Benchmarking Classical, Machine Learning, and Bayesian Survival Models for Clinical Prediction
Irving Gómez-Méndez, Sivakorn Phromsiri, Ittiphat Kijpaisansak, Settawut Chaithurdthum · 2025
Read Benchmarking Classical, Machine Learning, and Bayesian Survival Models for Clinical Prediction by Irving Gómez-Méndez, Sivakorn Phromsiri, Ittiphat Kijpaisansak, Settawut Chaithurdthum (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-10073.txt.
Introduction
Irving Gómez-Méndez and 3 co-authors benchmark six survival models in a 2025 clinical paper. Survival analysis gauges how long it takes for an event to occur. Such events include discharge from care or relapse. The line-up spans three eras of method. It holds classic Weibull and Cox models. It holds a random survival forest. It holds a deep learning model called DeepSurv. It also holds Bayesian forms with frailty terms. Those terms soak up hidden gaps between patients.
The test used a public breast cancer data set. Scores tracked agreement and error. DeepSurv predicted best overall. Readable rivals stayed close behind it. The frailty models earned praise for honest doubt. Standard libraries barely support them, so they were hand-built. The authors frame the choice plainly. It is raw accuracy against bedside clarity.
Catalog record
| CREATOR | Irving Gómez-Méndez, Sivakorn Phromsiri, Ittiphat Kijpaisansak, Settawut Chaithurdthum |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 57 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-10073
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Benchmarking Classical, Machine Learning, and Bayesian Survival Models for Clinical Prediction Irving Gómez-Méndez, Sivakorn Phromsiri, Ittiphat Kijpaisansak, Settawut Chaithurdthum Abstract Survival analysis is a statistical framework for modeling time-to-event data, particularly valuable in healthcare for predicting outcomes like patient discharge or recurrence. This study implements and compares several survival models—including Weibull, Weibull AFT, Weibull AFT with Gamma Frailty, Cox Proportional Hazards (CoxPH), Random Survival Forest (RSF), and DeepSurv—using a publicly available breast cancer dataset. This study aims to benchmark classical, machine learning, and Bayesian survival models in terms of their predictive performance, interpretability, and suitability for clinical deployment. The models are evaluated using performance metrics such as the Concordance Index (C-index) and the Root Mean Squared Error (RMSE). DeepSurv showed the highest predictive performance, while interpretable models like RSF and Weibull AFT with Gamma Frailty offered competitive results. We also explored the implementation of statistical models from a Bayesian perspective, including frailty models, due to their ability to properly quantify uncertainty. Notably, frailty models are not readily available in standard survival analysis libraries, necessitating custom implementation. Our results demonstrate that interpretable statistical models, when correctly implemented using parameters that are effectively estimated using a Bayesian approach, can perform competitively with modern black-box models. These findings illustrate the trade-offs between model complexity, interpretability, and predictive power, highlighting the potential of Bayesian survival models in clinical decision-making settings. Index Terms Survival analysis, Bayesian modeling, DeepSurv, hospital length of stay, Cox proportional hazards, random survival forest, Weibull AFT, frailty models. 1 Introduction Statistical survival analysis attempts to predict the amount of time until the occurrence of a particular event, such as death, recurrence of illness or equipment failure. It is essential in the medical field for assessing treatment effectiveness, comprehending patient prognoses, and allocating healthcare resources as efficiently as possible. Precise survival forecasts allow medical professionals to make well-informed choices, guaranteeing prompt interventions and effective use of medical resources. The primary objective of this study is to evaluate and compare several survival analysis models—classical, machine learning, and Bayesian—for predicting hospital length of stay using censored medical data. We focus not only on predictive performance, but also on the interpretability and uncertainty quantification of each model. Our key contributions include: (1) implementing Bayesian frailty models not readily available in standard toolkits, (2) benchmarking a variety of models using consistent metrics, and (3) discussing the trade-offs between model complexity, interpretability, and clinical applicability. To guide the reader through this paper, Section 2 introduces the theoretical foundations of survival analysis, including key concepts such as the survival function, hazard function, and likelihood estimation. Section 3 outlines the methodology, describing the dataset, preprocessing steps, and the survival models implemented. Section 4 presents the results, comparing model performance using evaluation metrics like the C-index and RMSE. Section 5 discusses the findings, highlighting the trade-offs between interpretability and predictive power across models. The paper concludes with reflections on practical implications, limitations, and suggestions for future work. 1.1 Application in Hospital Resource Management Reliable survival forecasts are essential for optimizing the distribution of medical resources, including hospital beds, personnel, and equipment, in the context of hospital resource management kim2024determinants ; wen2022time . Healthcare administrators can better plan interventions, save resources, and enhance patient care by predicting patient outcomes. For example, hospitals can control capacity and optimize operations by forecasting readmission rates or length of hospital stays vanbelle2021survival . Utilizing survival analysis, hospitals can: • Allocate ICU beds based on predicted patient deterioration rates, • Optimize discharge planning by forecasting length of stay (LOS), • Identify high-risk patients for early interventions, and • Minimize resource waste by adjusting staff and equipment allocation dynamically. 1.2 Historical Context and Evolution Survival analysis has its roots in medical research and actuarial science, with early studies evaluating the effects of treatments and estimating life expectancies. The basis of contemporary survival analysis was established by the 1958 introduction of the Kaplan-Meier estimator, which offered a non-parametric technique for estimating survival functions from censored data collett2021modelling . The Kaplan–Meier is, however, a descriptive estimator. It is unable to forecast the survival probability for a new individual based on their specific characteristics (covariates), such as age, tumor size, or treatment type. In other words, it does not incorporate covariate information into its predictions. A semi-parametric approach to evaluating the impact of covariates on survival times without assuming a functional form for the baseline hazard was then provided by the Cox Proportional Hazards (CoxPH) model, which was first presented in 1972 cox1972regression . The baseline hazard represents the underlying risk of the event occurring over time for an individual with average or baseline covariates. Not specifying a functional form means that the model does not require this hazard to follow any particular distribution, such as exponential or Weibull, giving CoxPH greater flexibility. Unlike the Kaplan–Meier estimator, the CoxPH model allows for prediction of survival outcomes. As interpretable models that have influenced clinical judgments over the past few decades, these conventional approaches have played a significant role in medical statistics. As data availability and computational power have increased, survival analysis has expanded to include machine learning-based approaches. Methods like Random Survival Forests (RSF) ishwaran2008 and DeepSurv katzman2018 aim to improve predictive accuracy by capturing intricate, non-linear relationships in the data. In contrast, Bayesian approaches represent a separate modeling paradigm. They focus on quantifying prediction uncertainty and incorporating prior knowledge, offering a probabilistic framework that is particularly useful in settings with small sample sizes or complex hierarchical structures. These models have also become more accessible with libraries such as PyMC salvatier2016pymc3 . For example, traditional parametric models like the Weibull or Weibull AFT can be implemented using a Bayesian approach to estimate full posterior distributions and account for frailty or unobserved heterogeneity. 1.3 Challenges and Considerations Implementing survival analysis in healthcare settings involves several challenges: • Data Quality and Censoring: Handling incomplete or censored data—where the event of interest (e.g., death, recurrence, discharge) has not occurred during the observation period— requires robust statistical techniques to avoid biased results. • Interpretability: Models must be interpretable to ensure that healthcare professionals can trust and effectively use the predictions in clinical decision-making. • Computational Complexity: Advanced machine learning and Bayesian methods require substantial computational resources for model fitting and inference. • Ethical and Legal Aspects: The use of patient data necessitates strict adherence to ethical guidelines and data protection regulations to maintain confidentiality and trust. Survival analysis has a vital position in medical studies and hospital administration, and the insights it gives can be used to achieve improved patient outcomes and more efficient healthcare systems. The combination of traditional statistical thinking and modern machine learning and Bayesian approaches is ready to address current challenges and enhance the predictability of survival models. With medical practice increasingly improving, utilization of advanced survival analysis techniques will prove increasingly essential in optimizing resource use and ensuring good patient care. 1.4 Literature Review Existing research in survival analysis has primarily focused on: 1. Statistical Models – Cox Proportional Hazards (CoxPH) cox1972regression , Weibull regression, and parametric Accelerated Failure Time (AFT) models. 2. Machine Learning Models – Random Survival Forests (RSF) and Survival Support Vector Machines vanbelle2021survival . 3. Deep Learning Models – DeepSurv and DeepCox vanbelle2021survival . 4. Bayesian