Temporal analysis of clinical variant reclassification (2012–2025)

ClinVar is the primary archive for human genetic variant classifications, yet existing studies lack a longitudinal perspective of its evolution, often conflating clinical reinterpretation of variant pathogenicity with variant reclassification resulting from database schema evolution. We systematically characterized ClinVar’s 13-year history (2012–2025), analyzing 3.68 million variants across 148 monthly snapshots using both Reference ClinVar (RCV)- and Variation ClinVar (VCV)-formatted Extensible Markup Language (XML) files. An era-stratified heuristic framework was developed to distinguish schema-driven reclassification artifacts from evidence-based clinical reinterpretation. We additionally assessed inter-submitter concordance, disease-domain reclassification patterns, and American College of Medical Genetics and Genomics/Association for Molecular Pathology (ACMG/AMP) evidence documentation. Overall, 9.3% of variants underwent at least one change in aggregate classification category; however, only 0.68% (25,105 variants) ever underwent a medically significant change — that is, a change crossing the clinical actionability boundary between pathogenic/likely pathogenic (P/LP) and variant of uncertain significance/benign/likely benign (VUS/B/LB), including 6,340 variants that underwent a drastic pathogenic-to-benign (or vice versa) switch. Accounting for censoring, the cumulative incidence of first reclassification was 2.5% at one year and 15.5% at five years after a variant’s first appearance. We identified three distinct schema eras involving over 120 structural database changes that impact data interpretation, with 2.6% of all reclassification events attributable to the RCV-to-VCV format transition. Reclassification rates were significantly higher in cancer-related genes than in other genes (16.6% vs. 8.2%) and in higher- than in lower-prevalence Mendelian disease variants (19.7% vs. 9.2%). Among multi-submitter variants, 72.1% were concordant while 17.6% maintained conflicting interpretations. Of 697,698 reclassification events, 33.2% resulted from submitter reinterpretation and 18.3% from new-submitter-driven category shifts. Population-frequency releases (the Exome Aggregation Consortium [ExAC] and the Genome Aggregation Database [gnomAD]) and functional evidence from high-throughput saturation mutagenesis and targeted biochemical assays were identified as important contributors to reclassification. Our findings support the need for domain-specific variant reassessment intervals and reinforce the importance of diverse population reference datasets and high-throughput functional assays to minimize classification uncertainty. We also published an open-source pipeline and visualization framework to enable reproducible, era-aware temporal analysis of clinical variant reclassification.