Pharmacological and epidemiological considerations while constructing treatment episodes using observational data: A simulation study

Abstract Background The permissible gap method is an extensively used approach for defining episodes of continuous treatment use in pharmacoepidemiology. This method uses the amount of drug redeemed, when available, and researcher‐defined temporal gaps to fill the interval between the calculated end of coverage of a redeemed prescription and the date of redemption of the next prescription in the same treatment episode. The final scope is defining periods of continuous use of medications. There are strong pharmacological and epidemiological arguments for adding the gap at the end of each treatment episode. However, the evidence is scarce on the impact that such a practice has on measures of association. This study aims to compare the impact of adding or not adding the researcher‐defined gap time to the end of a treatment episode on the incidence of drug discontinuation and the incidence rate for a simulated outcome that occurred during an observational window. Additionally, the study aims at assessing the magnitude of misclassification of exposure time between the two methods. Methods A simulated dataset of 100 patients available in the R package AdhereR that contains 1080 redeemed prescriptions was used. A gap time of 90 days was used for constructing treatment episodes in an observational window of 365 days following the first redeemed prescription. Two approaches were used for defining treatment episodes that were named “gap+” and “gap–” and that respectively add and did not add the gap time at the end of a treatment episode. An outcome was simulated by using an exponential baseline hazard function with scale parameter λ = 0.5 and censoring at time t = 365 days. The incidence rate ratio for the simulated outcome between the two approaches was computed. Results The 100 patients were followed for a median time of 183 days (interquartile range, IQR 50–365 days) and a median time of 273 days (IQR 140–365 days), respectively using “gap–” and “gap+”. During the first 100 days of the follow‐up period, none of the patients was found to discontinue the treatment with the method “gap+” while 38 patients discontinued using the method “gap–”. The approach “gap+” exerted a higher incidence rate for the simulated outcome among the exposed (0.98 events/person‐years) when compared to the “gap–” (0.82 events/person‐years) during defined periods of continuous treatment use. When comparing the two approaches and using the method “gap–” as the reference group, the incidence rate ratio for the simulated outcome was 1.20 (95% confidence interval: CI 0.72–2.02) among the exposed. Conclusions This study showed that not adding the gap at the end of the treatment episodes leads to an overestimation of drug discontinuation, particularly at the beginning of the observational window, and an underestimation of the incidence rate of a hypothetical outcome during the period of exposure to the medication.