Could the Recent Zika Epidemic Have Been Predicted?

Abstract

Given knowledge at the time, the recent 2015-2016 zika virus (ZIKV) epidemic probably could not have been predicted. Without the prior knowledge of ZIKV being already present in South America, and given the lack of understanding of key epidemiologic processes and long-term records of ZIKV cases in the continent, the best related prediction could be carried out for the potential risk of a generic Aedes-borne disease epidemic. Here we use a recently published two-vector basic reproduction number model to assess the predictability of the conditions conducive to epidemics of diseases like zika, chikungunya, or dengue, transmitted by the independent or concurrent presence of Aedes aegypti and Aedes albopictus. We compare the potential risk of transmission forcing the model with the observed climate and with state-of-the-art operational forecasts from the North American Multi Model Ensemble (NMME), finding that the predictive skill of this new seasonal forecast system is highest for multiple countries in Latin America and the Caribbean during the December-February and March-May seasons, and slightly lower-but still of potential use to decision-makers-for the rest of the year. In particular, we find that above-normal suitable conditions for the occurrence of the zika epidemic at the beginning of 2015 could have been successfully predicted at least 1 month in advance for several zika hotspots, and in particular for Northeast Brazil: the heart of the epidemic. Nonetheless, the initiation and spread of an epidemic depends on the effect of multiple factors beyond climate conditions, and thus this type of approach must be considered as a guide and not as a formal predictive tool of vector-borne epidemics.

Keywords: Aedes-borne diseases; R0 model; chikungunya; climate; dengue; predictability; zika.

Figure 1

Daily vector mortality rate as a function of mean temperature (in Celsius).

Figure 2

Observed climatology of R₀ considering all months in the period 1982–2010. Only R₀ > 1 values are plotted. There is no data over the oceans.

Figure 3

Spatial evolution of standardized R₀ yearly anomalies for (A) 2013, (B) 2014, and (C) 2015. (D) Average evolution of standardized R₀ anomalies (in units of standard deviations, s) for Latin America and the Caribbean [domain in panel (A)] for the 1950–2015 period. Black empty curve and filled curve show the raw and linearly detrended standardized anomalies, respectively. A 12-month running average filter was applied to both curves to better capture the inter-annual variability. There is no data over the oceans.

Figure 4

2AFC skill score for the seasonal forecast system for each one of the four seasons selected: (A) DJF, (B) MAM, (C) JJA and (D) SON. Units in %. The 2AFC score is an indication of how often the forecasts are correct; it also measures how well the system can distinguish between the above-normal, normal, and below-normal categories.

Figure 5

(A) Observed terciles (above normal, normal, below normal) for the basic reproduction number (R₀), computed using observed climate data for DJF 2014–2015 and the model presented in section Two-Vector One-Host Ento-Epidemiological Model. (B) Forecast probabilities (in %) for R₀ for the same DJF season, computed using predicted climate data, the vector model presented in Section Two-Vector One-Host Ento-Epidemiological Model and the probabilistic Principal Component Regression model described in Section Data and Methods.