Radiographic Phenotype-Driven Clustering in Lumbar Decompression: Comparative Study of Outcome and Reoperation Risk - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Spine J . Author manuscript; available in PMC: 2026 Apr 16. Published in final edited form as: Spine J. 2025 Apr 16;26(1):49–62. doi: 10.1016/j.spinee.2025.04.015 Search in PMC Search in PubMed View in NLM Catalog Add to search Radiographic Phenotype-Driven Clustering in Lumbar Decompression: Comparative Study of Outcome and Reoperation Risk Tomoyuki Asada Tomoyuki Asada , MD PhD 1. Hospital for Special Surgery New York, NY USA Find articles by Tomoyuki Asada 1 , Sereen Halayqeh Sereen Halayqeh , MD 1. Hospital for Special Surgery New York, NY USA Find articles by Sereen Halayqeh 1 , Adrian Lui Adrian Lui , MD 1. Hospital for Special Surgery New York, NY USA Find articles by Adrian Lui 1 , Andrea Pezzi Andrea Pezzi , MD 1. Hospital for Special Surgery New York, NY USA Find articles by Andrea Pezzi 1 , Eric Zhao Eric Zhao , BS 1. Hospital for Special Surgery New York, NY USA 2. Weill Cornell Medical College New York, NY USA Find articles by Eric Zhao 1, 2 , Adin Ehrlich Adin Ehrlich , BA 1. Hospital for Special Surgery New York, NY USA Find articles by Adin Ehrlich 1 , Olivia Tuma Olivia Tuma , BS 1. Hospital for Special Surgery New York, NY USA Find articles by Olivia Tuma 1 , Kasra Araghi Kasra Araghi , BS 1. Hospital for Special Surgery New York, NY USA Find articles by Kasra Araghi 1 , Tarek Harhash Tarek Harhash , BS 1. Hospital for Special Surgery New York, NY USA Find articles by Tarek Harhash 1 , Rujvee Patel Rujvee Patel , MBBS 1. Hospital for Special Surgery New York, NY USA Find articles by Rujvee Patel 1 , Kyle Morse Kyle Morse , MD 1. Hospital for Special Surgery New York, NY USA Find articles by Kyle Morse 1 , James E Dowdell James E Dowdell , MD 1. Hospital for Special Surgery New York, NY USA Find articles by James E Dowdell 1 , Sheeraz A Qureshi Sheeraz A Qureshi , MD MBA 1. Hospital for Special Surgery New York, NY USA Find articles by Sheeraz A Qureshi 1 , Sravisht Iyer Sravisht Iyer , MD 1. Hospital for Special Surgery New York, NY USA Find articles by Sravisht Iyer 1 Author information Article notes Copyright and License information 1. Hospital for Special Surgery New York, NY USA 2. Weill Cornell Medical College New York, NY USA ✉ Corresponding author: Sravisht Iyer: [email protected] , Hospital for Special Surgery, 535 E. 70th St, New York, NY, 10021, USA. Issue date 2026 Jan. PMC Copyright notice PMCID: PMC12353170 NIHMSID: NIHMS2075245 PMID: 40250570 The publisher's version of this article is available at Spine J Abstract Background Context: Lumbar spinal canal stenosis (LSCS) presents with various radiographic findings, often including concurrent degenerative changes. Prior studies have investigated the effects of individual radiographic findings and parameters separately using conventional methods such as logistic regression. However, applying these independent effects to real-world patients remains challenging due to an unknown interaction effect among multiple degenerative radiographic findings. Purpose: To identify distinct patient phenotypes based on preoperative radiographic findings using unsupervised clustering and to evaluate their associations with postoperative patient-reported outcomes and reoperation rates. Study Design: Retrospective cohort study Patient Sample: Patients undergoing single-level lumbar decompression Outcome Measures: Oswestry Disability Index (ODI), Short Form-12 physical component scale (SF-12 PCS), reoperation rates Methods: Unsupervised clustering was performed using preoperative radiographic data from standing X-ray imaging and magnetic resonance imaging (MRI). Variable selection was optimized through preliminary correlation analysis, causal assessment using a directed acyclic graph, and expert review. A multivariable mixed-effects model was used to assess the impact of cluster membership on postoperative outcomes. Reoperation rates were compared using Kaplan-Meier survival analysis and Cox proportional hazards models. Results: Unsupervised clustering identified four distinct clusters base on 10 radiographic variables: cluster 1 as “Young and Less Degenerative Spine” (cluster Y), cluster 2 as “Combined Coronal and Sagittal Spondylosis” (cluster CS), cluster 3 as “Coronal Spondylosis Characterized by Laterolisthesis” (cluster C), and cluster 4 as “Sagittal Spondylosis Characterized by Degenerative Spondylolisthesis” (cluster S). Multivariable regression analysis, adjusting for comorbidity, sex, and body mass index have revealed cluster C demonstrated slower improvement in ODI (β = 5.4, SE = 2.7, p = 0.043) and SF-12 PCS (β = −2.9, SE = 1.4, p = 0.045) compared to cluster Y. Regarding reoperation, cluster CS showed the highest hazard ratio (24.3%, HR = 4.18, 95% CI: 1.48–13.07, p = 0.007) compared to cluster S with the lowest reoperation rate (6.8%). Conclusion: Unsupervised clustering based on preoperative radiographic findings identified four distinct degenerative phenotypes in LSCS. Patients with coronal spondylosis was associated with slower improvements in disability and function compared to those with minimal degeneration. Additionally, patients with combined sagittal and coronal degeneration exhibited the highest reoperation rates. These findings highlight the clinical relevance of coronal and sagittal degeneration in surgical decision-making. Keywords: Lumbar decompression, Unsupervised clustering, Laterolisthesis, Spondylolisthesis, reoperation rate Introduction Lumbar spinal canal stenosis (LSCS) is a common degenerative condition frequently requiring surgical decompression. Although many patients benefit from surgery, previous studies have identified individual radiographic features—such as disc degeneration, degenerative spondylolisthesis, and sagittal malalignment through radiograph and magnetic resonance imaging (MRI)—as potential risk factors, yet they have largely been evaluated in isolation 1 – 8 . However, determining whether to perform isolated decompression or decompression with adjunct fusion remains controversial. In reality, most LSCS patients present with overlapping degenerative changes whose combined impact on outcomes remains poorly understood. Conventional multivariable analyses attempt to capture these complexities but often struggle with the “curse of dimensionality,” given the large number of radiographic variables 9 , 10 . While large national databases can offer adequate sample sizes, they typically lack the level of detail needed for comprehensive imaging assessments. Conversely, single-institution studies that include thorough radiographic evaluations may be underpowered. Consequently, the interplay among these degenerative factors is frequently underexplored, limiting our ability to optimize preoperative planning and patient counseling. Unsupervised clustering is a data-driven method that automatically organizes patients into subgroups (clusters) based on shared characteristics, without relying on predefined labels. Widely used in fields such as genomics, marketing, and medical imaging, it is well-suited to exploring intricate relationships in high-dimensional data 11 , 12 . By considering multiple radiographic variables simultaneously, this approach can uncover latent patterns that traditional one-factor-at-a-time analyses often miss. In the surgical context, it offers the potential to identify typical patient phenotypes that surgeons commonly encounter, providing insight into distinct risk profiles and enabling more personalized treatment strategies. The purpose of this study is to categorize LSCS patients undergoing single-level decompression by applying unsupervised clustering to categorize them according to multiple radiographic characteristics. We then compared postoperative outcomes and reoperation rates across the resulting clusters, aiming to identify clinically meaningful phenotypes. We hypothesize that certain patient phenotypes will be linked to worse outcomes, thereby highlighting subgroups that we may need to focus on for outcome optimization. Materials and Methods Study Design and Patient Population This is a retrospective study utilizing a prospectively – collected multi-surgeon database approved by the institutional review board (IRB number 2018–1599). Patients who had primary single-level lumbar decompression for central or lateral recess stenosis between April 2017 and August 2024 were evaluated for eligibility. Informed consent was obtained for database participation. Inclusion criteria were: (1) primary single-level bilateral lumbar decompression; (2) minimum clinical follow-up of 6 months, based on prior evidence suggesting symptomatic change plateaus within this period to maximize the data availability 13 . Exclusion criteria were: (1) acute radiculopathy solely from disc herniation, (2) revision surgeries, (3) prior lumbar fusion at any level, (4) history of vertebral fracture, and (5) absence of preoperative MRI data. Decompression procedures included hemi- or bilateral laminectomy with bilateral ligamentum flavum removal and lateral recess decompression, as indicated. Data collection and management were conducted using REDCap (Research Electronic Data Capture), hosted at the Weill Cornell Medicine Clinical and Translational Science Center. This platform is supported by the National Center for Advancing Translational Science of the National Institutes of Health under award number UL1 TR002384 14 , 15 . Demographics and Radiographic findings Demographic data, including age, sex, race, body mass index (BMI), Charlson Comorbidity Index including the age component (CCI with age), smoking status, and the presence of depressed mood, were obtained from electronic medical records. Preoperative imaging assessments included spinopelvic alignment parameters, such as lumbar lordosis (LL), pelvic tilt (PT), sacral slope (SS), pelvic incidence (PI), coronal Cobb angle (Cobb), and the calculated PI–LL mismatch. The presence and direction of spondylolisthesis (anterior, posterior, or no slip) were recorded, with a slip percentage exceeding 3% considered presence of spondylolisthesis. Laterolisthesis was defined as a lateral displacement greater than 3 mm 6 . Disc wedge angle in the coronal plane was measured by calculating the angle between the lower endplate of the upper vertebral body and the upper endplate of the lower vertebral body, with pedicles referenced when endplates were unclear 6 . Facet joint characteristics were assessed by measuring the facet angle and the presence of facet effusion 16 – 18 , which was categorized as bilateral, unilateral, or absent. Facet effusion was defined as a fluid collection exceeding 2 mm in width with a signal intensity similar to cerebrospinal fluid 19 , 20 . Facet joint was calculated from the average of left and right side. The Pfirrmann grade at the surgical index level was determined on midsagittal MRI. Additional MRI-based assessments included the dural sac cross-sectional area (DSCSA) at the index level and the presence of foraminal stenosis 21 . The cross-sectional area of the psoas muscle at the L3/4 disc level was normalized to BMI (NTPA) to account for body size differences. Radiographic measurements, including spinopelvic parameters, spondylolisthesis, laterolisthesis, and disc wedge angle, were obtained from standing radiographs or EOS, while facet effusion, facet angles, Pfirrmann grade, DSCSA, and NTPA were assessed using preoperative MRI. All measurements were manually performed by board-certified spine surgeons or radiologists following standardized protocols to ensure consistency and reliability. Clinical Outcomes Patient-reported outcomes (PROMs) included the Oswestry Disability Index (ODI) and the 12-Item Short Form Health Survey physical component summary (SF-12 PCS). Data were collected preoperatively and at 2 weeks, 6 weeks, 12 weeks, and 6 months or later postoperatively. Reoperation events related to the index level were tracked for up to 2 years postoperatively, noting both the date from the primary surgery and whether the reoperation involved fusion at the index level. Statistical Analysis All statistical analyses were performed using R (ver. 4.4.0, R Core Team (2024), Vienna, Austria). Continuous variables were reported as mean ± standard deviation (SD), while categorical variables were presented as frequency (percentage, %). Missing baseline data was considered as missing at random and was imputed using a nonparametric random forest-based imputation method that can handle both continuous and categorical variables. The imputation performance was validated using the algorithm’s out-of-bag (OOB) error estimates with a normalized root mean squared error (NRMSE) and a proportion of falsely classified (PFC) 22 . To identify and manage multicollinearity, pairwise correlations among radiographic variables were examined using Pearson correlation for continuous pairs, polyserial for continuous–categorical pairs, and polychoric for ordinal–categorical pairs. A directed acyclic graph (DAG), guided by prior studies, further informed possible causal relationships ( Supplemental Figures 1 and 2 ) 23 . Variables for clustering were selected by excluding or consolidating highly collinear features (|r| > 0.5), prioritizing parameters implicated in plausible causal pathways (based on DAG and clinical significance), and making agreement by the authors from the perspective of clinical significance. Unsupervised clustering was performed using the K-medoids algorithm with Gower’s distance, which allows the inclusion of both numeric and categorical variables by appropriately scaling dissimilarities within a single clustering step 24 , 25 . The optimal number of clusters was determined using the Silhouette method, which evaluates cluster cohesion and separation. The number of clusters was limited to a range of 2 to 6, balancing model complexity with clinical interpretability. To visualize and validate the distinctness of these clusters, a t-distributed stochastic neighbor embedding (t-SNE) plot was generated. This non-linear dimensionality reduction technique provided an intuitive graphical representation of the clustering structure, highlighting the separation between identified patient subgroups. To quantify the contribution of each radiographic variable to the clustering process, we applied a random forest classification model with the clusters identified as the dependent variable. Variable importance was assessed using Mean Decrease in Accuracy and Mean Decrease in Gini Index, both of which provide insights into how strongly each variable influences the cluster assignment. After clustering, medoid patients (representative cases closest to each cluster center) were identified to facilitate clinical interpretation. The characteristics of these medoid cases were examined to understand the typical radiographic patterns associated with each cluster. Demographic and radiographic findings were compared across the identified clusters using univariate ANOVA or chi-square tests, as appropriate. Clinical outcomes, including ODI and SF12 PCS, were assessed via multivariable mixed-effects models. These models provided coefficient (β) and standard error (SE), adjusting for CCI with age, sex, and race. Timepoints were treated as ordered factor variables to account for potential nonlinear trends over time. Statistical significance was assessed using p-values for time effects (p time ), while differences between clusters were evaluated through the interaction term (p group ) Reoperation rates were evaluated using the log-rank test and Cox proportional hazards models with hazard ratio (HR), adjusting for the same covariates (CCI with age, sex, and race). Patients lost to follow-up before the 2-year postoperative period were censored at their last available clinical visit. Due to potentially low reoperation events, Firth’s penalized likelihood method was employed in Cox proportional hazards models to reduce small-sample bias 26 . Statistical significance was set at p < 0.05. No adjustments for multiple comparisons were made, given the exploratory nature of this study. Results Patient Demographic Data in Overall Cohort A total of 357 patients who underwent single-level lumbar decompression were screened for eligibility. 54 patients with previous lumbar surgery, two with previous vertebral fracture, one with laminectomy for tumor and 12 missing preoperative MRI, leaving 288 patients included in the analysis. Preoperative SF-12 PCS were missed in 23.2% of patients and standing lumbar Xray were missed in 15.2%. The OOB evaluation yielded NRMSE of 0.414 and PFC error of 0.058, indicating that the imputation had converged and performed robustly. Mean age was 67.0 ± 13.1 years, 58.7% were male and mean BMI was 27.2 ± 4.9 kg/m2. Mean NTPA was 81.4 ± 26.0. For spinopelvic alignment, the mean of LL, PI, PT, SS, PI-LL were 44.4 ± 16.3°, 52.0 ± 14.1°, 20.7 ± 9.6°, 6.9 ± 13.2°, respectively. The mean Cobb was 9.4 ± 8.8°. The most common Pfirrmann grade at the index level was 4 presented in 124 patients (43.1%), followed by 3 in 87 (30.2%), 5 in 66 (22.9%), and 2 in 11 (3.8%). There were no patients with grade 1 or 0. Mean DSCSA was 81.4 ± 26.0 mm2. For degenerative spondylolisthesis, 107 patients (37.2%) showed anterior vertebral slip while 10 patients (3.5%) had posterior slip. There were 79 patients (27.4%) with laterolisthesis. Disc wedge angle was 2.6 ± 3.6°. Facet effusion existed bilaterally in 25 patients (8.7%), and unilaterally in 34 (12.8%). Among the patients included in the analysis, 81 patients (30.9%) did not present any of laterolisthesis, facet effusion, spondylolisthesis, degenerative scoliosis, or PI-LL over 20°, and 79 (27.4%) presented solely with one of radiographic findings. Meanwhile, 128 (44.4%) presented at least two of these degenerative diagnoses above. Unsupervised Clustering Following careful variable selection, PI-LL, NTPA, facet effusion, laterolisthesis, Pfirrmann grade, DSCSA, disc wedge angle, sagittal vertebral slip, facet joint angle, and preoperative Cobb angle were included for calculating Gower distance. Foraminal stenosis was excluded due to possible multiple association with background degenerative pathologies. The optimal number of clusters was determined as four through Silhouette method ( Fig.1a ). It indicated the mean Silhouette width value was 0.37, which suggests moderate ability for cluster separation and consolidation. Silhouette widths in each cluster were shown in Fig.1b . Subsequent t-SNE plot suggested moderate separation among clusters ( Fig.1c ). From these analyses, we have decided to use these four clusters. Figure 1. Assessments of Unsupervised Clustering. Open in a new tab a. Silhouette method for determining the optimal number of clusters. Using selected variables, the highest average silhouette width (0.37) was achieved with four clusters. b. Silhouette widths for each cluster. Clusters 1, 2, 3, and 4 are represented in orange, blue, green, and brown, respectively. The red dashed line indicates the average silhouette width of 0.37. c. t-SNE plot illustrating cluster cohesion and separation. Each point represents a case, and the white centroid marks the medoid case in each cluster. The medoid case serves as the “median” representative of the cluster. Demographic analysis among the four clusters showed that cluster 1 was the youngest, and cluster 2 and 4 were the oldest clusters ( Table 1 ). Further analysis also suggested significant differences in both sex proportion and CCI with age. The NTPA was significantly smaller in cluster 2 and 4. Sagittal alignment showed relatively higher PI and PI-LL in cluster 2. Table 1. Demographic among clusters 1 2 3 4 p-value n 136 37 42 73 Age (mean (SD)) 62.5 (14.6) 73.0 (7.9) 65.8 (12.2) 72.9 (8.7) <0.001 Male (%) 98 (72.1) 12 (32.4) 23 (54.8) 36 (49.3) <0.001 BMI (mean (SD)) 27.8 (4.6) 25.5 (6.2) 27.3 (4.7) 27.1 (4.5) 0.09 Race (%) 0.52 White 110 (80.9) 33 (89.2) 38 (90.5) 62 (84.9) Asian 7 (5.1) 0 (0.0) 0 (0.0) 4 (5.5) Black 7 (5.1) 1 (2.7) 0 (0.0) 3 (4.1) Others 12 (8.8) 3 (8.1) 4 (9.5) 4 (5.5) Smoking status (%) 44 (32.4) 16 (43.2) 15 (35.7) 26 (35.6) 0.59 CCI with Age (%) <0.001 0 22 (16.2) 1 (2.7) 3 (7.1) 1 (1.4) 1 16 (11.8) 0 (0.0) 4 (9.5) 3 (4.1) 2 29 (21.3) 6 (16.2) 10 (23.8) 10 (13.7) >3 69 (50.7) 30 (81.1) 25 (59.5) 59 (80.8) Surgical levels (%) 0.001 L1L2 1 (0.7) 0 (0.0) 1 (2.4) 0 (0.0) L2L3 11 (8.1) 0 (0.0) 1 (2.4) 1 (1.4) L3L4 20 (14.7) 8 (21.6) 13 (31.0) 7 (9.6) L4L5 84 (61.8) 29 (78.4) 25 (59.5) 62 (84.9) L5S1 20 (14.7) 0 (0.0) 2 (4.8) 3 (4.1) Open in a new tab BMI, body mass index; CCI, Charlson Comorbidity Index p < 0.05 shown in bold Radiographic findings demonstrated that Pfirrmann grade was higher in cluster 2 and 3 compared to cluster 1 (p = 0.003) ( Table 2 ). There were significant differences in vertebral slip direction, indicating that most patients in cluster 2 and 4 have anterior slip (p < 0.001). All patients in cluster 2 and 3 presented with laterolisthesis (p < 0.001), with significantly greater Cobb angle and disc wedge angles (both p < 0.05). Bilateral facet effusion was less common in cluster 3 compared to other clusters ( Fig.2a and b ). Table 2. Radiographic findings among groups 1 2 3 4 p-value NTPA (mean (SD)) 45.1 (14.2) 43.1 (22.2) 40.4 (20.6) 45.2 (13.2) <0.001 Coronal Cobb angle (mean (SD)) 6.2 (6.1) 16.0 (7.3) 16.4 (13.4) 8.2 (6.6) <0.001 LL (mean (SD)) 27.1 (10.1) 22.6 (14.4) 25.0 (12.4) 24.0 (9.7) 0.31 PI (mean (SD)) 49.2 (12.8) 58.2 (10.9) 51.1 (14.7) 54.0 (16.2) 0.006 PT (mean (SD)) 18.1 (7.1) 25.0 (7.2) 22.2 (9.3) 22.4 (13.2) <0.001 SS (mean (SD)) 31.2 (9.7) 33.2 (10.5) 28.5 (10.6) 30.1 (11.5) 0.15 PI-LL (mean (SD)) 3.9 (11.8) 11.8 (12.7) 8.8 (14.4) 8.9 (14.1) 0.003 Pfirrmann grade (%) 0.003 2 7 (5.3) 1 (2.7) 3 (7.1) 0 (0.0) 3 51 (38.3) 6 (16.2) 9 (21.4) 21 (27.6) 4 56 (42.1) 14 (37.8) 17 (40.5) 37 (48.7) 5 19 (14.3) 16 (43.2) 13 (31.0) 18 (23.7) DSCSA (mean (SD)) 83.8 (48.0) 64.1 (47.3) 75.6 (44.2) 65.3 (49.1) 0.024 Sagital slip direction (%) <0.001 anterior 0 (0.0) 37 (100.0) 0 (0.0) 70 (95.9) posterior 5 (3.7) 0 (0.0) 2 (4.8) 3 (4.1) Laterolisthesis (%) 0 (0.0) 37 (100.0) 42 (100.0) 0 (0.0) <0.001 Disc wedge angle (mean (SD)) 1.1 (2.5) 5.0 (4.3) 4.8 (4.1) 2.8 (3.5) <0.001 Facet effusion (%) 0.006 0 112 (82.4) 29 (78.4) 38 (90.5) 47 (64.4) 1 11 (8.1) 4 (10.8) 4 (9.5) 18 (24.7) 2 13 (9.6) 4 (10.8) 0 (0.0) 8 (11.0) Facet angle Rt (mean (SD)) 51.3 (13.4) 57.7 (12.1) 52.5 (13.2) 53.6 (12.5) 0.06 Facet angle Lt (mean (SD)) 49.7 (13.2) 56.2 (13.9) 49.7 (11.7) 51.7 (11.5) 0.043 Open in a new tab NTPA, BMI-normalized total psoas area; LL, lumbar lordosis; PI, pelvic incidence; PT, pelvic tilt; SS, sacral slope; PI-LL, pelvic incidence minus lumbar lordosis; DSCSA, dural sac cross sectional area; p < 0.05 shown in bold Figure 2. Characteristics of clusters and contribution of each variable. Open in a new tab Characteristics of each cluster were visualized. Based on these findings, clusters 1 to 4 were identified as Young and less degenerative (cluster Y), combined sagittal and coronal spondylosis (cluster CS), coronal spondylosis (cluster C), and sagittal spondylosis (cluster S), respectively. a. Chord plot illustrating the distribution of categorical variables across clusters. Connections represent the relationships between categorical variables such as laterolisthesis, facet effusion, sagittal slip, and Pfirrmann grade with each cluster. Details were shown in tables 1 and 2 . b. Radar plot displaying the distribution of continuous variables across the four clusters. Variables include mean value of NTPA, DSCSA, Cobb angle, facet angle, PI-LL, and disc wedge angle. Each line represents a cluster, highlighting the distinct profiles of each group. Details were shown in tables 1 and 2 . c. Contribution of variables to clustering, determined by a random forest model. The importance of each variable is presented using Accuracy and Gini scores, with sagittal slip and laterolisthesis showing the highest contributions. Random forest classification model revealed that sagittal slip and laterolisthesis are the two most contributing factors to clustering process ( Fig.2c ). From these, we identified cluster 1 as “Young and Less Degenerative Spine” (cluster Y), cluster 2 as “Combined Coronal and Sagittal Spondylosis” (cluster CS), cluster 3 as “Coronal Spondylosis Characterized with Laterolisthesis” (cluster C), and cluster 4 as “Sagittal Spondylosis Characterized with Degenerative Spondylolisthesis” (cluster S). Medoid cases, meaning representing cases from each cluster, were presented in figure 3 , 4 , 5 , and 6 , respectively. Figure 3. Medoid Case in Cluster Y (Cluster 1). Open in a new tab A 75-year-old male presenting with bilateral leg pain and intermittent claudication. The Charlson Comorbidity Index with the age component was 3 preoperatively. The following measurements were recorded: lumbar lordosis = 43, pelvic incidence = 49, pelvic tilt = 20, sacral slope = 30, pelvic incidence – lumbar lordosis mismatch = 6, Cobb angle = 5, and normalized total psoas area = 78.7. The disc wedge angle was 0, with no evidence of laterolisthesis or sagittal vertebral slip. The white lines showed the posterior and lateral edge of endplates at L4 and L5. A laminectomy was performed at the L4/5 level. Figure 4. Medoid Case in Cluster CS (Cluster 2). Open in a new tab A 82-year-old male. The Charlson Comorbidity Index with the age component was 4 preoperatively. The following measurements were recorded: lumbar lordosis = 31, pelvic incidence = 48, pelvic tilt = 26, sacral slope = 22, pelvic incidence – lumbar lordosis mismatch = 17, Cobb angle = 16.5, and normalized total psoas area = 84.8. The disc wedge angle was 1.6, with evidence of both laterolisthesis and sagittal vertebral slip at L3/4. The white lines showed the posterior and lateral edge of endplates at L3 and L4. A laminectomy was performed at the L3/4 level. Figure 5. Medoid Case in Cluster C (Cluster 3). Open in a new tab A 62-year-old male. The Charlson Comorbidity Index with the age component was 2 preoperatively. The following measurements were recorded: lumbar lordosis = 40, pelvic incidence = 48, pelvic tilt = 21, sacral slope = 27, pelvic incidence – lumbar lordosis mismatch = 6, Cobb angle = 11, and normalized total psoas area = 94.5. The disc wedge angle was 5.5, with evidence of laterolisthesis at L4/5. The white lines showed the posterior and lateral edge of endplates at L4 and L5. A laminectomy was performed at the L4/5 level. Figure 6. Medoid Case in Cluster S (Cluster 4). Open in a new tab A 68-year-old male. The Charlson Comorbidity Index with the age component was 2 preoperatively. The following measurements were recorded: lumbar lordosis = 48, pelvic incidence = 51, pelvic tilt = 29, sacral slope = 22, pelvic incidence – lumbar lordosis mismatch = 3, Cobb angle = 6, and normalized total psoas area = 79.7. The disc wedge angle was 0, with evidence of anterior vertebral slip at L4/5. The white lines showed the posterior and lateral edge of endplates at L4 and L5. A laminectomy was performed at the L4/5 level. PROMs Analysis A linear mixed-effects model adjusting for CCI with age, BMI, and sex was fitted with a random intercept for each patient to account for repeated measures. No significant differences in baseline ODI values were observed among groups ( Table 3 ). The model showed a significant linear improvement over time (ß = −17.7, SE 1.3, p time < 0.001). In group comparisons, a significant difference was observed between group Y and C, indicating the group C with slower improvement (β = 5.4, SE = 2.7, p group = 0.043), while no other significant differences were noted (all p group > 0.05). ( Fig. 7a ) The random intercept variance at the patient level was 124.6 ± 11.2 and residual was 185.0 ± 13.6, indicating moderate to strong between-subject variability. This means that the random effect effectively captures a significant and meaningful proportion of the between-patient variability. Table 3. Linear mixed-effects model for ODI comparison across clusters Coefficient (ß) SE statistic p-value Preoperative ODI Cluster CS 1.0 2.7 0.4 0.71 Cluster C 0.3 2.4 0.1 0.89 Cluster S −2.0 2.1 −0.9 0.34 Time −17.7 1.3 −13.3 <0.001* Cluster CS: Time 1.3 2.9 0.4 0.66 Cluster C: Time 5.4 2.7 2.0 0.043* Cluster S: Time 1.1 2.2 0.5 0.63 Open in a new tab SE, standard error. Reference level, cluster Y. P < 0.05 shown in bold. Figure 7. Temporal Changes in clinical Outcomes Across Clusters. Open in a new tab a. Oswestry Disability Index (ODI) scores over time across clusters. The y-axis represents ODI scores, with lower values indicating better function. Cluster C showed significantly slower improvement compared to cluster Y (ß = 5.4, SE = 2.7, p = 0.043). Error bars represent standard error. b. Short Form-12 Physical Component Summary (SF-12 PCS) scores over time across clusters. Cluster C showed significantly slower improvement compared to cluster Y (ß = −2.9, SE = 1.4, p = 0.045). Error bars represent standard error. Clusters (Y, CS, C, and S) are color-coded, showing distinct recovery trajectories for functional outcomes following surgery. Regarding SF-12 PCS, baseline values did not significantly differ among groups ( Table 4 ). Similar to ODI, the model showed a significant linear improvement over time (ß = 8.9, SE = 0.7, p time < 0.001). In group comparisons, significantly slower improvements were observed in cluster C compared to cluster Y (β = −2.9, SE = 1.4, p group = 0.045), while no significant differences were found among the other clusters (all p group > 0.05). ( Fig. 7b ) The random effects variance at the patient level was 28.0 ± 5.3 and the residual was 49.2 ± 7.0, indicating moderate between-subject variability. Table 4. Linear mixed-effects model for SF-12 PCS comparison across clusters Coefficient (ß) SE statistic p-value Preoperative ODI Cluster CS 1.8 1.4 1.3 0.19 Cluster C −0.4 1.2 −0.4 0.72 Cluster S 0.3 1.0 0.3 0.78 Time 8.9 0.7 12.2 <0.001* Cluster CS: Time 0.1 1.6 0.0 0.96 Cluster C: Time −2.9 1.4 −2.0 0.045* Cluster S: Time −1.0 1.2 −0.8 0.40 Open in a new tab SE, standard error. Reference level, cluster Y. P < 0.05 shown in bold. Survival Analysis for Reoperation Survival analysis was conducted to evaluate differences in all reoperation rates among the four clusters. Kaplan-Meier survival curves were generated for each cluster, and significant differences in reoperation-free survival were observed (p = 0.038, log-rank test) ( Fig 8 ). Any reoperation at the index level was performed in 20 (14.7%) for cluster Y, 5 (6.8%) for cluster S, 6 (14.3%) for cluster C, and 9 (24.3%) for cluster CS. In a Cox proportional hazards model adjusting for CCI with age, BMI, and sex, cluster Y showed a significantly higher hazard of reoperation compared with cluster S (HR = 2.71, 95% CI: 1.04–8.21, p = 0.042), and cluster CS demonstrated an even higher hazard (HR = 4.18, 95% CI: 1.48–13.07, p = 0.007). Although cluster C had an elevated hazard relative to cluster S (HR = 2.63, 95% CI: 0.81–8.84, P = 0.10), the difference did not reach statistical significance. Revision with fusion surgery was observed in 10 (7.3%) for cluster Y, 4 (5.5%) for cluster S, 5 (11.9%) for cluster C, and 5 (13.5%) for cluster CS (p = 0.40, log-rank test), indicating no significant differences among groups. Figure 8. Kaplan Meier Survival Curve Across Clusters. Open in a new tab Kaplan-Meier survival curves illustrating differences in survival probability among clusters (S, Y, CS, and C) over time. The log-rank test indicates a statistically significant difference between groups (p = 0.038). The adjusted hazard ratios (HR) for each cluster, with Cluster S as the reference, are as follows: Cluster Y: HR = 2.71 (95% CI: 1.04–8.21, p = 0.042); Cluster CS: HR = 4.18 (95% CI: 1.48–13.07, p = 0.007); Cluster C: HR = 2.63 (95% CI: 0.81–8.84, p = 0.10). Survival probabilities decrease at different rates across clusters, with Cluster CS demonstrating the highest adjusted hazard ratio. Cross marks indicate censored data points. Discussion This study identified four distinct clusters of patients with LSCS undergoing single-level lumbar decompression using unsupervised clustering based on preoperative radiographic characteristics. This approach yielded clinically relevant degenerative phenotypes, each exhibiting unique profiles of sagittal slip, laterolisthesis, and spinopelvic parameters, and facet changes. Notably, we labeled these clusters as: cluster Y (“Young and Less Degenerative Spine”), cluster S (“Sagittal Spondylosis Characterized with Degenerative Spondylolisthesis”), cluster C (“Coronal Spondylosis Characterized with Laterolisthesis”), and cluster CS (“Combined Coronal and Sagittal Spondylosis”). Compared with cluster Y, cluster C exhibited slower ODI and SF-12 PCS. Meanwhile, cluster S had the lowest reoperation rate, and cluster CS showed the highest hazard (HR 4.3) for reoperation over 2-year period. These findings underscore the potential value of radiographic phenotyping to refine surgical decision-making and provide valuable insights into postoperative prognosis in patients with LSCS. Detecting patient-specific risk factors is a long-standing goal in spine surgery. Conventional methods, such as logistic regression or even randomized controlled trials (RCTs), typically focus on a limited set of variables in isolation. However, degenerative spine conditions frequently involve overlapping radiographic factors, including spondylolisthesis, scoliosis, and sagittal malalignment, making real-world presentations far more complex. Indeed, in our cohort of single-level lumbar decompression patients, 44% exhibited at least two of these degenerative markers, highlighting the limitations of a unidimensional analytical approach. Unsupervised clustering excels at revealing hidden patterns within high-dimensional data without imposing strict priori assumptions 27 . While this method has been applied to characterize spinal curve types in spinal deformity 28 – 32 , to our knowledge, this is the first study to leverage this approach to phenotype patients with LSCS across multiple radiographic parameters, including both categorical and continuous measurements. We selected the K-medoids algorithm with Gower’s distance for its ability to handle mixed data types and outliers, making it particularly suitable for complex radiographic datasets. A key advantage of this method is the identification of ‘medoid cases,’ which serve as representative examples of each cluster ( Fig. 3 to 6 ), allowing clinicians to recognize these phenotypes in real-world practice. Prior studies in oncology and neurology have successfully used K-medoids clustering to refine disease subtypes and improve treatment stratification, demonstrating its utility in heterogeneous patient populations 24 , 25 . Our findings suggest that a similar approach in degenerative LSCS could enhance patient stratification based on radiographic features commonly recognized by surgeons, ultimately supporting more personalized surgical decision-making. Cluster C, defined by coronal imbalance (higher Cobb angle, greater disc wedging, and laterolisthesis) without substantial sagittal malalignment, had a slower postoperative ODI and SF-12 PCS improvement compared cluster Y when controlling for age and comorbidities as a covariate. Although the difference was statistically significant (p = 0.043), this borderline value should be interpreted with caution. Since the comparison was part of our main hypothesis tested via a linear mixed-effects model, we did not apply multiple comparison correction. Further validation is warranted. While prior meta-analyses have concluded that mild coronal deformities do not necessarily impair outcomes 33 , 34 , more severe coronal Cobb angle or surgical levels can alter the mechanical load on the spine, potentially worsening symptomatology and delaying recovery 6 , 35 , 36 . However, no prospective studies have specifically investigated coronal degeneration in a manner comparable to those on degenerative spondylolisthesis. For instance, the SLIP trial did not include any criteria for coronal degeneration, whereas the NORDSTEN Study excluded patients with Cobb angle over 20 degrees 37 , 38 . Thus, while no definitive causal relationship has been established, our findings indicate that coronal degenerative change may be associated with poorer postoperative outcomes following isolated decompression. Previous studies suggest that short-segment fusion achieves better outcomes than isolated decompression in patients with severe coronal curves 39 . Building on these findings, future research should focus on better identifying this cohort and evaluating whether fusion surgery improves their outcomes through larger prospective or well-matched cohort studies. Cluster S presented similar PROMs to cluster Y but the lowest hazard of reoperation in this cohort. Degenerative spondylolisthesis is one of the most extensively studied degenerative conditions in spine surgery, with numerous retrospective studies. Well-designed RCTs and meta-analysis aimed to clarify the surgical indications in these patients 3 – 5 , 7 , 37 , 40 – 44 . While overall clinical outcomes are comparable, some studies have reported differences in reoperation rates, with a tendency to perform fusion in cases exhibiting intervertebral instability. Given that degenerative spondylolisthesis is carefully evaluated for instability before surgery, this may explain why cluster S had a lower reoperation hazard than cluster Y. This finding may reflect the impact of prior research, leading to more effective contemporary surgical indications. Cluster CS presented mixed degenerative patterns, encompassing both sagittal and coronal changes, and had the highest hazard ratio for reoperation compared to the cluster S with the lowest reoperation rate. The cluster CS was characterized with both sagittal and coronal vertebral slippage, along with higher proportion of female patients and relatively greater PI and PI-LL. Bilateral facet effusion was more common in cluster CS than in cluster C, which may indicate greater sagittal instability, along with the presence of laterolisthesis 16 . Although this specific pattern of degeneration has not been previously investigated and defined, it was not uncommon in this cohort (37 patients, 12.8%), suggesting the clinical relevance of this patient phenotype. These characteristics resembled those of adult spinal deformity patients, suggesting a potential association with deformity pathology or a preclinical deformity state 45 . Notably, patients with degenerative scoliosis are known to compensate in static standing posture, and experience worsening sagittal malalignment while walking 46 – 48 . This compensation mechanism makes it difficult to fully assess sagittal imbalance using standing radiographs alone, potentially leading to an underestimation of sagittal malalignment in cluster CS patients. This limitation in assessment may partly explain the higher reoperation hazard in cluster CS. Additionally, female sex and facet sagittalization have been reported as risk factors for slip progression in natural history studies 49 . Cluster CS had a facet joint angle closest to 58 degrees, which has been suggested as a critical threshold for degenerative spondylolisthesis 17 . These characteristics suggest an increased risk of progressive listhesis, potentially necessitating additional surgical intervention 18 . Future research should focus on identifying patients in Cluster C and CS in real-world clinical practice and determining whether fusion surgery could improve outcomes for these subgroups while considering cost-effectiveness 50 . This radiographic pattern may not yet have a universally accepted definition, but its recognition could help surgeons identify patients who are at higher risk of poor outcomes following decompression alone. Unsupervised clustering may serve as a reproducible method to detect such complex phenotypes in a clinically meaningful way. This study has several limitations. First, its retrospective design may introduce selection bias, despite efforts to systematically collect multi-surgeon data from a single institution. Although the inclusion of multi-surgeon data reflects current clinical practice, this variability may limit the generalizability of the findings. Additionally, while we presented demographic data of the overall cohort to illustrate the range of radiographic and clinical parameters in this cohort, the results remain confined to one institution’s patient demographics. As such, caution is warranted in extending these observations to other settings. Second, clustering was based solely on preoperative radiographic characteristics, without incorporating other relevant clinical variables such as symptom duration, physical activity levels, or specific surgical techniques. To mitigate this limitation, a multivariable mixed-effects model with patient-level random effects was employed. Third, while the sample size was sufficient to detect major clustering trends, it may still be underpowered to identify subtle inter-group differences in reoperation rates and long-term functional outcomes, raising the risk of type II error. Fourth, we did not include dynamic radiographs. Patients with severe pain may not reliably demonstrate segmental instability on flexion–extension views, which could affect cluster assignments. Thus, we employed facet effusion in clustering variables as an alternative finding of instability. Finally, our findings apply to patients undergoing decompression alone, and the applicability of these clusters to those receiving concomitant fusion or multilevel procedures remains uncertain. Conclusion Our unsupervised clustering analysis identified distinct degenerative phenotypes in single-level lumbar stenosis, each with unique postoperative risk profiles. Patients with coronal degenerative spondylosis, characterized by laterolisthesis and coronal curvature, were associated with a slower recovery trajectory in disability and physical function. Those with both coronal and sagittal structural degeneration appeared to have a relatively higher risk of adverse outcomes. These findings underscore the clinical relevance of sagittal and coronal degenerative changes in surgical planning and patient counseling. While our results suggest that patients with greater coronal degeneration or combined degenerative patterns may benefit from strategies beyond decompression alone, further prospective studies with larger cohorts and advanced imaging are needed to refine methods for identifying these patient subgroups and to determine whether such strategies truly improve outcomes. Supplementary Material 1 Supplemental figure 1. Directed Acyclic Graph (DAG) for Spine Surgery Variables Causal DAG illustrating hypothesized relationships among spine surgery-related variables. Nodes represent key clinical and radiographic parameters, while directed edges indicate assumed causal pathways. Supplemental figure 2. Correlation Matrix of Spine Surgery Variables Heatmap depicting the correlation coefficients between various spine surgery-related variables. The scale bar represents the strength and direction of correlations, ranging from −1 (strong negative correlation, deep blue) to +1 (strong positive correlation, deep red). NIHMS2075245-supplement-1.docx (334.2KB, docx) Acknowledgement Generative AI (ChatGPT-4o or o1) was used solely to enhance the language and readability of this manuscript. After AI-assisted editing, the authors, including native English speakers, carefully reviewed and ensured the accuracy and integrity of the content. Source of Funding: No direct funding was received for this study. However, this study used REDCap (Research Electronic Data Capture) hosted at Weill Cornell Medicine Clinical and Translational Science Center supported by the National Center For Advancing Translational Science of the National Institute of Health under award number: UL1 TR002384. Conflicts of Interest: Sheeraz Qureshi has the following disclosures: Tissue Differentiation Intelligence: Ownership/Equity/Investment; Stryker K2M: Royalties from Intellectual Property, Designer, Consultant; SpineGuard, Inc.: Consultant; Globus Medical, Inc.: Royalties from Intellectual Property, Speakers’ Bureau, Consultant; Simplify Medical, Inc.: Clinical Event Committee; AMOpportunities: Honoraria; Surgalign: Consultant; Viseon, Inc.: Research Support (either personally or through HSS), Consultant; HS2, LLC: Ownership/Equity/Investment; LifeLink.com Inc.: Medical or Scientific Advisory Board Membership; Spinal Simplicity, LLC: Medical or Scientific Advisory Board Membership; Contemporary Spine Surgery: Editorial Board; North American Spine Society (NASS): Political Engagement Committee member, Payor Policy Review Committee member, SpinePAC Advisory Committee member, CME Committee member; Annals of Translational Medicine (ATM): Editorial Board; Hospital Special Surgery Journal: Editorial Board, Senior Associate Editor (2021–2024); Society of Minimally Invasive Spine Surgery (SMISS): Program Committee member, 2018 Annual Meeting Program Chair, Board of Directors (2021–2024), Professional Society member of Directors/Trustees/Governors/Managers, Member at Large or Committee member; Lumbar Spine Research Society (LSRS): Website Committee member (2022–2022), Professional Society member of Directors/Trustees/Governors/Managers, Member at Large or Committee member; Cervical Spine Research Society (CSRS): Publications Committee member (2019–2022), Professional Society member; Minimally Invasive Spine Study Group: Board of Directors - Treasurer; Association of Bone and Joint Surgeons (ABJS): Program Committee member, Professional Society member; International Society for the Advancement of Spine Surgery (ISASS): Education Committee, Program Committee, 2021 Annual Meeting Program Chair, Professional Society member. Sravisht Iyer has the following disclosures: Globus Medical: Paid presenter or speaker; Stryker: Paid presenter or speaker; Vertebral Columns/International Society for the Advancement of Spine Surgery (ISASS): Editorial or governing board Footnotes Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. References 1. Pearson AM, Lurie JD, Blood EA, et al. Spine patient outcomes research trial: radiographic predictors of clinical outcomes after operative or nonoperative treatment of degenerative spondylolisthesis. Spine (Phila Pa 1976). 2008;33:2759–66. doi: 10.1097/BRS.0b013e31818e2d8b [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 2. Chan AK, Bisson EF, Bydon M, et al. A comparison of minimally invasive transforaminal lumbar interbody fusion and decompression alone for degenerative lumbar spondylolisthesis. Neurosurg Focus. 2019;46:E13. doi: 10.3171/2019.2.Focus18722 [ DOI ] [ PubMed ] [ Google Scholar ] 3. Austevoll IM, Gjestad R, Solberg T, et al. Comparative Effectiveness of Microdecompression Alone vs Decompression Plus Instrumented Fusion in Lumbar Degenerative Spondylolisthesis. JAMA Netw Open. 2020;3:e2015015. doi: 10.1001/jamanetworkopen.2020.15015 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 4. Kaiser R, Kantorova L, Langaufova A, et al. Decompression alone versus decompression with instrumented fusion in the treatment of lumbar degenerative spondylolisthesis: a systematic review and meta-analysis of randomised trials. J Neurol Neurosurg Psychiatry. 2023;94:657–666. doi: 10.1136/jnnp-2022-330158 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Alimi M, Hofstetter CP, Pyo SY, Paulo D, Härtl R. Minimally invasive laminectomy for lumbar spinal stenosis in patients with and without preoperative spondylolisthesis: clinical outcome and reoperation rates. J Neurosurg Spine. 2015;22:339–52. doi: 10.3171/2014.11.Spine13597 [ DOI ] [ PubMed ] [ Google Scholar ] 6. Kato M, Namikawa T, Matsumura A, Konishi S, Nakamura H. Radiographic Risk Factors of Reoperation Following Minimally Invasive Decompression for Lumbar Canal Stenosis Associated With Degenerative Scoliosis and Spondylolisthesis. Global Spine J. 2017;7:498–505. doi: 10.1177/2192568217699192 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 7. Chan AK, Bydon M, Bisson EF, et al. Minimally invasive versus open transforaminal lumbar interbody fusion for grade I lumbar spondylolisthesis: 5-year follow-up from the prospective multicenter Quality Outcomes Database registry. Neurosurg Focus. 2023;54:E2. doi: 10.3171/2022.10.FOCUS22602 [ DOI ] [ PubMed ] [ Google Scholar ] 8. Ulrich NH, Burgstaller JM, Valeri F, et al. Incidence of Revision Surgery After Decompression With vs Without Fusion Among Patients With Degenerative Lumbar Spinal Stenosis. JAMA Netw Open. 2022;5:e2223803. doi: 10.1001/jamanetworkopen.2022.23803 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 9. Salsburg D Hundreds of Patients, Thousands of Observations: The Curse of Dimensionality in Clinical Research. Drug Information Journal. 1993;27:597–609. doi: 10.1177/009286159302700303 [ DOI ] [ Google Scholar ] 10. Barbour DL. Precision medicine and the cursed dimensions. NPJ Digit Med. 2019;2:4. doi: 10.1038/s41746-019-0081-5 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 11. Vishnuvarthanan G, Rajasekaran MP, Subbaraj P, Vishnuvarthanan A. An unsupervised learning method with a clustering approach for tumor identification and tissue segmentation in magnetic resonance brain images. Applied Soft Computing. 2016;38:190–212. doi: 10.1016/j.asoc.2015.09.016 [ DOI ] [ Google Scholar ] 12. Lopez C, Tucker S, Salameh T, Tucker C. An unsupervised machine learning method for discovering patient clusters based on genetic signatures. J Biomed Inform. 2018;85:30–39. doi: 10.1016/j.jbi.2018.07.004 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 13. Shahi P, Subramanian T, Tuma O, et al. Temporal Trends of Improvement After Minimally Invasive Transforaminal Lumbar Interbody Fusion. Spine (Phila Pa 1976). 2025;50:81–87. doi: 10.1097/brs.0000000000005024 [ DOI ] [ PubMed ] [ Google Scholar ] 14. Harris PA, Taylor R, Minor BL, et al. The REDCap consortium: Building an international community of software platform partners. J Biomed Inform. 2019;95:103208. doi: 10.1016/j.jbi.2019.103208 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 15. Harris PA, Taylor R, Thielke R, Payne J, Gonzalez N, Conde JG. Research electronic data capture (REDCap)--a metadata-driven methodology and workflow process for providing translational research informatics support. J Biomed Inform. 2009;42:377–81. doi: 10.1016/j.jbi.2008.08.010 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Lattig F, Fekete TF, Grob D, Kleinstuck FS, Jeszenszky D, Mannion AF. Lumbar facet joint effusion in MRI: a sign of instability in degenerative spondylolisthesis? Eur Spine J. 2012;21:276–81. doi: 10.1007/s00586-011-1993-1 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 17. Samartzis D, Cheung JP, Rajasekaran S, et al. Critical Values of Facet Joint Angulation and Tropism in the Development of Lumbar Degenerative Spondylolisthesis: An International, Large-Scale Multicenter Study by the AOSpine Asia Pacific Research Collaboration Consortium. Global Spine J. 2016;6:414–21. doi: 10.1055/s-0035-1564417 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 18. Liu Z, Duan Y, Rong X, Wang B, Chen H, Liu H. Variation of facet joint orientation and tropism in lumbar degenerative spondylolisthesis and disc herniation at L4-L5: A systematic review and meta-analysis. Clin Neurol Neurosurg. 2017;161:41–47. doi: 10.1016/j.clineuro.2017.08.005 [ DOI ] [ PubMed ] [ Google Scholar ] 19. Li C, Liu W, Luo W, Zhang H, Zhao J, Gu R. Lumbar Facet Joint Effusion on Magnetic Resonance Imaging: Do Different Joint Effusion Images Have Different Clinical Values? World Neurosurgery. 2022;167:e406–e412. doi: 10.1016/j.wneu.2022.08.029 [ DOI ] [ PubMed ] [ Google Scholar ] 20. Yamada K, Toyoda H, Takahashi S, et al. Relationship between facet joint opening on CT and facet joint effusion on MRI in patients with lumbar spinal stenosis: analysis of a less invasive decompression procedure. J Neurosurg Spine. 2022;36:376–384. doi: 10.3171/2021.6.SPINE21721 [ DOI ] [ PubMed ] [ Google Scholar ] 21. Urakawa H, Sato K, Vaishnav AS, et al. Preoperative cross-sectional area of psoas muscle correlates with short-term functional outcomes after posterior lumbar surgery. Eur Spine J. 2023;32:2326–2335. doi: 10.1007/s00586-023-07533-7 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Stekhoven DJ, Buhlmann P. MissForest--non-parametric missing value imputation for mixed-type data. Bioinformatics. 2012;28:112–8. doi: 10.1093/bioinformatics/btr597 [ DOI ] [ PubMed ] [ Google Scholar ] 23. Rueegg CS. Directed acyclic graphs to foster transparency and scientific dialogue. The Lancet Oncology. 2024;25:693–694. doi: 10.1016/S1470-2045(24)00271-7 [ DOI ] [ PubMed ] [ Google Scholar ] 24. Mitelpunkt A, Galili T, Kozlovski T, et al. Novel Alzheimer’s disease subtypes identified using a data and knowledge driven strategy. Sci Rep. 2020;10:1327. doi: 10.1038/s41598-020-57785-2 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 25. White K, Connor K, Meylan M, et al. Identification, validation and biological characterisation of novel glioblastoma tumour microenvironment subtypes: implications for precision immunotherapy. Ann Oncol. 2023;34:300–314. doi: 10.1016/j.annonc.2022.11.008 [ DOI ] [ PubMed ] [ Google Scholar ] 26. Nagashima K, Sato Y. Information criteria for Firth’s penalized partial likelihood approach in Cox regression models. Statistics in Medicine. 2017;36:3422–3436. doi: 10.1002/sim.7368 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 27. Ceccato A, Forne C, Bos LD, et al. Clustering COVID-19 ARDS patients through the first days of ICU admission. An analysis of the CIBERESUCICOVID Cohort. Crit Care. 2024;28:91. doi: 10.1186/s13054-024-04876-5 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 28. Lafage R, Passias P, Sheikh Alshabab B, et al. Patterns of Lumbar Spine Malalignment Leading to Revision Surgery for Proximal Junctional Kyphosis: A Cluster Analysis of Over-Versus Under-Correction. Global Spine J. 2023;13:1737–1744. doi: 10.1177/21925682211047461 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 29. Lafage R, Fourman MS, Smith JS, et al. Can unsupervised cluster analysis identify patterns of complex adult spinal deformity with distinct perioperative outcomes? J Neurosurg Spine. 2023;38:547–557. doi: 10.3171/2023.1.SPINE221095 [ DOI ] [ PubMed ] [ Google Scholar ] 30. Jin L, Jiang C, Gu L, et al. Predictive Classification System for Low Back Pain Based on Unsupervised Clustering. Global Spine J. 2023;13:630–635. doi: 10.1177/21925682211001813 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 31. Durand WM, Lafage R, Hamilton DK, et al. Artificial intelligence clustering of adult spinal deformity sagittal plane morphology predicts surgical characteristics, alignment, and outcomes. Eur Spine J. 2021;30:2157–2166. doi: 10.1007/s00586-021-06799-z [ DOI ] [ PubMed ] [ Google Scholar ] 32. Kim HJ, Virk S, Elysee J, et al. The morphology of cervical deformities: a two-step cluster analysis to identify cervical deformity patterns. J Neurosurg Spine. 2020;32:353–359. doi: 10.3171/2019.9.SPINE19730 [ DOI ] [ PubMed ] [ Google Scholar ] 33. Echt M, Bakare AA, Varela JR, et al. Comparison of minimally invasive decompression alone versus minimally invasive short-segment fusion in the setting of adult degenerative lumbar scoliosis: a propensity score-matched analysis. J Neurosurg Spine. 2023;39:394–403. doi: 10.3171/2023.4.SPINE221047 [ DOI ] [ PubMed ] [ Google Scholar ] 34. Echt M, De la Garza Ramos R, Geng E, et al. Decompression Alone in the Setting of Adult Degenerative Lumbar Scoliosis and Stenosis: A Systematic Review and Meta-Analysis. Global Spine J. 2023;13:861–872. doi: 10.1177/21925682221127955 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 35. Asada T, Simon CZ, Singh N, et al. Limited Improvement with Minimally Invasive Lumbar Decompression Alone for Degenerative Scoliosis with Cobb Angle over 20 Degrees: The Impact of Decompression Location. Spine (Phila Pa 1976). 2024;doi: 10.1097/BRS.0000000000004968 [ DOI ] [ PubMed ] [ Google Scholar ] 36. Asada T, Simon CZ, Durbas A, et al. Influence of coronal lumbar Cobb angle and surgical level on short-segment lumbar surgery outcomes in degenerative scoliosis. Eur Spine J. 2025;34:773–781. doi: 10.1007/s00586-024-08599-7 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 37. Ghogawala Z, Dziura J, Butler WE, et al. Laminectomy plus Fusion versus Laminectomy Alone for Lumbar Spondylolisthesis. N Engl J Med. 2016;374:1424–34. doi: 10.1056/NEJMoa1508788 [ DOI ] [ PubMed ] [ Google Scholar ] 38. Austevoll IM, Hermansen E, Fagerland M, et al. Decompression alone versus decompression with instrumental fusion the NORDSTEN degenerative spondylolisthesis trial (NORDSTEN-DS); study protocol for a randomized controlled trial. BMC Musculoskelet Disord. 2019;20:7. doi: 10.1186/s12891-018-2384-0 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 39. Asada T, Simon CZ, Durbas A, et al. Short-segment fusion versus isolated decompression in lumbar spinal canal stenosis patients with Cobb angles over 20 degrees. Spine J. 2024;doi: 10.1016/j.spinee.2024.10.007 [ DOI ] [ PubMed ] [ Google Scholar ] 40. Kepler CK, Hilibrand AS, Sayadipour A, et al. Clinical and radiographic degenerative spondylolisthesis (CARDS) classification. Spine J. 2015;15:1804–11. doi: 10.1016/j.spinee.2014.03.045 [ DOI ] [ PubMed ] [ Google Scholar ] 41. Mummaneni PV, Bisson EF, Kerezoudis P, et al. Minimally invasive versus open fusion for Grade I degenerative lumbar spondylolisthesis: analysis of the Quality Outcomes Database. Neurosurg Focus. 2017;43:E11. doi: 10.3171/2017.5.FOCUS17188 [ DOI ] [ PubMed ] [ Google Scholar ] 42. Bovonratwet P, Samuel AM, Mok JK, et al. Minimally Invasive Lumbar Decompression Versus Minimally Invasive Transforaminal Lumbar Interbody Fusion for Treatment of Low-Grade Lumbar Degenerative Spondylolisthesis. Spine (Phila Pa 1976). 2022;47:1505–1514. doi: 10.1097/BRS.0000000000004432 [ DOI ] [ PubMed ] [ Google Scholar ] 43. Karamian BA, Levy HA, DiMaria SL, et al. Effect of Clinical and Radiographic Degenerative Spondylolisthesis Classification on Patient-reported Outcomes and Spinopelvic Parameters for Patients With Single-level L4-L5 Degenerative Spondylolisthesis After Lumbar Fusion. Clin Spine Surg. 2023;36:E345–E352. doi: 10.1097/BSD.0000000000001461 [ DOI ] [ PubMed ] [ Google Scholar ] 44. Indrekvam K, Banerud IF, Hermansen E, et al. The Norwegian degenerative spondylolisthesis and spinal stenosis (NORDSTEN) study: study overview, organization structure and study population. Eur Spine J. 2023;32:4162–4173. doi: 10.1007/s00586-023-07827-w [ DOI ] [ PubMed ] [ Google Scholar ] 45. Terran J, Schwab F, Shaffrey CI, et al. The SRS-Schwab adult spinal deformity classification: assessment and clinical correlations based on a prospective operative and nonoperative cohort. Neurosurgery. 2013;73:559–68. doi: 10.1227/NEU.0000000000000012 [ DOI ] [ PubMed ] [ Google Scholar ] 46. Asada T, Miura K, Kadone H, et al. The relationship between spinal alignment and activity of paravertebral muscle during gait in patients with adult spinal deformity: a retrospective study. BMC Musculoskeletal Disorders. 2023;24doi: 10.1186/s12891-022-06121-y [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 47. Miura K, Kadone H, Asada T, et al. The fatty degeneration of the lumbar erector spinae muscles affects dynamic spinal compensation ability during gait in adult spinal deformity. Sci Rep. 2021;11:18088. doi: 10.1038/s41598-021-97358-5 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 48. Miura K, Kadone H, Koda M, et al. Thoracic kyphosis and pelvic anteversion in patients with adult spinal deformity increase while walking: analyses of dynamic alignment change using a three-dimensional gait motion analysis system. Eur Spine J. 2020;29:840–848. doi: 10.1007/s00586-020-06312-y [ DOI ] [ PubMed ] [ Google Scholar ] 49. Enyo Y, Yoshimura N, Yamada H, Hashizume H, Yoshida M. Radiographic natural course of lumbar degenerative spondylolisthesis and its risk factors related to the progression and onset in a 15-year community-based cohort study: the Miyama study. J Orthop Sci. 2015;20:978–84. doi: 10.1007/s00776-015-0759-8 [ DOI ] [ PubMed ] [ Google Scholar ] 50. Kaidi AC, Du JY, Subramanian T, et al. Isolated decompression for degenerative spondylolisthesis is less costly than fusion even with high revision rates. The Spine Journal. 2024;24:1690–1696. doi: 10.1016/j.spinee.2024.05.004 [ DOI ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials 1 Supplemental figure 1. Directed Acyclic Graph (DAG) for Spine Surgery Variables Causal DAG illustrating hypothesized relationships among spine surgery-related variables. Nodes represent key clinical and radiographic parameters, while directed edges indicate assumed causal pathways. Supplemental figure 2. Correlation Matrix of Spine Surgery Variables Heatmap depicting the correlation coefficients between various spine surgery-related variables. The scale bar represents the strength and direction of correlations, ranging from −1 (strong negative correlation, deep blue) to +1 (strong positive correlation, deep red). NIHMS2075245-supplement-1.docx (334.2KB, docx) ACTIONS View on publisher site PDF (2.1 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top