3C-05 - Impact of the lookback period, age and gender on the estimation of the incidence of diabetes and heart failure on medico-administrative databases – FIREFLY study

Background: French National Health Data System (SNDS) and other medico-administrative databases, widely used to estimate the incidence of diseases, do not contain the date of clinical diagnosis. Among patients identified with the disease during the inclusion period (i.e. prevalent patients), the incident patients are those without any disease-identifying criterion during a certain period prior to the index date, the lookback period (LP).

Objectives: The objective was to estimate the impact of the LP duration on the number of incident patients with diabetes or heart failure (HF).

Methods: Prevalent patients with diabetes or HF in 2021 were identified using published algorithms combining hospitalizations, long-term disease conditions or specific treatments in the 2/100th representative beneficiary sample (ESND) of the SNDS , which covers 99% of the French population. The number of incident patients with diabetes or HF was calculated for each LP by progressively adding 3 months of LP and compared to a "gold standard" of 10-year LP. It was estimated overall, according to age and sex and according to data source, i.e hospital data to identify patients treated in hospital or outpatient dispensing data to identify patients treated outside the hospital.

Results: In 2021, 86,019 prevalent patients with diabetes were identified, including 16,446 (19.1%) patients hospitalized with diabetes or 72,605 (84.4%) patients with antidiabetics dispensing. With a LP of one-year, the overestimation of diabetes incidence was 15.8%, and a 4.5-year LP was required to achieve an overestimation of ≤ 5%. This differed according to age, with a lower overestimation in children, reaching 5% with a LP of one year, compared with the adults where it was 13% for patients of 18-74 years old to 28% for patients aged over 75. To achieve an overestimation of ≤ 5%, a LP of 5 years was required for men and patients aged 18-to 74, and 6 years for women and patients aged over 75. 5 years was necessary to achieve an overestimation of ≤ 5% to identify new users of antidiabetics while more than 8.5 years were needed to identify new hospitalized cases with diabetes.
In comparison, the overestimation of HF incidence was 36.7% using a LP of one year, and 5 years and 9 months LP was required to have a lesser overestimation (≤ 5%). A period of 6 years was also necessary to identify newly hospitalized patients with HF.

Conclusions: These results show that, depending on the diseases being studied, selecting data source (hospital or outpatient treatment) and age or gender of patients may impact drastically the overestimation of incidence thereby reducing the required LP. Considering data minimization, this is important to have in mind when choosing LP duration and discussing incidence results.