The Polynomial Progression Subtype Inference Algorithm.

Longitudinal assessments are currently the gold standard for modeling progression of diseases, but they delay prognosis and increase burden on patients and healthcare systems. Cross-sectional inference offers a valuable alternative, enabling earlier patients' stratification and broader accessibility. Initial success in this direction has been found with the SuStaIn algorithm (Young et al. 2018) but computational and conceptual shortcomings hamper its usefulness. Here we introduce a more effective algorithm, PPSI, which is orders of magnitude faster, easier to interpret, equally or more accurate, applicable to more complex bidirectional phenomena, and can be fitted with many more variables at once. We demonstrate PPSI's utility using longitudinal prediction in Alzheimer's disease (ADNI database), clinical subtype recovery in breast cancer (TCGA-BRCA), and measurement of robustness under simulated conditions. To promote broad usability, we provide an extensive plotting suite for model exploration and diagnostics, and a graphical user interface that allows non-programmers to use the tool. While PPSI and SuStaIn are both able to derive useful subtypes from data, PPSI dramatically improves computational efficiency, enabling the inclusion of thousands of features and reducing runtimes from hours to seconds. PPSI is a performant interactive tool which makes disease progression modeling accessible to any subject matter expert.