Machine Learning Outperforms Cox Model For ART
Researchers compared Cox proportional hazards and multiple machine learning models using longitudinal EMR data from 621,115 Kenyan patients who started ART between 2017 and 2023. Recursive partitioning achieved the highest concordance-index scores—0.81 for first treatment interruptions and 0.89 for multiple interruptions—versus the CPH model's 0.78 and 0.87. Model-agnostic explainers (SHAP, breakdown, ceteris paribus) provided global and individual risk insights to support targeted retention interventions.
Key Points
- 1Demonstrates recursive partitioning achieved c-index 0.81 for first and 0.89 for multiple interruptions
- 2Highlights recursive partitioning's ability to model nonlinear relationships and detect complex interactions, boosting accuracy
- 3Enables patient-specific risk explanations using SHAP, breakdown, and ceteris paribus for targeted interventions
Scoring Rationale
Large national cohort and explainable ML drive high impact, while methodological novelty is incremental compared with prior survival ML.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
