Adaptive Validation Design

An internal validation substudy compares an imperfect measurement of a variable with a gold-standard measurement in a subset of the study population. Validation data permit calculation of a bias-adjusted estimate, which has the same expected value as the association that would have been observed had the gold-standard measurement been available for the entire study population. Existing guidance on optimal sampling for validation substudies assumes complete enrollment and follow-up of the target cohort. No guidance exists for validation substudy design while cohort data are actively being collected. In this article, we use the framework of Bayesian monitoring methods to develop an adaptive approach to validation study design. This method monitors whether sufficient validation data have been collected to meet predefined criteria for estimation of the positive and negative predictive values. We demonstrate the utility of this method using the Study of Transition, Outcomes and Gender-a cohort study of transgender and gender nonconforming people. We demonstrate the method's ability to determine efficacy (when sufficient validation data have accumulated to obtain estimates of the predictive values that fall above a threshold value) and futility (when sufficient validation data have accumulated to conclude the mismeasured variable is an untenable substitute for the gold-standard measurement). This proposed method can be applied within the context of any parent epidemiologic study design and modified to meet alternative criteria given specific study or validation study objectives. Our method provides a novel approach to effective and efficient estimation of classification parameters as validation data accrue.