As of March 2017, GBS had been observed to be temporally associated with Zika outbreaks in 23 countries [1]. We aimed to estimate the risk of GBS related to ZIKV infection given limited ecological-scale data. Although GBS is severe, generally recognizable, and recommendations have been made to standardize case definitions and diagnostic criteria [21], reporting likely varies across countries. To make estimates based on data from 11 different locations, we therefore assumed that, while true ZIKV-associated GBS risk was likely similar across locations, location-specific variability in risk and reporting would lead to variations in the observed number of cases, especially in relation to the observed number of suspect cases.
We evaluated all of our estimates by sequentially removing data from individual locations and obtaining new model estimates. This evaluation revealed that our estimates were most sensitive to the data from French Polynesia. This did not come as a surprise since French Polynesia is the only location with both an estimate of infection incidence and a non-zero number of GBS cases. The Yap outbreak also had an estimate of infection incidence but without any accompanying GBS cases detected. The data from Yap was therefore helpful in estimating the upper bound of reported GBS risk following ZIKV infection, but not in further refining that risk.
Our overall estimate for the risk of reported GBS given ZIKV infection was 2.0 (95% CrI 0.5–4.5) GBS reported cases per 10,000 ZIKV infections, close to the point estimate of 2.4 GBS cases per 10,000 ZIKV infections estimated using only data from French Polynesia [9]. The estimate for French Polynesia was obtained by dividing the 42 reported GBS cases by the estimated total number of infections using an incidence of infection of 66%, as estimated in a serosurvey, and the population size of French Polynesia [9]. When we removed the French Polynesia data, our estimate was somewhat lower, underscoring the importance of the infection prevalence estimates and of considering risk across various locations. In addition, since our model does not account for the correlation between GBS risk and a number of factors such as age and sex [4], we expect discrepancies between our risk estimate and what is observed in realistic public health settings. As data with more detail and from more locations become available, these estimates can be further refined, identifying specific risk groups (e.g., older men [4]) and characterizing the relationship between time of ZIKV infection and GBS onset. Beyond Zika, the current estimates of risk are also of a similar magnitude to the estimated risk of GBS caused by Campylobacter jejuni infection – an infection that is well known to be associated with GBS – which is estimated to be between 2.5 and 6.5 GBS cases per 10,000 infections [22].
For each location analyzed, the estimated number of ZIKV-associated reported GBS cases was several times higher than the expected number of reported baseline GBS cases, though the magnitude of the difference varied substantially based on the estimated ZIKV infection incidence. With a high incidence of infection, as in French Polynesia and Yap, the estimated risk of GBS was 25 to 43 times higher during the outbreaks than under typical baseline conditions. For each other location, estimates of incidence over the respective study period were lower (2 to 17%, on average) and indicated significant uncertainty and variability across locations (Table 2 and Additional file 1: Figure S1). These lower estimated incidences may indicate differing epidemiology, geographical heterogeneity within the location, different population densities, varied reporting, and the fact that the data for these locations only represents part of the time period over which the outbreaks occurred. In these locations, Zika-associated reported GBS risk was 1–6 times higher than baseline risk over the time periods of the collected data.
Our mean estimates for the relationship between reported suspect Zika cases and infection incidence ranged from 0.01 to 0.18 suspect cases per ZIKV infection. Differences across locations reflect the relative numbers of GBS and suspect Zika cases reported. For example, the Dominican Republic, Venezuela, and El Salvador all had relatively high numbers of reported GBS cases per suspect Zika case. These differences may be explained by relatively low reporting of suspect Zika cases, high reporting of suspect GBS cases, or a combination of the two. Variation may reflect differences in care-seeking behavior, the availability of public health services, and reporting practices for both Zika and GBS. Across locations, we estimated that 0.11 (95% CrI 0.01–0.24) suspect cases were reported per ZIKV infection, encompassing point estimates from the previous outbreaks. For the outbreak in Yap, only 185 individuals sought care with Zika symptoms even though there were an estimated 5005 individuals infected, such that the number of reported suspect cases represented only approximately 3.7% of the total number of infections [13]. In French Polynesia, over 31,000 suspect cases were reported [23], with an estimated 185,000 infections (based on 66% seroprevalence and a population of approximately 280,000 people [19]). In this case, reported suspect cases represent approximately 17% of the total number of infections. Others have estimated a 94% incidence of infection in that outbreak, in which case suspect cases would be approximately 11.5% of the number of infections [19].
This analysis has a number of limitations, some of which have been noted above. The first is the great variability that exists in reporting practices of Zika and GBS cases across locations and likely over time as the outbreaks unfolded. Furthermore, misclassification of ZIKV infections, or of other illnesses as Zika, certainly occurred in many countries in the Americas with ZIKV outbreaks. An additional limitation was the lack of weekly Zika and GBS case data for most locations. This type of data would have provided many more data points with which to characterize the relationship between ZIKV infections and GBS risk. The limited data also did not allow us to assess the association between GBS risk and other variables such as age and sex. Finally, as noted above, the only outbreaks with data on infection incidence were Yap and French Polynesia and these were also the only two locations with data over the full outbreak. For each other location, estimates of incidence over the respective study period indicated significant uncertainty and variability across locations (Table 2 and Additional file 1: Figure S1).
Without serological data to confirm infection prevalence, it is difficult to identify the reasons for difference in reporting or risk across locations. For example, the relationship between reported Zika and GBS cases in Bahia is clear, but both may have been substantially underreported given the early initiation of that outbreak, leading to a low estimate of infection prevalence. Yet, the number of cases that may have been missed cannot be directly assessed without better estimates of overall infection prevalence. If population-level serological data become available, the data could be directly incorporated in the model, as was done for Yap and French Polynesia, allowing the estimates to be informed by these additional data and providing a further pathway to evaluate the estimates.