Selection and Collider Restriction Bias Due to Predictor Availability in Prognostic Models

Abstract

This methodological note investigates and discuss possible selection and collider restriction bias due to predictor availability in prognostic models.

Figures (1)

• Figure 1: Selection and collider restriction bias arising from conditioning on predictor availability in prognostic models. Directed acyclic graphs are used to illustrate causal dependencies between observed and unobserved variables. $U$ represents unobserved disease severity; $P_1$ and $P_2$ represent predictors; and $Y$ represents the outcome. Framed nodes indicate measurement-based restriction. Panel A depicts simple selection bias, where underlying disease severity triggers measurement of the predictor. Panel B depicts simple selection bias in which disease severity is captured through a measured proxy. Panel C illustrates collider restriction bias, where both underlying disease severity and its proxy influence measurement of $P_2$, making its availability a collider.