FIGURE 6.

Proposed data analyses plan using traditional statistics and ML approaches. Previous biomarker projects on uncommon diseases have been hindered by small sample sizes limiting the power to query relevant subgroups. We here recruit plan to recruit 800 subjects from established CA centers and consortia, including 200 already enrolled in pilot studies, whose plasma samples and clinical data are in hand (“docket cases”). We will deploy supervised ML algorithms, in addition to traditional statistics, to define the best clustering and weighing of combined diagnostic and prognostic biomarker of SH in CA patients. LASSO, least absolute shrinkage and selection operator; CART, Classification & Regression Trees; SVM, support-vector machines; RF, relative frequency: RMSE, root mean square error; AUC, area under the curve.