A new paper by Ragon Institute biostatisticians in the Ghebremichael Lab introduces and tests a flexible way to grade medical tests like blood-based biomarkers so scientists can more confidently tell which ones truly separate “sick” from “healthy.” That clearer yardstick matters for immunology, where research teams must decide which candidate markers deserve scarce time and funding.

Published in Pharmaceutical Statistics, the study evaluates two common estimation approaches—maximum likelihood and partial likelihood—for a “bi-Weibull” ROC model. ROC/AUC is the field’s go-to score of test accuracy. The bi-Weibull model earns attention because it generates smooth, stable curves, has a simple formula for AUC, and adapts to many real-world data shapes. It also plugs into Cox regression, making it easier to adjust for factors like age or treatment.

Across large simulations, both methods performed well when data truly followed a Weibull-like pattern; when data didn’t, both struggled. In smaller samples, the partial-likelihood approach tended to show lower bias and better confidence-interval coverage, trading a little efficiency to gain reliability.

To ground the work in practice, the authors applied the methods to HIV datasets from sub-Saharan Africa. Pre-treatment CD4 counts showed moderate power to predict children’s immune recovery on therapy (useful as part of a panel, not alone), while another candidate signal performed poorly.

This study could have significant implications for future testing. With a more flexible, transparent scoring model, scientists can make cleaner calls about which biomarkers truly help patients in the future.