How incidence is affected by outcome misclassification in the presence of conditional dependence

Background: Small imperfections in sensitivity and specificity can create large biases in the estimation of incidence of rare diseases. While methods exist to adjust for misclassification, these methods assume conditional independence of repeated assessments on the same person. Objectives: To explore algebraically the sensitivity of estimates of cumulative incidence to misclassification in the presence of conditional dependence, and to derive unbiased adjusted estimates of cumulative incidence. Methods: We obtained expressions defining estimates of the adjusted incidence in terms of observed incidence, misclassification and conditional dependence probabilities. Mean squared error and variance were estimated to measure bias and variation of observed and adjusted rates. Results: The specificity had the greatest impact on the magnitude of bias, while the remission probability was the most sensitive estimate to the presence and degree of misclassification probabilities. The lower the true prevalence rates, the higher the impact of imperfect specificity on the bias of remission estimates. In contrast, the higher true prevalence rates are, the higher the impact of imperfect sensitivity on the magnitude of incidence rate estimates bias. The higher the conditional dependence between the test results, the smaller the magnitude of the bias. The MSE of adjusted rates were generally much smaller than MSE of observed rates. Conclusion: To obtain the most accurate incidence estimates, it is of great importance to choose the appropriate diagnostic test at the early stages of study design, and to properly adjust for any possible misclassification and or conditional dependence in the analysis stage.