Cholera epidemic in Yemen, 2016-18: an analysis of surveillance data

Abstract

Background: In war-torn Yemen, reports of confirmed cholera started in late September, 2016. The disease continues to plague Yemen today in what has become the largest documented cholera epidemic of modern times. We aimed to describe the key epidemiological features of this epidemic, including the drivers of cholera transmission during the outbreak.

Methods: The Yemen Health Authorities set up a national cholera surveillance system to collect information on suspected cholera cases presenting at health facilities. Individual variables included symptom onset date, age, severity of dehydration, and rapid diagnostic test result. Suspected cholera cases were confirmed by culture, and a subset of samples had additional phenotypic and genotypic analysis. We first conducted descriptive analyses at national and governorate levels. We divided the epidemic into three time periods: the first wave (Sept 28, 2016, to April 23, 2017), the increasing phase of the second wave (April 24, 2017, to July 2, 2017), and the decreasing phase of the second wave (July 3, 2017, to March 12, 2018). We reconstructed the changes in cholera transmission over time by estimating the instantaneous reproduction number, R_t. Finally, we estimated the association between rainfall and the daily cholera incidence during the increasing phase of the second epidemic wave by fitting a spatiotemporal regression model.

Findings: From Sept 28, 2016, to March 12, 2018, 1 103 683 suspected cholera cases (attack rate 3·69%) and 2385 deaths (case fatality risk 0·22%) were reported countrywide. The epidemic consisted of two distinct waves with a surge in transmission in May, 2017, corresponding to a median R_t of more than 2 in 13 of 23 governorates. Microbiological analyses suggested that the same Vibrio cholerae O1 Ogawa strain circulated in both waves. We found a positive, non-linear, association between weekly rainfall and suspected cholera incidence in the following 10 days; the relative risk of cholera after a weekly rainfall of 25 mm was 1·42 (95% CI 1·31-1·55) compared with a week without rain.

Interpretation: Our analysis suggests that the small first cholera epidemic wave seeded cholera across Yemen during the dry season. When the rains returned in April, 2017, they triggered widespread cholera transmission that led to the large second wave. These results suggest that cholera could resurge during the ongoing 2018 rainy season if transmission remains active. Therefore, health authorities and partners should immediately enhance current control efforts to mitigate the risk of a new cholera epidemic wave in Yemen.

Funding: Health Authorities of Yemen, WHO, and Médecins Sans Frontières.

Figure 1

Weekly time series of key cholera indicators for Yemen between Sept 28, 2016, and March 12, 2018 (A) New suspected cholera cases. (B) Case fatality risk (CFR; number of deaths divided by number of suspected cases). (C) Proportion of severely dehydrated patients. (D) Proportion of cases in children younger than 5 years of age. (E) Proportion of female cases. (F) Percentage of suspected cases with a rapid diagnostic test (RDT; pink bars) and percentage positive (blue line). (G) Proportion of cases tested for culture confirmation (per 1000, pink bars) and percentage positive (purple line). Shaded areas correspond to exact binomial 95% CIs for proportions. The Ramadan period (May 26–June 24, 2017) is indicated by a grey rectangle. The first vertical dashed line defines the end of the first wave/start of the increasing phase of the second wave and the second vertical dashed line indicates the end of the increasing phase/start of the decreasing phase of the second wave of the epidemic.

Figure 2

Spatial distribution of suspected cholera cases (A) Attack rate by district between Sept 28, 2016, and March 12, 2018. (B) Attack rate ratio by district for the first wave (Sept 28, 2016, to April 23, 2017) and increasing phase (April 24 to July 2, 2017) and decreasing phase (July 3, 2017, to March 12, 2018) of the second wave of the epidemic.

Figure 3

Daily time series of incidence, reproduction number, and rainfall by governorate between July 1, 2016, and March 12, 2018 (A) National incidence. (B) Contribution of each governorate to the national incidence. (C) Time-varying instantaneous reproduction number R_t represented by the mean estimate for the country (black line) and 95% credible interval for each governorate (shaded areas). (D) Country-level rainfall (mm per day). (E) Contribution of each governorate to the country rainfall. To obtain meaningful rainfall time series for comparison with cholera incidence time series at the national and governorate levels, we used a weighted mean of the district level rainfall time series, with daily weights proportional to the number of cases reported in each district over the following 2 weeks. To smooth the high level of noise in the daily reporting of suspected cases, we performed a rolling average with a 5-day time window on both the incidence (A) and reproduction number (C) time series. The Ramadan period (May 26–June 24, 2017) is indicated by a grey rectangle.

Figure 4

Detailed analysis of the effect of rainfall on cholera incidence during the increasing phase of the second epidemic wave (April 15–June 24, 2017) (A) Daily incidence and (C) accumulated rainfall during the 7 previous days (AR7D) in mm. Solid lines represent the day-wise median over all districts. Dark and light shaded areas represent the IQR and 95% quantile intervals (centred on the median), respectively. The Ramadan period (May 26–June 24, 2017) is indicated by a grey rectangle. (B) Relative risk (the ratio of cholera risk for individuals, cumulated over 10 days after a given AR7D exposure, to the risk when unexposed). Shaded area represents 95% CI. (D) AR7D distribution for all districts and days. The proportion for 0 mm is equal to 46% (omitted for the sake of visibility). (E) Mean daily relative risk attributable to rainfall during the rainy season (districts with no cases reported are in grey). For each district, the baseline risk corresponds to a typical day following a week with no rain. See appendix p 10 for a map of the highest daily relative risk.