Effect of lookback period length on disease incidence estimates in claims databases

Claims databases are used to examine diseases' and comorbidities' occurrences. However, the date of diagnosis is not available in those databases. Among the population of French adults at-risk of pneumococcal or influenza infections (i.e. patients with immunocompromised status or chronic medical conditions), three chronic diseases were selected for this study. The aim of the study was to determine the effect of the length of the lookback period on the incidence estimates of heart failure (HF), chronic liver disease, and HIV/AIDS using a claims database.

This retrospective longitudinal observational study was performed using secondary pseudonymized data from the French National Health Data System (SNDS). Prevalent patients with the three diseases in 2020 were identified using algorithms published by the National Health Insurance. The incident patients were those without any disease-identifying criterion during the lookback period of interest -which varied for the purpose of the study from 3 to 57 months. The overestimation of the incident population was computed for each lookback period length.

With a 1-year lookback period, the incidences of patients with HF, chronic liver disease, and HIV/AIDS were overestimated by 20.3 %, 19.2 %, and 12.0 %, respectively. The 5 % overestimation threshold was reached with a lookback period of 3.75 years, 2.5 years and 1.75 years, for HF, chronic liver disease, and HIV/AIDS, respectively.

The choice of the length of the lookback period for estimating incidence is an important methodological point. Depending on the disease and the length of the lookback period, the overestimation of incidence estimates varies.