Benefits of Repeated Matched-Cohort and Nested Case–Control Analyses with Time-dependent Exposure in Observational Studies

Matched-cohort and nested case–control (NCC) analyses have been proposed as dynamic matching methods for exposure occurring over time. However, these methods might produce noisier estimates due to their matching mechanism. The literature suggests the consideration of multiple controls to reduce the variability in the effect estimates. In the present study, by using simulations and data from a retrospective cohort (with a time-dependent exposure being tuberculosis and the outcome being cardiovascular disease), we showed that the hazard ratio (HR) estimates can be unstable regardless of the number of controls. Our case study revealed that the HRs might range from 1.64 to 2.32 in matched-cohort and 1.53 to 2.30 in NCC analyses. With a true HR of 2.0, simulation results showed that HR estimates range from 1.51 to 2.93 in matched-cohort and 1.30 to 3.23 in NCC analyses. To reduce the noise in HR estimates, we compared the matched-cohort and NCC analyses while varying the number of repeated samplings (repeating the analyses several times and pooling the results) for different numbers of controls. We showed that repeated analyses yield stable HR estimates similar to a full-cohort analysis under the proportional hazard assumption. The pooled HR was approximately 1.90 in our case study and approximately 2.0 in simulations when we repeated NCC analysis ≥ 10 times for any number of controls, and repeated the matched-cohort analyses ≥ 20 times with four or fewer or ≥ 10 times with more than four controls. We recommend using matched-cohort and NCC analyses with repeated samplings due to the benefits of using these methods in dealing with time-dependent exposure.