The risk of idiopathic thrombocytopenic purpura (ITP) following measles, mumps, and rubella (MMR) vaccination: Attributable risk and a simulation study to evaluate four study designs

Using the Vaccine Safety Datalink databases---a multi-site project representing 8 managed care organizations from across the United States---I identified 216 cases of childhood idiopathic thrombocytopenic purpura (ITP). This case population was used to examine the risk of developing ITP in a 42-day risk period following MMR vaccination and to conduct an investigation of the natural history of ITP. Two study methods were used to analyze the association between MMR and ITP---a risk-interval and self-controlled case series (SCCS). The risk-interval analyses included a cohort of over 780,000 MMR-vaccinated children, while the SCCS method involved comparing time periods before and after vaccination in a fraction of the cases. Both analyses demonstrated a greater than 2-fold risk for developing ITP in the 42-day risk period following vaccination. The risk-interval analysis demonstrated an attributable risk of 1 case per 50,000 doses of MMR in children ages 12--15 months.; I also conducted a simulation study to empirically compare the validity and stability of incidence rate ratio (IRR) estimates and the ability to control for confounding across four study designs (cohort, risk-interval, SCCS and case-control) by using actual vaccine safety data and simulated cases of an acute illness. I simulated 250 case sets within the cohort and constructed the other three study designs from the cohort at four different IRRs (1.50, 2.00, 3.00 and 4.00). I compared the designs at 18 levels of decreasing prevalence (from 400 to less than 10 simulated cases), and two confounding levels (20%, 40%). Each of the design-specific study samples was analyzed with the appropriate regression model. The subsequent means and standard errors of the IRR estimates were compared across the study designs, using the cohort as the gold standard. The IRR estimates of the risk-interval, SCCS and case-control designs were within three percent of the cohort estimates. However, when compared to the other study designs, the estimates of the case-control design were less precise, less powerful and biased in the presence of confounding. The SCCS and risk-interval designs, in contrast, proved to be efficient and valid alternatives to the cohort study design.