Climate-driven mosquito-borne viral suitability index: measuring risk transmission of dengue, chikungunya and Zika in Mexico

Index P dynamics in three case studies

Before we present the trends of the index P and mosquito-borne diseases for the entire country, we pause to analyse in more detail three cities: Acapulco, Cancún and Mexico City. We focus on these three cities because they have important differences in climatic conditions, geographical location, arbovirus incidence and are subject to distinct and very high degrees of human migration patterns. These differences make these cities ideal for assessing the robustness of the index P in estimating transmission risks on a daily basis during 2000–2020.Footnote 2

Acapulco, on the east Pacific coast, has a tropical wet and dry climate characterised by high temperatures with minimal variation and variable levels of precipitation [35]. Cancún has a tropical savanna climate on the Yucatán Peninsula [36]. Both these cities are among the most important domestic and international tourism centres in the country, and have high dengue incidence, albeit with peaks in different seasons. Mexico City, the capital and most populated city in the country, has a subhumid mild climate. This city is a particularly interesting case to analyse because its high elevation, 2240 m above sea level, is above the elevation ceiling that typically allows the Aedes mosquitoes to proliferate [37]. Nonetheless, climate warming could over time place high-elevation cities at increased risk of dengue transmission; and in the Americas the Aedes have been found in other similar high-altitude areas of 2200 m [38].

For each of our three case studies, Fig. 7 displays the daily climate patterns (temperature, humidity, and precipitation), the distribution of the entomological priors and the estimated index P.Footnote 3 Following the literature, we assume the entomological priors of the mosquito lifespan and incubation period to be the same across all the three cases [18]. As expected, Acapulco and Cancún have consistently higher indices P than Mexico City’s (with averages of 1.35, 1.32 and 0.54, respectively). These indices P predict that if there was one female mosquito per human, Acapulco and Cancún would be more susceptible to outbreaks of mosquito-borne diseases than Mexico City, where the index is less than one for most of the year. Acapulco and Cancún present higher levels of humidity and temperature and are subject to more irregular trends in all climate conditions than Mexico City. These patterns suggest that the daily index P for Acapulco and Cancún have more irregularities in their seasonal patterns. In contrast, Mexico City displays a more stable seasonal pattern during 2000–2020. These risks of transmission are well in line with the epidemiological profiles of each city. Acapulco, Cancún and Mexico City have a markedly different incidence of dengue per 100,000 inhabitants, with an average of 12.06, 8.98, and 0.015 respectively during 2000–2020. In fact, Acapulco has one of the most severe and persistent dengue-incidence profiles in the country, whilst Cancún has in recent years increased from medium to high levels of dengue [39]. Mexico City’s low index P also corresponds with its low incidence of dengue, chikungunya and Zika, typical of local climate conditions of high-altitude areas, which are not conducive to the endemic presence of Aedes mosquitoes.

Fig. 7figure 7figure 7

Climate, entomological priors, index P and 95% confidence intervals for Mexico City, Acapulco, and Cancún 2000–2020

The timing of when the index P reaches its maximum can be used to determine the timing of the highest mosquito-borne disease transmission potential. For visual simplification, Fig. 8 shows the month for which the index P reached its peak across all the simulations for the years, 2000, 2005, 2010, 2015 and 2019. Over this sub-period analysed, the index P in Mexico City tended to peak in June, reflecting very marked seasonal behaviour. For Acapulco and Cancún, their indices P tended to reach their maximums in June and October, respectively. Nonetheless, the timing of when the index P peaked displayed a high degree of variance.

Fig. 8figure 8

Peak distribution of the index P

Another key feature of the MVSE R-package is that it offers a visual representation of the sensitivity of the estimated index P to changes in climate conditions. Figure 9 depicts the humidity and temperature on the x- and y-axis, respectively, for all the combinations in the climate data. The different colours in Fig. 9 represent different values of the index P. The dots that form a ring shape within the figure represent the average values of the climate data for each month. The floating circles with numbers ranging from one to twelve indicate what month the average values refer to. Figure 9 shows that suitability in Mexico City follows a clear and gradual trend along the months in a year, while Cancún presents abrupt changes reflecting lower stationarity. The lower stationarity hinders the identification of the month of highest transmission risk. Acapulco behaves somewhere in between the other two case studies.

Fig. 9figure 9

Each point represents a humidity-temperature combination recorded in the climate data used as input for the index estimation. Temperature is measured in degrees Celsius. The white dots over the black link mark the mean for each month, while the floating circles indicate which month it refers to

As mentioned earlier, the estimation of the index P relies on defining priors for eight parameters. To assess how robust the index P is to these priors, we performed sensitivity analysis. We changed the value of some of our initial priors, taken from Obolski et al. [18] for the prior of human incubation period (from mean = 5.8, sd = 1, to mean = 5.0, sd = 1), or the mosquito life expectancy (from mean = 12, sd = 2, to mean = 14, sd = 3), or both these priors simultaneously. We changed the value of these priors as there is also support in the literature for these parameters which have been used recently in the estimation of the index P for the Dominican Republic [15]. In our sensitivity analysis we found no statistically significant differences with respect to the set of priors we used originally, as shown in Fig. 13 in the Appendix. This figure shows the value of index P using our original priors and the value of the index P when changing the human incubation period (left-hand side panel), or the mosquito life expectancy (middle panel), or both these parameters simultaneously (right-hand side panel). Figa. 13 also shows the resulting 95% confidence intervals for the estimated indices P. In most cases, the only noticeable change, is a slight increase in the upper confidence interval, particularly when we change the mosquito life expectancy parameter.

Index P spatiotemporal characterisation across Mexico

In this section, we provide a broader picture of the distribution of the index P across all the Mexican territory. As mentioned earlier, automatic meteorological stations offer the key advantage of measuring on the ground local climate conditions daily. Our choice of using these automatic measurements comes at a cost. Unfortunately, not all regions in the country had an automatic meteorological station nearby during 2000–2010.Footnote 4 Nonetheless, the number of automatic stations sharply increased after 2010, achieving good national coverage during 2010–2020. For this reason, in this section and the next one, we restrict our analysis to 2010–2020. In Additional file 1: Table S1, we present the index P for all available automatic meteorological stations, aggregated at state level, during 2000–2020.

Figure 10a, shows the average index P across all the Mexican territory during 2010–2020. This figure depicts the values of the index P by quintiles. As mentioned earlier, the risk of transmission increases when the index P takes a value greater than one, and the risk decreases when the index P takes the value of less than one. The regions with the highest index P (index > 1.16, shown in red) are in the southeast (Tabasco and the Yucatán peninsula), the Pacific coast, and in some northern states (Sonora, Chihuahua and Coahuila). The intermediate values of the index P (index between 0.98 and 1.16, shown in yellow and orange) are located in the Tehuantepec Isthmus and in some northern states. The index P with its lowest value (index < 0.98, blue and green), hence with the lowest transmission potential, is found in the centre of the country and the peninsula of Baja California.

Fig. 10figure 10

Spatial–temporal characterisation of the index P in Mexico per pixel, during 2010–2020

The month where the index P reaches its maximum value is shown in Fig. 10b. It stands out that the maximum peak per month does not always occur in the same regions, as it is mainly influenced by temperature variability. The index P peaks in July in the centre and centre east. For most of the rest of the country, the index P reaches its maximum value in August or September, where the transition from summer to autumn begins.

Figure 11 shows the monthly average of the index P during 2010–2020 for selected months. During January, the southeast, Tehuantepec Isthmus and the coasts of Michoacán and Guerrero stand out with the highest transmission potential. In May, the highest transmission potential occurs in the northeast region. In July, the highest transmission risk shifts to the northwest of the country, standing out the states of Sonora and Chihuahua. In September, the highest transmission risk is for northeast states, particularly Coahuila, Nuevo León, and Tamaulipas, as well as Yucatán in the southeast.

Fig. 11figure 11

Mean of index P during 2010–2020 in January, May, July and September

Correlation between the index P and mosquito-borne diseases in Mexico

In addition to practicality, good suitability indices must also be correlated to the phenomenon they intend to measure. Thus, next we assess the degree of correlation between dengue, chikungunya and Zika incidence and the index P for all the 2469 municipalities in Mexico and for nine selected cities during 2010–2020. These nine cities have been selected to provide a granular view across the territory, given their varying levels of arboviral infections and differences in socio-economic characteristics. These cities are in the north of the country (Ciudad Mante, Mexicali, and Monterrey), centre (Mexico City), southeast (Campeche and Tuxtla Gutiérrez), and various coasts (Acapulco, Cancún, and Coatzacoalcos). The geographical location of these nine cities is shown in Fig. 10. For each of these nine cities we estimate the Pearson correlation between the index P and dengue, Zika and chikungunya during 2010–2020. Like Obolski et al. [18], for each city, we estimate its Pearson correlation coefficient between the average index P for each month during 2010–2020 and its monthly average incidence of each arboviral disease (measured in natural logarithm) during 2010–2020.

Figure 12 depicts for each of the nine cities its monthly average index P, its monthly incidence of dengue (measured in logarithm) during 2010–2020, as well as the Pearson correlation between these variables. This correlation index is positive and ranges between 0.25 (Cancún) and 0.86 (Campeche). The correlation is much higher for southern and coastal cities that tend to have a high incidence of dengue, with the exception of Cancún perhaps due its very high flow of international and domestic tourism. The correlation is lower for northern cities that typically have low levels of dengue (Mexicali and Monterrey). These results suggest there might be other relevant factors, such as population mobility, density and socio-economic characteristics that explain the low incidence of dengue in some cities, which the index P does not consider. Nonetheless, overall the correlation between the index P and dengue are similar to the results obtained by Obolski et al. [18] for several cities in Brazil.

Fig. 12figure 12

Correlation of index P and dengue incidence for selected nine cities. Monthly averages of the index P are shown in blue and average incidence is shown in the dotted pink line. Shaded areas correspond to the 95% confidence intervals of the index. Pearson’s correlation is shown in each subplot

To analyse further the transmission risk potential predicted by the index P, Figs. 14 and 15, in the Appendix, show the correlation between the index P and chikungunya and Zika for the nine selected cities since they appeared in the country in 2014 and 2015 respectively. The incidence of both diseases has rapidly declined and is more clustered in specific regions. However, in the cities where there is chikungunya or Zika or both, and where it is possible to estimate the correlation with the index P, the correlation is relatively strong. For chiungunya the correlation ranges from -0.44 (Ciudad Mante) to 0.86 (Tuxtla Gutiérrez). For Zika the correlation with the index P ranges from 0.01 (Monterrey) to 0.92 (Mexico City).

There is substantial variability in the presence of Zika and chikungunya across the nine cities as these diseases are not widespread but concentrated in certain parts of the country only. This partly explains why the correlation coefficient between the index P and the rate of Zika and chikungunya shows wider range than the correlation between the index P and dengue. Moreover, the sporadic and scant incidence of Zika and chikungunya provides us with insufficient power to detect a statistically significant correlation between the index P and these arboviruses for each of these nine case studies.

To provide a more global view, Table 1 presents the Pearson correlation between the index P, dengue, Zika and chikungunya for two different scenarios: one considering all the nine cities together, and another considering all the 2469 municipalities in the Mexican territory during 2010–2020. We find that the correlation between the index P and dengue is 0.46 when considering together the data from the nine case studies. This correlation is statistically significant, and higher than when estimating the correlation for all the 2469 municipalities in the country (0.29) as the selected nine case studies tend to have high incidence of dengue (with the exception of Mexico City).

Table 1 Correlation between Index P, dengue, Zika and chikungunya incidence for case study and all municipalities in the country 2010–2020

Table 1 also shows that the correlation between the index P and dengue for all the 2469 municipalities in the country (0.29) is of similar magnitude to that of Zika (0.23) and chikungunya (0.25). All these correlations between the index P, dengue, Zika and chikungunya are statistically significant, and sufficiently powered with large number of observations, when considering the data across all the municipalities in the country.

The nine selected cities in Mexico are: Acapulco, Campeche, Cancún, Ciudad Mante, Coatzacoalcos, Mexicali, Mexico City, Monterrey, and Tuxtla Gutiérrez. The number of observations is the number of areas considered multiplied by twelve (as the correlation considers the average rate of mosquito-borne disease for each of the 12 months over the period 2010–2020). Some areas do not report mosquito-borne diseases for some months. For these missing cases the number of observations is reduced, as it is particularly the case for Zika and Chikungunya.

Hurricane Manuel and index P

Our analysis suggests that the index P offers valuable information on the potential dynamics of mosquito-borne risk transmission during a given year or for a long-time series. Another possible application of the index P could be to assess changes in transmission potential due to sudden weather shocks such as hurricanes. Adult mosquitoes do not generally survive during the high wind speed associated with hurricanes. However, a disease outbreak of dengue, Zika, and chikungunya might follow as hurricanes might cause significant property damage and increase precipitation that makes it more likely for mosquitoes to breed [40,41,42]. To assess to what extent hurricanes affect the index P, Table 2 shows the changes in monthly average index P, temperature (Celsius), monthly average humidity (a percentage that ranges between 0 and 100), and the monthly average precipitation (in millimetres), associated with Hurricane Manuel that affected Mexico in September 2013. Manuel was the first eastern North Pacific tropical cyclone to make landfall in mainland Mexico, redevelop over water, and then become a hurricane. Manuel brought heavy rains and floods to large parts of the Pacific coast, resulting in 123 deaths and 4.2 billion US dollars in damage, with the biggest impacts in Guerrero [43]. Over 30,000 homes were damaged in that state alone, and 46 rivers overflowed. Table 2 shows the sharp increase in rainfall that Acapulco in Guerrero experienced during September 2013.Footnote 5 The index P, as a result, increased during September 2013. This increase in the index P also coincided with the rise of dengue incidence in Acapulco during that month. This evidence suggests that the index P reasonably predicts how climate changes can lead to changes in potential transmission. It is worth noting that the index P seems to be more sensitive to changes in temperature and humidity. For instance, in September 2012, Acapulco did not experience a hurricane in the previous year. Nonetheless, there were statistically significant higher levels of humidity and temperature reflected in a higher index P and higher dengue incidence than the ones experienced in September 2013 where Hurricane Manuel affected Guerrero.

Table 2 Climate conditions, Index P and dengue incidence in Acapulco during July-November 2012 and 2013

Our evidence suggests that the index P is a good tool to assess increased risk of transmission which could alert policymakers which months, seasons, and areas could be at increased risk of mosquito-borne diseases due to changes in climatic factors. However, some cities and months can have higher correlation between the index P and incidence of arbovirus disease that will not necessarily be expected in other periods or space. That is, although increases in the index P suggest a rise in the risk of mosquito-borne transmission, such an increase in risk of transmission is not necessarily linear. In Table 2, for instance, the index P increased from 2.23 to 2.24, that is 0.82%, between September and October of 2012. During that period the incidence of dengue increased by 79.4%. A year later, when hurricane Manuel hit Guerrero, the index P increased from 1.84 to 1.95 between September and October of 2013. This increase of 5.74% in the index P was reflected in a substantial increase in the incidence of dengue of 186.8%. The rise in dengue was not as high as the one we would have expected had the index P and risk transmission followed a linear relationship. There are many reasons for this finding. As mentioned earlier, the incidence of mosquito-borne diseases depends on more factors not considered in the index P such as population mobility, opportunities for the mosquito population to breed, deforestation, etc. Still, the index P provides a good tool to assess increased risk of mosquito-borne disease transmission.

留言 (0)

沒有登入
gif