Molecular Diagnosis as a Probability of Necessity

Abstract

Genetic diagnosis requires attributing a patients disease to variant(s). We formalize this attribution through the Probability of Necessity (PN), a counterfactual estimand modeling the probability that disease would not have occurred absent a germline variant given that both are observed, and we use PN to examine the logic and limitations of genetic diagnosis. PN is a property of a variant in a population and disease context, not of a variant alone. PN dissociates from penetrance: identical penetrance yields different PN depending on baseline disease risk and competing causes. Case selection biases PN estimation, favoring population-scale cohorts for estimation over ascertained case series. PN falls as the prevalence of competing causes rises, and among carriers it can fall with increasing polygenic liability even as penetrance rises. We adopt the mediated probability of causation as the Probability of Mediated Necessity (PMN), which updates PN using a biomarker proxying the variants operative pathway. Using UK Biobank exome data, we estimate PN and PMN for rare LDLR variants in ischemic heart disease (IHD, ICD-10 I25), with hyperlipidemia (E78) as mediator, stratifying by ClinVar class and deleteriousness scores. PN resolves heterogeneity within the variant of uncertain significance (VUS) class beyond variant classification alone. In GBA1 carriers with Parkinsons disease, a mechanistically distinct, low-penetrance system, the same PN derived ordering of VUS holds, and PN is high despite modest penetrance. These results support causal necessity as a framework for genetic diagnosis, distinct from and complementary to variant classification, with implications for clinical and policy decisions in monogenic and common complex diseases.