Abstract
PURPOSE: Given the pace of predictive biomarker and targeted therapy development, it is unknown whether repeat annotation of the same next-generation sequencing data can identify additional clinically actionable targets that could be therapeutically leveraged. In this study, we sought to determine the predictive yield of serial reanalysis of clinical tumor sequencing data. EXPERIMENTAL DESIGN: Using artificial intelligence (AI)-assisted variant annotation, we retrospectively reanalyzed sequencing data from 2,219 patients with cancer from a single academic medical center at 3-month intervals totaling 9 months in 2020. The yield of serial reanalysis was assessed by the proportion of patients with improved strength of therapeutic recommendations. RESULTS: A total of 1,775 patients (80%) had ≥1 potentially clinically actionable mutation at baseline, including 243 (11%) patients who had an alteration targeted by an FDA-approved drug for their cancer type. By month 9, the latter increased to 458 (21%) patients mainly due to a single pan-cancer agent directed against tumors with high tumor mutation burden. Within this timeframe, 67 new therapies became available and 45 were no longer available. Variant pathogenicity classifications also changed leading to changes in treatment recommendations for 124 patients (6%). CONCLUSIONS: Serial reannotation of tumor sequencing data improved the strength of treatment recommendations (based on level of evidence) in a mixed cancer cohort and showed substantial changes in available therapies and variant classifications. These results suggest a role for repeat analysis of tumor sequencing data in clinical practice, which can be streamlined with AI support.