Heart transplant prediction models made transparent and easy to audit
Translation of Black-Box Clinical Prediction Models into Standalone Transparent Nomograms: Temporal External Validation in Heart Transplantation
Machine Learning
Summary
Many clinical prediction models are like 'black boxes'—they give results but don’t explain how. The authors created a method called PRiSM that turns these complex models into clear, straightforward charts called nomograms. These nomograms show how each factor affects the prediction, making them easier to understand and check. They tested this system on heart transplant data and found the transparent models worked just as well as the original ones. They also shared the method as an open-source tool anyone can use.
What this means in practice
- •For hospital data teams: Create easy-to-interpret clinical risk charts from existing complex heart transplant prediction models for better auditing and decision making.
- •For medical software developers: Build clinical prediction tools with transparent, standalone nomograms derived from machine learning models, improving clinician trust and use.$Commercial implications: This enables developers to sell interpretable clinical decision support software that transparently explains risk scores to clinicians.
Authors
Henry Pigot, Paulo J. G. Lisboa, Sandra Ortega-Martorell, Ivan Olier, Joseph Mahon, Johan Nilsson
Abstract
We convert black-box clinical prediction models for tabular data into standalone nomograms that can be audited term by term. PRiSM (Partial Responses in Structured Models) takes the shape of each effect and interaction from the source model, not merely which variables mattered, and lets the outcome select and weight them. We tested this in 50,356 heart transplant recipients, with validation in a later era than training. Nomograms from all 5 source models - a public clinical risk score, logistic regression, neural networks, random forests and extreme gradient boosting - met a prespecified noninferiority criterion for discrimination before any further simplification, and generally preserved calibration and clinical net benefit. Those from the 3 machine-learning models showed no detectable difference in discrimination from de novo generalized additive and explainable boosting models, exceeded neural additive models, and carried fewer terms than the explainable boosting model. PRiSM is released as an open-source Python package.