Analyzing the effects of drought at different time scales on cause-specific mortality in South Africa

South Africa (SA) is highly vulnerable to the effects of drought on the environment, economy, and society. However, its effect on human health remains unclear. Understanding the mortality risk associated with different types of droughts in different population groups and by specific causes would help clarify the potential mechanisms involved. The study aims to comprehensively assess the effect of droughts of varying time scales on cause-specific mortality (all; infectious and parasitic; endocrine, nutritional, and metabolic; cardiovascular; respiratory) in SA (from 2009-2016) and identify more vulnerable profiles based on sex and age. We also evaluated the urbanicity and district-level socioeconomic deprivation as potential risk modifiers. We used a two-stage time-series study design, with the weekly standardized precipitation-evapotranspiration index (SPEI) calculated at 1, 6, 12, and 15 months of accumulation to identify droughts of different duration (SPEI1, 6, 12, 15, respectively). We applied a quasi-Poisson regression adjusted by mean temperature to assess the association between each type of drought and weekly mortality in all district municipalities of SA, and then pooled the estimates in a meta-regression model. We reported relative risks (RRs) for one unit increase of drought severity. Overall, we found a positive association between droughts (regardless the time scale) and all causes of death analyzed. The strongest associations were found for the drought events more prolonged (RR [95%CI]: 1.027 [1.018, 1.036] (SPEI1); 1.035 [1.021, 1.050] (SPEI6); 1.033 [1.008, 1.058] (SPEI12); 1.098 [1.068, 1.129] (SPEI15)) and respiratory mortality (RRs varied from 1.037 [1.021, 1.053] (SPEI1) to 1.189 [1.14, 1.241] (SPEI15)). An indication of greater vulnerability was found in younger adults for the shortest droughts, in older adults for medium-term and long-term droughts, and children for very long-term droughts. However, differences were not significant. Further evidence of the relevance of urbanicity and demographic and socioeconomic conditions as potential risk modifiers is needed.