🤖 Generated with AI assistance · Does not replace individualized medical advice.
Explainable ML for Prostate Cancer PFS: RSF Outperforms Cox in Brunei Cohort
AI Transparency Notice: This post was generated by an AI assistant to summarize and structure scientific literature for educational purposes. It does not constitute medical advice, diagnosis, or treatment recommendations. Always consult a qualified oncologist for clinical decisions.
Lead
A retrospective cohort study from Brunei Darussalam demonstrates that Random Survival Forests (RSF) significantly outperform traditional Cox models in predicting progression-free survival (PFS) in prostate cancer. With a C-index of 0.906, RSF offers superior discriminative power, while SHAP analysis identifies specific laboratory markers as key risk drivers PMID: 42832781.
Key Facts
| Metric | Value | Context |
|---|---|---|
| Cohort Size | 212 patients | 478 longitudinal observations PMID: 42832781 |
| Top Model | Random Survival Forest (RSF) | C-index 0.906 PMID: 42832781 |
| Baseline Model | Cox Proportional Hazards (CPH) | C-index 0.531 PMID: 42832781 |
| Interpretability | SHAP Analysis | Identified sodium, ALT, MCH, platelets PMID: 42832781 |
Why It Matters / Context
Progression-free survival (PFS) is a critical endpoint in oncology, yet the translation of individualized, explainable machine learning (ML) predictions into real-world clinical practice remains limited PMID: 42832781. Traditional statistical models, such as the Cox proportional hazards (CPH) model, have long been the standard for survival analysis but often fail to capture complex, non-linear interactions between clinical variables PMID: 42832781.
This study addresses a significant gap by developing and validating explainable ML models using retrospective data from a national prostate cancer cohort in Brunei Darussalam PMID: 42832781. The integration of Shapley additive explanations (SHAP) is particularly noteworthy, as it moves beyond “black box” predictions to provide clinicians with interpretable insights into why a patient is classified as high-risk PMID: 42832781. By identifying specific laboratory markers—such as sodium, alanine aminotransferase (ALT), mean corpuscular hemoglobin (MCH), and platelet count—as key drivers of progression risk, the model offers actionable biological insights that may inform monitoring strategies or early intervention PMID: 42832781.
The superior performance of tree-based ensemble methods, specifically RSF, over traditional CPH models suggests that current standard-of-care risk stratification tools may be underestimating the predictive potential of available clinical data PMID: 42832781. If validated externally, such models could enhance clinical decision-making by providing more accurate, personalized risk assessments for prostate cancer patients PMID: 42832781.
Details
The study analyzed a retrospective cohort of 212 patients (478 longitudinal observations) treated at the Brunei Cancer Centre between January 2018 and December 2024 PMID: 42832781. Clinical, laboratory, and treatment data were harmonized, with missing values imputed using extremely randomized trees PMID: 42832781. To capture longitudinal patterns, a recurrent autoencoder was employed to generate latent representations of the patient data PMID: 42832781.
Four modeling approaches were compared: Cox proportional hazards (CPH), random survival forest (RSF), gradient boosting survival (GBS), and deep neural network survival models PMID: 42832781. Performance was evaluated using time-dependent area under the receiver operating characteristic curve (AUC), Harrell C-index, and integrated Brier score (IBS) PMID: 42832781.
| Model | C-index | AUC (4 Years) | AUC (5 Years) | IBS |
|---|---|---|---|---|
| Random Survival Forest (RSF) | 0.906 | 0.941 | 0.917 | 0.0698 |
| Gradient Boosting Survival (GBS) | 0.844 | 0.765 | 0.833 | 0.0590 |
| Deep Neural Network Survival | 0.719 | 0.941 | 0.917 | 0.0887 |
| Cox Proportional Hazards (CPH) | 0.531 | 0.706 | 0.833 | N/A |
Note: IBS values for CPH were not explicitly reported in the abstract summary provided. Lower IBS indicates better calibration.
The RSF model demonstrated improved discriminative performance and balanced calibration, achieving a C-index of 0.906 and AUCs of 0.941 and 0.917 at 4 and 5 years, respectively PMID: 42832781. In contrast, the traditional CPH model performed poorly, with a C-index of 0.531 and AUCs of 0.706 at 4 years and 0.833 at 5 years PMID: 42832781. The deep survival model showed high AUCs (0.941 at 4 years, 0.917 at 5 years) but a lower C-index (0.719) and higher IBS (0.0887) compared to RSF PMID: 42832781. The GBS model showed moderate performance with a C-index of 0.844 and the lowest IBS (0.0590), indicating good calibration despite lower discriminative power than RSF PMID: 42832781.
SHAP analysis identified sodium, alanine aminotransferase, mean corpuscular hemoglobin, platelet count, and specific treatment categories as key drivers of increased progression risk PMID: 42832781. These findings highlight the potential of explainable ML to enhance clinical decision-making by linking predictive signals to measurable clinical parameters PMID: 42832781.
Limitations
Critical Limitations
* Single-Institution Cohort: The study relies on data from a single national center in Brunei Darussalam, limiting generalizability to other populations or healthcare systems PMID: 42832781.
* Sample Size: With only 212 patients, the cohort is relatively small for training complex deep learning models, increasing the risk of overfitting PMID: 42832781.
* Lack of External Validation: The authors explicitly state that external validation in a larger multi-institutional, multiomics dataset is required before routine clinical implementation PMID: 42832781.
* Retrospective Design: The retrospective nature of the data may introduce selection biases and missing data issues, although imputation methods were used PMID: 42832781.
Clinical Takeaway
Clinicians should view this study as a proof-of-concept for integrating explainable ML into prostate cancer risk stratification, but routine adoption is premature until external validation confirms the robustness of the RSF model across diverse, multi-institutional cohorts PMID: 42832781.
Full Reference
Tun HM, Naing L, Malik OA, Abdullah MS, Ta T, Rahman HA. Explainable Machine Learning-Based Prediction of Progression-Free Survival in Prostate Cancer: Retrospective Cohort Study. JMIR Cancer. 2026. PMID: 42832781 [DOI: 10.2196/91510]