Observational studies evaluating the effect of a drug versus “non-use” are challenging, mainly when defining cohort entry for non-users. The approach using successive monthly cohorts to emulate the randomized trial can be perceived as somewhat opaque and complex. Alternatively, the prevalent new-user design can provide a potentially simpler more transparent emulation. This design is illustrated in the context of statins and cancer incidence.
We used the Clinical Practice Research Datalink to identify a cohort of subjects with low-density lipoprotein cholesterol level <5 mmol/L. We used a prevalent new-user design, matching each statin initiator to a non-user from the same time-based exposure set on time-conditional propensity scores with all subjects followed for 10 years for cancer incidence. We estimated the hazard ratio and 95% confidence interval (CI) of cancer incidence with statin use versus non-use using a Cox proportional hazards model, and the results were compared with those using the method of successive monthly cohorts.
The study cohort included 182,073 statin initiators and 182,073 matched non-users. The hazard ratio of any cancer after statin initiation versus non-use was 1.01 (95% CI = 0.98, 1.04), compared with 1.04 (95% CI = 1.02, 1.06) under the successive monthly cohorts approach. We estimated similar effects for specific cancers.
Using the prevalent new-user design to emulate a randomized trial when compared to “non-use” led to results comparable with the more complex successive monthly cohorts approach. The prevalent new-user design emulates the trial in a potentially more intuitive and palpable manner, providing simpler data presentations in line with those portrayed in a classical trial while producing comparable results.