Identifying the most effective and cost-effective colorectal cancer screening schedule:

Colorectal cancer is the second leading cause of cancer deaths in the United States. Screening for colorectal cancer (CRC) in asymptomatic people can reduce the incidence and mortality of CRC by as much as 50%. Current guidelines for colonoscopy surveillance recommend repeated screening at an interval of ten years starting at age 50 for the general population. In this project, we aim to empirically identify the most cost-effective screening interval for general population from observational data. We have started with a very simple scenario, developing and applying methods to identify the optimal covariate independent screening strategy. We first compared three regimes, (1) never screening; (2) starting screen at 50, after that screening every 10 years; (3) starting screen at 60, after that screening every 10 years. We generated three duplicates for each subject. One copy corresponds to one regime, but the copy is censored at the time point when the subject violated the corresponding regime. The artificial censorship would introduce biases in parameter estimation. We used marginal structure models to correct for the bias. The marginal structure models were adjusted for time-varying CRC risk factors, such as smoking status, alcohol intake, red-meat intake, and current multivitamin use. The outcome we currently used is the mortality. Each participant in Harvard’s Nurses’ Health Study (NHS) may never be screened, or screened several times but with different screening intervals. The artificial censorship significantly decreased the number of cases in NHS. For example, there were only 9 cases and 55 cases for regimes 2 and 3, respectively. The data analysis was underpowered, and we found that neither regime 2 nor regime 3 is significantly better than regime 1. We will consider the CRC incidence as the outcome to increase the number of cases for comparison.