(173) Short-term mortality in the Danish Better Health in Late Life Cohort - A study on cohort representativeness

Background: Cohort populations obtained from surveys aim to mirror the broader population (source population).
However, even with high response rates, the validity may be compromised by systematic differences in participation related to socioeconomic position (SEP) and chronic health conditions. The Danish Better Health in Late Life (BHLL) cohort was designed to facilitate pharmacoepidemiological studies. Linking to Danish registries enables extensive confounder control.

Objectives: We examined the representativeness of the cohort in terms of mortality compared with the general population. Additionally, we investigated the potential for achieving balance by incorporating covariates from health and social registries.

Methods: Among the 1,220,000 Danish residents aged 50-65 years, 25% were randomly invited to BHLL in the Fall 2021. Both respondents and the eligible general population were linked using a unique personal identifier to Danish registries. Data on covariates such as age, sex, recent hospitalization, Charlson Comorbidity Index score (CCI), prescription drug use, mortality, and SEP (education, income, employment status, cohabitation) were obtained. We calculated the one-year mortality (weighted on age and sex), and the crude and stepwise adjusted relative risk (RR) and 95% confidence intervals (CI) of one-year mortality using Poisson-regression. A RR=1 indicates balance (regarding mortality), while a deviation from 1 suggests residual or uncontrolled confounding in the model. We conducted stratified analyses by age, sex, CCI, recent hospitalization and drug-use.

Results: The BHLL cohort comprised 113,833 respondents (response rate 37%). Respondents differed from the general population, particularly in terms of higher education (42% vs. 34%), sex (females: 55% vs 50%), cohabitation (56% vs 50%), being employed (85% vs 79%), and income level (high: 36% vs 30%). The cohorts had similar age (median 58 vs 57 years), comorbidity (CCI=0: 84% vs. 83%), and drug use. One-year mortality was lower in the BHLL cohort than in the general population (0.31% vs 0.58%) yielding a crude RR of 0.54 (95%CI: 0.48-0.60). Fully adjusted RR was 0.71 (95%CI: 0.64-0.79). Improved representativeness was mainly achieved by including SEP in the model. The most representative stratum (RR closest to 1) was found in the group of 50-57 years old, with an adjusted RR at 0.81 (95%CI: 0.67-0.98) and among users of anti-psychotics with an adjusted RR at 0.83 (95%CI: 0.58-1.17).

Conclusions: Our findings showed a lower mortality rate among BHLL respondents than in the general population, suggesting presence of uncontrolled confounding. This confounding can be reduced, but not eliminated, by adjusting for SEP and health conditions, or by stratifying to subgroups.