Background/Aim:
Temporal autocorrelations of health outcomes occur because the data relate to largely
the same population in consecutive time periods. In addition, spatial patterns occur
either because of neighborhood effect or the omission of spatially patterned risk
factors for the response variable. Aim here is to (a) estimate the risk of depression
at the district level using data of the population- based Heinz Nixdorf Recall Study
(HNRS), (b) analyze temporal trends and spatial variations in the model and (c) further
test the impact of greenness on the risk of depression.
Methods:
Data of 4,814 participants (51% women) are aggregated in the 108 non-overlapping districts
(spatial unit) of Bochum, Essen and Mülheim, with known adjacency structure. Repeated
measurements of depressive symptoms (CES-D, n = 9 within 12 years) were used. Greenness
is defined as mean normalized difference vegetation index at the district level. We
computed spatio-temporal Poisson models to estimate the risk of depression in each
district accounting for covariate effects (greenness, age, socio-economic risk factors).
The spatio-temporal autocorrelation is modelled by the latent random effect, using
Conditional Autoregressive (CAR) type prior distributions and spatio-temporal extensions
thereof.
Results:
The results show low spatial correlation in the data after adjusting for covariate
effects, while the temporal correlation is very strong. The risk of depression for
1 unit increase in green is 0.90 [0.85,0.96] for the basic model, with no change when
spatial residual variation is considered. The effect of green is 0.98 for the fully
adjusted model, both for the model that considers and neglects spatial effects 95%
CI [0.91, 1.06].
Conclusion:
The low spatial correlation at the district level is indicative for neglecting it.
Some covariate effects may explain spatial variation in the model.
