Ability of ecological deprivation indices to measure social inequalities in a French cohort

Background: Despite the increasing interest in place effect to explain health inequalities, there is currently no consensus on which kind of area-based socioeconomic measures researchers should use to assess neighborhood socioeconomic position (SEP). The study aimed to evaluate the reliability of different area-based deprivation indices (DIs) in capturing socioeconomic residential conditions of French elderly women cohort. Methods: We assessed area-based SEP using 3 DIs: Townsend Index, French European Deprivation Index (FEDI) and French Deprivation index (FDep), among women from E3N (Etude epidemiologique aupres des femmes de la Mutuelle Generale de l'Education Nationale). DIs were derived from the 2009 French census at IRIS level (smallest geographical units in France). Educational level was used to evaluate individual-SEP. To evaluate external validity of the 3 DIs, associations between two well-established socially patterned outcomes among French elderly women (smoking and overweight) and SEP, were compared. Odd ratios were computed with generalized estimating equations to control for clustering effects from participants within the same IRIS. Results: The analysis was performed among 63,888 women (aged 64, 47% ever smokers and 30% overweight). Substantial agreement was observed between the two French DIs (Kappa coefficient = 0.61) and between Townsend and FEDI (0.74) and fair agreement between Townsend and FDep (0.21). As expected among French elderly women, those with lower educational level were significantly less prone to be ever smoker (Low vs. High; OR [95% CI] = 0.43 [0.40-0.46]) and more prone to being overweight (1.89 [1.77-2.01]) than women higher educated. FDep showed expected associations at area-level for both smoking (most deprived vs. least deprived quintile; 0.77 [0.73-0.81]) and overweight (1.52 [1.44-1.62]). For FEDI opposite associations with smoking (1.13 [1.07-1.19]) and expected association with overweight (1.20 [1.13-1.28]) were observed. Townsend showed opposite associations to those expected for both smoking and overweight (1.51 [1.43-1.59]; 0.93 [0.88-0.99], respectively). Conclusion: FDep seemed reliable to capture socioeconomic residential conditions of the E3N women, more educated in average than general French population. Results varied strongly according to the DI with unexpected results for some of them, which suggested the importance to test external validity before studying social disparities in health in specific populations.