium and large regimes. Across the data-size regimes, classification fine-tuning becomes more competitive relative to zero-shot inference as data sets grow, but remains weaker than the strongest survival heads for IBS. 5.3 Transfer to Competing Risks We finally examine whether the single-risk ordering persists when multiple mutually exclusive event types must be modeled. Figure 3 compares the three survival-head interfaces with competing-risk baselines, with per-cause results reported in Table 6. Cause-specific MTLR is the strongest TabFM interface on both IBS and Ctd , performing comparably to the strongest tree-based baseline, while joint DeepHit ranks in the middle of the field and cause-specific Cox is generally weaker. This differs from the single-risk ordering in Section 5.2, where Cox is the strongest interface overall. We interpret this ordering cautiously because only four competing-risk data sets were available for evaluation, making the mean ranks more sensitive to individual data sets and insufficient to establish a stable global ordering among the heads. The strong MTLR result also does not arise from joint probability normalization, since the cause-specific MTLR models are fitted independently; joint normalization is enforced only by the competing-risk DeepHit softmax over cause–time outcomes. The observed ranking therefore suggests an 14

TabFMs for Time-to-Event Prediction

Attention Baseline

Deep Finetune-CE

Finetune-Cox Finetune-DH

Small Datasets (N=40)

C-index RSF TabPFN-ZS TabPFN-Cox Cox PH TabDPT-ZS TabPFN-MTLR TabDPT-MTLR GBSA TabICL-CE TabPFN-CE TabICL-ZS DeepSurv TabDPT-Cox TabDPT-DH TabICL-MTLR MLP-MTLR TabICL-Cox TabDPT-CE DySurv TabPFN-DH TabICL-DH SurvTrace MLP-DH 5

Zero-shot

IBS

RSF TabPFN-Cox TabDPT-Cox TabICL-Cox DeepSurv GBSA TabPFN-ZS TabDPT-MTLR Cox PH TabICL-ZS TabPFN-MTLR TabDPT-ZS TabICL-MTLR MLP-MTLR MLP-DH TabDPT-DH TabICL-CE TabPFN-CE SurvTrace TabICL-DH TabPFN-DH TabDPT-CE DySurv

8.0±1.0 9.4±0.9 9.8±0.9 9.9±0.9 10.2±0.9 10.2±0.9 10.3±1.0 10.4±1.1 10.6±1.0 10.7±0.9 11.3±1.2 11.8±1.1 11.8±1.0 12.7±1.0 13.2±1.1 13.3±1.1 13.7±1.1 14.1±1.0 14.6±0.9 14.6±0.9 14.8±1.1 15.2±1.0 15.7±1.1

Finetune-MTLR Tree

3.7±0.6 5.0±0.6 6.1±0.8 7.3±0.8 7.8±0.9 8.3±0.7 9.3±0.9 9.7±0.7 9.9±1.1 10.4±0.8 11.2±0.7 11.5±0.8 11.9±0.6 14.1±0.8 15.1±0.8 15.6±0.7 15.7±1.0 15.9±0.9 16.1±1.0 16.6±0.7 16.8±0.8 17.1±0.8

20.9±0.5

Medium Datasets (N=25) TabDPT-MTLR TabDPT-DH TabICL-DH TabDPT-Cox TabICL-MTLR TabPFN-Cox TabICL-Cox TabICL-CE TabDPT-CE TabICL-ZS TabPFN-MTLR TabPFN-DH TabPFN-ZS DySurv DeepSurv RSF TabPFN-CE Cox PH TabDPT-ZS GBSA MLP-DH MLP-MTLR SurvTrace

6.4±0.7 6.4±1.0 6.7±0.9 6.9±0.9 7.1±0.9 8.7±1.0 9.3±1.1 9.8±1.2 10.0±1.4 10.1±1.1 10.8±0.9 11.6±1.0 12.0±1.0 14.3±1.2 14.3±1.3 14.5±1.0 15.1±1.5 16.2±1.1 16.4±1.1 16.9±1.2 17.0±1.1 17.6±1.1 17.9±1.1

TabDPT-Cox TabPFN-Cox TabICL-Cox RSF DeepSurv GBSA TabDPT-MTLR TabICL-MTLR TabPFN-MTLR Cox PH TabPFN-ZS TabICL-ZS TabDPT-ZS MLP-MTLR SurvTrace TabDPT-DH MLP-DH DySurv TabICL-DH TabPFN-DH TabPFN-CE TabICL-CE TabDPT-CE

4.4±0.5 4.4±0.5 5.1±0.5 6.0±0.9 7.5±0.7 7.8±0.8 8.7±1.1 9.2±1.0 10.3±0.9 10.6±1.0 10.6±1.1 11.6±1.0 12.6±1.0 12.6±1.3 13.5±1.3 15.5±0.9 15.8±0.8 16.0±1.2 16.0±1.0 17.6±0.9 18.6±0.6 20.6±0.6 20.9±0.5

Large Datasets (N=9) TabPFN-Cox TabICL-DH TabDPT-DH TabICL-MTLR TabICL-Cox TabICL-CE TabPFN-DH TabDPT-Cox TabDPT-CE TabPFN-MTLR TabDPT-MTLR DeepSurv DySurv MLP-DH GBSA MLP-MTLR TabPFN-CE RSF TabPFN-ZS TabICL-ZS Cox PH SurvTrace TabDPT-ZS

6.6±1.1 6.9±1.3 7.2±2.1 7.8±2.0 7.9±1.1 8.2±2.4 9.3±1.7 9.3±2.5 9.6±2.2 10.0±1.8 11.1±2.2 11.8±2.6 11.9±1.4 12.4±1.8 12.7±1.8 14.3±1.6 14.8±3.2 16.0±2.3 16.8±0.9 17.1±0.9 17.3±2.2 17.4±1.5 19.6±0.7

TabPFN-Cox TabDPT-Cox SurvTrace GBSA DeepSurv TabICL-Cox RSF TabPFN-ZS MLP-MTLR TabICL-ZS TabDPT-ZS Cox PH TabDPT-MTLR TabPFN-MTLR MLP-DH TabICL-MTLR DySurv TabICL-DH TabDPT-DH TabPFN-DH TabPFN-CE TabICL-CE TabDPT-CE

3.2±0.7

Mean Rank (C-index) ( better)

6.2±1.0 6.9±1.2 7.0±1.2 7.2±0.9 7.3±1.2 8.0±1.8 8.4±1.8 9.1±2.3 9.2±2.0 10.0±2.0 11.8±1.3 12.2±1.6 12.3±1.9 13.0±2.2 13.6±1.4 13.9±1.9 17.1±1.0 17.6±1.1 18.0±1.1 18.9±0.9 22.2±0.1 22.8±0.1

Mean Rank (IBS) ( better)

Figure 2: Size-stratified mean-rank comparison for single-risk survival analysis. Lower rank is better. Results are aggregated across 40 small, 25 medium, and 9 large data sets and compare discrimination using time-dependent Ctd with probabilistic prediction using IBS. interaction between objective, discretization, backbone, and competing-risk formulation under a limited benchmark rather than a definitive ordering of the three interfaces.

6 Discussion Overall, the results support viewing survival transfer as an adaptation interface problem rather than simply a choice of downstream loss. Native zero-shot TabFM inference is most competitive on smaller data sets, whereas supervised adaptation becomes increasingly useful 15

Minh-Khoi Pham et al.

Attention Baseline TabDPT-MTLR SurvBoost TabPFN-MTLR TabICL-MTLR DeepSurv TabDPT-DH Cox PH TabPFN-DH MLP-DH TabICL-DH TabPFN-Cox TabDPT-Cox SurvTrace TabICL-Cox DySurv

Deep Finetune-Cox

4.5±1.6 4.5±0.5 4.8±1.3 5.0±2.0 5.5±3.3 6.2±2.4 7.5±2.2 7.5±0.6 8.8±2.3 8.8±1.5 9.8±2.0 11.0±1.7 11.0±2.1 11.8±1.6 13.5±1.2

Mean Rank (C-index) ( better)

Finetune-DH Finetune-MTLR

Tree

TabDPT-MTLR SurvBoost TabPFN-MTLR TabICL-MTLR DeepSurv TabPFN-DH TabPFN-Cox TabDPT-DH SurvTrace Cox PH MLP-DH TabICL-DH TabDPT-Cox DySurv TabICL-Cox

4.0±1.1 4.0±1.4 4.5±1.8 5.5±1.3 6.2±3.2 7.8±1.2 7.8±2.5 8.5±2.2 8.5±2.1 8.8±2.0 8.8±2.9 9.8±0.9 11.2±1.4 12.0±2.1 12.8±1.6

Mean Rank (IBS) ( better)

Figure 3: Mean-rank comparison for competing-risk survival across four data sets. Lower rank is better. The comparison includes the three survival-head interfaces and classical, tree-based, and non-pretrained neural baselines; horizon-wise classification interfaces are excluded from this arm. as data sets grow. This pattern is consistent with broader TabFM studies that report strong native in-context performance on smaller data sets and a reduced advantage as data sets become larger or more complex (Purucker et al., 2026). Importantly, weaker zero-shot performance on larger data sets does not imply that the pretrained representation is no longer useful. Keeping the same frozen backbone while changing the downstream interface can substantially improve performance. Classification fine-tuning also becomes more competitive with zero-shot inference as data sets grow, but the stronger results from survival heads, particularly for IBS, suggest that learning a new head alone is not sufficient; the survival structure represented by that head also matters. The survival heads show different performance profiles. Cox is the most consistently strong interface in the single-risk benchmark, particularly for IBS and for concordance on larger data sets. One possible explanation is that the pretrained representation already captures nonlinear structure, allowing the comparatively simple Cox head to provide an effective mapping to relative risk, although its proportional-hazards assumption remains restrictive. DeepHit is relatively stronger for Ctd than for IBS, which is consistent with the ranking component in its single-risk objective. This illustrates that strong risk ordering does not necessarily translate into equally accurate survival probabilities. MTLR remains competitive, particularly in the medium and large regimes, but the broader benchmark shows that its advantage is not consistent across backbones, metrics, and data regimes. Together, these results suggest that Cox, MTLR, DeepHit, and CE emphasize different aspects of survival prediction rather than following one fixed ordering. The competing-risk experiments suggest that the single-risk ordering does not necessarily carry over when several mutually exclusive event types are modeled. Across the four evaluated 16

TabFMs for Time-to-Event Prediction

data sets, cause-specific MTLR ranks highest among the TabFM survival heads on both metrics, while cause-specific Cox ranks lower. Because this analysis contains only four data sets, the ordering should be treated as preliminary and may be sensitive to individual data sets. The stronger MTLR result also cannot be explained simply by joint probability normalization, since the reported MTLR models are fitted independently for each cause, whereas joint normalization is enforced by the competing-risk DeepHit formulation. The results therefore suggest that the choice of survival objective interacts with the event structure and modeling formulation, rather than identifying one head as generally preferable. Several limitations should be considered when interpreting these results. Mean ranks can hide important differences between individual data sets, and the competing-risk benchmark is much smaller than the single-risk benchmark. The competing-risk comparison is also limited to survival-head adaptation, so its conclusions apply only to the comparison among survival heads. In the single-risk analysis, CE and survival-head adaptation differ in both target construction and objective, so their comparison should not be interpreted as a pure loss-function ablation. Similarly, the data-size analysis compares different data sets and therefore does not isolate sample size from dimensionality, censoring, event prevalence, or domain. Explanations for the behavior of individual heads should therefore be treated as hypotheses consistent with the observed results rather than causal conclusions. Finally, the competing-risk comparison should not be interpreted as isolating the effect of the survival head alone. Cox and MTLR use cause-specific formulations, whereas DeepHit uses a jointly normalized cause–time distribution. The observed ordering may therefore reflect both the head family and the way competing risks are represented. Future work could use more controlled experiments to separate the effects of target construction, censoring treatment, ranking supervision, and sample size. Larger competingrisk benchmarks would also be needed to determine whether the observed ordering among survival heads is stable. It would also be useful to combine context adaptation with survivalspecific heads and to compare adaptation of generic TabFMs directly with survival-aware pretraining.

7 Conclusion We study how generic tabular foundation models can be transferred to censored time-to-event prediction through different adaptation interfaces. Across 74 single-risk data sets, native zero-shot inference remains competitive in the small-data regime, while supervised adaptation becomes increasingly useful on medium and large data sets. This pattern cannot be explained simply by a loss of value in the pretrained representation: with task-specific heads, frozen TabFM backbones support several of the strongest methods in the larger data regimes. Relative to the MTLR adaptation established in our earlier clinical study (Pham et al., 2026a), the broader benchmark shows that Cox is the most consistently strong interface for single-risk prediction, particularly for IBS and for concordance on larger data sets, while DeepHit is relatively stronger for discrimination than for probabilistic prediction. In the competing-risk analysis, based on four data sets, cause-specific MTLR performs best among the TabFM survival heads, although this ordering should be interpreted cautiously given the limited benchmark. Overall, the results indicate that successful TabFM transfer depends 17

Minh-Khoi Pham et al.

on the amount of available data, the evaluation target, and how directly the downstream interface represents survival structure.

Acknowledgments and Disclosure of Funding This research was conducted with the financial support of Taighde Éireann-Research Ireland under Grant Agreement No. 13/RC/2106_P2 at ADAPT, the Research Ireland Centre for AI-Driven Digital Content Technology at DCU funded through the Research Ireland Research Centres Programme. For the purpose of Open Access, the author has applied a CC BY public copyright licence to any author-accepted manuscript version arising from this submission. We declare no competing interests.

TabFMs for Time-to-Event Prediction

Appendix A. Data Set Summary Table 1: Data set characteristics. N : number of subjects. Event rate: proportion of uncensored observations. Bins: number of adaptive time intervals used by the single-risk discrete-time formulations. SurvSet data sets are collected from Drysdale (2022). ID

Data set

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

ACATH ACTG AIDS AIDS2 AML BULL BERGAMASCHI BREAST BURN CANCER CGD CHOP COLON COST CSL D.OROPHA.REC DATADIVAT1 DATADIVAT2 DATADIVAT3 DATAOVARIAN1 DBCD DIABETES DIALYSIS DIVORCE DLBCL E1684 EPILEPTIC FLCHAIN FRAMINGHAM FRTCS GBSG2 GLIOMA GRACE GSE1992 GSE3143 GSE4335 HDFAIL

Feat.

Event rate

2,230 1,151 938 2,810 115 82 100 154 169 128 412 911 518 2,035 192 5,943 1,837 4,267 910 295 394 6,805 3,371 240 284 681 6,521 4,658 1,391 686 37 1,000 123 158 113 52,422

3 17 5 12 6283 10 4 13 28 23 3833 12 13 6 24 16 4 16 162 4919 4 72 4 7399 3 5 32 17 5 9 4 5 15541 8663 12815 36

0.66 0.08 0.15 0.62 0.57 0.34 0.26 0.31 0.72 0.34 0.40 0.48 0.78 0.12 0.72 0.16 0.32 0.06 0.60 0.27 0.39 0.24 0.31 0.57 0.69 0.16 0.30 0.31 0.05 0.44 0.62 0.32 0.28 0.32 0.34 0.06

Bins

Domain

Regime

10 10 10 50 10 10 10 10 20 10 20 50 50 30 20 50 50 30 20 10 30 10 100 20 30 20 100 100 10 50 10 10 10 10 10 100

Cardiology Infectious disease Infectious disease Infectious disease Haematology Oncology Oncology Critical care Oncology Immunology Haematology Oncology Oncology Hepatology Oncology Nephrology Nephrology Nephrology Oncology Oncology Endocrinology Nephrology Social/Economics Haematology Oncology Neurology Haematology Cardiology Cardiology Oncology Oncology Cardiology Oncology Oncology Oncology Social/Economics

medium medium medium medium small small small small small small small medium medium medium small large medium large medium small small large medium small small medium large large medium medium small medium small small small large

Continued on next page 19

Minh-Khoi Pham et al.

Table continued ID

Data set

Feat.

Event rate

37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74

HEART 69 HEARTVALVE 444 HEPATOCELLULAR 101 LEUKSURV 1,043 MCLCLEANED 92 MELANOMA 205 MICRO.CENSURE 117 NKI70 144 NSBCD 115 NWTCO 4,028 OLDMORT 6,495 OVA 358 OVARIAN 26 PBC 312 PBC3 327 PHARMACOSMOKING 113 PHPL04K8A 442 PROSTATE 467 PROSTATESURVIVAL14,065 RDATA 1,040 RETINOPATHY 394 RHC 634 ROSSI 973 ROTT2 2,982 SCANIA 1,931 SMARTO 1,021 STAGEC 136 SUPPORT2 530 TRACE 1,878 UIS 579 UNEMPDUR 3,241 UNEMPLOYMENT 450 VDV 78 VETERAN 137 VLBW 257 WPBC 194 Z243 100 ZINC 431

4 23 43 29 574 5 81 76 549 9 13 11 4 6 19 16 21 25 6 6 11 73 27 12 8 34 18 65 6 14 6 7 4705 8 41 32 23 20

26 52,422 2,027

3 15541 901

Min Max Mean

Bins

Domain

Regime

0.65 0.05 0.45 0.84 0.70 0.28 0.22 0.33 0.33 0.14 0.30 0.74 0.46 0.40 0.17 0.68 0.53 0.70 0.06 0.53 0.39 0.42 0.07 0.43 0.56 0.11 0.38 0.53 0.51 0.81 0.61 0.56 0.44 0.93 0.14 0.24 1.00 0.19

10 10 10 30 10 10 10 10 10 50 10 50 10 20 10 10 30 20 50 50 20 30 10 100 100 10 10 30 100 50 10 10 10 20 10 10 10 10

Cardiology Cardiology Oncology Haematology Haematology Oncology Oncology Oncology Oncology Oncology All-cause mortality Oncology Oncology Hepatology Hepatology Medicine Oncology Urology Urology Oncology Ophthalmology Critical care Social/Economics Oncology All-cause mortality Cardiology Urology Critical care Cardiology Psychiatry Social/Economics Social/Economics Oncology Oncology Neonatology Oncology Medicine Oncology

small small small medium small small small small small large large small small small small small small small large medium small medium medium medium medium medium small medium medium medium medium small small small small small small small

0.05 1.00 0.41

10 100 28 Continued on next page

TabFMs for Time-to-Event Prediction

Table continued ID

Data set

Feat.

Event rate

Bins

IQR25–75

[146, 1,331]

[6, 36]

[0.24, 0.57]

[10, 30]

Domain

Regime

Table 2: Data set characteristics for data sets with competing risks. N : number of subjects. Event rate: proportion of uncensored observations. Bins: number of discrete time intervals used by discrete-time models; interval construction depends on the formulation, with cause-specific MTLR using the corresponding adaptive single-risk discretization and joint DeepHit using equal-width intervals over the observed event-time range. ID

Data set

Feat.

Event rate

Bins

Domain

Regime

Data Sets with Competing Risks 1 2 3 4

FRAMINGHAM PBC2 SUPPORT2CR SYNTHETIC Min Max Mean IQR25–75

4,273 1,945 9,105

18 16 37

0.42 (K=2) 0.45 (K=2) 0.68 (K=2)

100 50 30

Cardiology Hepatology Critical care

large medium large

30,000

0.50 (K=2)

Synthetic

large

1,945 30,000 11,331 [3,691, 14,329]

12 37 21 [15, 23]

0.42 0.68 0.51 [0.44, 0.55]

10 100 48 [25, 62]

Appendix B. Model Details This section summarizes the non-TabFM models evaluated in the main experiments. Unless otherwise stated, classical and deep baselines are implemented using scikit-survival or pycox, while DySurv and SurvTRACE are adapted from their original codebases. The three TabFM backbones and exact versions used in the benchmark are described in Section 2.2. Cox Proportional Hazards (Cox PH) assumes h(t | x) = h0 (t) exp(β ⊤ x), where h0 (t) is an unspecified baseline hazard and β ⊤ x defines relative risk. Parameters are estimated by maximizing the regularized partial likelihood; the benchmark implementation uses fixed regularization rather than an Optuna search. Random Survival Forests (RSF) extends random forests to censored outcomes using log-rank splitting criteria and ensemble averaging. Each tree estimates cumulative hazard via the Nelson–Aalen estimator, and predictions are aggregated across trees. Gradient Boosting Survival Analysis (GBSA) GBSA applies gradient boosting to survival objectives such as Cox partial likelihood, sequentially adding weak learners through functional gradient descent. SurvBoost is the competing-risk gradient-boosting baseline. For each cause it fits a gradient-boosted survival model with the remaining observed causes treated as censoring, and evaluates the resulting cause-specific survival functions on a shared time grid. It is the 21

Minh-Khoi Pham et al.

strongest non-pretrained competitor in the competing-risk benchmark and is reported only there. DeepSurv generalizes Cox PH by replacing the linear predictor with a neural network, modeling h(t | x) = h0 (t) exp(fθ (x)). Training minimizes the negative Cox partial loglikelihood. DeepHit models a discrete event-time distribution. In the single-risk experiments we use likelihood plus ranking supervision; in the competing-risk TabFM implementation, the reported joint DeepHit objective is the normalized cause–time likelihood described in Section 3.5. MLP-MTLR and MLP-DH are non-pretrained counterparts to the TabFM survivalhead variants. Both use the same multilayer-perceptron trunk as DeepSurv over the raw standardized covariates, replacing the Cox head with the MTLR and DeepHit objectives respectively and using the same adaptive event-balanced discretization as the corresponding TabFM heads. They isolate the contribution of the pretrained representation: comparing TabPFN-MTLR with MLP-MTLR, or TabPFN-DH with MLP-DH, holds the survival objective and the time grid fixed and varies only whether the representation comes from a pretrained tabular prior or a trunk trained from scratch. DeepSurv plays the same role for the Cox head. SurvTRACE applies self-attention to tabular features and is pretrained using masked feature modeling before survival fine-tuning. DySurv combines neural survival modeling with latent-variable inference to capture nonlinear and heterogeneous risk dynamics.

TabFMs for Time-to-Event Prediction

Appendix C. Training and Hyperparameter Details No hyperparameter search is performed for the TabFM variants; all three interfaces use fixed settings across data sets. Table 3 summarizes these settings. The non-pretrained baselines with model-specific search spaces are tuned independently within each training fold using 20 Optuna trials, with the search spaces reported in Table 4. Table 3: Strategy-level settings used for the TabFM interfaces. A dash denotes a setting that is not applicable.

Setting

Zero-shot

Classification FT

Survival-head

Backbone Gradient updates Trainable head Head architecture

None Native classifier –

None (frozen) None (frozen) Classification Survival 1 hidden layer, 64 units, dropout 0.1

Optimization Optimizer Learning rate Schedule Gradient clipping Epochs Batch size Min. events/batch

– – – – – – –

AdamW; weight decay 10−3 10−5 10−5 Cosine annealing to 0.01× initial LR Max norm 1.0 5 ≤ 50; patience 5 128 128 16 16

Context Max. support size Context resampling Inference contexts (R)

512 Per horizon 5

512 Per step 5

Protocol CV folds Hyperparameter search

5 None

C.1 Hyperparameter Search for Non-Pretrained Baselines The non-pretrained baselines with model-specific search configurations are tuned independently within each training fold using 20 Optuna trials. Table 4 reports the exact search spaces defined in the experimental configuration files. Continuous parameters marked as log-uniform are sampled on a logarithmic scale; integer ranges are sampled with unit step unless otherwise noted; and categorical parameters are sampled from the listed values.

Appendix D. Ablation Studies The following supplementary analyses probe complementary aspects of the main benchmark rather than introducing separate performance claims. Together they ask when additional su23

Minh-Khoi Pham et al.

Table 4: Hyperparameter search spaces for the tuned non-pretrained baselines. Each model is optimized using 20 Optuna trials within each training fold. Model

Hyperparameter

Search space

RSF

Number of trees Minimum samples to split Minimum samples per leaf

[50, 200] (integer) [5, 20] (integer) [5, 20] (integer)

GBSA

Learning rate Number of estimators Maximum depth

[10−2 , 2 × 10−1 ] (log-uniform) [50, 200] (integer) [2, 5] (integer)

DeepSurv

Learning rate Dropout Number of hidden layers Hidden width Epochs

[10−4 , 5 × 10−2 ] (log-uniform) [0, 0.5] (uniform) {1, 2, 3} {16, 32, 64} {10, 20, . . . , 100}

MLP-MTLR

Learning rate Dropout Number of hidden layers Hidden width

[10−4 , 10−1 ] (log-uniform) [0, 0.5] (uniform) [1, 3] (integer) {32, 64, 128}

MLP-DH

Learning rate Dropout Number of hidden layers Hidden width DeepHit α DeepHit σ

[10−4 , 10−1 ] (log-uniform) [0, 0.5] (uniform) [1, 3] (integer) {32, 64, 128} [0, 0.5] (uniform) [10−2 , 1] (log-uniform)

SurvTRACE

Learning rate Hidden size Hidden layers Attention heads Intermediate size

[10−4 , 10−2 ] (log-uniform) {8, 16, 32} {1, 2, 3, 4} {2, 4, 8} {32, 64, 128}

DySurv

Learning rate Encoded features Batch size

[10−4 , 10−2 ] (log-uniform) {16, 32, 64, 128} {64, 128, 256}

pervision becomes useful, how sensitive the discrete-time interfaces are to temporal resolution, and how the three adaptation regimes differ in computational and qualitative behavior. D.1 Label Efficiency Under Increasing Supervision Figure 4 provides a controlled complement to the data-regime analysis in Section 5.2. In the extreme low-label regime, zero-shot inference and DeepHit adaptation begin at essentially the same discrimination, but their trajectories diverge as more labeled supervision becomes available: zero-shot improves quickly and then saturates, whereas DeepHit continues to gain through the full-label setting. Classification fine-tuning remains lower throughout this experiment. The result is consistent with the main benchmark without duplicating its claim: 24

TabFMs for Time-to-Event Prediction

0.8

C-index

0.7

0.6

0.5

0.4 1%

Training Label Fraction

Zero-shot Finetune-DH Finetune-CE Other methods 10% 100%

Figure 4: Time-dependent Ctd as the available labeled training fraction increases from 1% to 100%. native in-context inference can be useful when supervision is scarce, while DeepHit continues to improve as additional labeled data become available. D.2 Sensitivity to Temporal Discretization

0.74

C-index

0.72 0.70 0.68 0.66

Zero-shot Finetune-DH Finetune-CE Other methods

Number of Bins

100

Figure 5: Sensitivity of the classification and discrete-time survival interfaces to the number of temporal bins. 25

Minh-Khoi Pham et al.

Ideal

Performance (higher is better)

Model Groups

Attention Baseline Deep Finetune-CE Finetune-Cox

Finetune-DH Finetune-MTLR Tree Zero-shot Ideal (Pareto Front)

Training Time (lower is better)

Model Groups

Attention Baseline Deep Finetune-CE Finetune-Cox

Finetune-DH Finetune-MTLR Tree Zero-shot Ideal (Pareto Front)

Inference Time (lower is better)

(a) Training time vs. performance

(b) Inference time vs. performance

Figure 6: Efficiency–performance trade-offs. Performance versus computational cost on a log scale. Training time for the tuned non-pretrained baselines includes their per-fold Optuna search, which the TabFM variants do not perform, so these values describe the cost of the reported experimental pipelines rather than a direct comparison of optimization speed. Figure 5 tests whether the discrete-time conclusions depend strongly on a particular temporal resolution. DeepHit is most sensitive when moving from a very coarse grid to a moderate number of bins and then largely plateaus, whereas CE changes less in discrimination. The main implication is therefore robustness rather than a new ranking: temporal resolution matters most when the grid is too coarse, while increasing it beyond a moderate level produces diminishing returns. This supports the adaptive event-balanced discretization used in the benchmark as a way to avoid bins with too few events rather than as an explanation for the relative advantage of any particular interface. D.3 Efficiency–Performance Trade-Off The benchmark also exposes a computational distinction between the three interfaces. Zeroshot inference avoids task-specific optimization but performs native in-context prediction separately across time horizons and context samples. Classification adaptation pays an optimization cost on the temporally expanded data, whereas survival-head adaptation optimizes directly on subject-level representations and then averages predictions over five training-context samples at inference. Figure 6 adds a practical perspective to the accuracy results rather than changing their ordering. The absence of optimization makes zero-shot attractive when fitting cost is the primary constraint, but repeated horizon-wise in-context prediction shifts part of that cost to inference. Supervised adaptation trades additional fitting time for the performance gains seen in the main benchmark, with the survival-head route avoiding the full subject–horizon expansion used by CE. Because the tuned baselines include their hyperparameter-search cost whereas the TabFM variants use fixed settings, the figure should be interpreted as a comparison of the full reported pipelines, not as evidence that one model family is intrinsically faster to optimize. 26

TabFMs for Time-to-Event Prediction

D.4 Risk Stratification and Survival Curve Separation The aggregate benchmark evaluates ranking and probabilistic accuracy, but it does not show how those differences appear when predictions are used to form risk groups. Figures 7 and 8 therefore provide a qualitative view on two data sets from different size regimes: COLON (N = 911, medium) and DIABETES (N = 394, small). Across these examples, the adaptation strategies differ not only in scalar performance but also in the degree to which predicted risk tiers translate into ordered Kaplan–Meier curves. Survival-head variants generally produce clearer separation, while zero-shot and classification-based predictions show more overlap in some panels. These examples are intended to illustrate the practical manifestation of the benchmark-wide discrimination results, not to establish a separate ranking of interfaces or to imply that the same degree of separation holds on every data set.

Minh-Khoi Pham et al.

Other methods DeepSurv

DySurv

1.0

0.8

0.6

0.4

0.2

0.0

Survival probability

1.0

Survival probability

DeepHit

0.6

0.4

0.2

500

1000

1500

2000

Time (hours)

2500

0.0

3000

0.6

0.4

0.2

500

1000

1500

2000

Time (hours)

2500

0.0

3000

500

1000

1500

2000

Time (hours)

2500

3000

Tabular Foundation Models TabDPT

TabICL

1.0

0.8

0.6

0.4

0.2

0.0

Survival probability

1.0

Survival probability

TabPFN

0.6

0.4

0.2

500

1000

1500

2000

Time (hours)

2500

0.0

3000

0.6

0.4

0.2

500

1000

1500

2000

Time (hours)

2500

0.0

3000

500

1000

500

1000

500

1000

1500

2000

2500

3000

1500

2000

2500

3000

1500

2000

2500

3000

Time (hours)

1.0

0.8

0.6

0.4

0.2

0.0

Survival probability

1.0

Survival probability

(a) Zero-shot

0.6

0.4

0.2

500

1000

1500

2000

Time (hours)

2500

0.0

3000

0.6

0.4

0.2

500

1000

1500

2000

Time (hours)

2500

0.0

3000

Time (hours)

1.0

0.8

0.6

0.4

0.2

0.0

Survival probability

1.0

Survival probability

(b) Classification fine-tuning

0.6

0.4

0.2

500

1000

1500

2000

Time (hours)

2500

3000

0.0

0.6

0.4

0.2

500

1000

1500

2000

Time (hours)

2500

3000

0.0

Time (hours)

(c) Survival head adaptation (Cox; frozen backbone) Figure 7: Risk-stratified Kaplan–Meier curves on the COLON data set. 28

TabFMs for Time-to-Event Prediction

Other methods Gradient Boosting

DeepSurv

1.0

0.8

0.6

0.4

0.2

0.0

Survival probability

1.0

Survival probability

CoxPH

0.6

0.4

0.2

Time (hours)

0.0

0.6

0.4

0.2

Time (hours)

0.0

Time (hours)

Tabular Foundation Models TabDPT

TabICL

1.0

0.8

0.6

0.4

0.2

0.0

Survival probability

1.0

Survival probability

TabPFN

0.6

0.4

0.2

Time (hours)

0.0

0.6

0.4

0.2

Time (hours)

0.0

Time (hours)

1.0

0.8

0.6

0.4

0.2

0.0

Survival probability

1.0

Survival probability

(a) Zero-shot)

0.6

0.4

0.2

Time (hours)

0.0

0.6

0.4

0.2

Time (hours)

0.0

Time (hours)

1.0

0.8

0.6

0.4

0.2

0.0

Survival probability

1.0

Survival probability

(b) Classification fine-tuning

0.6

0.4

0.2

Time (hours)

0.0

0.6

0.4

0.2

Time (hours)

0.0

Time (hours)

(c) Survival head adaptation (MTLR; frozen backbone) Figure 8: Risk-stratified Kaplan–Meier curves on the DIABETES data set. 29

Minh-Khoi Pham et al.

Appendix E. Full Results Table 5: Detailed Single-risk datasets Performance Metrics (mean ± std across 5 folds). Ctd : Antolini time-dependent concordance; AUC: mean dynamic AUC across q25/q50/q75. IBS: Integrated Brier Score. AML-BULL Model

Ctd ↑

AIDS2 IBS ↓

AUC ↑

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.549±0.078 0.242±0.077 0.519±0.132 RSF 0.602±0.100 0.184±0.046 0.642±0.134 GBSA 0.600±0.121 0.210±0.078 0.593±0.167 DeepSurv 0.568±0.044 0.206±0.044 0.572±0.078 MLP-MTLR 0.445±0.128 0.217±0.055 0.473±0.071 MLP-DH 0.489±0.080 0.235±0.158 0.497±0.082 SurvTRACE 0.494±0.073 0.305±0.108 0.496±0.106 DySurv 0.518±0.105 0.287±0.079 0.521±0.165 TabPFN-ZS 0.513±0.195 0.225±0.089 0.554±0.134 TabDPT-ZS 0.517±0.101 0.232±0.053 0.475±0.131 TabICL-ZS 0.439±0.126 0.217±0.105 0.497±0.091 TabPFN-CE 0.519±0.085 0.228±0.108 0.513±0.117 TabDPT-CE 0.507±0.090 0.223±0.103 0.514±0.153 TabICL-CE 0.586±0.045 0.248±0.042 0.612±0.072 TabPFN-Cox 0.540±0.121 0.195±0.040 0.523±0.154 TabDPT-Cox 0.538±0.125 0.195±0.047 0.557±0.191 TabICL-Cox 0.575±0.088 0.203±0.042 0.589±0.126 TabPFN-DH 0.556±0.076 0.223±0.085 0.537±0.124 TabDPT-DH 0.512±0.138 0.224±0.084 0.547±0.131 TabICL-DH 0.481±0.039 0.219±0.086 0.519±0.065 TabPFN-MTLR 0.564±0.051 0.209±0.084 0.571±0.052 TabDPT-MTLR 0.512±0.179 0.204±0.072 0.530±0.225 TabICL-MTLR 0.543±0.097 0.207±0.071 0.551±0.177

Cox PH 0.547±0.019 0.138±0.015 0.579±0.025 RSF 0.542±0.004 0.136±0.015 0.590±0.019 GBSA 0.458±0.053 0.138±0.014 0.558±0.021 DeepSurv 0.550±0.012 0.139±0.013 0.582±0.017 MLP-MTLR 0.537±0.031 0.146±0.023 0.570±0.034 MLP-DH 0.549±0.011 0.146±0.023 0.574±0.017 SurvTRACE 0.524±0.016 0.154±0.027 0.572±0.021 DySurv 0.549±0.016 0.157±0.034 0.583±0.022 TabPFN-ZS 0.533±0.020 0.142±0.018 0.587±0.027 TabDPT-ZS 0.527±0.007 0.144±0.019 0.585±0.021 TabICL-ZS 0.526±0.009 0.143±0.018 0.585±0.021 TabPFN-CE 0.505±0.010 0.171±0.018 0.541±0.010 TabDPT-CE 0.538±0.019 0.176±0.020 0.581±0.024 TabICL-CE 0.532±0.011 0.174±0.018 0.571±0.020 TabPFN-Cox 0.553±0.023 0.138±0.015 0.589±0.030 TabDPT-Cox 0.538±0.013 0.138±0.016 0.568±0.015 TabICL-Cox 0.536±0.022 0.138±0.015 0.565±0.031 TabPFN-DH 0.560±0.016 0.139±0.017 0.573±0.027 TabDPT-DH 0.552±0.017 0.139±0.017 0.571±0.022 TabICL-DH 0.552±0.020 0.139±0.017 0.574±0.027 TabPFN-MTLR 0.555±0.023 0.138±0.016 0.590±0.036 TabDPT-MTLR 0.546±0.011 0.138±0.017 0.580±0.025 TabICL-MTLR 0.543±0.017 0.138±0.016 0.569±0.026

BERGAMASCHI

DBCD

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH SurvTRACE DySurv TabPFN-ZS TabDPT-ZS TabICL-ZS TabPFN-CE TabDPT-CE TabICL-CE TabPFN-Cox TabDPT-Cox TabICL-Cox TabPFN-DH

Ctd ↑

IBS ↓

AUC ↑

0.602±0.064 0.167±0.040 0.645±0.143 0.663±0.155 0.187±0.060 0.801±0.092 0.626±0.242 0.229±0.149 0.784±0.149 0.486±0.114 0.223±0.064 0.521±0.053 0.451±0.076 0.285±0.167 0.484±0.227 0.502±0.030 0.405±0.216 0.446±0.056 0.625±0.113 0.247±0.066 0.781±0.095 0.551±0.029 0.412±0.192 0.566±0.057 0.615±0.071 0.336±0.230 0.664±0.144 0.539±0.110 0.313±0.222 0.617±0.133 0.608±0.077 0.322±0.231 0.625±0.066 0.597±0.135 0.356±0.273 0.605±0.216 0.547±0.082 0.407±0.245 0.505±0.133 0.584±0.097 0.397±0.255 0.551±0.195 0.565±0.128 0.193±0.053 0.651±0.210 0.431±0.089 0.209±0.047 0.539±0.161 0.421±0.216 0.216±0.071 0.439±0.310 0.570±0.054 0.406±0.247 0.563±0.202

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH SurvTRACE DySurv TabPFN-ZS TabDPT-ZS TabICL-ZS TabPFN-CE TabDPT-CE TabICL-CE TabPFN-Cox TabDPT-Cox TabICL-Cox TabPFN-DH

Ctd ↑ 0.690±0.056 0.737±0.077 0.695±0.084 0.723±0.080 0.722±0.062 0.733±0.044 0.647±0.042 0.703±0.127 0.544±0.104 0.731±0.040 0.473±0.054 0.697±0.077 0.555±0.029 0.619±0.022 0.716±0.091 0.702±0.111 0.648±0.098 0.673±0.025

IBS ↓

AUC ↑

0.236±0.092 0.756±0.027 0.146±0.019 0.794±0.066 0.156±0.025 0.773±0.067 0.163±0.032 0.782±0.059 0.209±0.043 0.785±0.018 0.210±0.065 0.747±0.044 0.181±0.042 0.681±0.066 0.314±0.074 0.747±0.149 0.237±0.058 0.507±0.121 0.244±0.078 0.766±0.048 0.248±0.072 0.464±0.077 0.253±0.061 0.755±0.102 0.239±0.059 0.586±0.023 0.208±0.032 0.669±0.041 0.152±0.025 0.784±0.064 0.160±0.032 0.757±0.081 0.163±0.028 0.690±0.087 0.306±0.060 0.714±0.055

TabFMs for Time-to-Event Prediction

TabDPT-DH TabICL-DH TabPFN-MTLR TabDPT-MTLR TabICL-MTLR

0.505±0.079 0.387±0.147 0.580±0.161 0.547±0.085 0.554±0.097

0.397±0.241 0.397±0.243 0.294±0.172 0.344±0.239 0.282±0.152

0.411±0.154 0.369±0.157 0.633±0.133 0.561±0.133 0.510±0.173

TabDPT-DH 0.740±0.079 TabICL-DH 0.675±0.063 TabPFN-MTLR 0.724±0.027 TabDPT-MTLR 0.694±0.083 TabICL-MTLR 0.680±0.061

DLBCL Model

Ctd ↑

0.255±0.082 0.265±0.072 0.203±0.060 0.178±0.043 0.194±0.057

0.768±0.075 0.691±0.066 0.743±0.039 0.724±0.103 0.696±0.065

DIALYSIS IBS ↓

AUC ↑

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.468±0.158 0.479±0.181 0.518±0.175 RSF 0.620±0.083 0.202±0.012 0.672±0.083 GBSA 0.600±0.040 0.248±0.059 0.623±0.046 DeepSurv 0.579±0.111 0.226±0.037 0.603±0.128 MLP-MTLR 0.620±0.054 0.265±0.056 0.653±0.056 MLP-DH 0.597±0.051 0.215±0.021 0.614±0.123 SurvTRACE 0.504±0.050 0.299±0.072 0.559±0.100 DySurv 0.558±0.081 0.380±0.086 0.589±0.085 TabPFN-ZS 0.517±0.066 0.230±0.050 0.585±0.074 TabDPT-ZS 0.552±0.062 0.240±0.073 0.643±0.091 TabICL-ZS 0.479±0.047 0.235±0.059 0.515±0.094 TabPFN-CE 0.547±0.024 0.279±0.086 0.640±0.045 TabDPT-CE 0.484±0.036 0.370±0.043 0.522±0.087 TabICL-CE 0.514±0.053 0.380±0.001 0.549±0.088 TabPFN-Cox 0.555±0.066 0.214±0.033 0.583±0.083 TabDPT-Cox 0.562±0.128 0.217±0.036 0.549±0.146 TabICL-Cox 0.523±0.066 0.220±0.037 0.523±0.105 TabPFN-DH 0.535±0.052 0.268±0.081 0.563±0.081 TabDPT-DH 0.544±0.078 0.263±0.072 0.552±0.136 TabICL-DH 0.499±0.031 0.265±0.084 0.490±0.054 TabPFN-MTLR 0.555±0.057 0.243±0.060 0.601±0.044 TabDPT-MTLR 0.522±0.064 0.235±0.063 0.523±0.109 TabICL-MTLR 0.505±0.076 0.233±0.063 0.530±0.102

Cox PH 0.726±0.011 0.176±0.002 0.741±0.011 RSF 0.725±0.009 0.157±0.003 0.744±0.013 GBSA 0.711±0.014 0.154±0.005 0.719±0.016 DeepSurv 0.731±0.014 0.147±0.002 0.746±0.013 MLP-MTLR 0.710±0.017 0.152±0.007 0.721±0.013 MLP-DH 0.722±0.007 0.154±0.014 0.734±0.018 SurvTRACE 0.557±0.013 0.155±0.006 0.730±0.013 DySurv 0.709±0.033 0.202±0.028 0.728±0.043 TabPFN-ZS 0.702±0.006 0.175±0.013 0.782±0.011 TabDPT-ZS 0.680±0.006 0.188±0.013 0.751±0.008 TabICL-ZS 0.695±0.006 0.176±0.013 0.776±0.009 TabPFN-CE 0.675±0.021 0.238±0.028 0.713±0.018 TabDPT-CE 0.638±0.022 0.478±0.022 0.697±0.018 TabICL-CE 0.713±0.010 0.440±0.023 0.766±0.015 TabPFN-Cox 0.730±0.013 0.144±0.002 0.759±0.010 TabDPT-Cox 0.644±0.014 0.165±0.003 0.651±0.006 TabICL-Cox 0.696±0.017 0.157±0.005 0.710±0.015 TabPFN-DH 0.724±0.005 0.212±0.010 0.747±0.014 TabDPT-DH 0.641±0.018 0.215±0.010 0.651±0.013 TabICL-DH 0.711±0.016 0.212±0.012 0.734±0.013 TabPFN-MTLR 0.729±0.008 0.151±0.003 0.768±0.010 TabDPT-MTLR 0.635±0.017 0.177±0.011 0.676±0.011 TabICL-MTLR 0.706±0.010 0.167±0.007 0.740±0.021

FRTCS

FRAMINGHAM

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH SurvTRACE DySurv TabPFN-ZS TabDPT-ZS TabICL-ZS TabPFN-CE TabDPT-CE TabICL-CE TabPFN-Cox TabDPT-Cox TabICL-Cox TabPFN-DH TabDPT-DH TabICL-DH

Ctd ↑

IBS ↓

AUC ↑

0.635±0.094 0.033±0.003 0.644±0.127 0.509±0.115 0.035±0.002 0.501±0.080 0.353±0.120 0.038±0.004 0.521±0.080 0.592±0.118 0.036±0.004 0.612±0.157 0.430±0.161 0.036±0.013 0.403±0.138 0.513±0.161 0.055±0.035 0.473±0.163 0.614±0.130 0.033±0.004 0.606±0.170 0.560±0.081 0.045±0.019 0.542±0.125 0.575±0.095 0.042±0.013 0.544±0.102 0.533±0.084 0.044±0.016 0.476±0.092 0.563±0.105 0.049±0.024 0.547±0.101 0.568±0.102 0.046±0.020 0.570±0.100 0.524±0.056 0.097±0.047 0.529±0.089 0.592±0.077 0.096±0.031 0.581±0.080 0.574±0.054 0.034±0.004 0.599±0.112 0.535±0.033 0.036±0.004 0.558±0.088 0.533±0.075 0.038±0.005 0.525±0.070 0.591±0.091 0.088±0.033 0.576±0.115 0.619±0.110 0.077±0.033 0.607±0.151 0.569±0.057 0.086±0.033 0.529±0.056

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH SurvTRACE DySurv TabPFN-ZS TabDPT-ZS TabICL-ZS TabPFN-CE TabDPT-CE TabICL-CE TabPFN-Cox TabDPT-Cox TabICL-Cox TabPFN-DH TabDPT-DH TabICL-DH

Ctd ↑

IBS ↓

AUC ↑

0.705±0.013 0.120±0.003 0.768±0.018 0.689±0.004 0.112±0.003 0.766±0.018 0.696±0.022 0.114±0.005 0.763±0.029 0.709±0.012 0.112±0.004 0.774±0.021 0.695±0.016 0.116±0.007 0.760±0.017 0.692±0.018 0.116±0.001 0.756±0.030 0.700±0.006 0.114±0.002 0.765±0.016 0.707±0.011 0.133±0.015 0.772±0.019 0.697±0.009 0.110±0.003 0.776±0.015 0.690±0.008 0.111±0.003 0.771±0.014 0.698±0.009 0.110±0.003 0.773±0.014 0.668±0.013 0.145±0.004 0.734±0.017 0.705±0.011 0.242±0.005 0.773±0.016 0.701±0.013 0.243±0.010 0.769±0.015 0.707±0.010 0.111±0.002 0.773±0.019 0.708±0.012 0.111±0.003 0.774±0.017 0.704±0.007 0.112±0.004 0.770±0.014 0.696±0.011 0.141±0.003 0.755±0.018 0.706±0.010 0.132±0.003 0.768±0.018 0.703±0.013 0.137±0.002 0.766±0.021

Minh-Khoi Pham et al.

TabPFN-MTLR TabDPT-MTLR TabICL-MTLR

0.601±0.086 0.608±0.099 0.557±0.046

0.055±0.020 0.066±0.026 0.068±0.029

0.588±0.130 0.590±0.143 0.519±0.093

TabPFN-MTLR TabDPT-MTLR TabICL-MTLR

GBSG2 Model

Ctd ↑

0.698±0.012 0.706±0.011 0.704±0.007

0.113±0.003 0.111±0.002 0.113±0.002

0.763±0.019 0.772±0.018 0.769±0.015

LEUKSURV IBS ↓

AUC ↑

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.680±0.040 0.181±0.017 0.740±0.061 RSF 0.670±0.030 0.167±0.027 0.750±0.048 GBSA 0.643±0.049 0.179±0.019 0.723±0.051 DeepSurv 0.654±0.072 0.171±0.014 0.702±0.090 MLP-MTLR 0.659±0.042 0.184±0.021 0.724±0.047 MLP-DH 0.633±0.069 0.224±0.035 0.665±0.060 SurvTRACE 0.671±0.048 0.199±0.070 0.722±0.080 DySurv 0.678±0.038 0.193±0.028 0.732±0.059 TabPFN-ZS 0.686±0.032 0.199±0.034 0.755±0.055 TabDPT-ZS 0.654±0.049 0.208±0.038 0.717±0.067 TabICL-ZS 0.687±0.044 0.205±0.040 0.753±0.063 TabPFN-CE 0.665±0.044 0.206±0.051 0.698±0.067 TabDPT-CE 0.677±0.057 0.245±0.036 0.718±0.094 TabICL-CE 0.683±0.056 0.237±0.038 0.724±0.093 TabPFN-Cox 0.681±0.038 0.165±0.017 0.742±0.060 TabDPT-Cox 0.694±0.031 0.164±0.017 0.753±0.064 TabICL-Cox 0.687±0.051 0.171±0.017 0.740±0.086 TabPFN-DH 0.676±0.042 0.233±0.038 0.728±0.050 TabDPT-DH 0.691±0.038 0.222±0.038 0.744±0.065 TabICL-DH 0.693±0.049 0.221±0.036 0.743±0.079 TabPFN-MTLR 0.679±0.046 0.187±0.019 0.741±0.072 TabDPT-MTLR 0.684±0.042 0.185±0.023 0.739±0.062 TabICL-MTLR 0.690±0.050 0.188±0.023 0.743±0.080

Cox PH 0.673±0.018 0.114±0.023 0.742±0.019 RSF 0.672±0.022 0.114±0.023 0.746±0.022 GBSA 0.658±0.033 0.114±0.024 0.736±0.022 DeepSurv 0.637±0.038 0.117±0.017 0.690±0.053 MLP-MTLR 0.642±0.019 0.125±0.019 0.704±0.034 MLP-DH 0.639±0.015 0.138±0.024 0.686±0.027 SurvTRACE 0.544±0.018 0.131±0.048 0.705±0.022 DySurv 0.624±0.086 0.137±0.034 0.715±0.027 TabPFN-ZS 0.567±0.021 0.117±0.027 0.751±0.030 TabDPT-ZS 0.544±0.012 0.119±0.030 0.730±0.029 TabICL-ZS 0.563±0.020 0.118±0.029 0.747±0.030 TabPFN-CE 0.550±0.020 0.126±0.030 0.707±0.028 TabDPT-CE 0.581±0.021 0.128±0.032 0.748±0.023 TabICL-CE 0.577±0.017 0.127±0.032 0.745±0.021 TabPFN-Cox 0.662±0.015 0.114±0.019 0.726±0.011 TabDPT-Cox 0.677±0.018 0.114±0.021 0.745±0.021 TabICL-Cox 0.671±0.019 0.113±0.019 0.740±0.020 TabPFN-DH 0.652±0.021 0.128±0.030 0.704±0.024 TabDPT-DH 0.681±0.019 0.125±0.029 0.744±0.024 TabICL-DH 0.677±0.022 0.125±0.029 0.737±0.019 TabPFN-MTLR 0.652±0.028 0.116±0.025 0.716±0.028 TabDPT-MTLR 0.677±0.016 0.114±0.026 0.744±0.021 TabICL-MTLR 0.677±0.022 0.114±0.025 0.743±0.025

MCLCLEANED

MELANOMA

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.656±0.112 0.180±0.060 0.724±0.090 RSF 0.712±0.074 0.136±0.037 0.747±0.130 GBSA 0.719±0.068 0.140±0.053 0.767±0.108 DeepSurv 0.546±0.200 0.165±0.058 0.561±0.238 MLP-MTLR 0.668±0.151 0.161±0.049 0.725±0.158 MLP-DH 0.662±0.083 0.145±0.037 0.730±0.150 SurvTRACE 0.708±0.068 0.137±0.024 0.778±0.133 DySurv 0.663±0.115 0.156±0.052 0.690±0.158 TabPFN-ZS 0.686±0.101 0.143±0.049 0.745±0.149 TabDPT-ZS 0.679±0.104 0.156±0.056 0.725±0.134 TabICL-ZS 0.679±0.093 0.142±0.042 0.712±0.176 TabPFN-CE 0.704±0.095 0.161±0.059 0.706±0.178 TabDPT-CE 0.640±0.108 0.155±0.050 0.667±0.178 TabICL-CE 0.597±0.132 0.195±0.023 0.595±0.143 TabPFN-Cox 0.718±0.090 0.141±0.044 0.773±0.114 TabDPT-Cox 0.634±0.144 0.153±0.055 0.667±0.181 TabICL-Cox 0.541±0.078 0.152±0.035 0.549±0.111 TabPFN-DH 0.637±0.066 0.154±0.036 0.658±0.115 TabDPT-DH 0.526±0.111 0.154±0.039 0.562±0.057 TabICL-DH 0.581±0.054 0.154±0.036 0.567±0.058 TabPFN-MTLR 0.713±0.118 0.130±0.021 0.736±0.175 TabDPT-MTLR 0.624±0.112 0.148±0.035 0.639±0.130

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.698±0.035 0.160±0.039 0.763±0.097 RSF 0.705±0.046 0.153±0.041 0.783±0.090 GBSA 0.657±0.029 0.183±0.070 0.726±0.050 DeepSurv 0.648±0.088 0.158±0.043 0.684±0.128 MLP-MTLR 0.629±0.101 0.165±0.063 0.678±0.132 MLP-DH 0.609±0.159 0.262±0.208 0.646±0.138 SurvTRACE 0.657±0.127 0.238±0.175 0.697±0.138 DySurv 0.637±0.115 0.365±0.404 0.640±0.104 TabPFN-ZS 0.713±0.070 0.158±0.081 0.771±0.103 TabDPT-ZS 0.716±0.067 0.137±0.052 0.766±0.093 TabICL-ZS 0.712±0.098 0.147±0.086 0.759±0.120 TabPFN-CE 0.666±0.054 0.307±0.281 0.659±0.064 TabDPT-CE 0.568±0.111 0.314±0.295 0.534±0.146 TabICL-CE 0.695±0.056 0.318±0.300 0.686±0.036 TabPFN-Cox 0.704±0.027 0.154±0.043 0.777±0.089 TabDPT-Cox 0.646±0.084 0.164±0.053 0.691±0.143 TabICL-Cox 0.559±0.128 0.163±0.042 0.583±0.186 TabPFN-DH 0.724±0.076 0.310±0.225 0.785±0.079 TabDPT-DH 0.628±0.129 0.295±0.216 0.703±0.105 TabICL-DH 0.612±0.110 0.323±0.231 0.595±0.138 TabPFN-MTLR 0.716±0.082 0.201±0.150 0.759±0.100 TabDPT-MTLR 0.686±0.101 0.212±0.147 0.716±0.116

TabFMs for Time-to-Event Prediction

TabICL-MTLR

0.543±0.098

0.148±0.040

0.598±0.170

TabICL-MTLR

NSBCD Model

Ctd ↑

0.680±0.041

0.261±0.212

0.688±0.064

IBS ↓

AUC ↑

PBC3 IBS ↓

AUC ↑

Model

Ctd ↑

Cox PH 0.645±0.105 0.171±0.046 0.683±0.115 RSF 0.692±0.092 0.131±0.026 0.762±0.080 GBSA 0.591±0.119 0.136±0.016 0.636±0.091 DeepSurv 0.623±0.108 0.181±0.045 0.682±0.146 MLP-MTLR 0.646±0.060 0.174±0.051 0.622±0.092 MLP-DH 0.569±0.141 0.167±0.075 0.550±0.157 SurvTRACE 0.665±0.123 0.190±0.089 0.722±0.108 DySurv 0.652±0.089 0.242±0.119 0.666±0.120 TabPFN-ZS 0.661±0.085 0.166±0.098 0.672±0.142 TabDPT-ZS 0.584±0.056 0.240±0.103 0.606±0.067 TabICL-ZS 0.715±0.068 0.164±0.107 0.754±0.049 TabPFN-CE 0.688±0.023 0.141±0.059 0.699±0.067 TabDPT-CE 0.568±0.097 0.163±0.085 0.617±0.108 TabICL-CE 0.619±0.033 0.131±0.049 0.654±0.098 TabPFN-Cox 0.592±0.221 0.154±0.038 0.572±0.229 TabDPT-Cox 0.477±0.202 0.170±0.014 0.491±0.216 TabICL-Cox 0.529±0.156 0.168±0.017 0.554±0.126 TabPFN-DH 0.612±0.154 0.185±0.068 0.562±0.143 TabDPT-DH 0.580±0.163 0.191±0.078 0.611±0.243 TabICL-DH 0.420±0.094 0.191±0.078 0.396±0.122 TabPFN-MTLR 0.577±0.220 0.166±0.055 0.587±0.228 TabDPT-MTLR 0.644±0.184 0.149±0.045 0.696±0.161 TabICL-MTLR 0.466±0.089 0.163±0.048 0.579±0.119

Cox PH 0.774±0.119 0.106±0.024 0.853±0.038 RSF 0.724±0.145 0.110±0.015 0.835±0.050 GBSA 0.662±0.124 0.124±0.020 0.770±0.100 DeepSurv 0.731±0.127 0.115±0.029 0.804±0.064 MLP-MTLR 0.815±0.039 0.130±0.038 0.845±0.083 MLP-DH 0.766±0.063 0.156±0.027 0.820±0.072 SurvTRACE 0.805±0.052 0.111±0.018 0.818±0.081 DySurv 0.752±0.120 0.270±0.204 0.809±0.114 TabPFN-ZS 0.803±0.037 0.135±0.029 0.825±0.056 TabDPT-ZS 0.817±0.028 0.155±0.046 0.866±0.050 TabICL-ZS 0.807±0.044 0.156±0.046 0.851±0.063 TabPFN-CE 0.786±0.060 0.163±0.049 0.788±0.043 TabDPT-CE 0.759±0.056 0.164±0.051 0.765±0.055 TabICL-CE 0.800±0.067 0.179±0.050 0.796±0.040 TabPFN-Cox 0.743±0.127 0.119±0.019 0.841±0.051 TabDPT-Cox 0.774±0.102 0.111±0.012 0.870±0.062 TabICL-Cox 0.765±0.114 0.123±0.016 0.845±0.073 TabPFN-DH 0.696±0.099 0.289±0.064 0.744±0.133 TabDPT-DH 0.827±0.032 0.270±0.059 0.867±0.059 TabICL-DH 0.778±0.089 0.278±0.052 0.815±0.074 TabPFN-MTLR 0.791±0.058 0.210±0.048 0.803±0.110 TabDPT-MTLR 0.823±0.022 0.168±0.035 0.859±0.068 TabICL-MTLR 0.812±0.045 0.188±0.034 0.840±0.071

ROSSI

TRACE

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.590±0.116 0.087±0.007 0.578±0.114 RSF 0.584±0.084 0.084±0.007 0.566±0.087 GBSA 0.527±0.132 0.086±0.007 0.576±0.132 DeepSurv 0.577±0.077 0.087±0.008 0.542±0.073 MLP-MTLR 0.556±0.097 0.106±0.005 0.580±0.110 MLP-DH 0.512±0.118 0.178±0.080 0.473±0.113 SurvTRACE 0.596±0.078 0.088±0.007 0.580±0.095 DySurv 0.517±0.088 0.142±0.053 0.486±0.081 TabPFN-ZS 0.620±0.078 0.110±0.029 0.579±0.137 TabDPT-ZS 0.661±0.056 0.109±0.030 0.621±0.073 TabICL-ZS 0.724±0.053 0.112±0.034 0.713±0.046 TabPFN-CE 0.556±0.072 0.176±0.078 0.578±0.073 TabDPT-CE 0.424±0.067 0.241±0.088 0.457±0.093 TabICL-CE 0.556±0.063 0.266±0.073 0.597±0.082 TabPFN-Cox 0.574±0.080 0.087±0.008 0.566±0.101 TabDPT-Cox 0.641±0.065 0.085±0.009 0.616±0.084 TabICL-Cox 0.681±0.118 0.084±0.008 0.639±0.125 TabPFN-DH 0.534±0.083 0.199±0.042 0.539±0.105 TabDPT-DH 0.663±0.054 0.161±0.053 0.656±0.089 TabICL-DH 0.677±0.120 0.166±0.053 0.639±0.141 TabPFN-MTLR 0.588±0.096 0.139±0.043 0.594±0.115 TabDPT-MTLR 0.639±0.061 0.123±0.032 0.621±0.096 TabICL-MTLR 0.682±0.108 0.129±0.037 0.647±0.132

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.736±0.014 0.177±0.004 0.789±0.023 RSF 0.717±0.017 0.162±0.006 0.791±0.026 GBSA 0.740±0.015 0.160±0.008 0.796±0.027 DeepSurv 0.734±0.015 0.163±0.008 0.788±0.024 MLP-MTLR 0.702±0.045 0.181±0.031 0.757±0.051 MLP-DH 0.728±0.012 0.179±0.020 0.781±0.019 SurvTRACE 0.700±0.019 0.171±0.014 0.774±0.032 DySurv 0.734±0.016 0.164±0.007 0.788±0.026 TabPFN-ZS 0.713±0.017 0.172±0.006 0.792±0.026 TabDPT-ZS 0.704±0.017 0.176±0.006 0.787±0.026 TabICL-ZS 0.710±0.017 0.173±0.007 0.789±0.031 TabPFN-CE 0.632±0.017 0.242±0.013 0.704±0.029 TabDPT-CE 0.711±0.016 0.272±0.009 0.788±0.020 TabICL-CE 0.707±0.012 0.277±0.009 0.785±0.022 TabPFN-Cox 0.737±0.012 0.161±0.006 0.790±0.025 TabDPT-Cox 0.741±0.014 0.160±0.006 0.795±0.025 TabICL-Cox 0.736±0.012 0.161±0.005 0.789±0.023 TabPFN-DH 0.737±0.010 0.217±0.002 0.785±0.019 TabDPT-DH 0.741±0.017 0.220±0.007 0.791±0.025 TabICL-DH 0.739±0.012 0.216±0.006 0.790±0.021 TabPFN-MTLR 0.735±0.015 0.163±0.006 0.789±0.024 TabDPT-MTLR 0.739±0.015 0.162±0.006 0.794±0.027 TabICL-MTLR 0.736±0.017 0.163±0.006 0.792±0.027

Minh-Khoi Pham et al.

UNEMPDUR Model

Ctd ↑

IBS ↓

UNEMPLOYMENT AUC ↑

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.687±0.013 0.158±0.013 0.731±0.020 RSF 0.692±0.010 0.156±0.017 0.737±0.014 GBSA 0.664±0.009 0.156±0.014 0.740±0.015 DeepSurv 0.689±0.014 0.158±0.014 0.734±0.017 MLP-MTLR 0.689±0.009 0.160±0.020 0.732±0.012 MLP-DH 0.692±0.014 0.162±0.020 0.736±0.023 SurvTRACE 0.399±0.008 0.163±0.014 0.734±0.019 DySurv 0.667±0.061 0.156±0.019 0.736±0.021 TabPFN-ZS 0.694±0.002 0.161±0.022 0.740±0.014 TabDPT-ZS 0.689±0.007 0.163±0.022 0.733±0.013 TabICL-ZS 0.694±0.009 0.161±0.022 0.739±0.010 TabPFN-CE 0.654±0.015 0.174±0.024 0.679±0.021 TabDPT-CE 0.691±0.014 0.215±0.028 0.736±0.017 TabICL-CE 0.685±0.003 0.210±0.025 0.728±0.015 TabPFN-Cox 0.695±0.008 0.158±0.014 0.740±0.014 TabDPT-Cox 0.693±0.012 0.158±0.015 0.738±0.018 TabICL-Cox 0.693±0.008 0.158±0.014 0.736±0.016 TabPFN-DH 0.693±0.012 0.167±0.019 0.729±0.020 TabDPT-DH 0.692±0.007 0.167±0.018 0.732±0.013 TabICL-DH 0.694±0.009 0.168±0.018 0.739±0.017 TabPFN-MTLR 0.691±0.009 0.154±0.020 0.736±0.013 TabDPT-MTLR 0.695±0.004 0.155±0.020 0.737±0.013 TabICL-MTLR 0.694±0.006 0.155±0.019 0.737±0.015

Cox PH 0.515±0.024 0.184±0.033 0.552±0.028 RSF 0.527±0.040 0.188±0.043 0.557±0.028 GBSA 0.479±0.036 0.187±0.036 0.523±0.056 DeepSurv 0.496±0.065 0.187±0.026 0.535±0.078 MLP-MTLR 0.509±0.031 0.197±0.033 0.552±0.035 MLP-DH 0.522±0.020 0.196±0.062 0.557±0.061 SurvTRACE 0.397±0.051 0.205±0.047 0.521±0.053 DySurv 0.516±0.030 0.215±0.071 0.545±0.036 TabPFN-ZS 0.504±0.038 0.195±0.063 0.554±0.037 TabDPT-ZS 0.515±0.045 0.195±0.060 0.547±0.037 TabICL-ZS 0.527±0.038 0.195±0.063 0.559±0.032 TabPFN-CE 0.512±0.027 0.241±0.066 0.556±0.026 TabDPT-CE 0.516±0.020 0.239±0.068 0.543±0.045 TabICL-CE 0.519±0.034 0.235±0.063 0.549±0.051 TabPFN-Cox 0.485±0.034 0.188±0.033 0.510±0.029 TabDPT-Cox 0.488±0.031 0.184±0.039 0.510±0.026 TabICL-Cox 0.493±0.031 0.183±0.047 0.516±0.031 TabPFN-DH 0.524±0.035 0.180±0.041 0.531±0.055 TabDPT-DH 0.534±0.032 0.182±0.049 0.541±0.043 TabICL-DH 0.531±0.042 0.182±0.046 0.544±0.035 TabPFN-MTLR 0.510±0.024 0.184±0.037 0.523±0.030 TabDPT-MTLR 0.529±0.033 0.184±0.047 0.540±0.024 TabICL-MTLR 0.524±0.031 0.185±0.053 0.544±0.023

Z243

ACATH

Model

Ctd ↑

IBS ↓

AUC ↑

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.908±0.036 0.033±0.017 0.952±0.028 RSF 0.816±0.033 0.062±0.021 0.900±0.054 GBSA 0.909±0.034 0.033±0.012 0.958±0.018 DeepSurv 0.747±0.117 0.053±0.026 0.788±0.126 MLP-MTLR 0.828±0.055 0.049±0.018 0.929±0.049 MLP-DH 0.784±0.107 0.056±0.018 0.892±0.104 SurvTRACE 0.676±0.144 0.049±0.018 0.911±0.046 DySurv 0.569±0.204 0.111±0.044 0.607±0.250 TabPFN-ZS 0.918±0.024 0.035±0.019 0.958±0.027 TabDPT-ZS 0.908±0.059 0.032±0.014 0.958±0.042 TabICL-ZS 0.871±0.100 0.036±0.015 0.938±0.068 TabPFN-CE 0.926±0.017 0.034±0.016 0.969±0.009 TabDPT-CE 0.920±0.016 0.037±0.020 0.960±0.012 TabICL-CE 0.912±0.033 0.043±0.019 0.957±0.018 TabPFN-Cox 0.637±0.223 0.070±0.013 0.654±0.278 TabDPT-Cox 0.534±0.144 0.076±0.019 0.545±0.192 TabICL-Cox 0.597±0.083 0.075±0.019 0.648±0.110 TabPFN-DH 0.598±0.120 0.082±0.017 0.681±0.105 TabDPT-DH 0.646±0.137 0.084±0.014 0.659±0.133 TabICL-DH 0.518±0.087 0.082±0.017 0.576±0.116 TabPFN-MTLR 0.575±0.183 0.080±0.008 0.606±0.208 TabDPT-MTLR 0.609±0.073 0.080±0.019 0.657±0.131 TabICL-MTLR 0.449±0.075 0.082±0.021 0.425±0.178

Cox PH 0.600±0.014 0.092±0.010 0.616±0.028 RSF 0.594±0.016 0.098±0.011 0.609±0.028 GBSA 0.579±0.028 0.093±0.011 0.617±0.024 DeepSurv 0.605±0.010 0.093±0.013 0.619±0.024 MLP-MTLR 0.604±0.016 0.093±0.009 0.621±0.029 MLP-DH 0.596±0.023 0.093±0.011 0.610±0.032 SurvTRACE 0.524±0.016 0.094±0.009 0.610±0.038 DySurv 0.572±0.032 0.093±0.010 0.603±0.027 TabPFN-ZS 0.596±0.021 0.092±0.011 0.623±0.022 TabDPT-ZS 0.598±0.012 0.092±0.011 0.620±0.019 TabICL-ZS 0.593±0.017 0.092±0.011 0.614±0.019 TabPFN-CE 0.541±0.022 0.114±0.015 0.553±0.021 TabDPT-CE 0.596±0.019 0.115±0.014 0.623±0.030 TabICL-CE 0.590±0.014 0.116±0.015 0.615±0.025 TabPFN-Cox 0.610±0.014 0.091±0.010 0.625±0.024 TabDPT-Cox 0.607±0.011 0.091±0.010 0.621±0.018 TabICL-Cox 0.602±0.012 0.092±0.010 0.614±0.021 TabPFN-DH 0.599±0.017 0.095±0.010 0.618±0.024 TabDPT-DH 0.599±0.010 0.096±0.010 0.618±0.022 TabICL-DH 0.599±0.017 0.095±0.011 0.617±0.025 TabPFN-MTLR 0.606±0.017 0.091±0.010 0.627±0.026 TabDPT-MTLR 0.601±0.020 0.091±0.011 0.622±0.027 TabICL-MTLR 0.602±0.015 0.091±0.010 0.620±0.023

ACTG

AIDS

TabFMs for Time-to-Event Prediction

Model

Ctd ↑

IBS ↓

AUC ↑

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.735±0.063 0.060±0.003 0.742±0.083 RSF 0.717±0.044 0.057±0.005 0.737±0.042 GBSA 0.691±0.059 0.059±0.002 0.730±0.076 DeepSurv 0.698±0.037 0.058±0.003 0.716±0.061 MLP-MTLR 0.710±0.058 0.112±0.036 0.727±0.073 MLP-DH 0.721±0.034 0.161±0.050 0.731±0.067 SurvTRACE 0.693±0.087 0.060±0.008 0.669±0.140 DySurv 0.724±0.046 0.197±0.108 0.713±0.087 TabPFN-ZS 0.710±0.058 0.061±0.010 0.747±0.041 TabDPT-ZS 0.700±0.064 0.062±0.009 0.705±0.060 TabICL-ZS 0.721±0.042 0.062±0.008 0.758±0.031 TabPFN-CE 0.686±0.064 0.119±0.037 0.678±0.102 TabDPT-CE 0.673±0.100 0.176±0.081 0.677±0.091 TabICL-CE 0.684±0.046 0.189±0.054 0.698±0.066 TabPFN-Cox 0.728±0.044 0.058±0.004 0.718±0.075 TabDPT-Cox 0.738±0.047 0.058±0.004 0.758±0.062 TabICL-Cox 0.731±0.031 0.058±0.003 0.756±0.037 TabPFN-DH 0.730±0.076 0.288±0.054 0.702±0.080 TabDPT-DH 0.736±0.052 0.184±0.031 0.741±0.066 TabICL-DH 0.751±0.054 0.215±0.046 0.763±0.052 TabPFN-MTLR 0.725±0.070 0.097±0.011 0.698±0.073 TabDPT-MTLR 0.739±0.049 0.090±0.010 0.746±0.045 TabICL-MTLR 0.741±0.044 0.091±0.016 0.755±0.042

Cox PH 0.701±0.033 0.181±0.023 0.539±0.040 RSF 0.664±0.050 0.181±0.023 0.535±0.050 GBSA 0.567±0.084 0.188±0.017 0.536±0.041 DeepSurv 0.703±0.032 0.178±0.022 0.551±0.041 MLP-MTLR 0.679±0.033 0.167±0.032 0.537±0.053 MLP-DH 0.650±0.038 0.214±0.044 0.544±0.048 SurvTRACE 0.335±0.084 0.191±0.037 0.554±0.051 DySurv 0.701±0.030 0.215±0.112 0.552±0.039 TabPFN-ZS 0.673±0.024 0.247±0.089 0.534±0.052 TabDPT-ZS 0.654±0.020 0.249±0.094 0.532±0.059 TabICL-ZS 0.691±0.018 0.255±0.102 0.540±0.051 TabPFN-CE 0.586±0.045 0.327±0.128 0.513±0.038 TabDPT-CE 0.675±0.013 0.405±0.130 0.538±0.044 TabICL-CE 0.672±0.010 0.374±0.111 0.529±0.053 TabPFN-Cox 0.698±0.034 0.179±0.020 0.542±0.040 TabDPT-Cox 0.681±0.022 0.184±0.022 0.528±0.048 TabICL-Cox 0.690±0.039 0.185±0.020 0.529±0.036 TabPFN-DH 0.673±0.038 0.217±0.066 0.530±0.049 TabDPT-DH 0.685±0.028 0.214±0.084 0.539±0.043 TabICL-DH 0.697±0.028 0.222±0.066 0.533±0.032 TabPFN-MTLR 0.683±0.034 0.171±0.044 0.540±0.048 TabDPT-MTLR 0.690±0.019 0.178±0.050 0.541±0.043 TabICL-MTLR 0.689±0.029 0.179±0.049 0.530±0.047

BREAST

BURN

Model

Ctd ↑

IBS ↓

AUC ↑

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.761±0.066 0.107±0.019 0.805±0.120 RSF 0.655±0.102 0.103±0.016 0.741±0.102 GBSA 0.706±0.089 0.103±0.014 0.755±0.148 DeepSurv 0.736±0.077 0.109±0.024 0.773±0.121 MLP-MTLR 0.624±0.142 0.128±0.030 0.640±0.171 MLP-DH 0.562±0.121 0.217±0.105 0.637±0.091 SurvTRACE 0.661±0.083 0.226±0.205 0.731±0.131 DySurv 0.397±0.221 0.426±0.387 0.420±0.215 TabPFN-ZS 0.674±0.088 0.113±0.023 0.756±0.074 TabDPT-ZS 0.632±0.111 0.115±0.026 0.737±0.109 TabICL-ZS 0.583±0.144 0.125±0.029 0.666±0.167 TabPFN-CE 0.731±0.074 0.254±0.065 0.789±0.110 TabDPT-CE 0.704±0.095 0.320±0.091 0.770±0.145 TabICL-CE 0.740±0.068 0.302±0.059 0.789±0.126 TabPFN-Cox 0.627±0.173 0.116±0.032 0.653±0.173 TabDPT-Cox 0.414±0.197 0.120±0.026 0.507±0.214 TabICL-Cox 0.728±0.057 0.113±0.026 0.768±0.118 TabPFN-DH 0.559±0.288 0.219±0.041 0.583±0.279 TabDPT-DH 0.604±0.135 0.273±0.047 0.727±0.174 TabICL-DH 0.649±0.189 0.242±0.044 0.643±0.249 TabPFN-MTLR 0.695±0.076 0.138±0.050 0.737±0.072 TabDPT-MTLR 0.609±0.170 0.224±0.080 0.650±0.205 TabICL-MTLR 0.629±0.169 0.148±0.056 0.645±0.190

Cox PH 0.637±0.100 0.202±0.090 0.671±0.139 RSF 0.523±0.035 0.223±0.063 0.506±0.086 GBSA 0.566±0.125 0.212±0.067 0.623±0.192 DeepSurv 0.514±0.140 0.240±0.138 0.478±0.175 MLP-MTLR 0.447±0.058 0.288±0.101 0.477±0.102 MLP-DH 0.460±0.065 0.242±0.096 0.450±0.068 SurvTRACE 0.569±0.096 0.379±0.155 0.648±0.068 DySurv 0.534±0.073 0.283±0.191 0.569±0.114 TabPFN-ZS 0.541±0.060 0.264±0.144 0.608±0.102 TabDPT-ZS 0.608±0.115 0.268±0.147 0.611±0.112 TabICL-ZS 0.585±0.120 0.256±0.105 0.614±0.125 TabPFN-CE 0.553±0.036 0.319±0.154 0.560±0.087 TabDPT-CE 0.529±0.143 0.295±0.152 0.557±0.141 TabICL-CE 0.586±0.040 0.296±0.129 0.609±0.124 TabPFN-Cox 0.542±0.108 0.202±0.051 0.536±0.142 TabDPT-Cox 0.581±0.131 0.198±0.036 0.572±0.138 TabICL-Cox 0.565±0.173 0.200±0.039 0.551±0.234 TabPFN-DH 0.584±0.075 0.248±0.110 0.562±0.063 TabDPT-DH 0.591±0.137 0.245±0.110 0.615±0.206 TabICL-DH 0.489±0.065 0.253±0.117 0.464±0.108 TabPFN-MTLR 0.526±0.166 0.220±0.118 0.546±0.138 TabDPT-MTLR 0.631±0.086 0.211±0.101 0.675±0.065 TabICL-MTLR 0.547±0.063 0.244±0.119 0.562±0.098

CANCER

CGD

Model

Ctd ↑

IBS ↓

AUC ↑

Model

Ctd ↑

IBS ↓

AUC ↑

Minh-Khoi Pham et al.

Cox PH 0.589±0.074 0.145±0.018 0.606±0.089 RSF 0.588±0.075 0.144±0.021 0.642±0.100 GBSA 0.537±0.036 0.163±0.050 0.571±0.082 DeepSurv 0.515±0.064 0.160±0.024 0.515±0.092 MLP-MTLR 0.488±0.084 0.243±0.055 0.541±0.090 MLP-DH 0.545±0.127 0.210±0.071 0.581±0.157 SurvTRACE 0.596±0.059 0.193±0.043 0.618±0.077 DySurv 0.596±0.069 0.204±0.041 0.630±0.092 TabPFN-ZS 0.615±0.058 0.151±0.023 0.595±0.078 TabDPT-ZS 0.603±0.070 0.149±0.023 0.624±0.087 TabICL-ZS 0.566±0.042 0.162±0.036 0.558±0.055 TabPFN-CE 0.522±0.069 0.161±0.011 0.515±0.106 TabDPT-CE 0.579±0.039 0.157±0.029 0.584±0.069 TabICL-CE 0.578±0.072 0.150±0.018 0.574±0.065 TabPFN-Cox 0.572±0.093 0.153±0.010 0.584±0.125 TabDPT-Cox 0.578±0.055 0.145±0.021 0.602±0.083 TabICL-Cox 0.569±0.051 0.146±0.023 0.581±0.101 TabPFN-DH 0.559±0.082 0.148±0.022 0.544±0.096 TabDPT-DH 0.559±0.033 0.148±0.024 0.554±0.073 TabICL-DH 0.498±0.046 0.150±0.022 0.513±0.062 TabPFN-MTLR 0.565±0.072 0.155±0.009 0.573±0.093 TabDPT-MTLR 0.616±0.031 0.146±0.024 0.664±0.061 TabICL-MTLR 0.536±0.085 0.147±0.021 0.556±0.093

Cox PH 0.554±0.085 0.163±0.012 0.558±0.109 RSF 0.508±0.130 0.160±0.008 0.548±0.212 GBSA 0.571±0.163 0.174±0.035 0.602±0.167 DeepSurv 0.618±0.082 0.155±0.016 0.625±0.084 MLP-MTLR 0.516±0.068 0.239±0.084 0.478±0.098 MLP-DH 0.423±0.084 0.326±0.210 0.431±0.165 SurvTRACE 0.454±0.086 0.453±0.248 0.486±0.069 DySurv 0.418±0.094 0.393±0.222 0.385±0.170 TabPFN-ZS 0.538±0.097 0.191±0.031 0.545±0.147 TabDPT-ZS 0.527±0.107 0.199±0.036 0.517±0.147 TabICL-ZS 0.497±0.113 0.196±0.028 0.530±0.167 TabPFN-CE 0.518±0.051 0.213±0.030 0.540±0.111 TabDPT-CE 0.434±0.069 0.242±0.041 0.408±0.087 TabICL-CE 0.562±0.083 0.241±0.039 0.554±0.121 TabPFN-Cox 0.521±0.070 0.166±0.021 0.538±0.075 TabDPT-Cox 0.528±0.063 0.161±0.016 0.519±0.108 TabICL-Cox 0.539±0.043 0.162±0.017 0.507±0.066 TabPFN-DH 0.471±0.126 0.242±0.023 0.459±0.180 TabDPT-DH 0.461±0.084 0.230±0.030 0.452±0.155 TabICL-DH 0.478±0.113 0.238±0.026 0.488±0.152 TabPFN-MTLR 0.461±0.050 0.221±0.037 0.432±0.025 TabDPT-MTLR 0.415±0.081 0.183±0.017 0.388±0.112 TabICL-MTLR 0.463±0.069 0.210±0.033 0.475±0.075

CHOP

COLON

Model

Ctd ↑

IBS ↓

AUC ↑

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.651±0.044 0.289±0.243 0.665±0.052 RSF 0.675±0.054 0.216±0.119 0.702±0.053 GBSA 0.630±0.039 0.245±0.185 0.648±0.040 DeepSurv 0.646±0.078 0.237±0.156 0.679±0.094 MLP-MTLR 0.657±0.036 0.298±0.173 0.687±0.058 MLP-DH 0.617±0.058 0.251±0.134 0.634±0.077 SurvTRACE 0.569±0.063 0.236±0.075 0.585±0.084 DySurv 0.655±0.066 0.354±0.211 0.678±0.061 TabPFN-ZS 0.599±0.077 0.221±0.154 0.607±0.062 TabDPT-ZS 0.611±0.054 0.279±0.250 0.653±0.043 TabICL-ZS 0.520±0.039 0.264±0.223 0.577±0.044 TabPFN-CE 0.572±0.045 0.307±0.251 0.618±0.069 TabDPT-CE 0.619±0.084 0.324±0.268 0.672±0.066 TabICL-CE 0.562±0.038 0.381±0.002 0.605±0.053 TabPFN-Cox 0.620±0.042 0.213±0.112 0.631±0.062 TabDPT-Cox 0.657±0.040 0.229±0.127 0.681±0.054 TabICL-Cox 0.579±0.053 0.210±0.110 0.573±0.079 TabPFN-DH 0.548±0.106 0.262±0.214 0.579±0.096 TabDPT-DH 0.648±0.090 0.264±0.220 0.643±0.134 TabICL-DH 0.563±0.051 0.267±0.220 0.576±0.078 TabPFN-MTLR 0.604±0.050 0.278±0.241 0.618±0.072 TabDPT-MTLR 0.648±0.079 0.229±0.163 0.655±0.087 TabICL-MTLR 0.564±0.068 0.270±0.227 0.558±0.089

Cox PH 0.651±0.023 0.188±0.018 0.704±0.046 RSF 0.641±0.014 0.182±0.018 0.692±0.040 GBSA 0.645±0.026 0.185±0.022 0.684±0.049 DeepSurv 0.625±0.050 0.193±0.021 0.675±0.072 MLP-MTLR 0.616±0.024 0.207±0.019 0.660±0.040 MLP-DH 0.620±0.023 0.223±0.039 0.655±0.063 SurvTRACE 0.616±0.043 0.345±0.179 0.646±0.084 DySurv 0.637±0.025 0.315±0.138 0.684±0.046 TabPFN-ZS 0.646±0.034 0.232±0.055 0.699±0.037 TabDPT-ZS 0.641±0.037 0.231±0.055 0.686±0.043 TabICL-ZS 0.653±0.036 0.230±0.054 0.695±0.039 TabPFN-CE 0.606±0.021 0.256±0.033 0.608±0.024 TabDPT-CE 0.631±0.015 0.322±0.045 0.654±0.020 TabICL-CE 0.642±0.025 0.323±0.049 0.667±0.035 TabPFN-Cox 0.642±0.025 0.184±0.019 0.694±0.050 TabDPT-Cox 0.649±0.023 0.182±0.021 0.694±0.044 TabICL-Cox 0.646±0.023 0.182±0.021 0.684±0.045 TabPFN-DH 0.638±0.024 0.261±0.042 0.666±0.042 TabDPT-DH 0.652±0.028 0.248±0.046 0.684±0.053 TabICL-DH 0.646±0.025 0.244±0.039 0.684±0.038 TabPFN-MTLR 0.632±0.031 0.208±0.038 0.680±0.046 TabDPT-MTLR 0.647±0.015 0.205±0.039 0.694±0.036 TabICL-MTLR 0.647±0.024 0.206±0.042 0.686±0.044

COST

CSL

Model

Ctd ↑

IBS ↓

AUC ↑

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH RSF

0.676±0.033 0.670±0.028

0.178±0.007 0.172±0.008

0.730±0.042 0.733±0.031

Cox PH RSF

0.751±0.048 0.746±0.054

0.124±0.020 0.137±0.023

0.792±0.027 0.805±0.039

TabFMs for Time-to-Event Prediction

GBSA 0.647±0.026 0.179±0.012 0.696±0.034 DeepSurv 0.650±0.043 0.174±0.010 0.696±0.062 MLP-MTLR 0.632±0.042 0.224±0.055 0.668±0.073 MLP-DH 0.628±0.047 0.211±0.027 0.652±0.059 SurvTRACE 0.623±0.053 0.332±0.100 0.695±0.074 DySurv 0.678±0.023 0.214±0.051 0.733±0.027 TabPFN-ZS 0.656±0.030 0.168±0.010 0.734±0.030 TabDPT-ZS 0.643±0.044 0.170±0.014 0.722±0.048 TabICL-ZS 0.663±0.033 0.167±0.012 0.735±0.027 TabPFN-CE 0.637±0.026 0.213±0.027 0.695±0.035 TabDPT-CE 0.667±0.025 0.185±0.013 0.735±0.031 TabICL-CE 0.671±0.028 0.181±0.012 0.747±0.020 TabPFN-Cox 0.681±0.032 0.169±0.011 0.737±0.039 TabDPT-Cox 0.683±0.029 0.168±0.013 0.739±0.039 TabICL-Cox 0.678±0.025 0.170±0.009 0.733±0.032 TabPFN-DH 0.679±0.028 0.207±0.013 0.726±0.041 TabDPT-DH 0.690±0.027 0.197±0.010 0.743±0.038 TabICL-DH 0.683±0.020 0.199±0.007 0.738±0.024 TabPFN-MTLR 0.669±0.026 0.175±0.015 0.722±0.036 TabDPT-MTLR 0.691±0.024 0.164±0.008 0.747±0.033 TabICL-MTLR 0.681±0.019 0.167±0.008 0.737±0.022

GBSA 0.744±0.041 DeepSurv 0.750±0.052 MLP-MTLR 0.707±0.030 MLP-DH 0.737±0.025 SurvTRACE 0.757±0.021 DySurv 0.743±0.059 TabPFN-ZS 0.768±0.041 TabDPT-ZS 0.773±0.046 TabICL-ZS 0.806±0.032 TabPFN-CE 0.783±0.030 TabDPT-CE 0.738±0.033 TabICL-CE 0.735±0.042 TabPFN-Cox 0.752±0.050 TabDPT-Cox 0.767±0.060 TabICL-Cox 0.781±0.051 TabPFN-DH 0.745±0.031 TabDPT-DH 0.775±0.045 TabICL-DH 0.802±0.038 TabPFN-MTLR 0.761±0.026 TabDPT-MTLR 0.778±0.027 TabICL-MTLR 0.798±0.035

D.OROPHA.REC Model

Ctd ↑

IBS ↓

0.138±0.037 0.133±0.030 0.154±0.052 0.158±0.041 0.135±0.032 0.143±0.031 0.172±0.080 0.178±0.085 0.161±0.070 0.153±0.052 0.295±0.081 0.295±0.091 0.128±0.023 0.128±0.023 0.125±0.024 0.206±0.098 0.205±0.092 0.210±0.095 0.169±0.067 0.173±0.075 0.171±0.075

0.786±0.020 0.786±0.028 0.741±0.045 0.773±0.018 0.779±0.022 0.763±0.069 0.809±0.030 0.835±0.031 0.849±0.031 0.813±0.025 0.763±0.045 0.754±0.060 0.790±0.032 0.806±0.038 0.825±0.029 0.779±0.026 0.806±0.040 0.836±0.032 0.789±0.029 0.810±0.026 0.833±0.028

DATADIVAT1 AUC ↑

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.648±0.053 0.176±0.057 0.718±0.049 RSF 0.684±0.058 0.167±0.043 0.747±0.036 GBSA 0.640±0.015 0.181±0.044 0.715±0.046 DeepSurv 0.509±0.096 0.196±0.063 0.524±0.155 MLP-MTLR 0.575±0.045 0.253±0.088 0.633±0.073 MLP-DH 0.584±0.024 0.228±0.089 0.649±0.057 SurvTRACE 0.633±0.069 0.278±0.109 0.719±0.064 DySurv 0.628±0.053 0.252±0.069 0.677±0.044 TabPFN-ZS 0.655±0.042 0.181±0.068 0.726±0.067 TabDPT-ZS 0.655±0.043 0.184±0.068 0.729±0.058 TabICL-ZS 0.660±0.026 0.180±0.070 0.735±0.047 TabPFN-CE 0.616±0.031 0.212±0.069 0.668±0.049 TabDPT-CE 0.594±0.033 0.214±0.053 0.663±0.043 TabICL-CE 0.629±0.015 0.196±0.062 0.673±0.040 TabPFN-Cox 0.646±0.071 0.179±0.057 0.723±0.072 TabDPT-Cox 0.637±0.063 0.178±0.054 0.701±0.055 TabICL-Cox 0.605±0.061 0.185±0.052 0.669±0.081 TabPFN-DH 0.625±0.029 0.187±0.051 0.681±0.061 TabDPT-DH 0.631±0.063 0.188±0.054 0.692±0.057 TabICL-DH 0.582±0.060 0.190±0.054 0.636±0.103 TabPFN-MTLR 0.621±0.046 0.179±0.056 0.689±0.048 TabDPT-MTLR 0.649±0.037 0.176±0.050 0.727±0.037 TabICL-MTLR 0.614±0.042 0.189±0.055 0.654±0.075

Cox PH 0.648±0.019 0.137±0.023 0.707±0.029 RSF 0.615±0.024 0.185±0.034 0.672±0.041 GBSA 0.638±0.027 0.146±0.025 0.693±0.037 DeepSurv 0.642±0.021 0.143±0.028 0.684±0.034 MLP-MTLR 0.608±0.041 0.139±0.044 0.706±0.038 MLP-DH 0.621±0.033 0.141±0.043 0.716±0.024 SurvTRACE 0.609±0.021 0.141±0.033 0.731±0.030 DySurv 0.619±0.019 0.210±0.057 0.676±0.055 TabPFN-ZS 0.610±0.024 0.136±0.045 0.747±0.034 TabDPT-ZS 0.587±0.015 0.134±0.044 0.733±0.031 TabICL-ZS 0.601±0.018 0.136±0.046 0.742±0.030 TabPFN-CE 0.538±0.018 0.232±0.056 0.629±0.020 TabDPT-CE 0.620±0.029 0.355±0.071 0.736±0.032 TabICL-CE 0.613±0.027 0.355±0.082 0.737±0.029 TabPFN-Cox 0.642±0.020 0.138±0.027 0.704±0.026 TabDPT-Cox 0.650±0.026 0.139±0.030 0.714±0.036 TabICL-Cox 0.649±0.017 0.146±0.034 0.711±0.023 TabPFN-DH 0.642±0.023 0.207±0.063 0.688±0.032 TabDPT-DH 0.654±0.030 0.230±0.068 0.714±0.027 TabICL-DH 0.639±0.031 0.230±0.067 0.698±0.044 TabPFN-MTLR 0.646±0.024 0.262±0.076 0.724±0.027 TabDPT-MTLR 0.655±0.024 0.240±0.068 0.725±0.033 TabICL-MTLR 0.653±0.019 0.241±0.067 0.726±0.027

DATADIVAT2

DATADIVAT3

Model Cox PH RSF GBSA DeepSurv

Ctd ↑

IBS ↓

AUC ↑

0.604±0.022 0.554±0.050 0.571±0.013 0.602±0.023

0.184±0.012 0.194±0.017 0.192±0.017 0.189±0.013

0.636±0.021 0.618±0.031 0.632±0.012 0.633±0.029

Model Cox PH RSF GBSA DeepSurv

Ctd ↑

IBS ↓

AUC ↑

0.700±0.065 0.576±0.042 0.642±0.067 0.658±0.051

0.070±0.002 0.073±0.004 0.071±0.004 0.072±0.004

0.701±0.062 0.640±0.034 0.710±0.043 0.664±0.065

Minh-Khoi Pham et al.

MLP-MTLR 0.573±0.033 MLP-DH 0.592±0.015 SurvTRACE 0.545±0.027 DySurv 0.602±0.028 TabPFN-ZS 0.555±0.012 TabDPT-ZS 0.552±0.015 TabICL-ZS 0.555±0.031 TabPFN-CE 0.478±0.030 TabDPT-CE 0.562±0.029 TabICL-CE 0.559±0.031 TabPFN-Cox 0.604±0.025 TabDPT-Cox 0.609±0.022 TabICL-Cox 0.604±0.018 TabPFN-DH 0.605±0.028 TabDPT-DH 0.603±0.023 TabICL-DH 0.610±0.015 TabPFN-MTLR 0.600±0.021 TabDPT-MTLR 0.611±0.025 TabICL-MTLR 0.610±0.020

0.193±0.014 0.194±0.013 0.196±0.015 0.194±0.029 0.220±0.029 0.221±0.029 0.221±0.032 0.379±0.043 0.402±0.052 0.408±0.050 0.185±0.011 0.188±0.014 0.192±0.017 0.220±0.030 0.215±0.027 0.218±0.027 0.196±0.019 0.199±0.023 0.201±0.026

0.615±0.035 0.634±0.022 0.618±0.034 0.635±0.028 0.646±0.019 0.634±0.026 0.638±0.021 0.530±0.033 0.632±0.028 0.631±0.020 0.636±0.023 0.637±0.023 0.631±0.017 0.633±0.017 0.638±0.026 0.634±0.018 0.639±0.021 0.640±0.034 0.635±0.021

MLP-MTLR MLP-DH SurvTRACE DySurv TabPFN-ZS TabDPT-ZS TabICL-ZS TabPFN-CE TabDPT-CE TabICL-CE TabPFN-Cox TabDPT-Cox TabICL-Cox TabPFN-DH TabDPT-DH TabICL-DH TabPFN-MTLR TabDPT-MTLR TabICL-MTLR

DATAOVARIAN1 Model

Ctd ↑

IBS ↓

0.590±0.101 0.605±0.129 0.604±0.044 0.649±0.107 0.637±0.056 0.600±0.054 0.632±0.057 0.582±0.044 0.594±0.041 0.594±0.051 0.666±0.062 0.694±0.053 0.671±0.046 0.657±0.065 0.690±0.057 0.668±0.040 0.658±0.062 0.675±0.057 0.661±0.068

0.069±0.009 0.604±0.112 0.119±0.060 0.632±0.100 0.068±0.005 0.677±0.045 0.200±0.181 0.667±0.123 0.066±0.014 0.728±0.063 0.067±0.018 0.677±0.048 0.064±0.014 0.732±0.055 0.118±0.023 0.647±0.041 0.458±0.089 0.650±0.049 0.439±0.086 0.651±0.050 0.070±0.005 0.672±0.058 0.071±0.003 0.684±0.051 0.073±0.005 0.648±0.067 0.264±0.089 0.691±0.049 0.210±0.075 0.682±0.060 0.231±0.076 0.643±0.050 0.144±0.053 0.671±0.058 0.157±0.053 0.665±0.057 0.182±0.069 0.639±0.103

DIABETES AUC ↑

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.621±0.015 0.172±0.050 0.666±0.022 RSF 0.606±0.051 0.135±0.042 0.674±0.051 GBSA 0.621±0.023 0.155±0.075 0.664±0.031 DeepSurv 0.646±0.015 0.135±0.046 0.689±0.017 MLP-MTLR 0.596±0.031 0.200±0.010 0.647±0.027 MLP-DH 0.616±0.039 0.175±0.045 0.648±0.031 SurvTRACE 0.586±0.039 0.170±0.050 0.631±0.055 DySurv 0.648±0.013 0.210±0.078 0.694±0.019 TabPFN-ZS 0.658±0.022 0.145±0.062 0.700±0.014 TabDPT-ZS 0.625±0.023 0.152±0.068 0.675±0.031 TabICL-ZS 0.656±0.027 0.148±0.067 0.699±0.029 TabPFN-CE 0.638±0.029 0.163±0.065 0.692±0.038 TabDPT-CE 0.639±0.021 0.155±0.065 0.691±0.015 TabICL-CE 0.635±0.020 0.154±0.064 0.672±0.025 TabPFN-Cox 0.658±0.014 0.132±0.050 0.707±0.018 TabDPT-Cox 0.637±0.017 0.144±0.057 0.689±0.019 TabICL-Cox 0.639±0.007 0.138±0.056 0.684±0.017 TabPFN-DH 0.630±0.010 0.149±0.063 0.683±0.013 TabDPT-DH 0.625±0.019 0.150±0.064 0.673±0.020 TabICL-DH 0.645±0.016 0.148±0.062 0.687±0.027 TabPFN-MTLR 0.629±0.030 0.142±0.058 0.674±0.045 TabDPT-MTLR 0.651±0.009 0.143±0.059 0.698±0.010 TabICL-MTLR 0.641±0.015 0.140±0.062 0.680±0.026

Cox PH 0.594±0.032 0.202±0.018 0.593±0.036 RSF 0.570±0.030 0.204±0.024 0.591±0.067 GBSA 0.613±0.022 0.200±0.021 0.638±0.026 DeepSurv 0.585±0.019 0.205±0.019 0.598±0.060 MLP-MTLR 0.572±0.013 0.229±0.050 0.575±0.054 MLP-DH 0.529±0.084 0.304±0.095 0.547±0.067 SurvTRACE 0.568±0.029 0.362±0.156 0.585±0.045 DySurv 0.551±0.070 0.374±0.224 0.539±0.100 TabPFN-ZS 0.575±0.029 0.244±0.061 0.612±0.026 TabDPT-ZS 0.599±0.043 0.251±0.071 0.618±0.057 TabICL-ZS 0.595±0.036 0.247±0.070 0.617±0.059 TabPFN-CE 0.600±0.058 0.337±0.077 0.610±0.085 TabDPT-CE 0.594±0.037 0.410±0.091 0.603±0.077 TabICL-CE 0.583±0.026 0.406±0.096 0.619±0.045 TabPFN-Cox 0.605±0.022 0.196±0.017 0.613±0.038 TabDPT-Cox 0.643±0.045 0.193±0.028 0.649±0.057 TabICL-Cox 0.619±0.020 0.196±0.021 0.630±0.062 TabPFN-DH 0.598±0.032 0.294±0.054 0.612±0.030 TabDPT-DH 0.606±0.076 0.262±0.061 0.633±0.089 TabICL-DH 0.632±0.039 0.279±0.066 0.654±0.071 TabPFN-MTLR 0.602±0.016 0.224±0.051 0.597±0.049 TabDPT-MTLR 0.621±0.048 0.208±0.049 0.637±0.060 TabICL-MTLR 0.633±0.032 0.221±0.055 0.659±0.085

DIVORCE

E1684

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH

Ctd ↑

IBS ↓

AUC ↑

0.445±0.018 0.426±0.016 0.434±0.019 0.442±0.012 0.426±0.014 0.433±0.010

0.200±0.012 0.200±0.013 0.205±0.012 0.203±0.010 0.199±0.017 0.223±0.045

0.553±0.025 0.552±0.007 0.549±0.026 0.552±0.015 0.535±0.022 0.546±0.024

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH

Ctd ↑

IBS ↓

AUC ↑

0.538±0.051 0.507±0.034 0.531±0.055 0.520±0.047 0.513±0.053 0.485±0.043

0.227±0.045 0.230±0.042 0.259±0.089 0.227±0.036 0.258±0.034 0.293±0.086

0.562±0.080 0.581±0.061 0.573±0.051 0.514±0.064 0.573±0.061 0.510±0.026

TabFMs for Time-to-Event Prediction

SurvTRACE 0.420±0.041 DySurv 0.422±0.028 TabPFN-ZS 0.435±0.015 TabDPT-ZS 0.413±0.010 TabICL-ZS 0.439±0.021 TabPFN-CE 0.452±0.017 TabDPT-CE 0.450±0.027 TabICL-CE 0.449±0.029 TabPFN-Cox 0.441±0.026 TabDPT-Cox 0.444±0.025 TabICL-Cox 0.439±0.023 TabPFN-DH 0.448±0.031 TabDPT-DH 0.443±0.028 TabICL-DH 0.441±0.030 TabPFN-MTLR 0.450±0.018 TabDPT-MTLR 0.444±0.030 TabICL-MTLR 0.449±0.031

0.214±0.019 0.291±0.122 0.299±0.066 0.297±0.067 0.298±0.066 0.463±0.064 0.466±0.066 0.466±0.068 0.201±0.013 0.201±0.012 0.205±0.015 0.287±0.062 0.276±0.061 0.282±0.064 0.224±0.040 0.222±0.042 0.227±0.051

0.521±0.040 0.523±0.030 0.552±0.016 0.523±0.024 0.548±0.017 0.562±0.023 0.558±0.027 0.558±0.024 0.549±0.033 0.550±0.027 0.549±0.027 0.549±0.024 0.553±0.024 0.542±0.027 0.563±0.021 0.554±0.024 0.553±0.030

SurvTRACE 0.482±0.041 0.246±0.043 0.484±0.045 DySurv 0.465±0.020 0.319±0.160 0.442±0.016 TabPFN-ZS 0.500±0.032 0.259±0.089 0.557±0.058 TabDPT-ZS 0.516±0.046 0.262±0.088 0.567±0.077 TabICL-ZS 0.552±0.039 0.259±0.090 0.590±0.076 TabPFN-CE 0.550±0.034 0.321±0.110 0.564±0.064 TabDPT-CE 0.557±0.036 0.306±0.095 0.585±0.055 TabICL-CE 0.572±0.032 0.304±0.093 0.593±0.043 TabPFN-Cox 0.533±0.045 0.230±0.044 0.565±0.054 TabDPT-Cox 0.523±0.029 0.234±0.051 0.544±0.043 TabICL-Cox 0.543±0.069 0.227±0.045 0.558±0.091 TabPFN-DH 0.526±0.026 0.274±0.085 0.520±0.054 TabDPT-DH 0.514±0.066 0.264±0.084 0.536±0.080 TabICL-DH 0.559±0.038 0.270±0.088 0.584±0.046 TabPFN-MTLR 0.527±0.028 0.252±0.077 0.549±0.046 TabDPT-MTLR 0.543±0.044 0.243±0.070 0.546±0.084 TabICL-MTLR 0.563±0.033 0.245±0.073 0.567±0.067

EPILEPTIC Model

Ctd ↑

FLCHAIN IBS ↓

AUC ↑

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.535±0.061 0.217±0.062 0.546±0.100 RSF 0.601±0.052 0.239±0.107 0.623±0.111 GBSA 0.498±0.157 0.217±0.053 0.579±0.111 DeepSurv 0.503±0.071 0.227±0.091 0.495±0.124 MLP-MTLR 0.609±0.133 0.278±0.111 0.644±0.165 MLP-DH 0.614±0.084 0.237±0.071 0.660±0.087 SurvTRACE 0.563±0.080 0.216±0.063 0.590±0.136 DySurv 0.574±0.025 0.223±0.059 0.592±0.051 TabPFN-ZS 0.560±0.055 0.281±0.084 0.624±0.114 TabDPT-ZS 0.658±0.130 0.275±0.086 0.706±0.129 TabICL-ZS 0.591±0.054 0.282±0.089 0.655±0.089 TabPFN-CE 0.598±0.051 0.328±0.074 0.596±0.059 TabDPT-CE 0.570±0.063 0.379±0.091 0.585±0.101 TabICL-CE 0.545±0.053 0.382±0.088 0.560±0.076 TabPFN-Cox 0.537±0.097 0.217±0.052 0.563±0.141 TabDPT-Cox 0.566±0.047 0.213±0.057 0.615±0.071 TabICL-Cox 0.514±0.104 0.210±0.056 0.548±0.112 TabPFN-DH 0.574±0.098 0.253±0.078 0.599±0.130 TabDPT-DH 0.601±0.078 0.244±0.066 0.619±0.077 TabICL-DH 0.522±0.053 0.246±0.061 0.550±0.082 TabPFN-MTLR 0.580±0.094 0.253±0.074 0.609±0.142 TabDPT-MTLR 0.578±0.064 0.249±0.099 0.608±0.080 TabICL-MTLR 0.566±0.050 0.250±0.085 0.592±0.082

Cox PH 0.916±0.005 0.059±0.004 0.917±0.011 RSF 0.922±0.005 0.044±0.001 0.945±0.005 GBSA 0.930±0.001 0.044±0.001 0.944±0.003 DeepSurv 0.928±0.001 0.045±0.002 0.942±0.002 MLP-MTLR 0.926±0.006 0.050±0.006 0.942±0.006 MLP-DH 0.926±0.002 0.071±0.018 0.940±0.004 SurvTRACE 0.922±0.003 0.045±0.002 0.943±0.004 DySurv 0.922±0.014 0.052±0.012 0.941±0.004 TabPFN-ZS 0.920±0.006 0.045±0.002 0.942±0.004 TabDPT-ZS 0.913±0.006 0.049±0.001 0.935±0.003 TabICL-ZS 0.920±0.005 0.045±0.001 0.942±0.003 TabPFN-CE 0.918±0.005 0.060±0.003 0.939±0.005 TabDPT-CE 0.922±0.005 0.081±0.006 0.944±0.004 TabICL-CE 0.923±0.004 0.079±0.004 0.945±0.003 TabPFN-Cox 0.929±0.001 0.045±0.001 0.943±0.003 TabDPT-Cox 0.930±0.002 0.045±0.001 0.944±0.004 TabICL-Cox 0.930±0.002 0.045±0.001 0.944±0.003 TabPFN-DH 0.929±0.002 0.064±0.012 0.941±0.002 TabDPT-DH 0.930±0.002 0.066±0.011 0.944±0.004 TabICL-DH 0.930±0.002 0.058±0.012 0.944±0.003 TabPFN-MTLR 0.928±0.002 0.045±0.002 0.942±0.003 TabDPT-MTLR 0.929±0.003 0.046±0.001 0.943±0.004 TabICL-MTLR 0.930±0.003 0.045±0.001 0.945±0.004

GLIOMA

GRACE

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH SurvTRACE DySurv

Ctd ↑

IBS ↓

AUC ↑

0.841±0.085 0.836±0.114 0.718±0.101 0.760±0.095 0.756±0.143 0.631±0.168 0.733±0.163 0.504±0.282

0.105±0.059 0.126±0.026 0.144±0.067 0.153±0.067 0.123±0.053 0.179±0.096 0.167±0.097 0.401±0.094

0.869±0.083 0.848±0.114 0.740±0.048 0.824±0.110 0.854±0.131 0.787±0.157 0.791±0.162 0.522±0.278

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH SurvTRACE DySurv

Ctd ↑

IBS ↓

AUC ↑

0.708±0.029 0.750±0.020 0.708±0.024 0.694±0.057 0.751±0.019 0.750±0.017 0.679±0.023 0.663±0.088

0.182±0.003 0.165±0.008 0.173±0.006 0.170±0.010 0.170±0.010 0.227±0.064 0.190±0.037 0.198±0.050

0.732±0.038 0.795±0.026 0.739±0.017 0.721±0.054 0.799±0.013 0.771±0.019 0.766±0.037 0.700±0.093

Minh-Khoi Pham et al.

TabPFN-ZS 0.885±0.101 TabDPT-ZS 0.769±0.155 TabICL-ZS 0.591±0.297 TabPFN-CE 0.773±0.112 TabDPT-CE 0.771±0.096 TabICL-CE 0.751±0.115 TabPFN-Cox 0.837±0.084 TabDPT-Cox 0.689±0.193 TabICL-Cox 0.699±0.174 TabPFN-DH 0.787±0.171 TabDPT-DH 0.436±0.339 TabICL-DH 0.601±0.149 TabPFN-MTLR 0.755±0.117 TabDPT-MTLR 0.364±0.264 TabICL-MTLR 0.680±0.191

0.106±0.040 0.115±0.044 0.175±0.057 0.170±0.067 0.189±0.052 0.187±0.036 0.120±0.028 0.191±0.020 0.196±0.033 0.232±0.053 0.245±0.057 0.239±0.052 0.176±0.063 0.265±0.090 0.196±0.054

0.887±0.116 0.911±0.025 0.682±0.238 0.778±0.162 0.798±0.128 0.736±0.173 0.838±0.119 0.731±0.163 0.685±0.210 0.818±0.225 0.433±0.325 0.597±0.108 0.747±0.190 0.484±0.241 0.658±0.256

TabPFN-ZS 0.748±0.010 0.162±0.010 0.811±0.024 TabDPT-ZS 0.749±0.013 0.164±0.009 0.806±0.021 TabICL-ZS 0.745±0.016 0.166±0.007 0.806±0.026 TabPFN-CE 0.726±0.031 0.198±0.017 0.724±0.037 TabDPT-CE 0.738±0.017 0.325±0.031 0.797±0.025 TabICL-CE 0.737±0.028 0.317±0.039 0.768±0.041 TabPFN-Cox 0.721±0.037 0.165±0.009 0.753±0.039 TabDPT-Cox 0.726±0.029 0.166±0.008 0.764±0.031 TabICL-Cox 0.720±0.030 0.169±0.006 0.750±0.033 TabPFN-DH 0.755±0.020 0.266±0.039 0.780±0.028 TabDPT-DH 0.768±0.023 0.267±0.042 0.793±0.027 TabICL-DH 0.740±0.022 0.262±0.027 0.761±0.037 TabPFN-MTLR 0.732±0.017 0.193±0.011 0.791±0.030 TabDPT-MTLR 0.736±0.015 0.202±0.016 0.790±0.025 TabICL-MTLR 0.721±0.016 0.196±0.009 0.768±0.031

GSE1992 Model

Ctd ↑

GSE3143 IBS ↓

AUC ↑

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.604±0.137 0.384±0.305 0.664±0.112 RSF 0.645±0.184 0.200±0.102 0.727±0.100 GBSA 0.521±0.199 0.269±0.234 0.657±0.119 DeepSurv 0.639±0.176 0.236±0.111 0.754±0.066 MLP-MTLR 0.652±0.073 0.321±0.261 0.714±0.105 MLP-DH 0.480±0.186 0.269±0.103 0.531±0.079 SurvTRACE 0.404±0.245 0.435±0.390 0.486±0.210 DySurv 0.484±0.088 0.396±0.354 0.527±0.090 TabPFN-ZS 0.676±0.140 0.280±0.265 0.636±0.148 TabDPT-ZS 0.610±0.156 0.296±0.219 0.658±0.131 TabICL-ZS 0.412±0.133 0.318±0.345 0.428±0.154 TabPFN-CE 0.656±0.118 0.342±0.372 0.675±0.174 TabDPT-CE 0.478±0.053 0.327±0.341 0.477±0.046 TabICL-CE 0.490±0.127 0.257±0.167 0.519±0.179 TabPFN-Cox 0.581±0.213 0.214±0.105 0.653±0.158 TabDPT-Cox 0.577±0.122 0.230±0.123 0.623±0.118 TabICL-Cox 0.411±0.172 0.234±0.118 0.524±0.226 TabPFN-DH 0.500±0.109 0.303±0.280 0.549±0.123 TabDPT-DH 0.447±0.172 0.305±0.281 0.492±0.119 TabICL-DH 0.585±0.213 0.300±0.267 0.566±0.257 TabPFN-MTLR 0.533±0.183 0.279±0.198 0.581±0.200 TabDPT-MTLR 0.538±0.133 0.270±0.232 0.552±0.123 TabICL-MTLR 0.483±0.134 0.283±0.245 0.505±0.169

Cox PH 0.670±0.077 0.240±0.082 0.695±0.122 RSF 0.608±0.058 0.171±0.019 0.645±0.095 GBSA 0.632±0.061 0.191±0.043 0.700±0.086 DeepSurv 0.586±0.098 0.210±0.081 0.578±0.136 MLP-MTLR 0.501±0.139 0.217±0.072 0.583±0.151 MLP-DH 0.536±0.146 0.221±0.079 0.591±0.126 SurvTRACE 0.549±0.199 0.467±0.247 0.665±0.159 DySurv 0.535±0.120 0.369±0.119 0.591±0.159 TabPFN-ZS 0.574±0.145 0.224±0.042 0.551±0.171 TabDPT-ZS 0.626±0.039 0.237±0.094 0.675±0.076 TabICL-ZS 0.455±0.110 0.204±0.051 0.472±0.130 TabPFN-CE 0.632±0.080 0.236±0.092 0.673±0.136 TabDPT-CE 0.688±0.120 0.211±0.062 0.727±0.129 TabICL-CE 0.526±0.047 0.193±0.015 0.534±0.049 TabPFN-Cox 0.550±0.122 0.174±0.024 0.575±0.108 TabDPT-Cox 0.556±0.144 0.174±0.017 0.589±0.109 TabICL-Cox 0.496±0.171 0.173±0.026 0.468±0.233 TabPFN-DH 0.435±0.151 0.230±0.078 0.457±0.155 TabDPT-DH 0.581±0.126 0.228±0.069 0.629±0.168 TabICL-DH 0.463±0.100 0.225±0.078 0.402±0.094 TabPFN-MTLR 0.534±0.160 0.230±0.089 0.556±0.187 TabDPT-MTLR 0.565±0.131 0.210±0.064 0.634±0.093 TabICL-MTLR 0.588±0.085 0.211±0.050 0.588±0.147

GSE4335

HDFAIL

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH SurvTRACE DySurv TabPFN-ZS TabDPT-ZS

Ctd ↑

IBS ↓

AUC ↑

0.728±0.083 0.194±0.111 0.726±0.162 0.653±0.105 0.172±0.039 0.701±0.168 0.694±0.156 0.212±0.164 0.751±0.170 0.653±0.153 0.175±0.038 0.682±0.188 0.681±0.113 0.223±0.123 0.666±0.198 0.687±0.080 0.206±0.089 0.609±0.131 0.556±0.114 0.302±0.178 0.584±0.108 0.578±0.068 0.321±0.086 0.614±0.059 0.560±0.158 0.269±0.110 0.539±0.115 0.654±0.132 0.240±0.098 0.619±0.184

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH SurvTRACE DySurv TabPFN-ZS TabDPT-ZS

Ctd ↑ 0.852±0.008 0.831±0.021 0.824±0.029 0.739±0.132 0.853±0.013 0.866±0.014 0.844±0.020 0.858±0.013 0.833±0.011 0.819±0.006

IBS ↓

AUC ↑

0.185±0.212 0.855±0.011 0.210±0.289 0.880±0.008 0.178±0.226 0.871±0.009 0.174±0.202 0.738±0.150 0.107±0.147 0.894±0.008 0.141±0.112 0.893±0.011 0.102±0.123 0.904±0.006 0.117±0.107 0.886±0.009 0.192±0.147 0.919±0.005 0.194±0.147 0.899±0.007

TabFMs for Time-to-Event Prediction

TabICL-ZS TabPFN-CE TabDPT-CE TabICL-CE TabPFN-Cox TabDPT-Cox TabICL-Cox TabPFN-DH TabDPT-DH TabICL-DH TabPFN-MTLR TabDPT-MTLR TabICL-MTLR

0.576±0.094 0.601±0.084 0.563±0.132 0.579±0.086 0.492±0.139 0.562±0.183 0.470±0.110 0.556±0.097 0.507±0.118 0.508±0.127 0.647±0.117 0.410±0.063 0.559±0.083

0.274±0.140 0.311±0.144 0.316±0.160 0.217±0.098 0.205±0.053 0.207±0.044 0.201±0.038 0.295±0.131 0.294±0.143 0.301±0.135 0.261±0.139 0.269±0.127 0.290±0.158

0.623±0.099 0.599±0.099 0.585±0.147 0.569±0.068 0.424±0.165 0.532±0.224 0.469±0.131 0.558±0.121 0.491±0.142 0.459±0.081 0.599±0.131 0.393±0.135 0.504±0.121

TabICL-ZS 0.831±0.009 TabPFN-CE 0.808±0.007 TabDPT-CE 0.843±0.007 TabICL-CE 0.856±0.005 TabPFN-Cox 0.863±0.007 TabDPT-Cox 0.819±0.010 TabICL-Cox 0.862±0.008 TabPFN-DH 0.876±0.006 TabDPT-DH 0.833±0.005 TabICL-DH 0.867±0.003 TabPFN-MTLR 0.872±0.006 TabDPT-MTLR 0.829±0.012 TabICL-MTLR 0.870±0.008

HEART Model

Ctd ↑

0.193±0.150 0.299±0.255 0.412±0.219 0.405±0.217 0.159±0.175 0.189±0.187 0.184±0.258 0.242±0.161 0.275±0.159 0.276±0.134 0.146±0.109 0.140±0.092 0.192±0.109

0.917±0.006 0.882±0.010 0.902±0.009 0.917±0.005 0.891±0.008 0.829±0.017 0.858±0.013 0.891±0.010 0.830±0.012 0.854±0.005 0.896±0.005 0.847±0.015 0.880±0.014

HEARTVALVE IBS ↓

AUC ↑

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.654±0.100 0.114±0.057 0.692±0.133 RSF 0.563±0.105 0.117±0.051 0.610±0.124 GBSA 0.574±0.127 0.121±0.080 0.615±0.182 DeepSurv 0.641±0.117 0.111±0.059 0.674±0.149 MLP-MTLR 0.575±0.113 0.164±0.043 0.652±0.169 MLP-DH 0.415±0.244 0.140±0.105 0.523±0.116 SurvTRACE 0.542±0.159 0.177±0.107 0.625±0.159 DySurv 0.481±0.184 0.155±0.113 0.484±0.227 TabPFN-ZS 0.544±0.155 0.115±0.066 0.600±0.158 TabDPT-ZS 0.577±0.190 0.122±0.078 0.620±0.193 TabICL-ZS 0.541±0.069 0.119±0.080 0.523±0.186 TabPFN-CE 0.593±0.142 0.131±0.082 0.660±0.162 TabDPT-CE 0.557±0.175 0.129±0.075 0.574±0.217 TabICL-CE 0.555±0.154 0.136±0.087 0.585±0.205 TabPFN-Cox 0.636±0.126 0.118±0.054 0.675±0.149 TabDPT-Cox 0.540±0.185 0.117±0.064 0.543±0.231 TabICL-Cox 0.493±0.088 0.125±0.078 0.479±0.134 TabPFN-DH 0.591±0.227 0.118±0.073 0.623±0.159 TabDPT-DH 0.471±0.159 0.118±0.071 0.493±0.216 TabICL-DH 0.568±0.089 0.116±0.072 0.578±0.178 TabPFN-MTLR 0.636±0.128 0.118±0.060 0.686±0.143 TabDPT-MTLR 0.508±0.123 0.120±0.060 0.527±0.146 TabICL-MTLR 0.525±0.109 0.117±0.070 0.578±0.206

Cox PH 0.766±0.147 0.057±0.033 0.819±0.141 RSF 0.640±0.100 0.058±0.024 0.761±0.202 GBSA 0.602±0.286 0.060±0.025 0.732±0.220 DeepSurv 0.657±0.210 0.065±0.026 0.692±0.216 MLP-MTLR 0.773±0.138 0.060±0.034 0.788±0.139 MLP-DH 0.855±0.135 0.089±0.041 0.865±0.147 SurvTRACE 0.804±0.057 0.062±0.025 0.803±0.066 DySurv 0.703±0.040 0.258±0.153 0.730±0.057 TabPFN-ZS 0.810±0.093 0.050±0.022 0.832±0.152 TabDPT-ZS 0.909±0.039 0.060±0.026 0.920±0.023 TabICL-ZS 0.887±0.062 0.058±0.029 0.880±0.105 TabPFN-CE 0.746±0.090 0.091±0.035 0.719±0.194 TabDPT-CE 0.496±0.274 0.270±0.058 0.510±0.281 TabICL-CE 0.598±0.094 0.270±0.055 0.537±0.089 TabPFN-Cox 0.500±0.105 0.070±0.022 0.535±0.101 TabDPT-Cox 0.789±0.153 0.069±0.022 0.894±0.022 TabICL-Cox 0.727±0.151 0.070±0.022 0.846±0.115 TabPFN-DH 0.525±0.130 0.351±0.128 0.575±0.144 TabDPT-DH 0.785±0.116 0.314±0.131 0.768±0.176 TabICL-DH 0.782±0.126 0.301±0.138 0.807±0.164 TabPFN-MTLR 0.550±0.080 0.236±0.117 0.488±0.108 TabDPT-MTLR 0.850±0.050 0.210±0.142 0.864±0.077 TabICL-MTLR 0.832±0.032 0.208±0.109 0.870±0.089

HEPATOCELLULAR

MICRO.CENSURE

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH SurvTRACE DySurv TabPFN-ZS TabDPT-ZS TabICL-ZS TabPFN-CE

Ctd ↑

IBS ↓

AUC ↑

0.752±0.088 0.179±0.065 0.817±0.142 0.759±0.073 0.186±0.053 0.825±0.115 0.733±0.079 0.175±0.065 0.788±0.124 0.611±0.123 0.231±0.083 0.618±0.165 0.650±0.117 0.271±0.152 0.727±0.096 0.613±0.059 0.261±0.194 0.702±0.113 0.642±0.098 0.206±0.115 0.758±0.172 0.649±0.166 0.323±0.198 0.676±0.263 0.723±0.109 0.160±0.092 0.790±0.101 0.720±0.142 0.197±0.095 0.809±0.136 0.699±0.118 0.176±0.070 0.774±0.100 0.748±0.121 0.182±0.079 0.816±0.143

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH SurvTRACE DySurv TabPFN-ZS TabDPT-ZS TabICL-ZS TabPFN-CE

Ctd ↑

IBS ↓

0.486±0.173 0.228±0.030 0.589±0.135 0.151±0.027 0.545±0.081 0.161±0.044 0.563±0.103 0.159±0.047 0.598±0.064 0.191±0.038 0.658±0.191 0.186±0.055 0.591±0.129 0.250±0.070 0.553±0.132 0.275±0.073 0.646±0.185 0.130±0.027 0.579±0.186 0.138±0.029 0.672±0.171 0.140±0.033 0.669±0.081 0.171±0.023

AUC ↑ 0.509±0.200 0.619±0.174 0.643±0.068 0.589±0.166 0.590±0.086 0.628±0.224 0.607±0.157 0.530±0.158 0.560±0.208 0.567±0.228 0.628±0.209 0.633±0.086

Minh-Khoi Pham et al.

TabDPT-CE TabICL-CE TabPFN-Cox TabDPT-Cox TabICL-Cox TabPFN-DH TabDPT-DH TabICL-DH TabPFN-MTLR TabDPT-MTLR TabICL-MTLR

0.706±0.087 0.676±0.104 0.679±0.179 0.493±0.178 0.531±0.096 0.614±0.083 0.500±0.134 0.607±0.117 0.703±0.111 0.507±0.141 0.584±0.196

0.192±0.083 0.203±0.080 0.170±0.034 0.199±0.037 0.200±0.037 0.287±0.150 0.309±0.179 0.284±0.150 0.219±0.100 0.295±0.118 0.241±0.126

0.765±0.153 0.739±0.155 0.706±0.249 0.483±0.275 0.538±0.217 0.638±0.178 0.438±0.160 0.602±0.233 0.748±0.146 0.594±0.174 0.638±0.205

TabDPT-CE TabICL-CE TabPFN-Cox TabDPT-Cox TabICL-Cox TabPFN-DH TabDPT-DH TabICL-DH TabPFN-MTLR TabDPT-MTLR TabICL-MTLR

NKI70 Model

0.583±0.043 0.656±0.081 0.405±0.196 0.516±0.187 0.578±0.096 0.490±0.277 0.556±0.220 0.556±0.170 0.669±0.122 0.582±0.151 0.415±0.165

0.161±0.009 0.154±0.029 0.150±0.029 0.154±0.026 0.148±0.032 0.163±0.016 0.154±0.020 0.160±0.017 0.154±0.029 0.146±0.023 0.163±0.035

0.534±0.090 0.586±0.150 0.411±0.283 0.535±0.198 0.661±0.171 0.509±0.238 0.611±0.086 0.488±0.165 0.632±0.117 0.577±0.203 0.405±0.192

NWTCO

Ctd ↑

IBS ↓

AUC ↑

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.722±0.057 0.271±0.258 0.808±0.058 RSF 0.728±0.065 0.193±0.099 0.804±0.073 GBSA 0.665±0.053 0.239±0.170 0.723±0.084 DeepSurv 0.663±0.082 0.182±0.112 0.679±0.082 MLP-MTLR 0.664±0.090 0.272±0.197 0.691±0.106 MLP-DH 0.765±0.044 0.264±0.182 0.793±0.039 SurvTRACE 0.607±0.118 0.315±0.151 0.631±0.147 DySurv 0.631±0.060 0.451±0.337 0.652±0.079 TabPFN-ZS 0.700±0.052 0.262±0.269 0.803±0.056 TabDPT-ZS 0.741±0.048 0.266±0.286 0.833±0.078 TabICL-ZS 0.756±0.034 0.248±0.269 0.837±0.070 TabPFN-CE 0.677±0.030 0.276±0.264 0.732±0.094 TabDPT-CE 0.583±0.078 0.248±0.138 0.623±0.140 TabICL-CE 0.693±0.061 0.230±0.119 0.745±0.087 TabPFN-Cox 0.676±0.107 0.203±0.120 0.771±0.056 TabDPT-Cox 0.672±0.091 0.222±0.176 0.725±0.065 TabICL-Cox 0.628±0.101 0.209±0.109 0.747±0.058 TabPFN-DH 0.655±0.040 0.281±0.231 0.730±0.084 TabDPT-DH 0.697±0.064 0.267±0.219 0.736±0.112 TabICL-DH 0.704±0.071 0.281±0.227 0.708±0.063 TabPFN-MTLR 0.652±0.070 0.239±0.200 0.676±0.045 TabDPT-MTLR 0.698±0.055 0.230±0.195 0.719±0.085 TabICL-MTLR 0.685±0.047 0.258±0.209 0.769±0.057

Cox PH 0.702±0.043 0.115±0.003 0.737±0.062 RSF 0.659±0.038 0.104±0.009 0.741±0.056 GBSA 0.639±0.027 0.107±0.012 0.743±0.065 DeepSurv 0.708±0.040 0.109±0.011 0.735±0.052 MLP-MTLR 0.711±0.054 0.158±0.074 0.749±0.064 MLP-DH 0.705±0.036 0.294±0.091 0.719±0.046 SurvTRACE 0.691±0.017 0.112±0.010 0.712±0.027 DySurv 0.710±0.050 0.245±0.066 0.742±0.068 TabPFN-ZS 0.703±0.040 0.109±0.024 0.744±0.060 TabDPT-ZS 0.695±0.046 0.109±0.023 0.743±0.060 TabICL-ZS 0.698±0.036 0.110±0.023 0.740±0.055 TabPFN-CE 0.611±0.013 0.315±0.053 0.633±0.039 TabDPT-CE 0.694±0.030 0.663±0.042 0.718±0.047 TabICL-CE 0.692±0.022 0.640±0.056 0.725±0.042 TabPFN-Cox 0.717±0.044 0.106±0.008 0.746±0.061 TabDPT-Cox 0.715±0.042 0.107±0.010 0.747±0.062 TabICL-Cox 0.713±0.033 0.106±0.008 0.745±0.059 TabPFN-DH 0.712±0.047 0.308±0.043 0.743±0.056 TabDPT-DH 0.717±0.041 0.294±0.035 0.749±0.059 TabICL-DH 0.716±0.038 0.290±0.039 0.745±0.054 TabPFN-MTLR 0.705±0.041 0.152±0.019 0.739±0.056 TabDPT-MTLR 0.707±0.040 0.133±0.015 0.739±0.058 TabICL-MTLR 0.717±0.036 0.155±0.014 0.749±0.058

OLDMORT

OVA

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH SurvTRACE DySurv TabPFN-ZS TabDPT-ZS TabICL-ZS TabPFN-CE TabDPT-CE TabICL-CE

Ctd ↑

IBS ↓

AUC ↑

0.640±0.023 0.740±0.015 0.718±0.027 0.698±0.022 0.702±0.025 0.698±0.021 0.284±0.038 0.708±0.021 0.681±0.029 0.691±0.021 0.685±0.028 0.729±0.022 0.682±0.022 0.669±0.033

0.043±0.002 0.048±0.003 0.040±0.002 0.040±0.003 0.108±0.012 0.121±0.042 0.044±0.002 0.088±0.003 0.067±0.002 0.060±0.002 0.069±0.003 0.087±0.006 0.179±0.006 0.184±0.008

0.877±0.014 0.818±0.013 0.813±0.014 0.847±0.010 0.857±0.014 0.855±0.014 0.772±0.027 0.875±0.009 0.900±0.009 0.895±0.010 0.902±0.009 0.848±0.007 0.897±0.012 0.892±0.013

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH SurvTRACE DySurv TabPFN-ZS TabDPT-ZS TabICL-ZS TabPFN-CE TabDPT-CE TabICL-CE

Ctd ↑

IBS ↓

AUC ↑

0.613±0.035 0.623±0.035 0.599±0.062 0.565±0.083 0.598±0.060 0.563±0.100 0.613±0.028 0.613±0.035 0.626±0.028 0.607±0.035 0.618±0.031 0.613±0.048 0.626±0.032 0.625±0.037

0.184±0.030 0.182±0.028 0.185±0.031 0.188±0.043 0.198±0.044 0.208±0.020 0.220±0.027 0.224±0.079 0.196±0.042 0.195±0.039 0.194±0.040 0.236±0.045 0.237±0.046 0.237±0.045

0.681±0.054 0.691±0.061 0.690±0.056 0.617±0.131 0.666±0.084 0.594±0.111 0.681±0.044 0.677±0.064 0.676±0.077 0.680±0.046 0.694±0.057 0.682±0.064 0.679±0.068 0.695±0.053

TabFMs for Time-to-Event Prediction

TabPFN-Cox 0.699±0.027 0.039±0.002 TabDPT-Cox 0.729±0.025 0.039±0.002 TabICL-Cox 0.726±0.024 0.039±0.002 TabPFN-DH 0.729±0.015 0.057±0.004 TabDPT-DH 0.747±0.019 0.056±0.004 TabICL-DH 0.747±0.022 0.056±0.003 TabPFN-MTLR 0.681±0.025 0.094±0.008 TabDPT-MTLR 0.702±0.030 0.081±0.007 TabICL-MTLR 0.679±0.028 0.082±0.007

0.878±0.012 0.856±0.014 0.858±0.008 0.857±0.009 0.827±0.010 0.836±0.012 0.893±0.010 0.891±0.006 0.892±0.008

TabPFN-Cox 0.619±0.028 TabDPT-Cox 0.629±0.030 TabICL-Cox 0.625±0.021 TabPFN-DH 0.598±0.053 TabDPT-DH 0.618±0.039 TabICL-DH 0.614±0.024 TabPFN-MTLR 0.621±0.040 TabDPT-MTLR 0.619±0.038 TabICL-MTLR 0.622±0.020

OVARIAN Model

Ctd ↑

0.184±0.032 0.182±0.033 0.184±0.031 0.216±0.045 0.206±0.040 0.213±0.046 0.187±0.040 0.184±0.036 0.186±0.034

0.697±0.047 0.701±0.053 0.700±0.047 0.650±0.068 0.685±0.057 0.682±0.034 0.710±0.064 0.690±0.066 0.696±0.041

IBS ↓

AUC ↑

PBC IBS ↓

AUC ↑

Model

Ctd ↑

Cox PH 0.799±0.140 0.172±0.041 0.869±0.211 RSF 0.559±0.317 0.188±0.047 0.770±0.254 GBSA 0.585±0.247 0.306±0.239 0.689±0.426 DeepSurv 0.920±0.078 0.155±0.068 0.937±0.108 MLP-MTLR 0.701±0.206 0.320±0.191 0.706±0.412 MLP-DH 0.636±0.272 0.368±0.204 0.658±0.409 SurvTRACE 0.569±0.203 0.261±0.151 0.856±0.149 DySurv 0.594±0.356 0.472±0.290 0.670±0.310 TabPFN-ZS 0.719±0.121 0.239±0.103 0.781±0.393 TabDPT-ZS 0.670±0.309 0.287±0.127 0.667±0.416 TabICL-ZS 0.744±0.243 0.188±0.071 0.839±0.188 TabPFN-CE 0.688±0.167 0.335±0.208 0.847±0.150 TabDPT-CE 0.342±0.125 0.358±0.142 0.237±0.226 TabICL-CE 0.654±0.146 0.352±0.174 0.906±0.146 TabPFN-Cox 0.619±0.242 0.208±0.062 0.739±0.300 TabDPT-Cox 0.323±0.245 0.208±0.048 0.245±0.244 TabICL-Cox 0.573±0.320 0.202±0.041 0.583±0.383 TabPFN-DH 0.311±0.302 0.322±0.128 0.330±0.370 TabDPT-DH 0.603±0.297 0.319±0.127 0.803±0.274 TabICL-DH 0.434±0.291 0.324±0.130 0.486±0.441 TabPFN-MTLR 0.475±0.347 0.378±0.143 0.567±0.413 TabDPT-MTLR 0.591±0.258 0.273±0.128 0.541±0.455 TabICL-MTLR 0.385±0.234 0.317±0.129 0.419±0.387

Cox PH 0.766±0.037 0.168±0.028 0.816±0.043 RSF 0.745±0.060 0.162±0.026 0.812±0.050 GBSA 0.753±0.049 0.173±0.033 0.804±0.056 DeepSurv 0.776±0.040 0.157±0.018 0.817±0.036 MLP-MTLR 0.694±0.070 0.182±0.025 0.762±0.049 MLP-DH 0.706±0.101 0.178±0.030 0.783±0.087 SurvTRACE 0.739±0.040 0.202±0.076 0.801±0.036 DySurv 0.742±0.062 0.185±0.050 0.787±0.060 TabPFN-ZS 0.734±0.040 0.185±0.059 0.807±0.051 TabDPT-ZS 0.742±0.049 0.188±0.059 0.806±0.053 TabICL-ZS 0.736±0.040 0.183±0.057 0.807±0.067 TabPFN-CE 0.722±0.043 0.222±0.061 0.772±0.055 TabDPT-CE 0.726±0.052 0.275±0.082 0.792±0.053 TabICL-CE 0.751±0.036 0.262±0.085 0.812±0.048 TabPFN-Cox 0.759±0.047 0.172±0.024 0.800±0.048 TabDPT-Cox 0.739±0.067 0.178±0.034 0.767±0.090 TabICL-Cox 0.622±0.135 0.191±0.030 0.664±0.158 TabPFN-DH 0.668±0.066 0.227±0.061 0.729±0.088 TabDPT-DH 0.718±0.113 0.224±0.063 0.741±0.118 TabICL-DH 0.527±0.129 0.239±0.066 0.538±0.181 TabPFN-MTLR 0.743±0.051 0.179±0.043 0.792±0.057 TabDPT-MTLR 0.748±0.030 0.182±0.049 0.792±0.043 TabICL-MTLR 0.607±0.111 0.236±0.079 0.679±0.122

PHARMACOSMOKING

PHPL04K8A

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH SurvTRACE DySurv TabPFN-ZS TabDPT-ZS TabICL-ZS TabPFN-CE TabDPT-CE TabICL-CE TabPFN-Cox TabDPT-Cox

Ctd ↑

IBS ↓

AUC ↑

0.572±0.036 0.236±0.014 0.577±0.089 0.570±0.041 0.223±0.010 0.551±0.064 0.489±0.058 0.280±0.046 0.457±0.083 0.544±0.039 0.232±0.019 0.529±0.075 0.517±0.086 0.320±0.066 0.509±0.144 0.468±0.112 0.287±0.043 0.530±0.041 0.480±0.031 0.353±0.060 0.519±0.060 0.600±0.076 0.300±0.083 0.643±0.135 0.561±0.081 0.226±0.013 0.514±0.042 0.545±0.070 0.227±0.014 0.536±0.067 0.521±0.072 0.272±0.013 0.527±0.092 0.524±0.079 0.260±0.035 0.527±0.122 0.510±0.082 0.283±0.062 0.482±0.100 0.584±0.045 0.238±0.026 0.569±0.101 0.604±0.026 0.217±0.015 0.604±0.072 0.600±0.073 0.225±0.014 0.609±0.134

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH SurvTRACE DySurv TabPFN-ZS TabDPT-ZS TabICL-ZS TabPFN-CE TabDPT-CE TabICL-CE TabPFN-Cox TabDPT-Cox

Ctd ↑

IBS ↓

AUC ↑

0.639±0.063 0.186±0.074 0.688±0.077 0.620±0.063 0.191±0.066 0.678±0.086 0.562±0.050 0.209±0.082 0.606±0.087 0.623±0.058 0.173±0.056 0.670±0.071 0.587±0.056 0.230±0.091 0.612±0.082 0.560±0.050 0.246±0.100 0.588±0.091 0.619±0.086 0.243±0.066 0.685±0.077 0.619±0.052 0.287±0.147 0.666±0.059 0.628±0.067 0.193±0.089 0.682±0.062 0.600±0.065 0.204±0.093 0.666±0.105 0.635±0.071 0.199±0.095 0.689±0.071 0.607±0.064 0.238±0.051 0.648±0.075 0.618±0.076 0.210±0.097 0.655±0.081 0.613±0.087 0.208±0.095 0.652±0.094 0.614±0.059 0.192±0.082 0.660±0.079 0.627±0.066 0.178±0.070 0.665±0.092

Minh-Khoi Pham et al.

TabICL-Cox TabPFN-DH TabDPT-DH TabICL-DH TabPFN-MTLR TabDPT-MTLR TabICL-MTLR

0.492±0.118 0.555±0.100 0.552±0.027 0.539±0.099 0.582±0.090 0.578±0.098 0.504±0.044

0.231±0.006 0.268±0.028 0.252±0.025 0.264±0.036 0.235±0.024 0.232±0.023 0.252±0.030

0.471±0.191 0.541±0.065 0.559±0.068 0.563±0.085 0.602±0.101 0.588±0.140 0.536±0.074

TabICL-Cox TabPFN-DH TabDPT-DH TabICL-DH TabPFN-MTLR TabDPT-MTLR TabICL-MTLR

PROSTATE Model

Ctd ↑

0.630±0.077 0.584±0.060 0.628±0.071 0.635±0.090 0.606±0.060 0.630±0.077 0.633±0.079

0.188±0.083 0.201±0.088 0.200±0.088 0.202±0.090 0.207±0.093 0.198±0.096 0.200±0.092

0.680±0.087 0.617±0.069 0.662±0.077 0.681±0.109 0.643±0.067 0.671±0.093 0.686±0.096

PROSTATESURVIVAL IBS ↓

AUC ↑

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.629±0.031 0.184±0.011 0.657±0.039 RSF 0.635±0.021 0.181±0.011 0.659±0.023 GBSA 0.621±0.046 0.184±0.016 0.642±0.048 DeepSurv 0.578±0.027 0.195±0.009 0.596±0.041 MLP-MTLR 0.555±0.046 0.250±0.041 0.581±0.038 MLP-DH 0.564±0.040 0.219±0.014 0.584±0.038 SurvTRACE 0.600±0.059 0.257±0.093 0.653±0.053 DySurv 0.583±0.089 0.291±0.088 0.603±0.121 TabPFN-ZS 0.609±0.054 0.183±0.011 0.637±0.044 TabDPT-ZS 0.606±0.041 0.187±0.011 0.628±0.064 TabICL-ZS 0.628±0.038 0.183±0.009 0.657±0.042 TabPFN-CE 0.608±0.039 0.221±0.022 0.625±0.058 TabDPT-CE 0.607±0.058 0.204±0.024 0.640±0.065 TabICL-CE 0.607±0.041 0.200±0.017 0.650±0.059 TabPFN-Cox 0.629±0.035 0.183±0.012 0.659±0.047 TabDPT-Cox 0.626±0.040 0.186±0.014 0.657±0.054 TabICL-Cox 0.626±0.039 0.192±0.015 0.663±0.051 TabPFN-DH 0.619±0.058 0.208±0.017 0.653±0.076 TabDPT-DH 0.620±0.033 0.199±0.016 0.653±0.066 TabICL-DH 0.621±0.017 0.202±0.013 0.654±0.039 TabPFN-MTLR 0.614±0.042 0.185±0.015 0.642±0.049 TabDPT-MTLR 0.620±0.036 0.185±0.013 0.651±0.051 TabICL-MTLR 0.613±0.035 0.188±0.012 0.651±0.042

Cox PH 0.726±0.017 0.068±0.002 0.714±0.016 RSF 0.710±0.017 0.067±0.002 0.722±0.014 GBSA 0.731±0.020 0.069±0.003 0.720±0.019 DeepSurv 0.595±0.173 0.074±0.008 0.603±0.152 MLP-MTLR 0.728±0.019 0.067±0.003 0.720±0.013 MLP-DH 0.733±0.019 0.084±0.029 0.721±0.014 SurvTRACE 0.730±0.022 0.067±0.002 0.718±0.020 DySurv 0.723±0.016 0.068±0.003 0.713±0.022 TabPFN-ZS 0.714±0.025 0.116±0.014 0.713±0.014 TabDPT-ZS 0.716±0.023 0.116±0.015 0.713±0.015 TabICL-ZS 0.723±0.020 0.117±0.015 0.721±0.016 TabPFN-CE 0.729±0.020 0.461±0.033 0.722±0.018 TabDPT-CE 0.728±0.022 0.490±0.030 0.720±0.021 TabICL-CE 0.727±0.021 0.480±0.032 0.720±0.022 TabPFN-Cox 0.731±0.022 0.067±0.002 0.719±0.018 TabDPT-Cox 0.728±0.022 0.068±0.002 0.717±0.022 TabICL-Cox 0.730±0.022 0.069±0.002 0.718±0.021 TabPFN-DH 0.726±0.017 0.147±0.039 0.718±0.017 TabDPT-DH 0.731±0.020 0.145±0.021 0.719±0.020 TabICL-DH 0.729±0.021 0.142±0.025 0.718±0.013 TabPFN-MTLR 0.731±0.019 0.112±0.014 0.722±0.018 TabDPT-MTLR 0.728±0.017 0.113±0.016 0.717±0.020 TabICL-MTLR 0.732±0.018 0.113±0.021 0.720±0.019

RDATA

RETINOPATHY

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH SurvTRACE DySurv TabPFN-ZS TabDPT-ZS TabICL-ZS TabPFN-CE TabDPT-CE TabICL-CE TabPFN-Cox TabDPT-Cox TabICL-Cox TabPFN-DH

Ctd ↑

IBS ↓

AUC ↑

0.682±0.021 0.175±0.007 0.709±0.029 0.636±0.023 0.179±0.010 0.697±0.021 0.662±0.027 0.175±0.007 0.710±0.030 0.679±0.026 0.175±0.010 0.707±0.037 0.637±0.041 0.180±0.014 0.674±0.038 0.643±0.045 0.196±0.022 0.666±0.049 0.670±0.019 0.179±0.007 0.699±0.028 0.674±0.017 0.177±0.010 0.701±0.032 0.675±0.018 0.167±0.009 0.705±0.025 0.661±0.031 0.171±0.010 0.694±0.030 0.673±0.028 0.169±0.010 0.698±0.027 0.599±0.010 0.253±0.018 0.615±0.025 0.664±0.010 0.296±0.016 0.693±0.025 0.661±0.020 0.292±0.013 0.690±0.015 0.675±0.024 0.178±0.008 0.701±0.031 0.679±0.019 0.177±0.007 0.709±0.028 0.681±0.023 0.175±0.008 0.706±0.030 0.662±0.018 0.213±0.013 0.686±0.034

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH SurvTRACE DySurv TabPFN-ZS TabDPT-ZS TabICL-ZS TabPFN-CE TabDPT-CE TabICL-CE TabPFN-Cox TabDPT-Cox TabICL-Cox TabPFN-DH

Ctd ↑

IBS ↓

AUC ↑

0.654±0.044 0.192±0.016 0.689±0.033 0.642±0.069 0.185±0.016 0.695±0.044 0.661±0.045 0.192±0.021 0.697±0.031 0.613±0.063 0.188±0.019 0.638±0.060 0.617±0.042 0.209±0.040 0.641±0.030 0.560±0.035 0.304±0.044 0.566±0.052 0.613±0.049 0.252±0.136 0.655±0.046 0.594±0.114 0.352±0.184 0.610±0.152 0.631±0.057 0.225±0.053 0.668±0.028 0.609±0.057 0.228±0.057 0.662±0.048 0.617±0.070 0.221±0.053 0.665±0.055 0.554±0.016 0.283±0.065 0.556±0.036 0.616±0.052 0.381±0.073 0.622±0.057 0.594±0.068 0.359±0.090 0.603±0.063 0.646±0.034 0.186±0.019 0.677±0.033 0.642±0.035 0.187±0.015 0.661±0.034 0.629±0.042 0.188±0.015 0.642±0.036 0.611±0.042 0.321±0.069 0.654±0.038

TabFMs for Time-to-Event Prediction

TabDPT-DH TabICL-DH TabPFN-MTLR TabDPT-MTLR TabICL-MTLR

0.675±0.014 0.676±0.024 0.667±0.022 0.668±0.013 0.668±0.018

0.210±0.012 0.209±0.013 0.178±0.010 0.181±0.008 0.180±0.008

0.705±0.020 0.700±0.033 0.700±0.029 0.696±0.019 0.694±0.028

TabDPT-DH TabICL-DH TabPFN-MTLR TabDPT-MTLR TabICL-MTLR

RHC Model

0.648±0.060 0.626±0.044 0.646±0.044 0.638±0.055 0.635±0.051

0.266±0.057 0.261±0.052 0.204±0.038 0.207±0.034 0.209±0.032

0.664±0.053 0.640±0.046 0.680±0.030 0.661±0.057 0.658±0.062

ROTT2

Ctd ↑

IBS ↓

AUC ↑

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.605±0.043 0.144±0.014 0.609±0.036 RSF 0.616±0.084 0.148±0.018 0.609±0.051 GBSA 0.554±0.044 0.153±0.013 0.556±0.048 DeepSurv 0.537±0.046 0.156±0.020 0.515±0.054 MLP-MTLR 0.549±0.114 0.235±0.072 0.579±0.069 MLP-DH 0.575±0.043 0.179±0.043 0.564±0.069 SurvTRACE 0.611±0.027 0.188±0.040 0.616±0.033 DySurv 0.570±0.061 0.291±0.058 0.556±0.047 TabPFN-ZS 0.615±0.047 0.157±0.018 0.608±0.039 TabDPT-ZS 0.608±0.048 0.157±0.017 0.614±0.038 TabICL-ZS 0.625±0.064 0.155±0.019 0.626±0.041 TabPFN-CE 0.604±0.048 0.171±0.027 0.596±0.050 TabDPT-CE 0.601±0.036 0.168±0.014 0.572±0.060 TabICL-CE 0.618±0.029 0.160±0.018 0.594±0.048 TabPFN-Cox 0.610±0.069 0.145±0.016 0.599±0.081 TabDPT-Cox 0.636±0.043 0.143±0.014 0.630±0.056 TabICL-Cox 0.636±0.033 0.143±0.015 0.620±0.049 TabPFN-DH 0.597±0.061 0.141±0.016 0.566±0.067 TabDPT-DH 0.619±0.065 0.142±0.016 0.609±0.069 TabICL-DH 0.643±0.040 0.141±0.014 0.616±0.063 TabPFN-MTLR 0.614±0.054 0.148±0.018 0.602±0.068 TabDPT-MTLR 0.638±0.044 0.143±0.015 0.627±0.062 TabICL-MTLR 0.645±0.047 0.143±0.013 0.625±0.060

Cox PH 0.691±0.028 0.182±0.016 0.727±0.029 RSF 0.701±0.032 0.173±0.019 0.774±0.038 GBSA 0.699±0.027 0.167±0.034 0.760±0.030 DeepSurv 0.703±0.026 0.175±0.017 0.742±0.034 MLP-MTLR 0.693±0.018 0.161±0.018 0.745±0.032 MLP-DH 0.693±0.031 0.177±0.021 0.740±0.035 SurvTRACE 0.709±0.029 0.169±0.027 0.757±0.038 DySurv 0.704±0.030 0.218±0.041 0.745±0.034 TabPFN-ZS 0.714±0.026 0.181±0.029 0.766±0.035 TabDPT-ZS 0.703±0.019 0.190±0.028 0.754±0.027 TabICL-ZS 0.714±0.030 0.182±0.027 0.770±0.036 TabPFN-CE 0.684±0.017 0.200±0.031 0.724±0.017 TabDPT-CE 0.716±0.029 0.300±0.035 0.773±0.040 TabICL-CE 0.709±0.027 0.303±0.036 0.763±0.039 TabPFN-Cox 0.706±0.029 0.173±0.027 0.748±0.035 TabDPT-Cox 0.722±0.030 0.172±0.028 0.767±0.037 TabICL-Cox 0.716±0.032 0.178±0.030 0.762±0.039 TabPFN-DH 0.707±0.029 0.227±0.029 0.746±0.030 TabDPT-DH 0.725±0.031 0.225±0.030 0.763±0.038 TabICL-DH 0.722±0.030 0.226±0.027 0.766±0.038 TabPFN-MTLR 0.705±0.032 0.197±0.030 0.749±0.041 TabDPT-MTLR 0.720±0.031 0.199±0.037 0.770±0.038 TabICL-MTLR 0.715±0.032 0.203±0.029 0.761±0.041

SCANIA

SMARTO

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH SurvTRACE DySurv TabPFN-ZS TabDPT-ZS TabICL-ZS TabPFN-CE TabDPT-CE TabICL-CE TabPFN-Cox TabDPT-Cox TabICL-Cox TabPFN-DH TabDPT-DH TabICL-DH

Ctd ↑ 0.550±0.022 0.550±0.021 0.546±0.026 0.534±0.022 0.508±0.037 0.539±0.028 0.538±0.027 0.544±0.026 0.556±0.022 0.543±0.022 0.557±0.025 0.535±0.016 0.569±0.023 0.566±0.014 0.547±0.024 0.562±0.026 0.557±0.020 0.542±0.032 0.560±0.021 0.555±0.021

IBS ↓

AUC ↑

0.172±0.003 0.534±0.025 0.173±0.006 0.581±0.029 0.172±0.005 0.584±0.029 0.173±0.003 0.528±0.036 0.181±0.015 0.508±0.028 0.176±0.003 0.519±0.037 0.175±0.008 0.548±0.062 0.230±0.076 0.543±0.031 0.163±0.005 0.634±0.036 0.168±0.005 0.594±0.034 0.164±0.004 0.628±0.033 0.251±0.010 0.573±0.016 0.235±0.014 0.621±0.029 0.237±0.016 0.615±0.031 0.172±0.003 0.546±0.033 0.168±0.004 0.595±0.040 0.170±0.003 0.585±0.029 0.176±0.005 0.539±0.041 0.172±0.004 0.613±0.038 0.172±0.003 0.606±0.041

Model Cox PH RSF GBSA DeepSurv MLP-MTLR MLP-DH SurvTRACE DySurv TabPFN-ZS TabDPT-ZS TabICL-ZS TabPFN-CE TabDPT-CE TabICL-CE TabPFN-Cox TabDPT-Cox TabICL-Cox TabPFN-DH TabDPT-DH TabICL-DH

Ctd ↑ 0.699±0.075 0.651±0.084 0.659±0.078 0.644±0.080 0.617±0.102 0.487±0.183 0.600±0.044 0.668±0.069 0.648±0.073 0.621±0.054 0.667±0.078 0.648±0.064 0.582±0.087 0.598±0.080 0.690±0.065 0.694±0.072 0.695±0.062 0.665±0.057 0.686±0.076 0.685±0.054

IBS ↓

AUC ↑

0.076±0.006 0.698±0.072 0.078±0.005 0.687±0.087 0.085±0.010 0.673±0.073 0.081±0.006 0.618±0.076 0.112±0.044 0.640±0.072 0.177±0.100 0.585±0.065 0.160±0.168 0.617±0.024 0.246±0.063 0.646±0.094 0.089±0.014 0.663±0.011 0.093±0.020 0.639±0.049 0.089±0.014 0.695±0.048 0.144±0.037 0.651±0.062 0.211±0.091 0.604±0.128 0.197±0.053 0.624±0.064 0.078±0.003 0.694±0.081 0.079±0.006 0.681±0.079 0.079±0.003 0.678±0.071 0.283±0.023 0.649±0.067 0.151±0.021 0.688±0.081 0.192±0.015 0.676±0.048

Minh-Khoi Pham et al.

TabPFN-MTLR 0.536±0.026 TabDPT-MTLR 0.574±0.027 TabICL-MTLR 0.564±0.021

0.175±0.005 0.168±0.007 0.168±0.005

0.533±0.036 0.607±0.034 0.601±0.033

TabPFN-MTLR 0.667±0.049 TabDPT-MTLR 0.687±0.091 TabICL-MTLR 0.701±0.055

STAGEC Model

Ctd ↑

0.117±0.021 0.102±0.021 0.107±0.011

0.666±0.046 0.671±0.099 0.682±0.061

SUPPORT2 IBS ↓

AUC ↑

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.729±0.076 0.199±0.063 0.762±0.094 RSF 0.614±0.116 0.203±0.046 0.660±0.112 GBSA 0.635±0.077 0.225±0.082 0.721±0.075 DeepSurv 0.703±0.104 0.206±0.071 0.750±0.131 MLP-MTLR 0.716±0.108 0.233±0.091 0.758±0.158 MLP-DH 0.677±0.138 0.297±0.163 0.709±0.153 SurvTRACE 0.599±0.120 0.309±0.213 0.682±0.122 DySurv 0.698±0.068 0.338±0.162 0.753±0.059 TabPFN-ZS 0.705±0.072 0.298±0.170 0.759±0.092 TabDPT-ZS 0.647±0.103 0.306±0.169 0.702±0.050 TabICL-ZS 0.696±0.046 0.297±0.134 0.706±0.096 TabPFN-CE 0.698±0.090 0.280±0.188 0.768±0.097 TabDPT-CE 0.630±0.052 0.379±0.225 0.626±0.078 TabICL-CE 0.728±0.094 0.357±0.214 0.783±0.111 TabPFN-Cox 0.687±0.064 0.203±0.064 0.719±0.083 TabDPT-Cox 0.657±0.070 0.200±0.061 0.703±0.078 TabICL-Cox 0.651±0.068 0.205±0.049 0.713±0.089 TabPFN-DH 0.663±0.124 0.344±0.196 0.765±0.091 TabDPT-DH 0.656±0.044 0.309±0.174 0.682±0.076 TabICL-DH 0.599±0.090 0.345±0.197 0.661±0.091 TabPFN-MTLR 0.683±0.085 0.255±0.147 0.729±0.153 TabDPT-MTLR 0.652±0.133 0.269±0.164 0.694±0.125 TabICL-MTLR 0.613±0.089 0.280±0.158 0.653±0.092

Cox PH 0.783±0.019 0.153±0.013 0.894±0.022 RSF 0.742±0.046 0.173±0.011 0.845±0.030 GBSA 0.771±0.035 0.167±0.011 0.883±0.024 DeepSurv 0.740±0.026 0.183±0.014 0.808±0.028 MLP-MTLR 0.744±0.056 0.198±0.018 0.879±0.033 MLP-DH 0.737±0.028 0.222±0.015 0.860±0.030 SurvTRACE 0.763±0.031 0.242±0.114 0.888±0.035 DySurv 0.700±0.043 0.399±0.122 0.769±0.070 TabPFN-ZS 0.784±0.026 0.164±0.008 0.907±0.018 TabDPT-ZS 0.755±0.039 0.193±0.014 0.882±0.028 TabICL-ZS 0.782±0.027 0.176±0.007 0.906±0.016 TabPFN-CE 0.784±0.026 0.223±0.031 0.876±0.017 TabDPT-CE 0.783±0.021 0.221±0.026 0.871±0.016 TabICL-CE 0.774±0.021 0.216±0.016 0.873±0.021 TabPFN-Cox 0.777±0.025 0.166±0.022 0.877±0.025 TabDPT-Cox 0.786±0.024 0.149±0.011 0.890±0.014 TabICL-Cox 0.756±0.052 0.162±0.019 0.835±0.043 TabPFN-DH 0.759±0.019 0.275±0.031 0.850±0.024 TabDPT-DH 0.790±0.025 0.233±0.016 0.884±0.008 TabICL-DH 0.768±0.047 0.248±0.022 0.841±0.044 TabPFN-MTLR 0.776±0.023 0.176±0.019 0.870±0.028 TabDPT-MTLR 0.787±0.026 0.150±0.007 0.895±0.014 TabICL-MTLR 0.769±0.020 0.165±0.011 0.870±0.017

UIS

VDV

Model

Ctd ↑

Cox PH 0.579±0.020 RSF 0.556±0.026 GBSA 0.573±0.019 DeepSurv 0.539±0.033 MLP-MTLR 0.511±0.023 MLP-DH 0.519±0.027 SurvTRACE 0.551±0.030 DySurv 0.546±0.040 TabPFN-ZS 0.548±0.025 TabDPT-ZS 0.536±0.028 TabICL-ZS 0.554±0.033 TabPFN-CE 0.537±0.044 TabDPT-CE 0.545±0.034 TabICL-CE 0.550±0.040 TabPFN-Cox 0.562±0.028 TabDPT-Cox 0.559±0.017 TabICL-Cox 0.557±0.012 TabPFN-DH 0.546±0.025 TabDPT-DH 0.543±0.019 TabICL-DH 0.557±0.021 TabPFN-MTLR 0.560±0.026 TabDPT-MTLR 0.548±0.015

IBS ↓

AUC ↑

0.125±0.016 0.123±0.015 0.135±0.028 0.128±0.023 0.159±0.042 0.140±0.021 0.153±0.035 0.151±0.042 0.124±0.027 0.125±0.025 0.124±0.027 0.153±0.027 0.138±0.025 0.138±0.025 0.123±0.019 0.124±0.018 0.122±0.019 0.129±0.027 0.124±0.024 0.124±0.026 0.125±0.023 0.122±0.022

0.613±0.022 0.597±0.028 0.607±0.028 0.560±0.042 0.526±0.017 0.513±0.033 0.571±0.056 0.563±0.059 0.581±0.030 0.570±0.045 0.590±0.054 0.556±0.047 0.570±0.027 0.580±0.055 0.584±0.040 0.583±0.025 0.586±0.018 0.559±0.043 0.563±0.040 0.570±0.022 0.580±0.045 0.567±0.031

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.580±0.115 0.268±0.085 0.571±0.141 RSF 0.693±0.122 0.190±0.048 0.726±0.150 GBSA 0.598±0.086 0.242±0.118 0.659±0.063 DeepSurv 0.616±0.135 0.199±0.033 0.609±0.184 MLP-MTLR 0.600±0.023 0.206±0.031 0.604±0.052 MLP-DH 0.584±0.129 0.368±0.212 0.543±0.082 SurvTRACE 0.453±0.189 0.381±0.165 0.444±0.209 DySurv 0.499±0.225 0.569±0.295 0.481±0.197 TabPFN-ZS 0.538±0.123 0.414±0.327 0.456±0.137 TabDPT-ZS 0.444±0.105 0.299±0.117 0.456±0.113 TabICL-ZS 0.406±0.127 0.395±0.319 0.354±0.106 TabPFN-CE 0.585±0.161 0.283±0.117 0.604±0.198 TabDPT-CE 0.581±0.085 0.383±0.278 0.543±0.082 TabICL-CE 0.582±0.213 0.339±0.278 0.600±0.226 TabPFN-Cox 0.584±0.150 0.200±0.050 0.589±0.180 TabDPT-Cox 0.418±0.173 0.225±0.067 0.406±0.160 TabICL-Cox 0.659±0.176 0.200±0.059 0.620±0.207 TabPFN-DH 0.522±0.157 0.404±0.255 0.503±0.226 TabDPT-DH 0.475±0.129 0.387±0.253 0.521±0.093 TabICL-DH 0.596±0.159 0.393±0.261 0.605±0.260 TabPFN-MTLR 0.587±0.057 0.291±0.120 0.670±0.018 TabDPT-MTLR 0.594±0.142 0.284±0.140 0.544±0.187

TabFMs for Time-to-Event Prediction

TabICL-MTLR

0.562±0.023

0.121±0.022

0.586±0.026

TabICL-MTLR

VETERAN Model

Ctd ↑

0.656±0.186

0.266±0.173

0.628±0.236

IBS ↓

AUC ↑

VLBW IBS ↓

AUC ↑

Model

Ctd ↑

Cox PH 0.710±0.023 0.068±0.017 0.792±0.035 RSF 0.676±0.019 0.070±0.020 0.781±0.057 GBSA 0.704±0.055 0.070±0.021 0.797±0.060 DeepSurv 0.707±0.059 0.072±0.017 0.791±0.083 MLP-MTLR 0.681±0.055 0.077±0.010 0.788±0.062 MLP-DH 0.675±0.057 0.081±0.019 0.785±0.074 SurvTRACE 0.604±0.086 0.084±0.031 0.703±0.100 DySurv 0.694±0.059 0.082±0.026 0.779±0.078 TabPFN-ZS 0.719±0.050 0.065±0.017 0.828±0.050 TabDPT-ZS 0.695±0.068 0.063±0.018 0.810±0.073 TabICL-ZS 0.671±0.035 0.069±0.019 0.767±0.080 TabPFN-CE 0.686±0.040 0.071±0.020 0.775±0.053 TabDPT-CE 0.712±0.043 0.070±0.022 0.808±0.067 TabICL-CE 0.724±0.050 0.068±0.022 0.809±0.067 TabPFN-Cox 0.706±0.025 0.065±0.017 0.790±0.045 TabDPT-Cox 0.722±0.033 0.067±0.020 0.806±0.045 TabICL-Cox 0.687±0.053 0.075±0.017 0.764±0.061 TabPFN-DH 0.699±0.068 0.081±0.018 0.758±0.093 TabDPT-DH 0.664±0.116 0.080±0.020 0.706±0.111 TabICL-DH 0.654±0.062 0.082±0.017 0.692±0.104 TabPFN-MTLR 0.712±0.057 0.070±0.020 0.786±0.069 TabDPT-MTLR 0.700±0.060 0.071±0.024 0.778±0.071 TabICL-MTLR 0.653±0.089 0.075±0.021 0.734±0.126

Cox PH 0.956±0.019 0.068±0.080 0.926±0.047 RSF 0.928±0.045 0.102±0.112 0.876±0.109 GBSA 0.940±0.044 0.064±0.090 0.903±0.093 DeepSurv 0.875±0.122 0.101±0.139 0.812±0.191 MLP-MTLR 0.805±0.119 0.057±0.074 0.926±0.044 MLP-DH 0.577±0.249 0.062±0.075 0.725±0.285 SurvTRACE 0.502±0.243 0.086±0.103 0.865±0.071 DySurv 0.805±0.121 0.058±0.098 0.781±0.136 TabPFN-ZS 0.567±0.209 0.077±0.101 0.917±0.103 TabDPT-ZS 0.555±0.193 0.076±0.100 0.909±0.096 TabICL-ZS 0.565±0.190 0.080±0.101 0.912±0.101 TabPFN-CE 0.731±0.260 0.044±0.061 0.863±0.104 TabDPT-CE 0.721±0.258 0.053±0.061 0.891±0.040 TabICL-CE 0.707±0.246 0.066±0.077 0.850±0.122 TabPFN-Cox 0.940±0.022 0.095±0.128 0.898±0.078 TabDPT-Cox 0.928±0.024 0.117±0.150 0.885±0.068 TabICL-Cox 0.917±0.045 0.105±0.149 0.871±0.082 TabPFN-DH 0.836±0.104 0.064±0.082 0.836±0.117 TabDPT-DH 0.901±0.069 0.062±0.083 0.896±0.059 TabICL-DH 0.916±0.043 0.065±0.085 0.885±0.053 TabPFN-MTLR 0.935±0.028 0.061±0.098 0.895±0.083 TabDPT-MTLR 0.921±0.031 0.064±0.115 0.885±0.061 TabICL-MTLR 0.918±0.036 0.069±0.100 0.880±0.084

WPBC

ZINC

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.664±0.065 0.181±0.029 0.683±0.065 RSF 0.633±0.121 0.171±0.013 0.702±0.099 GBSA 0.570±0.053 0.198±0.028 0.631±0.067 DeepSurv 0.484±0.144 0.197±0.038 0.471±0.140 MLP-MTLR 0.591±0.080 0.245±0.069 0.625±0.082 MLP-DH 0.528±0.188 0.328±0.138 0.522±0.188 SurvTRACE 0.507±0.179 0.201±0.021 0.543±0.158 DySurv 0.563±0.087 0.572±0.337 0.560±0.116 TabPFN-ZS 0.628±0.047 0.236±0.076 0.687±0.065 TabDPT-ZS 0.615±0.072 0.228±0.081 0.709±0.082 TabICL-ZS 0.634±0.042 0.219±0.071 0.714±0.043 TabPFN-CE 0.570±0.064 0.208±0.031 0.609±0.054 TabDPT-CE 0.611±0.094 0.270±0.103 0.638±0.092 TabICL-CE 0.622±0.105 0.264±0.090 0.638±0.099 TabPFN-Cox 0.634±0.105 0.167±0.013 0.649±0.109 TabDPT-Cox 0.619±0.051 0.176±0.018 0.648±0.053 TabICL-Cox 0.681±0.120 0.177±0.018 0.705±0.121 TabPFN-DH 0.547±0.091 0.381±0.136 0.598±0.068 TabDPT-DH 0.641±0.052 0.317±0.133 0.650±0.083 TabICL-DH 0.627±0.156 0.351±0.143 0.678±0.141 TabPFN-MTLR 0.622±0.032 0.215±0.058 0.670±0.028 TabDPT-MTLR 0.631±0.046 0.211±0.065 0.665±0.046 TabICL-MTLR 0.646±0.073 0.243±0.077 0.685±0.076

Model

Ctd ↑

IBS ↓

AUC ↑

Cox PH 0.725±0.059 0.096±0.010 0.728±0.052 RSF 0.742±0.066 0.090±0.013 0.770±0.059 GBSA 0.666±0.118 0.096±0.016 0.683±0.145 DeepSurv 0.755±0.042 0.096±0.012 0.751±0.054 MLP-MTLR 0.661±0.092 0.109±0.033 0.675±0.118 MLP-DH 0.662±0.091 0.112±0.019 0.689±0.104 SurvTRACE 0.745±0.058 0.095±0.021 0.738±0.066 DySurv 0.638±0.172 0.211±0.133 0.633±0.166 TabPFN-ZS 0.751±0.077 0.092±0.019 0.749±0.056 TabDPT-ZS 0.756±0.058 0.092±0.017 0.760±0.061 TabICL-ZS 0.753±0.049 0.091±0.016 0.750±0.051 TabPFN-CE 0.697±0.036 0.161±0.022 0.681±0.051 TabDPT-CE 0.643±0.086 0.218±0.023 0.631±0.108 TabICL-CE 0.671±0.059 0.214±0.026 0.663±0.066 TabPFN-Cox 0.766±0.032 0.090±0.011 0.762±0.049 TabDPT-Cox 0.760±0.029 0.089±0.011 0.747±0.036 TabICL-Cox 0.759±0.044 0.093±0.011 0.754±0.038 TabPFN-DH 0.697±0.065 0.289±0.023 0.723±0.046 TabDPT-DH 0.755±0.040 0.214±0.031 0.743±0.041 TabICL-DH 0.752±0.052 0.261±0.064 0.762±0.049 TabPFN-MTLR 0.777±0.053 0.128±0.023 0.781±0.064 TabDPT-MTLR 0.767±0.041 0.105±0.006 0.760±0.034 TabICL-MTLR 0.756±0.050 0.115±0.015 0.742±0.043

Minh-Khoi Pham et al.

Table 6: Competing-risk results per cause (mean ± std across 5 folds). Superscript (1) and (2) index the two competing causes. Best value per column within a data set is bold, second best underlined. IBS(1) ↓

AUC(1) ↑

Ctd ↑

(2)

IBS(2) ↓

AUC(2) ↑

Cox PH 0.744±0.009 SurvBoost 0.730±0.004 DeepSurv 0.691±0.057 MLP-DH 0.718±0.012 DySurv 0.720±0.010 SurvTrace 0.728±0.005 TabPFN-DH 0.713±0.007 TabDPT-DH 0.726±0.005 TabICL-DH 0.719±0.014 TabPFN-MTLR 0.728±0.008 TabDPT-MTLR 0.732±0.010 TabICL-MTLR 0.729±0.010 TabPFN-Cox 0.726±0.014 TabDPT-Cox 0.701±0.022 TabICL-Cox 0.712±0.009

0.177±0.007 0.180±0.007 0.185±0.010 0.188±0.008 0.175±0.011 0.176±0.010 0.181±0.005 0.182±0.004 0.185±0.007 0.165±0.007 0.163±0.009 0.162±0.007 0.166±0.007 0.198±0.058 0.203±0.079

0.639±0.017 0.623±0.025 0.606±0.030 0.621±0.010 0.608±0.026 0.615±0.017 0.611±0.019 0.612±0.011 0.596±0.017 0.634±0.018 0.627±0.022 0.631±0.021 0.639±0.008 0.622±0.015 0.622±0.020

0.713±0.020 0.701±0.018 0.568±0.138 0.673±0.019 0.679±0.029 0.702±0.027 0.697±0.017 0.709±0.025 0.706±0.032 0.701±0.026 0.710±0.014 0.706±0.023 0.666±0.013 0.625±0.048 0.631±0.046

0.146±0.005 0.146±0.005 0.159±0.010 0.152±0.005 0.147±0.005 0.144±0.005 0.148±0.005 0.150±0.006 0.150±0.003 0.135±0.005 0.133±0.005 0.134±0.003 0.133±0.007 0.146±0.015 0.147±0.016

0.592±0.042 0.591±0.046 0.546±0.068 0.597±0.040 0.537±0.050 0.569±0.045 0.575±0.030 0.574±0.049 0.566±0.052 0.594±0.045 0.590±0.043 0.589±0.041 0.588±0.047 0.594±0.029 0.573±0.038

PBC2-CR Cox PH 0.790±0.008 SurvBoost 0.837±0.014 DeepSurv 0.815±0.013 MLP-DH 0.802±0.026 DySurv 0.765±0.010 SurvTrace 0.801±0.015 TabPFN-DH 0.815±0.013 TabDPT-DH 0.846±0.016 TabICL-DH 0.828±0.021 TabPFN-MTLR 0.807±0.012 TabDPT-MTLR 0.837±0.011 TabICL-MTLR 0.833±0.013 TabPFN-Cox 0.735±0.047 TabDPT-Cox 0.765±0.119 TabICL-Cox 0.781±0.066

0.127±0.010 0.102±0.005 0.117±0.011 0.098±0.009 0.142±0.009 0.109±0.009 0.108±0.014 0.098±0.010 0.109±0.012 0.123±0.012 0.108±0.012 0.111±0.013 0.399±0.058 0.298±0.130 0.341±0.068

0.829±0.015 0.869±0.016 0.851±0.013 0.863±0.032 0.811±0.011 0.849±0.022 0.852±0.020 0.874±0.014 0.858±0.019 0.845±0.013 0.862±0.008 0.858±0.011 0.821±0.018 0.844±0.013 0.860±0.010

0.830±0.047 0.872±0.032 0.867±0.023 0.801±0.057 0.677±0.044 0.840±0.019 0.866±0.034 0.880±0.036 0.816±0.058 0.844±0.020 0.867±0.042 0.887±0.024 0.781±0.038 0.771±0.092 0.683±0.080

0.034±0.001 0.028±0.001 0.030±0.002 0.026±0.002 0.041±0.007 0.030±0.003 0.030±0.002 0.031±0.002 0.033±0.003 0.035±0.002 0.034±0.005 0.033±0.001 0.099±0.023 0.080±0.028 0.065±0.027

0.839±0.042 0.886±0.031 0.863±0.031 0.884±0.044 0.708±0.076 0.856±0.028 0.847±0.031 0.869±0.020 0.825±0.039 0.850±0.031 0.873±0.039 0.879±0.036 0.760±0.072 0.714±0.117 0.664±0.084

Supp-CR

Cox PH 0.776±0.013 SurvBoost 0.813±0.008 DeepSurv 0.844±0.009 MLP-DH 0.821±0.017 DySurv 0.480±0.015 SurvTrace 0.565±0.026 TabPFN-DH 0.816±0.016 TabDPT-DH 0.805±0.023 TabICL-DH 0.825±0.020 TabPFN-MTLR 0.840±0.010 TabDPT-MTLR 0.835±0.017 TabICL-MTLR 0.835±0.018 TabPFN-Cox 0.783±0.020 TabDPT-Cox 0.778±0.033 TabICL-Cox 0.783±0.033

0.154±0.005 0.131±0.006 0.120±0.009 0.129±0.003 0.173±0.005 0.137±0.006 0.139±0.005 0.144±0.013 0.136±0.013 0.131±0.007 0.131±0.006 0.136±0.006 0.133±0.006 0.133±0.013 0.127±0.015

0.795±0.011 0.834±0.015 0.861±0.010 0.827±0.015 0.759±0.019 0.805±0.008 0.782±0.014 0.777±0.026 0.797±0.018 0.841±0.012 0.843±0.011 0.840±0.025 0.816±0.013 0.813±0.031 0.821±0.029

0.804±0.005 0.815±0.006 0.813±0.005 0.745±0.007 0.357±0.013 0.363±0.010 0.804±0.004 0.804±0.005 0.800±0.008 0.816±0.003 0.821±0.007 0.819±0.005 0.715±0.005 0.716±0.008 0.708±0.009

0.175±0.009 0.170±0.008 0.170±0.008 0.202±0.007 0.201±0.006 0.179±0.007 0.172±0.007 0.173±0.007 0.176±0.008 0.171±0.009 0.172±0.011 0.172±0.010 0.175±0.008 0.177±0.007 0.183±0.009

0.846±0.009 0.855±0.007 0.850±0.006 0.798±0.010 0.803±0.013 0.834±0.007 0.836±0.008 0.841±0.011 0.839±0.010 0.850±0.008 0.853±0.008 0.850±0.009 0.841±0.005 0.840±0.005 0.836±0.005

Syn-CR

Cox PH SurvBoost

0.580±0.007 0.686±0.009

0.188±0.001 0.169±0.003

0.573±0.007 0.744±0.014

0.590±0.008 0.683±0.014

0.189±0.004 0.173±0.004

0.577±0.011 0.737±0.014

Dataset

Model

Fram-CR

(1)

Ctd ↑

TabFMs for Time-to-Event Prediction

Dataset

Model

(1)

Ctd ↑

DeepSurv 0.747±0.007 MLP-DH 0.739±0.005 DySurv 0.504±0.012 SurvTrace 0.496±0.013 TabPFN-DH 0.608±0.004 TabDPT-DH 0.538±0.009 TabICL-DH 0.504±0.007 TabPFN-MTLR 0.714±0.007 TabDPT-MTLR 0.593±0.011 TabICL-MTLR 0.569±0.013 TabPFN-Cox 0.714±0.006 TabDPT-Cox 0.669±0.013 TabICL-Cox 0.616±0.012

IBS(1) ↓

AUC(1) ↑

Ctd ↑

(2)

IBS(2) ↓

AUC(2) ↑

0.150±0.004 0.239±0.005 0.267±0.007 0.253±0.009 0.239±0.004 0.250±0.005 0.251±0.004 0.166±0.004 0.197±0.003 0.200±0.004 0.227±0.002 0.272±0.041 0.278±0.038

0.807±0.009 0.750±0.019 0.804±0.011 0.733±0.014 0.660±0.012 0.603±0.005 0.575±0.007 0.796±0.007 0.610±0.016 0.577±0.010 0.711±0.020 0.604±0.030 0.585±0.008

0.749±0.005 0.743±0.005 0.498±0.012 0.491±0.014 0.610±0.011 0.540±0.018 0.513±0.011 0.717±0.006 0.600±0.014 0.579±0.014 0.716±0.009 0.679±0.011 0.621±0.014

0.152±0.003 0.243±0.005 0.268±0.007 0.255±0.014 0.244±0.008 0.251±0.009 0.254±0.006 0.167±0.004 0.203±0.006 0.202±0.003 0.244±0.025 0.234±0.028 0.235±0.015

0.805±0.006 0.752±0.012 0.803±0.005 0.720±0.016 0.673±0.017 0.602±0.024 0.583±0.019 0.798±0.007 0.608±0.021 0.598±0.017 0.701±0.025 0.630±0.038 0.593±0.017

References Laura Antolini, Patrizia Boracchi, and Elia Biganzoli. A time-dependent discrimination index for survival data. Statistics in medicine, 24(24):3927–3944, 2005. doi: 10.1002/sim.2427. Huynh Thi Khanh Chi, Amonae Dabbs-Brown, Anna Jurek-Loughrey, James Mulhall, Tuan Dung Pham, Ngoc Phu Doan, Viet Hung Tran, Zichi Zhang, Xuan Hoang Nguyen, Yimeng An, et al. Obstructive sleep apnea prediction: a comprehensive review and comparative study. Machine Learning, 115(2):29, 2026. D. R. Cox. Regression models and life-tables. Journal of the Royal Statistical Society: Series B, 34(2):187–220, 1972. doi: 10.1111/j.2517-6161.1972.tb00899.x. URL https://rss. onlinelibrary.wiley.com/doi/pdfdirect/10.1111/j.2517-6161.1972.tb00899.x. Erik Drysdale. Survset: An open-source time-to-event dataset repository. arXiv preprint arXiv:2203.03094, 2022. doi: 10.48550/arxiv.2203.03094. URL https://arxiv.org/pdf/ 2203.03094. E. Graf, C. Schmoor, W. Sauerbrei, et al. Assessment and comparison of prognostic classification schemes for survival data. Statistics in Medicine, 18(17–18):2529–2545, 1999. doi: 10.1002/(sici)1097-0258(19990915/30)18:17/18<2529::aid-sim274>3.0.co;2-5. N. Hollmann, S. Müller, K. Eggensperger, et al. Tabpfn: A transformer that solves small tabular classification problems in a second. In International Conference on Learning Representations (ICLR), 2023. doi: 10.48550/arxiv.2207.01848. URL https://arxiv. org/pdf/2207.01848. N. Hollmann, S. Müller, L. Purucker, et al. Tabpfn v2: Improved in-context learning for tabular data with advanced pre-training. arXiv preprint arXiv:2501.02945, 2025. doi: 10.48550/arxiv.2501.02945. URL https://arxiv.org/pdf/2501.02945. H. Ishwaran, U. B. Kogalur, E. H. Blackstone, et al. Random survival forests. The Annals of Applied Statistics, 2(3):841–860, 2008. doi: 10.1214/08-aoas169. URL 49

Minh-Khoi Pham et al.

https://projecteuclid.org/journals/annals-of-applied-statistics/volume-2/ issue-3/Random-survival-forests/10.1214/08-AOAS169.pdf. J. D. Kalbfleisch and R. L. Prentice. The statistical analysis of failure time data. Wiley series in probability and statistics, 2002. doi: 10.1002/9781118032985. Edward L Kaplan and Paul Meier. Nonparametric estimation from incomplete observations. Journal of the American statistical association, 53(282):457–481, 1958. doi: 10.1080/ 01621459.1958.10501452. Jared L Katzman, Uri Shaham, Alexander Cloninger, et al. Deepsurv: personalized treatment recommender system using a cox proportional hazards deep neural network. BMC medical research methodology, 18(1):24, 2018. doi: 10.1186/s12874-018-0482-1. URL https:// bmcmedresmethodol.biomedcentral.com/counter/pdf/10.1186/s12874-018-0482-1. S. Kim, H. Lai, and X. Zhang. Tabular foundation models can do survival analysis. arXiv preprint arXiv:2601.22259, 2026. doi: 10.48550/arxiv.2601.22259. Weihao Kong and Abhimanyu Das. Introducing tabfm: A zero-shot foundation model for tabular data. Google Research Blog, 2026. URL https://research.google/blog/ introducing-tabfm-a-zero-shot-foundation-model-for-tabular-data/. H. Kvamme, Ø. Borgan, and I. Scheel. Time-to-event prediction with neural networks and cox regression. Journal of Machine Learning Research, 20(129):1–30, 2019. doi: 10.48550/arxiv.1907.00825. URL https://arxiv.org/pdf/1907.00825. C. Lee, W. R. Zame, J. Yoon, et al. Deephit: A deep learning approach to survival analysis with competing risks. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 32, 2018. doi: 10.1609/aaai.v32i1.11842. URL https://ojs.aaai.org/index. php/AAAI/article/download/11842/11701. Si-Yang Liu and Han-Jia Ye. Tabpfn unleashed: A scalable and effective solution to tabular classification problems. ArXiv.org, 2025. doi: 10.48550/arxiv.2502.02527. URL https://arxiv.org/pdf/2502.02527. Junwei Ma, Valentin Thomas, Rasa Hosseinzadeh, et al. Tabdpt: Scaling tabular foundation models on real data. arXiv (Cornell University), 2025. doi: 10.48550/arxiv.2410.18164. URL https://arxiv.org/pdf/2410.18164. Munib Mesinovic, Peter Watkinson, and Tingting Zhu. Dysurv: dynamic deep learning model for survival analysis with conditional variational inference. Journal of the American Medical Informatics Association, 33(1):112–122, 2026. doi: 10.1093/ jamia/ocae271. URL https://academic.oup.com/jamia/advance-article-pdf/doi/ 10.1093/jamia/ocae271/60775529/ocae271.pdf. Chirag Nagpal, Xinyu Li, and Artur Dubrawski. Deep survival machines: Fully parametric survival regression and representation learning for censored data with competing risks. IEEE Journal of Biomedical and Health Informatics, 25(8):3163–3175, 2021. doi: 10.1109/ jbhi.2021.3052441. 50

TabFMs for Time-to-Event Prediction

Minh-Khoi Pham, Luca Cotugno, Alina Sîrbu, Tai Tan Mai, Martin Crane, and Marija Bezbradica. Tabular foundation models for clinical survival analysis via survival-aware adaptation. In International Conference on AI in Healthcare, pages 315–328. Springer, 2026a. doi: 10.1007/978-3-032-35387-0_23. Minh-Khoi Pham, Thang-Long Nguyen Ho, Thao Thi Phuong Dao, et al. Retrieval-aligned tabular foundation models enable robust clinical risk prediction in electronic health records under real-world constraints. arXiv preprint arXiv:2604.01841, 2026b. doi: 10.21203/rs.3. rs-9085469/v1. URL https://doi.org/10.21203/rs.3.rs-9085469/v1. Lennart Purucker, Andrej Tschalzev, Nick Erickson, et al. Beyond iid: How general are tabular foundation models, really? arXiv preprint arXiv:2606.30410, 2026. URL https: //arxiv.org/abs/2606.30410. Shi-ang Qi, Vahid Balazadeh, Michael Cooper, et al. Survivalpfn: Amortizing survival prediction via in-context bayesian inference. arXiv preprint arXiv:2605.15488, 2026. URL https://arxiv.org/pdf/2605.15488. Jingang Qu, David Holzmüller, Gaël Varoquaux, and Marine Le Morvan. Tabicl: A tabular foundation model for in-context learning on large data. In Proceedings of the 42nd International Conference on Machine Learning, ICML’25. JMLR.org, 2025. Dmitrii Seletkov, Paul Hager, Rickmer Braren, et al. Survival in-context: Prior-fitted in-context learning tabular foundation model for survival analysis. arXiv preprint arXiv:2603.29475, 2026. doi: 10.48550/arxiv.2603.29475. URL https://doi.org/10. 48550/arxiv.2603.29475. Z. Wang, J. Sun, and A. Zhan. Survtrace: Transformers for survival analysis with competing events. In Proceedings of the ACM Conference on Health, Inference, and Learning (CHIL), pages 176–190, 2022. doi: 10.1145/3535508.3545521. URL https://arxiv.org/pdf/2110. 00855. C. N. Yu, R. Greiner, H. C. Lin, et al. Learning patient-specific cancer survival distributions as a sequence of dependent regressors. In Advances in Neural Information Processing Systems, volume 24, 2011. doi: 10.5555/2986459.2986665.

Record · ID 660838 · SHA-256 60559f2ff8fefbfd
Retrieved via Conceptio — every document is proof-bundled with source, license, and retrieval metadata.