OBJECTIVES : Randomized controlled trials (RCT) are uncommon in precision oncology. In the absence of RCTs, quasi-experimental matching can be used to mitigate selection bias while estimating the health impacts of omics-guided care. In this study, we introduce and compare two matching methods for estimating the effects of precision oncology in an illustrative case study.

METHODS : Our case study focuses on British Columbia’s (BC) Personalized OncoGenomics (POG) program, which applies whole-genome and transcriptome analysis (WGTA) to guide advanced cancer care. Our cohort comprises patients who participated in POG between 2014 and 2015 and matched controls. We generated our matched cohort using BC Cancer Registry data combined with 1:1 propensity score matching (PSM) and genetic matching. To explore survival differences, we estimated Kaplan-Meier survival functions and Weibull regression models.

RESULTS : During our study period, 230 patients participated in POG and 93,736 patients were identified as possible controls. Of these, 5,224 control patients were eligible for matching. Final weighted matched cohorts each included 230 controls. Genetic matching outperformed PSM when balancing covariates of interest. Survival analyses on unmatched and matched cohorts indicated that overall survival did not significantly differ across POG and control patients (p>0.05). Stratification by WGTA-informed treatment revealed differences in estimated survival. In all cohorts patients whose WGTA information led to treatment change were at a statistically significantly reduced hazard of death compared to controls. Estimated hazard ratios ranged from 0.33 (95% CI: 0.13, 0.81) in propensity score matched patients, to 0.34 (95% CI: 0.14, 0.86) in genetic matched patients, to 0.41 (95% CI: 0.17, 0.98) in unmatched patients.

CONCLUSIONS : Matching combined with real-world data offers a solution to the challenges of non-randomized enrollment observed in many precision oncology applications. Yet validity relies on strong underlying assumptions. Careful study design and balance assessment are essential to produce reliable effect estimates.