The seasonality of cholera in sub-Saharan Africa: a statistical modelling study

Abstract

Background: Cholera remains a major threat in sub-Saharan Africa (SSA), where some of the highest case-fatality rates are reported. Knowing in what months and where cholera tends to occur across the continent could aid in improving efforts to eliminate cholera as a public health concern. However, largely due to the absence of unified large-scale datasets, no continent-wide estimates exist. In this study, we aimed to estimate cholera seasonality across SSA and explore the correlation between hydroclimatic variables and cholera seasonality.

Methods: Using the global cholera database of the Global Task Force on Cholera Control, we developed statistical models to synthesise data across spatial and temporal scales to infer the seasonality of excess (defined as incidence higher than the 2010-16 mean incidence rate) suspected cholera occurrence in SSA. We developed a Bayesian statistical model to infer the monthly risk of excess cholera at the first and second administrative levels. Seasonality patterns were then grouped into spatial clusters. Finally, we studied the association between seasonality estimates and hydroclimatic variables (mean monthly fraction of area flooded, mean monthly air temperature, and cumulative monthly precipitation).

Findings: 24 (71%) of the 34 countries studied had seasonal patterns of excess cholera risk, corresponding to approximately 86% of the SSA population. 12 (50%) of these 24 countries also had subnational differences in seasonality patterns, with strong differences in seasonality strength between regions. Seasonality patterns clustered into two macroregions (west Africa and the Sahel vs eastern and southern Africa), which were composed of subregional clusters with varying degrees of seasonality. Exploratory association analysis found most consistent and positive correlations between cholera seasonality and precipitation and, to a lesser extent, between cholera seasonality and temperature and flooding.

Interpretation: Widespread cholera seasonality in SSA offers opportunities for intervention planning. Further studies are needed to study the association between cholera and climate.

Funding: US National Aeronautics and Space Administration Applied Sciences Program and the Bill & Melinda Gates Foundation.

Figure 1

Seasonality patterns and strength of cholera occurrence in sub-Saharan Africa (A) Estimates of monthly odds ratios of excess cholera occurrence, defined as monthly incidence greater than the 2010–16 baseline; estimates are given for the best-performing model at the first or second administrative levels, depending on data availability (appendix pp 1–3), as opposed to countries for which the non-seasonal null model was selected (dark grey) and countries that were excluded from analyses due to absence of data (light grey); for illustrative purposes, the odds ratios colour scale was truncated at ±2·5 on the log scale (0·08–12·00 on the natural scale); country-level odds ratios time-series are shown in the appendix (p 11). (B) Strength of cholera seasonality quantified as the proportion of risk occurring in the 3-month window around the seasonal cholera peak. Administrative units for which the majority (at least 50%) of the probability of excess cholera occurred in the 3-month window around the peak are highlighted (black borders).

Figure 2

Seasonality grouping in sub-Saharan Africa (A) Map of cholera seasonality clusters (top) and proportion of population in each cluster for different categories of mean annual cholera incidence (bottom); clustering results are shown for models with three (thick borders) and five (colour fill) clusters, along with countries for which seasonality was not retained (dark grey) and excluded because of no data (light grey); macroregions were outlined with use of the convex hull of the corresponding administrative units; these macroregions were based on a three-cluster model where Sudan formed its own cluster. (B) Seasonality of cholera excess risk in each cluster for each administrative unit and overall trend estimated by a generalised additive model of odds ratio as a function of the month of the year.

Figure 3

Excess cholera seasonality and climatology Maps show the Spearman correlation between odds ratios of excess cholera and the mean monthly values of hydroclimatic variables at lags of 0 months, 1 month, and 2 months. Hydroclimatic variables include the mean monthly flooded area, the mean monthly air temperature, and the monthly cumulative precipitation. Correlation is shown for significant coefficients (p value <0·05, full colour, black border), with non-significant values (p value >0·05) given for indication (transparent, no border). Associations with other hydroclimatic variables are presented in the appendix (p 18).