Research on machine learning-based clinical prediction models: a bibliometric analysis.

Research on machine learning-based clinical prediction models: a bibliometric analysis - 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 Front Oncol . 2026 Apr 1;16:1786176. doi: 10.3389/fonc.2026.1786176 Search in PMC Search in PubMed View in NLM Catalog Add to search Research on machine learning-based clinical prediction models: a bibliometric analysis Qinshan Li Qinshan Li 1 West China School of Nursing, Sichuan University, Chengdu, China 2 Cancer Center, Division of Head and Neck Tumor Multimodality Treatment, West China Hospital, Sichuan University, Chengdu, China Data curation, Writing – review & editing, Writing – original draft, Formal analysis, Conceptualization, Visualization Find articles by Qinshan Li 1, 2, † , Mingli Zeng Mingli Zeng 3 Cancer Center, West China Hospital, Sichuan University, Chengdu, China 4 Cancer Center, Shangjin Nanfu Hospital, West China Hospital, Chengdu, China Writing – original draft, Data curation, Conceptualization, Visualization, Writing – review & editing, Formal analysis Find articles by Mingli Zeng 3, 4, † , Danmei Liang Danmei Liang 1 West China School of Nursing, Sichuan University, Chengdu, China Writing – review & editing Find articles by Danmei Liang 1 , Xuemei Zheng Xuemei Zheng 2 Cancer Center, Division of Head and Neck Tumor Multimodality Treatment, West China Hospital, Sichuan University, Chengdu, China Writing – review & editing Find articles by Xuemei Zheng 2 , Xiaoxia Zhang Xiaoxia Zhang 1 West China School of Nursing, Sichuan University, Chengdu, China 3 Cancer Center, West China Hospital, Sichuan University, Chengdu, China Writing – review & editing, Writing – original draft, Supervision Find articles by Xiaoxia Zhang 1, 3, * , Ruben Martin-Payo Ruben Martin-Payo 5 Facultad de Medicina y Ciencias de La Salud, Universidad de Oviedo, Oviedo, Spain 6 Precam Research Group, Instituto de Investigación Sanitaria del Principado de Asturias (ISPA), Oviedo, Spain Writing – review & editing Find articles by Ruben Martin-Payo 5, 6 Author information Article notes Copyright and License information 1 West China School of Nursing, Sichuan University, Chengdu, China 2 Cancer Center, Division of Head and Neck Tumor Multimodality Treatment, West China Hospital, Sichuan University, Chengdu, China 3 Cancer Center, West China Hospital, Sichuan University, Chengdu, China 4 Cancer Center, Shangjin Nanfu Hospital, West China Hospital, Chengdu, China 5 Facultad de Medicina y Ciencias de La Salud, Universidad de Oviedo, Oviedo, Spain 6 Precam Research Group, Instituto de Investigación Sanitaria del Principado de Asturias (ISPA), Oviedo, Spain * Correspondence: Xiaoxia Zhang, [email protected] †These authors have contributed equally to this work Roles Qinshan Li : Data curation, Writing – review & editing, Writing – original draft, Formal analysis, Conceptualization, Visualization Mingli Zeng : Writing – original draft, Data curation, Conceptualization, Visualization, Writing – review & editing, Formal analysis Danmei Liang : Writing – review & editing Xuemei Zheng : Writing – review & editing Xiaoxia Zhang : Writing – review & editing, Writing – original draft, Supervision Ruben Martin-Payo : Writing – review & editing Received 2026 Jan 12; Accepted 2026 Mar 16; Revised 2026 Feb 18; Collection date 2026. Copyright © 2026 Li, Zeng, Liang, Zheng, Zhang and Martin-Payo. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY) . The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms. PMC Copyright notice PMCID: PMC13078998  PMID: 41994643 Abstract Background Machine learning (ML) has emerged as a transformative approach for developing high-performance clinical prediction models (CPMs). By leveraging multidimensional patient data, ML enables more accurate disease risk stratification, prognostic assessment, and clinical decision-making. In recent years, research on CPMs has expanded rapidly, with nearly 250,000 publications indexed as of 2024. Despite this remarkable growth, a comprehensive bibliometric analysis of the field is currently lacking. Objective This study aimed to analyze the global research status, evolutionary trends, and thematic hotspots of machine learning-based clinical prediction models (ML-CPMs) through bibliometric and visualization techniques. Methods Publications related to ML-CPMs were retrieved from the Web of Science Core Collection and the Scopus database (up to May 9, 2025). Bibliometric analyses were performed using various tools, including R, VOSviewer, and CiteSpace, to generate annual publication trends, collaboration networks, and journal distributions, as well as co-citation, clustering, and keyword analyses. Results A total of 8,619 publications (8,000 original articles and 619 reviews) from 118 countries were identified. Since 2015, annual publications have grown exponentially ( R² = 0.9919). While China led in total publication volume, the United States maintained the highest academic influence (H-index = 105; Total Citations = 66,788). Harvard University and BMC Medical Informatics and Decision Making emerged as the most productive institution and journal, respectively. Tian J from the Chinese Academy of Sciences led in publication count, while Wynants L from KU Leuven in Belgium recorded the highest citation frequency. Key research hotspots include algorithm optimization, multimodal data integration, and model interpretability, with clinical applications primarily focused on oncology, cardiovascular diseases, and critical care medicine. Conclusion Research on ML-CPMs has experienced rapid global growth over the past decade, forming extensive international collaboration networks. However, challenges such as limited interpretability, data heterogeneity, and privacy concerns persist. Future studies should prioritize external validation, clinical applicability, and the integration of human-AI collaborative decision-making to ensure robust implementation in real-world clinical settings. Keywords: bibliometric analysis, CiteSpace, clinical prediction models, machine learning, VOSviewer 1. Introduction Clinical Prediction Models (CPMs) quantify disease risk or prognosis by integrating multidimensional patient data and serve as essential tools for supporting clinical decision-making and optimizing resource allocation, thereby improving healthcare quality and efficiency ( 1 ). In recent years, machine learning (ML), a widely adopted branch of artificial intelligence (AI), has become a key approach for developing high-performance predictive models ( 2 ). ML can automatically extract complex features and capture nonlinear relationships, demonstrating remarkable advantages in various medical domains such as image analysis, electronic health record (EHR) mining, and multi-omics data integration ( 3 – 5 ). ML-based clinical prediction models (ML-CPMs) have not only facilitated the clinical translation of predictive modeling in oncology, cardiovascular medicine, and critical care but have also shown great potential in prognostic assessment, risk stratification, and early warning applications ( 6 – 8 ). Recent estimates suggest that by 2024, the total number of publications related to CPMs will have reached approximately 250,000 ( 9 ), reflecting the substantial research volume in this field. Given the rapid growth of the literature, several bibliometric analyses have been conducted focusing on specific disease areas. For instance, research on cervical cancer has shown that risk prediction models commonly employ random forest and support vector machine algorithms, yet they often lack external validation and remain limited in interpretability ( 10 ). In the intensive care field, research hotspots have primarily centered on early warning of clinical deterioration, mortality risk prediction, and patient phenotyping, with machine learning and deep learning approaches being the dominant methodologies ( 11 ). Studies in diabetes have reported a continuous increase in publication volume, with research themes mainly focused on diabetic retinopathy and diabetic foot, predominantly utilizing machine learning and deep learning techniques ( 12 , 13 ). At a broader disciplinary scale, bibliometric research has also addressed AI and ML in healthcare as comprehensive technological domains, encompassing diverse applications such as medical imaging, clinical decision support, disease diagnosis, and risk prediction, and highlighting the expanding role of data-driven technologies in medicine ( 14 – 16 ). However, ML-driven CPMs have not been examined as a distinct research domain in the bibliometric literature. Bibliometric analysis, which applies statistical methods to quantitatively evaluate academic publications, serves as a valuable tool for revealing research trends, collaboration networks, and scholarly influence within a field ( 17 ). Therefore, this study aimed to integrate statistical and visualization techniques to characterize the research landscape, developmental trends, and emerging hotspots in the field of ML-CPMs. 2. Materials and methods 2.1. Data sources and search strategies Web of Science Core Collection (WoSCC) and Scopus were selected because they offer broad multidisciplinary coverage and provide rich citation information ( 18 ). Compared with PubMed, which is primarily designed for biomedical indexing and lacks complete citation network data, WoSCC and Scopus are more suitable for bibliometric mapping and citation analysis. On May 9, 2025, literature retrieval was conducted independently in WoSCC and Scopus. A topic-based search was performed using keyword combinations such as “machine learning,” “artificial intelligence,” “predictive models,” and “diagnostic models.” To ensure thematic relevance to the medical domain, subject-based filters were applied at the database level in both WoSCC and Scopus. The complete search strategy is provided in Supplementary Table 1 . 2.2. Literature screening and eligibility criteria Two researchers independently screened titles and abstracts, with duplicate and non-eligible records removed through manual verification according to the predefined inclusion and exclusion criteria. In cases of disagreement regarding study selection, data extraction, or quality assessment, a third researcher was consulted to resolve discrepancies through discussion and consensus. Inclusion criteria: (1) development of clinical prediction models using machine learning methods; (2) human subjects; (3) original articles or reviews; (4) written in English; and (5) published from database inception to May 9, 2025. Exclusion criteria: (1) animal or basic laboratory experiments; (2) letters, comments, replies, or corrections; (3) editorials or commentaries; (4) retracted publications; and (5) news reports. 2.3. Data processing Bibliographic records from WoSCC were exported in BibTeX format with full records and cited references, while records from Scopus were exported in CSV format with the full available metadata to improve compatibility for multi-database integration ( 19 ). The two datasets were then imported into R 4.5.0 and processed using the bibliometrix 5.0.1 package, which converts records from different sources into a common data structure. Records were subsequently cleaned, standardized, and merged into a unified dataset for downstream analyses across multiple bibliometric tools. Prior to data analysis, manual normalization was performed for keywords, author names, and country names, including synonym merging, spelling correction, and abbreviation standardization, to ensure consistency and accuracy across different software tools. For instance, “People’s Republic of China,” “Taiwan,” and “Hong Kong” were standardized as “China,” while “Northern Ireland,” “Wales,” “England,” and “Scotland” were grouped under “United Kingdom”. 2.4. Data analysis VOSviewer 1.6.20 was used to analyze and visualize international collaboration among countries, institutions, and authors, as well as journal co-occurrence and keyword co-occurrence networks. In VOSviewer, inclusion thresholds were tailored to each analytical task to maintain interpretability while preserving informative network structure. CiteSpace 6.4.R1 was applied to generate dual-map overlays of journals, keyword timeline views, burst detection maps, and co-citation networks, thereby revealing the evolutionary pathways and emerging frontiers of research themes. For CiteSpace analyses, the time span was set to 1994–2025 with one-year slices and g-index node selection. Pruning strategies were selected according to the analytical task: Pathfinder was used for cluster visualization when appropriate, whereas no pruning was applied to co-citation networks used for burst detection. In addition, Microsoft Excel 2019 and ArcMap 10.8 were employed for data processing and visualization. 3. Result 3.1. Overall analysis A total of 11,187 records were initially retrieved from WoSCC, and 13,403 records were retrieved from Scopus. After merging records from the two databases, removing duplicates, and applying the inclusion criteria, 8,619 eligible publications were retained for analysis, comprising 8,000 original research articles (92.8%) and 619 review articles (7.2%). Figure 1 illustrates the detailed flowchart of the literature screening process. Figure 1. Open in a new tab Flowchart of the literature screening process. 3.2. Annual publication trends The analysis revealed that as early as 1994, studies had attempted to apply machine learning methods to clinical prediction modeling. Woolery LK and Lowell WE independently developed neural network-based models for predicting preterm birth risk and hospitalization duration in patients with mental illness, respectively ( 20 , 21 ). These studies marked the initial exploration of machine learning in clinical risk prediction. Figure 2 presents the annual and cumulative publication trends in the ML-CPMs field from 1994 to 2025. Over the past three decades, there has been significant growth, reaching a total of 8,619 by May 2025. Polynomial trend fitting demonstrated a strong correlation between publication volume and time ( R² = 0.9919), indicating a statistically significant upward trajectory in publication growth. Figure 2. Open in a new tab Annual and cumulative number of publications. Further analysis of the annual publication trends suggests that research on ML-CPMs has evolved through four distinct developmental phases. The initial phase, from 1994 to 2006 represented the exploratory stage, with only 1–10 publications per year. The second phase (2007–2014) was characterized by slow growth, peaking in 2012 before a slight decline over the following two years. The third phase (2015–2020) marked a period of accelerated development, during which annual publications increased sharply from 81 to 685. The most recent phase (2021–2025) reflects a period of high research activity, reaching a peak of 1,965 publications in 2024 and projected to surpass 2,000 by 2025. 3.3. Analysis of countries A total of 118 countries have contributed to research on ML-CPMs. As shown in Figure 3A , China (n = 3,548), the United States (n = 2,423), and the United Kingdom (n = 661) ranked as the top three countries in publication output. Collaboration network analysis ( Figure 3B ) revealed that the United States (TLS = 2,003), the United Kingdom (TLS = 1,195), and China (TLS = 942) occupied central positions within the global cooperation network. Temporal evolution analysis reveals that countries such as the United States, the United Kingdom, Germany, the Netherlands, and Canada had earlier average publication years (around 2021), representing the core research forces in this field. In subsequent years, research activity gradually expanded to emerging regions across Asia and the Middle East, with countries such as China, India, Saudi Arabia, Iran, Turkey, Thailand, and South Africa showing more recent average publication years (after 2022), reflecting an increasing trend toward geographical diversification in global research collaboration ( Figure 3C ). Figure 3. Open in a new tab (A) Global distribution of publications by country. Darker colors indicate a higher number of publications. (B) Collaboration network among countries with ≥5 publications. Node size represents publication volume, and connecting lines denote collaborative relationships between countries. Different colors indicate distinct collaboration clusters. Total collaboration strength is quantified by total link strength, with higher values reflecting closer international cooperation. (C) Temporal evolution of international collaboration. Node colors indicate the average publication year for each country, illustrating the temporal distribution of research activity from 2021 to 2023. The publication performance of the top ten countries is summarized in Table 1 . China ranked first with 3,548 publications (41.2%), although it placed second in total citations (43,208) and H-index (77). The United States, despite a slightly lower publication count, achieved the highest total citations (66,788) and H-index (105), reflecting its greater academic influence in this field. The United Kingdom ranked third with 661 papers (7.7%) and an average of 33.86 citations per article. Notably, the Netherlands, though contributing 352 papers (4.1%), recorded the highest average citation rate (42.60), indicating the high impact of its research output. Table 1. Top 10 productive countries ranked by numbers of publications. Rank Country NP(%) NC AC TLS H-index 1 China 3,548(41.2%) 43,208 12.18 942 77 2 USA 2,423(28.1%) 66,788 27.58 2,003 105 3 United Kingdom 661(7.7%) 22,381 33.86 1,195 63 4 South Korea 448(5.2%) 6,123 13.67 233 38 5 Italy 410(4.8%) 8,741 21.32 691 44 6 Canada 387(4.5%) 9,848 25.45 643 44 7 Germany 379(4.4%) 10,145 26.77 706 46 8 Netherlands 352(4.1%) 15,037 42.60 706 57 9 Australia 263(3.1%) 7,449 28.32 461 38 10 Spain 259(3.0%) 4,724 18.24 435 38 Open in a new tab NP, Number of Publications. NC, Number of Citations. AC, Average Citations. TLS, Total Link Strength. USA, The United States of America. 3.4. Analysis of institutions A total of 9,903 institutions have contributed to publications in the ML-CPMs field. As shown in Figure 4A , Harvard University (TLS = 730), Harvard Medical School (TLS = 330), Sun Yat-sen University (TLS = 237), Capital Medical University (TLS = 237), and Fudan University (TLS = 185) hold central positions in the institutional collaboration network. This indicates a high level of research connectivity and cooperative activity. Temporally, as depicted in Figure 4B , institutions such as Harvard University, Stanford University, and the University of Toronto were prominent in research before 2021. In contrast, after 2022, Chinese universities including Fudan University, Shanghai Jiao Tong University, and Sun Yat-sen University significantly increased their publication output, reflecting the growing research momentum in Asia. Figure 4. Open in a new tab (A) Collaboration network of the top 100 research institutions, each contributing at least 19 publications. Each node represents a research institution, with node size indicating publication volume and line thickness reflecting collaboration strength. Different colors denote clusters of closely collaborating institutions. (B) Temporal evolution of institutional collaboration. Node colors indicate the average publication year for each institution, illustrating the temporal distribution of research activity from 2021 to 2023. The top ten most productive institutions are listed in Table 2 . Harvard University ranked first, with the largest number of publications (n = 312) and the highest H-index (49). The top four institutions were all located in the United States, which highlights the strong academic influence of American research organizations in this field. Institutions ranked fifth and below were predominantly from China, including Sun Yat-sen University, Capital Medical University, and Fudan University. Although Chinese institutions have shown rapid growth in publication output, their average citation counts and H-index values remain generally lower than those of their US counterparts, indicating that there is still room to strengthen their international research impact. Table 2. Top 10 productive institutions with publications. Rank Institution Country NP NC AC H-index 1 Harvard University USA 312 12,868 41.24 49 2 Harvard University Medical Affiliates USA 249 9,800 39.36 45 3 University of California System USA 223 6,844 30.69 42 4 Harvard Medical School USA 202 8,217 40.68 39 5 Sun Yat-sen University China 196 3,614 18.44 31 6 Capital Medical University China 191 2,101 11.00 24 7 Fudan University China 190 3,145 16.55 29 8 Chinese Academy of Sciences China 189 3,948 20.89 36 9 Chinese Academy of Medical Sciences Peking Union Medical College China 184 2,433 13.22 28 10 Shanghai Jiao Tong University China 183 2,254 12.32 22 Open in a new tab NP, Number of Publications. NC, Number of Citations. AC, Average Citations. 3.5. Journal productivity and citation patterns The bibliometric analysis identified 677 journals contributing to research on ML-CPMs. The journal collaboration network formed six major clusters, reflecting the field’s diverse research directions ( Figure 5A ). These clusters are medical imaging and radiotherapy (red cluster), medical informatics and digital health (green cluster), brain imaging and neuroscience (yellow), interdisciplinary health informatics (purple cluster), biomedical engineering and computational modeling (light blue cluster), and comprehensive clinical research (dark blue cluster). Figure 5B illustrates citation relationships between journals and cocited journals. Two prominent green citation paths were apparent, showing that clinical medicine journals frequently cited nursing and public health ( f = 19,945, z = 5.06) and molecular biology and genetics ( f = 19,882, z = 5.33). Figure 5. Open in a new tab (A) Journal collaboration network. Colors denote different clusters, node size indicates publication volume, and line thickness reflects collaboration strength. (B) Dual-map overlay of journals. The left side represents citing journals, and the right side represents cited journals. Colored paths illustrate citation relationships across disciplines, where f denotes citation frequency, indicating the intensity of interdisciplinary knowledge flow, and z represents standardized significance, indicating the prominence of each citation path. Table 3 lists the top ten journals by publication volume, which together contributed 2,269 papers, accounting for 26.3% of the total output. Among them, BMC Medical Informatics and Decision Making (n = 328), Diagnostics (n = 317), and Journal of Clinical Medicine (n = 239) ranked among the top three. Six of the ten journals were classified in JCR Q1, with impact factors ranging from 2.9 to 6.8. IEEE Journal of Biomedical and Health Informatics had the highest impact factor (6.8), whereas European Radiology achieved the highest citation count (6,289 citations, H-index = 43). Table 3. Top 10 productive journals with publications. Rank Journal Country NP NC JCR Quartile IF (2025) H-index 1 BMC Med Inform Decis Mak United Kingdom 328 5,532 Q2 3.8 32 2 Diagnostics Switzerland 317 2,508 Q2 3.3 23 3 J Clin Med Switzerland 239 2,420 Q1 2.9 25 4 Front Med Switzerland 222 1,645 Q1 3 21 5 Eur Radiol Germany 213 6,289 Q1 4.7 43 6 Int J Med Inform Netherlands 209 3,966 Q1 4.1 34 7 J Med Internet Res Canada 202 3,804 Q1 6 31 8 Comput Methods Programs Biomed Ireland 185 3,651 Q1 4.8 33 9 JMIR Med Inform Canada 178 2,371 Q2 3.8 23 10 IEEE J Biomed Health Inform USA 176 2,902 Q1 6.8 28 Open in a new tab IF, Impact Factor, accessed from JCR 2025. 3.6. Author productivity and impact A total of 50,425 authors were identified across the 8,619 included publications. As shown in Figure 6A , the collaboration network of the top 100 most productive authors showed that Slomka PJ (TLS = 116), Berman DS (TLS = 105), Dey D (TLS = 105), and Tian J (TLS = 90) held central roles, indicating significant research activity and strong collaborative ties. In contrast, the author co-citation network in Figure 6B illustrated the scholarly influence of researchers through citation relationships. In this network, Collins GS (TLS = 6,287), Lambin P (TLS = 5,447), and Breiman L (TLS = 4,872) exhibited the highest co-citation strength, underscoring their substantial academic impact in the ML-CPMs domain. According to publication statistics in Table 4 , Tian J from China ranked first with 42 papers and the highest H-index of 28. Slomka PJ from the United States and Dekker A from the Netherlands followed in second and third places, respectively. Notably, although Collins GS from the United Kingdom had a relatively modest publication count, his total citations (5,955) and average citations (270.68) were significantly higher than those of other authors, reflecting the broad recognition and high influence of his scientific contributions. Figure 6. Open in a new tab (A) Author collaboration network (authors with ≥10 publications). Each node represents an author, with node size indicating publication volume and line thickness reflecting the strength of collaboration. Colors denote different collaborative clusters. (B) Author co-citation network. Each node represents a cited author, with node size indicating citation frequency and connecting lines representing co-citation relationships between authors. Different colors indicate clusters of authors frequently cited together. Table 4. Top 10 productive authors with publications. Rank Author NP NC AC H-Index Country Institution 1 Tian J 42 1,999 47.60 28 China Chinese Academy of Sciences 2 Slomka PJ 33 1,295 39.24 17 USA Cedars-Sinai Medical Center 3 Dekker A 30 1,200 40.00 17 Netherlands Maastricht University 4 Berman DS 29 1,331 45.90 17 USA Cedars-Sinai Medical Center 5 Dey D 29 1,141 39.34 16 USA Cedars-Sinai Medical Center 6 Saba L 27 880 32.59 18 Italy University of Cagliari 7 Ong MEH 24 620 25.83 13 Singapore Singapore General Hospital 8 Collins GS 22 5,955 270.68 16 United Kingdom University of Oxford 9 Miller RJH 21 401 19.1 12 Canada University of Calgary 10 Churpek MM 19 778 40.95 11 USA University of Wisconsin-Madison Open in a new tab NP, Number of Publications. NC, Number of Citations. AC, Average Citations. USA, the United States of America. 3.7. Cited and co-cited references Regarding highly cited publications, the top ten papers identified in this study were primarily published in BMJ, Nature Medicine, and NPJ Digital Medicine , encompassing the fields of clinical medicine, epidemiology, and medical informatics ( Table 5 ). Among these, the most frequently cited publication was the systematic review by Wynants L, published in BMJ in 2020, titled “Prediction models for diagnosis and prognosis of COVID-19 infection: systematic review and critical appraisal” ( 22 ). This study systematically evaluated diagnostic and prognostic prediction models related to COVID-19 and assessed their methodological quality, accumulating 2,049 citations to date. Table 5. Top 10 highly cited publications. Rank Publications title Journal First author Year NC Document Type 1 Prediction models for diagnosis and prognosis of covid-19infection: systematic review and critical appraisal BMJ Wynants L ( 22 ) 2020 2,005 Review 2 Diffuse large B-cell lymphoma outcome prediction by gene-expression profiling and supervised machine learning Nat Med Shipp MA ( 23 ) 2002 1,856 Article 3 Scalable and accurate deep learning with electronic health records NPJ Digit Med Rajkomar A ( 24 ) 2018 1,407 Article 4 Advantages and disadvantages of using artificial neural networks versus logistic regression for predicting medical outcome J Clin Epidemiol Tu JV ( 25 ) 1996 1,320 Review 5 A systematic review shows no performance benefit of machine learning over logistic regression for clinical prediction models J Clin Epidemiol Christodoulou E ( 26 ) 2019 1,119 Review 6 Calibration: the Achilles heel of predictive analytics BMC Med Van Calster B ( 27 ) 2019 980 Article 7 Comparing different supervised machine learning algorithms for disease prediction BMC Med Inform Decis Mak Uddin S ( 28 ) 2019 801 Article 8 A five-gene signature and clinical outcome in non-small-cell lung cancer N Engl J Med Chen HY ( 29 ) 2007 753 Article 9 A radiomics model from joint FDG-PET and MRI texture features for the prediction of lung metastases in soft-tissue sarcomas of the extremities Phys Med Biol Vallières M ( 30 ) 2015 704 Article 10 Machine learning in medicine: a practical introduction BMC Med Res Methodol Sidey-Gibbons JAM ( 31 ) 2019 685 Article Open in a new tab NC, Number of Citations. Based on 8,619 original publications, a total of 1,637 co-cited references were identified to construct the co-citation network ( Figure 7A ). The most frequently co-cited paper was “Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries” ( 32 ), with 345 co-citations. This study provided a comprehensive assessment of the global cancer burden using data from GLOBOCAN 2020. Other highly co-cited papers among the top ten addressed topics such as radiomics standardization, explainable artificial intelligence (XAI), methodological comparisons of prediction models, and ML algorithms. Subsequent clustering analysis of co-cited references identified ten major thematic clusters ( Figure 7B ), with Modularity Q = 0.7537 and Weighted Mean Silhouette S = 0.8648. Both values exceeding 0.75 indicated that the clustering structure was well defined and internally consistent. The clusters covered topics including clinical prediction model development methods, applications of electronic health records, and disease-specific studies such as non-small cell lung cancer, acute pancreatitis, hepatocellular carcinoma, cardiovascular disease, and COVID-19. Moreover, Figure 7C presents the top 25 publications with the strongest citation bursts. The review by Gillies RJ, published in 2016, exhibited the highest burst intensity (Strength = 72.06) and remained active until 2021. This paper introduced the concept and research framework of radiomics and provided a detailed description of the complete workflow from image acquisition and feature extraction to model construction and clinical application ( 33 ). Seven publications showed ongoing citation bursts extending through 2025, most of which were published within the past five years and focused on cancer epidemiology, model interpretability, reporting standards for prediction models, and radiomics modeling. Figure 7. Open in a new tab (A) Co-citation network of the 10 most-cited references. The g-index (k = 25) was used as the scaling factor, node size corresponds to co-citation frequency, and labels indicate the first author and publication year. (B) Clustering of cited references. Clusters are numbered based on reference keywords, smaller numbers denote larger clusters. (C) Top 25 references with the strongest citation bursts. Blue bars indicate the overall time span, and red segments mark the burst-active periods. 3.8. Keyword co-occurrence and cluster analysis A total of 20,030 keywords were extracted from the analyzed publications. Based on their semantic features and contextual relevance within ML-CPM research, high-frequency keywords were grouped into five categories: technical methods, model development, disease applications, clinical settings, and clinical outcomes. Table 6 lists the top three high-frequency keywords in each category. To further clarify the thematic landscape of ML-CPM research, we conducted a keyword co-occurrence analysis. As shown in Figure 8A , the co-occurrence network consisted of three clusters: medical imaging and oncology-related research (red cluster), AI methodologies and their application in hospital settings (green cluster), and prediction models and health outcomes (blue cluster). Overall, machine learning (TLS = 12,787) occupied a central position in the network, showing strong linkages with prediction model (TLS = 4,989), artificial intelligence (TLS = 4,934), and risk (TLS = 3,980). Table 6. High-frequency keywords and their classifications. Classification Top 3 Count TSL Technical methods machine learning 3,328 12,787 artificial intelligence 1,199 4,934 radiomics 1,133 4,940 Model development prediction model 1,247 4,989 risk 886 3,980 classification 886 3,473 Disease applications cancer 580 2,537 cardiovascular diseases 326 1,443 COVID-19 312 1,123 Clinical settings electronic health records 363 1,611 intensive care unit 214 1,078 clinical decision support 99 476 Clinical outcomes outcomes 405 1,877 mortality prediction 144 539 hospital readmission 120 538 Open in a new tab TLS, Total Link Strength. Figure 8. Open in a new tab (A) Keyword co-occurrence network. This map highlights clusters of interrelated terms, each representing a distinct research domain. (B) Timeline view of keyword clusters. Horizontal lines indicate clusters, with labels displayed on the right side of the figure. Different colors denote separate clusters, and smaller cluster numbers correspond to larger cluster sizes. The position of nodes along the timeline reflects the temporal evolution of research topics. (C) Keyword burst analysis. Red line segments indicate burst periods, representing time intervals when the corresponding keywords received heightened research attention. The keyword timeline view revealed the dynamic evolution of research themes within the ML-CPM field ( Figure 8B ), identifying ten major clusters with high structural clarity and internal consistency (Modularity Q = 0.863; Weighted Mean Silhouette S = 0.9596). Cluster #0 “breast cancer” primarily involved studies related to support vector machines and early diagnosis in breast cancer, whereas #3 “machine learning” focused on diagnostic modeling, prognostic analysis, and hospitalization risk assessment. In contrast, #7 “artificial intelligence” and #9 “prediction model” emerged primarily over the past five years, representing frontier hotspots in the field. Figure 8C summarizes the top 25 keywords with the strongest citation bursts. Early emerging hotspots included neural networks (Strength = 29.77), logistic regression (Strength = 12.33), and decision tree (Strength = 19.24), which were most active between 2000 and 2015. More recent bursts were observed for keywords such as graph neural networks, bladder cancer, traditional Chinese medicine, ultrasound imaging, and older adults, beginning in or after 2023 and indicating their growing prominence in the recent literature. 4. Discussion Since Breiman L introduced the random forest algorithm in 2001, a solid methodological foundation has been established for the development of clinical prediction models ( 34 ). Research on prediction model construction based on machine learning has since gained widespread attention, with the number of related publications increasing exponentially. Early studies were predominantly conducted in the United States, the United Kingdom, and Germany, where well-established research infrastructures and access to large-scale medical databases fostered international collaboration and academic leadership. Over the past decade, China has rapidly emerged as a major contributor, driven by increased research investment, data-sharing initiatives, and supportive AI policies. This shift has given rise to a global pattern characterized by leadership from developed countries and rapid progress among developing nations. At the institutional level, Harvard University and the University of California System in the influence over time. Meanwhile, Chinese institutions such as Sun Yat-sen University, Fudan University, and Capital Medical University have demonstrated sustained growth in publication output, reflecting a notable rise in research activity. Collectively, these developments have fueled the global expansion of ML-CPMs research, closely linked to the rise of deep learning, the increasing availability of open-access medical databases, and the policy momentum toward precision medicine. Journals, authors, and their representative publications collectively constitute the core network of knowledge dissemination and academic accumulation in this field, serving as key carriers of research productivity and scholarly influence. BMC Medical Informatics and Decision Making published the largest number of papers, consistently focusing on clinical data mining, decision support, and model validation, and has maintained high visibility within the domain of medical informatics. Diagnostics and the Journal of Clinical Medicine have also made notable contributions in radiomics, disease risk prediction, and model performance evaluation, indicating a growing trend toward the integration of ML-CPMs research with clinical practice. Among the leading researchers, Professor Tian J from the Chinese Academy of Sciences stands out as the most prolific contributor in this field, having authored 42 publications. In a 2022 article published in The Lancet Digital Health, his team demonstrated the clinical potential of deep learning for predicting chemoradiotherapy response in rectal cancer ( 35 ). In contrast, British scholar Collins GS is widely recognized for his methodological contributions, leading the development of the TRIPOD-AI reporting guideline that has greatly improved transparency and reproducibility in model development and validation ( 36 ). In parallel, several highly cited publications have further shaped the methodological foundations of ML-CPMs. Rajkomar A introduced deep learning frameworks for electronic health records ( 24 ), showcasing the power of machine learning to capture complex clinical patterns. Christodoulou E conducted a systematic review comparing the performance of machine learning and traditional statistical models ( 26 ), providing important insights into their relative advantages and limitations. Lambin P, through his pioneering research in radiomics ( 37 ), advanced the integration of imaging features into predictive modeling. Taken together, these studies have focused on algorithmic optimization, data integration, and multimodal modeling, reflecting a strong methodological and thematic alignment among leading contributors within the field. By integrating the visualized results of keyword and co-citation analyses, research on clinical prediction models has shown a clear methodological evolution from conventional statistical approaches for risk prediction to intelligent model development driven by ML algorithms. Early investigations, characterized by keywords such as “logistic regression” and “decision tree,” primarily focused on variable selection, model calibration, and individualized risk prediction, emphasizing the feasibility and internal validation of traditional statistical methods ( 27 , 38 – 40 ). With advances in computational capacity and the widespread adoption of electronic health records, keywords including “support vector machine,” “random forest,” and “data mining” appeared with increasing frequency, suggesting the increasing application of ML techniques in model development, comparison, and optimization. Notably, these terms were most prominent before 2019. Since then, research priorities have gradually shifted toward the construction and refinement of AI-driven CPMs. High-frequency keywords such as “machine learning,” “prediction model,” “artificial intelligence,” “radiomics,” and “deep learning” reflect the transition within the field toward multimodal feature integration and model performance enhancement. The combination of radiomics and deep learning represents one of the most active research frontiers, where multimodal feature extraction based on CT radiomics and 2.5D deep learning frameworks has demonstrated superior diagnostic accuracy in predicting tumor grade, recurrence, and metastasis risk ( 41 – 43 ). Across major clinical disease domains, research on ML-CPMs has primarily concentrated on conditions with high prevalence and substantial disease burden, including chronic illnesses, acute and critical conditions, and infectious diseases. Among these, oncology has become the most mature and rapidly advancing research field, with studies mainly emphasizing cancer risk prediction, recurrence assessment, and treatment response modeling. The incorporation of deep learning has markedly improved model performance by enabling the automated extraction and representation of complex imaging and multimodal features, thereby driving the transition from algorithmic development toward precision medicine and individualized clinical decision-making ( 44 – 47 ). In comparison, cardiovascular research has placed greater emphasis on population-level risk stratification and long-term outcome prediction. ML-based models that combine imaging, pathologic, and physiologic data have demonstrated strong predictive stability and high accuracy across various clinical settings ( 48 – 50 ). In the field of infectious diseases, the COVID-19 pandemic has acted as a major catalyst for the rapid development of predictive modeling. The growing availability of large-scale clinical and imaging datasets has extended the application of ML-CPMs from single-disease risk prediction to more comprehensive assessments of overall health status ( 22 ). Furthermore, the application of ML-CPMs among critically ill and older adults presents unique challenges due to disease complexity, multiple comorbidities, and rapidly changing clinical conditions. These challenges have highlighted the need for improved model robustness, real-time adaptability, and interpretability. Consequently, XAI approaches have been increasingly adopted to enhance model transparency, improve interpretability, and promote greater confidence among clinicians in AI-assisted decision-making ( 51 , 52 ). Despite considerable advances in ML-CPMs, several unresolved challenges in the current evidence base continue to hinder clinical translation. A persistent imbalance remains between rapid model development and insufficient attention to comprehensive evaluation. Although discrimination is frequently reported, calibration, clinical utility, and performance heterogeneity across patient subgroups and care settings are less consistently examined, despite established methodological guidance emphasizing their importance for clinical decision-making ( 53 , 54 ). External validation and assessment of transportability across institutions and health systems are also less common than model development, raising concerns about robustness and applicability in routine clinical practice ( 54 ). Reporting and appraisal frameworks have strengthened expectations for transparency, reproducibility, and risk-of-bias assessment, yet adherence remains variable ( 36 , 55 ). Differences in data processing and predictor modeling across studies further limit comparability and hinder cumulative evidence synthesis. While explainable AI techniques aim to enhance interpretability, their reliability and clinical usefulness across diverse settings remain uncertain ( 56 , 57 ). Addressing these limitations requires shifting research priorities from model construction alone toward rigorous external validation, transportability-oriented performance evaluation, and stronger adherence to reporting standards. Growing emphasis has also been placed on model lifecycle management, including post-deployment monitoring and systematic model updating to address distributional shift and maintain performance and reliability over time ( 58 , 59 ). The broader implementation of ML-CPMs increasingly depends on secure and privacy-preserving data infrastructures that enable cross-institutional collaboration while protecting patient confidentiality. Data sharing across institutions is frequently constrained by regulatory, legal, and organizational requirements ( 60 ). Even distributed learning approaches such as federated learning do not fully eliminate privacy risks, as sensitive information may still be inferred from model parameters or updates ( 61 ). Together, these constraints and residual privacy risks underscore the need for rigorous security design. This includes secure aggregation, differential privacy methods, and clearly defined governance frameworks to ensure accountability in data stewardship and model oversight ( 62 ). Current evidence suggests that large-scale clinical deployment of privacy-preserving machine learning remains limited relative to methodological development ( 63 , 64 ). This highlights the need to evaluate privacy–utility trade-offs and security risks in addition to predictive performance. This study presents several limitations. Our dataset was retrieved from WoSCC and Scopus, and relevant publications indexed exclusively in other databases may have been missed. Compared with prevalent bibliometric workflows that rely on a single database and a single mapping tool, our dual-database and multi-tool pipeline broadens coverage and enables cross-tool triangulation. The strengths of this approach include retrieval breadth and cross-tool validation. A key weakness is that citation-lag effects may underrepresent emerging topics, as recently published studies have had limited time to accumulate citations. Looking ahead, opportunities lie in pairing bibliometric mapping with systematic model evaluation, thereby linking identified hotspots to evidence quality and clinical relevance. Meanwhile, threats include evolving database indexing policies and software updates that may affect longitudinal comparability. Despite these limitations, this study offers a multidimensional overview of the development and research landscape of ML-CPMs over the past three decades, providing valuable insights for researchers in the field. 5. Conclusion From 1994 to 2025, global research on ML-CPMs has shown sustained growth, reflecting a clear evolution from methodological innovation toward clinical implementation. Over time, this field has expanded to encompass diverse diseases and clinical applications. The United States continues to lead in academic influence and methodological standardization, whereas China has exhibited rapid growth in publication output, underscoring regional differences in research productivity and quality. Current research hotspots primarily focus on algorithmic optimization, integration of imaging and multi-omics data, model interpretability, and clinical validation. Overall, ML has played a pivotal role in advancing the development and optimization of CPMs, marking a transition from model development to a more clinically oriented phase of integrated innovation. As the field continues to evolve, with the increasing integration of multi-source data and the advancement of intelligent decision-support systems, ML-CPMs are expected to deliver broader and more sustainable clinical value in disease prediction, prognostic assessment, and precision intervention. Funding Statement The author(s) declared that financial support was not received for this work and/or its publication. Footnotes Edited by: Sharon R Pine , University of Colorado Anschutz Medical Campus, United States Reviewed by: C V Guru Rao , Gayatri Vidya Parishad College of Engineering, India Anamika Sharma , Thapar Institute of Engineering and Technology (Deemed to be University), India Data availability statement The original contributions presented in the study are included in the article/ Supplementary Material , further inquiries can be directed to the corresponding author/s. Author contributions QL: Data curation, Writing – review & editing, Writing – original draft, Formal analysis, Conceptualization, Visualization. MZ: Writing – original draft, Data curation, Conceptualization, Visualization, Writing – review & editing, Formal analysis. DL: Writing – review & editing. XMZ: Writing – review & editing. XXZ: Writing – review & editing, Writing – original draft, Supervision. RM: Writing – review & editing. Conflict of interest The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Generative AI statement The author(s) declared that generative AI was not used in the creation of this manuscript. Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us. Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. Supplementary material The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc.2026.1786176/full#supplementary-material Table1.docx (672.7KB, docx) Table2.docx (673.5KB, docx) References 1. Van Smeden M, Reitsma JB, Riley RD, Collins GS, Moons KG. Clinical prediction models: diagnosis versus prognosis. J Clin Epidemiol. (2021) 132:142–5. doi:  10.1016/j.jclinepi.2021.01.009. PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 2. Shamout F, Zhu T, Clifton DA. Machine learning for clinical outcome prediction. IEEE Rev BioMed Eng. (2020) 14:116–26. doi:  10.1109/rbme.2020.3007816. PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 3. Chen X, Wang X, Zhang K, Fung K-M, Thai TC, Moore K, et al. Recent advances and clinical applications of deep learning in medical image analysis. Med Image Anal. (2022) 79:102444. doi:  10.1016/j.media.2022.102444. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 4. Sammut S-J, Crispin-Ortuzar M, Chin S-F, Provenzano E, Bardwell HA, Ma W, et al. Multi-omic machine learning predictor of breast cancer therapy response. Nature. (2022) 601:623–9. doi:  10.1038/s41586-021-04278-5. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Vasudevan L, Kibria MG, Kucirka LM, Shieh K, Wei M, Masoumi S, et al. Machine learning models to predict risk of maternal morbidity and mortality from electronic medical record data: Scoping review. J Med Internet Res. (2025) 27:e68225. doi:  10.2196/68225. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 6. Gautam N, Mueller J, Alqaisi O, Gandhi T, Malkawi A, Tarun T, et al. Machine learning in cardiovascular risk prediction and precision preventive approaches. Curr Atheroscl Rep. (2023) 25:1069–81. doi:  10.1007/s11883-023-01174-3. PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 7. Poursaeed R, Mohammadzadeh M, Safaei AA. Survival prediction of glioblastoma patients using machine learning and deep learning: a systematic review. BMC Cancer. (2024) 24:1581. doi:  10.1186/s12885-024-13320-4. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 8. Thiele D, Rodseth R, Friedland R, Berger F, Mathew C, Maslo C, et al. Machine learning models for the early real-time prediction of deterioration in intensive care units—a novel approach to the early identification of high-risk patients. J Clin Med. (2025) 14:350. doi:  10.3390/jcm14020350. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 9. Arshi B, Wynants L, Rijnhart E, Reeve K, Cowley LE, Smits LJ. Number of publications on new clinical prediction models: a bibliometric review. JMIR Med Inf. (2025) 13:e62710. doi:  10.2196/62710. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 10. Ngema F, Mdhluli B, Mmileng P, Shungube P, Makgaba M, Hossana T. A bibliometric review of predictive modelling for cervical cancer risk. Front Res Metrics Anal. (2024) 9:1493944. doi:  10.3389/frma.2024.1493944. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 11. ang R, Zhang S, Ding C, Zhu M, Gao Y. Artificial intelligence in intensive care medicine: bibliometric analysis. J Med Internet Res. (2022) 24:e42185. doi:  10.2196/42185. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 12. Abumalloh RA, Nilashi M, Samad S, Ahmadi H, Alghamdi A, Alrizq M, et al. Parkinson’s disease diagnosis using deep learning: a bibliometric analysis and literature review. Ageing Res Rev. (2024) 96:102285. doi:  10.1016/j.arr.2024.102285. PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 13. Tao Y, Hou J, Zhou G, Zhang D. Artificial intelligence applied to diabetes complications: a bibliometric analysis. Front Artif Intell. (2025) 8:1455341. doi:  10.3389/frai.2025.1455341. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 14. Yang Z, Xu Y, Wang J, Li W, Zhang X, Pan H. Application of artificial intelligence in medical risk prediction: bibliometric analysis. Digital Health. (2025) 11:20552076251380652. doi:  10.1177/20552076251380652. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 15. Senthil R, Anand T, Somala CS, Saravanan KM. Bibliometric analysis of artificial intelligence in healthcare research: trends and future directions. Future Healthcare J. (2024) 11:100182. doi:  10.1016/j.fhj.2024.100182. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Xie Y, Zhai Y, Lu G. Evolution of artificial intelligence in healthcare: a 30-year bibliometric study. Front Med. (2025) 11:1505692. doi:  10.3389/fmed.2024.1505692. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 17. Hicks D, Wouters P, Waltman L, De Rijcke S, Rafols I. Bibliometrics: the Leiden manifesto for research metrics. Nature. (2015) 520:429–31. doi:  10.1038/520429a. PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 18. Gusenbauer M. Beyond Google Scholar, Scopus, and Web of Science: an evaluation of the backward and forward citation coverage of 59 databases' citation indices. Res Synth Methods. (2024) 15:802–17. doi:  10.1002/jrsm.1729. PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 19. Lim WM, Kumar S, Donthu N. How to combine and clean bibliometric data and use bibliometric tools synergistically: guidelines using metaverse research. J Business Res. (2024) 182:114760. doi:  10.1016/j.jbusres.2024.114760. PMID: 41847267 [ DOI ] [ Google Scholar ] 20. Lowell WE, Davis GE. Predicting length of stay for psychiatric diagnosis-related groups using neural networks. J Am Med Inf Assoc. (1994) 1:459–66. doi:  10.1136/jamia.1994.95153435. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 21. Woolery LK, Grzymala-Busse J. Machine learning for an expert system to predict preterm birth risk. J Am Med Inf Assoc. (1994) 1:439–46. doi:  10.1136/jamia.1994.95153433. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Wynants L, Van Calster B, Collins GS, Riley RD, Heinze G, Schuit E, et al. Prediction models for diagnosis and prognosis of covid-19: systematic review and critical appraisal. Bmj. (2020) 369:m1328. doi:  10.1136/bmj.m1328. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 23. Shipp MA, Ross KN, Tamayo P, et al. Diffuse large B-cell lymphoma outcome prediction by gene-expression profiling and supervised machine learning. Nat Med.. (2002) 8:68–74. doi:  10.1038/nm0102-68, PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 24. Rajkomar A, Oren E, Chen K, Dai AM, Hajaj N, Hardt M, et al. Scalable and accurate deep learning with electronic health records. NPJ Digital Med. (2018) 1:18. doi:  10.1038/s41746-018-0029-1. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 25. Tu JV. Advantages and disadvantages of using artificial neural networks versus logistic regression for predicting medical outcomes. J Clin Epidemiol.. (1996) 49:1225–1231. doi:  10.1016/s0895-4356(96)00002-9, PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 26. Christodoulou E, Ma J, Collins GS, Steyerberg EW, Verbakel JY, Van Calster B. A systematic review shows no performance benefit of machine learning over logistic regression for clinical prediction models. J Clin Epidemiol. (2019) 110:12–22. doi:  10.1016/j.jclinepi.2019.02.004. PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 27. Van Calster B, McLernon DJ, Van Smeden M, Wynants L, Steyerberg EW, tests TGEd et al. Calibration: the Achilles heel of predictive analytics. BMC Med. (2019) 17:230. doi:  10.1186/s12916-019-1466-7. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 28. Uddin S, Khan A, Hossain ME, Moni MA. Comparing different supervised machine learning algorithms for disease prediction. BMC Med Inform Decis Mak.. (2019) 19:281. doi:  10.1186/s12911-019-1004-8, PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 29. Chen HY, Yu SL, Chen CH, et al. A five-gene signature and clinical outcome in non-small-cell lung cancer. N Engl J Med.. (2007) 356:11–20. doi:  10.1056/NEJMoa060096, PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 30. Valli�res M, Freeman CR, Skamene SR, El Naqa I, et al. A radiomics model from joint FDG-PET and MRI texture features for the prediction of lung metastases in soft-tissue sarcomas of the extremities. Phys Med Biol.. (2015) 60:5471–5496. doi:  10.1088/0031-9155/60/14/5471, PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 31. Sidey-Gibbons JAM, Sidey-Gibbons CJ. Machine learning in medicine: a practical introduction. BMC Med Res Methodol.. (2019) 19:64. doi:  10.1186/s12874-019-0681-4, PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 32. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA: A Cancer J For Clin. (2021) 71:209–49. doi:  10.3322/caac.21660. PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 33. Gillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures, they are data. Radiology. (2016) 278:563–77. doi:  10.1148/radiol.2015151169. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 34. Breiman L. Random forests. Mach Learn. (2001) 45:5–32. doi:  10.1023/a:1010933404324. PMID: 40177105 [ DOI ] [ Google Scholar ] 35. Feng L, Liu Z, Li C, Li Z, Lou X, Shao L, et al. Development and validation of a radiopathomics model to predict pathological complete response to neoadjuvant chemoradiotherapy in locally advanced rectal cancer: a multicentre observational study. Lancet Digital Health. (2022) 4:e8–e17. doi:  10.1016/s2589-7500(21)00215-6. PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 36. Collins GS, Moons KG, Dhiman P, Riley RD, Beam AL, Van Calster B, et al. TRIPOD+ AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. Bmj. (2024) 385:e078378. doi:  10.1136/bmj-2023-078378. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 37. Lambin P, Leijenaar RT, Deist TM, Peerlings J, De Jong EE, Van Timmeren J, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol. (2017) 14:749–62. doi:  10.1038/nrclinonc.2017.141. PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 38. Podgorelec V, Kokol P, Stiglic B, Rozman I. Decision trees: an overview and their use in medicine. J Med Syst. (2002) 26:445–63. doi:  10.1023/a:1016409317640. PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 39. Royston P, Moons KG, Altman DG, Vergouwe Y. Prognosis and prognostic research: developing a prognostic model. Bmj. (2009) 338:b604. doi:  10.1136/bmj.b604. PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 40. Steyerberg EW, Harrell FE, Borsboom GJ, Eijkemans M, Vergouwe Y, Habbema JDF. Internal validation of predictive models: efficiency of some procedures for logistic regression analysis. J Clin Epidemiol. (2001) 54:774–83. doi:  10.1016/s0895-4356(01)00341-9. PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 41. Huang X, Huang X, Wang K, Bai H, Lu X, Jin G. 2.5 D deep learning radiomics and clinical data for predicting occult lymph node metastasis in lung adenocarcinoma. BMC Med Imaging. (2025) 25:225. doi:  10.1186/s12880-025-01759-1. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 42. Xie H, Zhang Y, Dong L, Lv H, Li X, Zhao C, et al. Deep learning driven diagnosis of Malignant soft tissue tumors based on dual-modal ultrasound images and clinical indexes. Front Oncol. (2024) 14:1361694. doi:  10.3389/fonc.2024.1361694. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 43. Zhou Y, Zhan Y, Zhao J, Zhong L, Zou F, Zhu X, et al. CT-based radiomics deep learning signatures for non-invasive prediction of metastatic potential in pheochromocytoma and paraganglioma: a multicohort study. Insights into Imaging. (2025) 16:1–15. doi:  10.1186/s13244-025-01952-4. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 44. Hussain S, Ali M, Naseem U, Nezhadmoghadam F, Jatoi MA, Gulliver TA, et al. Breast cancer risk prediction using machine learning: a systematic review. Front Oncol. (2024) 14:1343627. doi:  10.3389/fonc.2024.1343627. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 45. Kourou K, Exarchos TP, Exarchos KP, Karamouzis MV, Fotiadis DI. Machine learning applications in cancer prognosis and prediction. Comput Struct Biotechnol J. (2015) 13:8–17. doi:  10.1016/j.csbj.2014.11.005. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 46. Placido D, Yuan B, Hjaltelin JX, Zheng C, Haue AD, Chmura PJ, et al. A deep learning algorithm to predict risk of pancreatic cancer from disease trajectories. Nat Med. (2023) 29:1113–22. doi:  10.1038/s41591-023-02332-5. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 47. Unger M, Kather JN. Deep learning in cancer genomics and histopathology. Genome Med. (2024) 16:44. doi:  10.1186/s13073-024-01315-6. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 48. Yang H, Yang M, Chen J, Yao G, Zou Q, Jia L. Multimodal deep learning approaches for precision oncology: a comprehensive review. Briefings Bioinf. (2025) 26:bbae699. doi:  10.1093/bib/bbae699. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 49. Hannun AY, Rajpurkar P, Haghpanahi M, Tison GH, Bourn C, Turakhia MP, et al. Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network. Nat Med. (2019) 25:65–9. doi:  10.1038/s41591-018-0268-3. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 50. Martin SS, Aday AW, Allen NB, Almarzooq ZI, Anderson CA, Arora P, et al. 2025 heart disease and stroke statistics: a report of US and global data from the American Heart Association. Circulation. (2025) 151:e41–e660. doi:  10.1161/cir.0000000000001303. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 51. Benovic S, Ajlani AH, Leinert C, Fotteler M, Wolf D, Steger F, et al. Introducing a machine learning algorithm for delirium prediction—the Supporting SURgery with GEriatric Co-Management and AI project (SURGE-Ahead). Age Ageing. (2024) 53:afae101. doi:  10.1093/ageing/afae101. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 52. Pieszko K, Shanbhag AD, Singh A, Hauser MT, Miller RJ, Liang JX, et al. Time and event-specific deep learning for personalized risk assessment after cardiac perfusion imaging. NPJ Digital Med. (2023) 6:78. doi:  10.1038/s41746-023-00806-x. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 53. Collins GS, Dhiman P, Ma J, Schlussel MM, Archer L, Van Calster B, et al. Evaluation of clinical prediction models (part 1): from development to external validation. Bmj. (2024) 384:e074819. doi:  10.1136/bmj-2023-074819. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 54. Riley RD, Archer L, Snell KI, Ensor J, Dhiman P, Martin GP, et al. Evaluation of clinical prediction models (part 2): how to undertake an external validation study. Bmj. (2024) 384:e074820. doi:  10.1136/bmj-2023-074820. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 55. Moons KG, Damen JA, Kaul T, Hooft L, Navarro CA, Dhiman P, et al. PROBAST+ AI: an updated quality, risk of bias, and applicability assessment tool for prediction models using regression or artificial intelligence methods. Bmj. (2025) 388:e082505. doi:  10.1136/bmj-2024-082505. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 56. Abgrall G, Holder AL, Chelly Dagdia Z, Zeitouni K, Monnet X. Should AI models be explainable to clinicians? Crit Care. (2024) 28:301. doi:  10.1186/s13054-024-05005-y. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 57. Muhammad D, Bendechache M. Unveiling the black box: A systematic review of explainable artificial intelligence in medical image analysis. Comput Struct Biotechnol J. (2024) 24:542–60. doi:  10.1016/j.csbj.2024.08.005. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 58. Koch LM, Baumgartner CF, Berens P. Distribution shift detection for the postmarket surveillance of medical AI algorithms: A retrospective simulation study. NPJ Digital Med. (2024) 7:120. doi:  10.1038/s41746-024-01085-w. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 59. Tanner KT, Diaz-Ordaz K, Keogh RH. Implementation of a dynamic model updating pipeline provides a systematic process for maintaining performance of prediction models. J Clin Epidemiol. (2024) 175:111531. doi:  10.1016/j.jclinepi.2024.111531. PMID: [ DOI ] [ PubMed ] [ Google Scholar ] 60. Van Scharen A, Cruyt K, Colon J, De Sutter S, Duerinck J, Forsyth R, et al. Unlocking health data for research: Legal, technical, and organisational lessons from a Belgian interdisciplinary case study. J Healthcare Inf Res. (2025) 10:179–208. doi:  10.1007/s41666-025-00220-w. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 61. Pati S, Kumar S, Varma A, Edwards B, Lu C, Qu L, et al. Privacy preservation for federated learning in health care. Patterns. (2024) 5:100974. doi:  10.1016/j.patter.2024.100974. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 62. Gupta S, Kumar S, Chang K, Lu C, Singh P, Kalpathy-Cramer J. Collaborative privacy-preserving approaches for distributed deep learning using multi-institutional data. RadioGraphics. (2023) 43:e220107. doi:  10.1148/rg.220107. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 63. Choudhury A, Volmer L, Martin F, Fijten R, Wee L, Dekker A, et al. Advancing privacy-preserving health care analytics and implementation of the personal health train: Federated deep learning study. JMIR AI. (2025) 4:e60847. doi:  10.2196/60847. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 64. Teo ZL, Jin L, Liu N, Li S, Miao D, Zhang X, et al. Federated machine learning in healthcare: A systematic review on clinical applications and technical architecture. Cell Rep Med. (2024) 5:101419. doi:  10.1016/j.xcrm.2024.101481. PMID: [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Table1.docx (672.7KB, docx) Table2.docx (673.5KB, docx) Data Availability Statement The original contributions presented in the study are included in the article/ Supplementary Material , further inquiries can be directed to the corresponding author/s. Articles from Frontiers in Oncology are provided here courtesy of Frontiers Media SA ACTIONS View on publisher site PDF (21.3 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