Introduction

Snakebite envenoming is a neglected tropical disease which mainly affects marginalized people living in rural areas of the tropics. Snakebites can result in mild, moderate or severe envenomation, psychological disturbances, long-term disability and deaths. Annually, 4.5–5.4 million people are bitten by snakes globally, with 1.8–2.7 million envenomings, and 81 000–138 000 deaths [1,2]. Snakebite incidence demonstrates both spatial variation and seasonal patterns in many countries [3,4], but there has been no previous in-depth evaluations of spatiotemporal patterns of snakebite risk at a country level.

Sri Lanka is an agricultural tropical island in the Indian Ocean, extending from 5°55′ to 9°51′N and from 79°42′ to 81°53′E. The geography varies from low flat plains (mean elevation 228 meters) to South-Central highland regions (mean elevation 2400 meters). Annual rainfall varies from less than 1000 mm in some areas to over 4500 mm in others, and the country is divided into three climate zones based on rainfall; wet (South-West and Central parts), dry (Northern and South-Eastern parts) and intermediate (the areas in between). Central, southern and western parts receive rain mainly from Southwest monsoon (May to September) and northern and eastern parts receive from Northeast monsoon (November to February) [5,6].

Sri Lanka has a high snakebite incidence [7]. The country has over 100 snake species. Six are medically important; Naja naja (cobra), Bungarus ceylonicus (Ceylon krait), Bungarus caeruleus (Common krait), Daboia russelii (Russell’s viper), Echis carinatus (Saw scaled viper) and Hypnale hypnale (Hump nosed viper). Naja naja and Daboia russelii are widely found throughout the country. Hypnale hypnale can also be found throughout the country, particularly in the wet zone in rubber, tea, coconut and cocoa plantations, and are responsible for the majority of bites in the country. Bungarus caeruleus are found in the dry zone and Bungarus ceylonicus in wet mountains. Saw scaled vipers are confined to the dry arid zones of the Northern and Eastern parts of the country [8–10].

There is a clear geographical variation in the annual snakebite incidence in Sri Lanka [9]. Snakebite incidence rates have also shown seasonal patterns [10]. Therefore, snakebite is likely to demonstrate spatiotemporal variations in Sri Lanka and, to date, no study has undertaken an in-depth evaluation of spatiotemporal patterns of snakebite in the country. Identifying the high-risk geographical areas with their seasonal patterns will help to implement targeted preventive measures and allocation of resources, including antivenom, to manage the snakebite burden in Sri Lanka [11,12]. The aim of this work was to understand the spatiotemporal dynamics of snakebite in Sri Lanka in order to identify both persistent and time-varying snakebite hotspots.

Methods

Results

Snakebite incidence

Different combinations of spatial variables (elevation from sea level, population density, percentage of agricultural workers in the cluster, average rainfall and temperature) appeared to be important in different provinces. Differences in harmonic terms (i.e. sine and cosine terms) showed different annual, biannual, triannual and quadrennial seasonal snakebite patterns operating in different provinces. Estimates of the recall adjustment factor corresponded to under-reporting of bites in seven of the nine provinces and two provinces where over-reporting was estimated. The two provinces in which the estimated recall factor corresponded to over-reporting had the highest and fourth highest snakebite incidences. The fitted spatiotemporal model is summarized in Table 1 (details of the parameter estimates of the fitted model is provided in S1 Table). The goodness of fit tests (S3 Appendix) and observed versus predicted snakebite incidence for the survey sample (S4 Appendix) did not show evidence of over-dispersion relative to the assumed Poisson process model.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Spatiotemporal model for snakebites.

https://doi.org/10.1371/journal.pntd.0009447.t001

The spatiotemporal variation in estimated monthly snakebite incidence in the whole country is shown in Fig 1. Snakebite hotspots (i.e. high-risk areas) were mainly scattered in North-Central, North-Western, South-Western and Eastern parts of the country. Most hotspots showed biannual seasonal patterns except in the South-Western inlands where triannual seasonal patterns were noticed. The hotspots were scattered over the country during January. These hotspots then developed along the North-Western and South-Western inlands during February. North-Western hotspots migrated to North-Central regions during April. South-Western hotspots geographically expanded until May. The country as a whole had a relatively low snakebite risk during June and July. In August, hotspots appeared as a band circling the central mountain regions of the country and faded away during the following two months. Thereafter, hotspots appeared in the North-Central inlands in November and then migrated to South-Western inlands during December. Hotspots in the Eastern parts appeared during September and December.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Monthly snakebite incidence in Sri Lanka (bites per 100 000 people).

https://doi.org/10.1371/journal.pntd.0009447.g001

Average monthly snakebite incidence across all the provinces in Sri Lanka was 39 (95% CI: 38 – 40) bites per 100 000 people per month. This translates into 110 000 (95% CI: 107 500 – 112 500) bites per year in the whole country. The probability of exceeding 39 bites per 100 000 people in each month at any given location is shown in Fig 2. The red islands indicate the most likely places to exceed the average monthly incidence in Sri Lanka. In these areas, the probability of observing more than 39 bites per 100 000 people is at least 0.95. The blue areas have the lowest probability of exceeding the average monthly incidence. In these areas, the probability of observing more than 39 bites per 100 000 people is at most 0.05. Map areas with probability close to 0.5 indicate high uncertainty.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Maps of exceedance probability of more than 39 snakebites per 100 000 people per month.

https://doi.org/10.1371/journal.pntd.0009447.g002

Envenoming bite incidence

Different provinces showed different combinations of spatial variables (elevation from sea level, population density, percentage of agricultural workers in the cluster, average rainfall and temperature). The harmonic terms (i.e. sine and cosine terms) showed differences in provinces reflecting different annual, biannual, triannual and quadrennial seasonal snakebite patterns. Recall adjustment indicated under-reporting and over-reporting of bites in different provinces. The provinces with over-reporting had either high envenoming incidences or were located adjacent to provinces with high envenoming incidences. The fitted spatiotemporal model is summarized in Table 2 (details of the parameter estimates of the fitted model is provided in S2 Table). The goodness of fit tests (S5 Appendix) and observed versus predicted snakebite incidence for the survey sample (S6 Appendix) did not show evidence of over-dispersion relative to the assumed Poisson process model.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Spatiotemporal model for envenoming bites.

https://doi.org/10.1371/journal.pntd.0009447.t002

The spatiotemporal variation in monthly envenoming incidence in Sri Lanka is shown in Fig 3. Envenoming hotspots were mainly confined to the North-Central, Eastern and South-Western inlands of the country. Hotspots in North-Central parts showed triannual seasonal patterns that geographically expanded during March to May, July to September and November to December. Hotspots in South-Eastern regions appeared in June and then expanded towards South-Eastern regions until November. Eastern parts of the country showed high envenoming risk throughout the year.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Monthly envenoming bite incidence in Sri Lanka (bites per 100 000 people).

https://doi.org/10.1371/journal.pntd.0009447.g003

Average monthly envenoming across all the provinces in Sri Lanka was 19 (95% CI: 13 – 30) envenomings per 100 000 people. This translates into 45 000 (95% CI: 32 000 – 73 000) envenomings in a year for the whole country. The probability of exceeding 19 bites per 100 000 people (i.e. the average monthly envenoming) in each month at any given location is shown in Fig 4. The red islands indicate the places most likely to exceed the national average monthly envenoming bite incidence. In these areas, the probability of observing more than 19 envenomings per 100 000 people is at least 0.95. The blue areas have the lowest probability to exceed the national average monthly incidence. In these areas the probability of observing more than 19 envenomings per 100 000 people is at most 0.05. Areas with probability close to 0.5 indicates areas of high uncertainty. Here, the North-Western province shows intermediate risk compared to surrounding provinces and appears as a distinct block. This is a by-product of the separate parameter estimation in each province.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Maps of exceedance probability of more than 19 envenoming bites per 100 000 people per month.

https://doi.org/10.1371/journal.pntd.0009447.g004

Discussion

Our data showed complex spatiotemporal patterns for both overall snakebites and envenomings in Sri Lanka. Snakebite hotspots were mainly observed in North-Central, North-Western, South-Western and Eastern parts of the country. These hotspots showed biannual seasonal patterns except for South-Western inlands, which had triannual patterns. Envenoming bite hotspots were mainly confined to the North-Central, Eastern and South-Western inlands of the country. Hotspots in North-Central regions showed triannual seasonal patterns. Hotspots in South-Eastern regions had annual seasonality. Hotspots in Eastern regions remained throughout the year. Western and Central provinces did not show clear spatiotemporal variation in the pattern of snakebites or envenomings. This could be due to relatively low bite incidences masking such variation. The majority of recorded bites in the South-Western region were non-envenoming. These patterns are compatible with previously reported local [15–17] and regional data [18–22]. The differences between overall snakebites and envenoming bites are likely due to the differences in the distribution of snake species [23–28], their biological behaviour [29–33], rainfall patterns, operating agricultural seasons [6, 34] and human behaviour [35,36]. The results of our work can be used to find solutions to problems such as snakebite using a bioethical perspective. Such an approach would include discussion among the general public, academia and decision makers to identify vulnerabilities of local communities and to reach consensus on a roadmap to minimize snakebites.

We developed our spatiotemporal models at the provincial level. According to our estimates, national average snakebite and envenoming incidences were 39 and 19 per 100 000 per month, respectively. This translates into 110 000 snakebites and 45 000 envenomings in a year. The previous purely spatial analysis estimated the burden in the country to be 80 500 snakebites and 30 500 envenoming bites per year [9], whereas a purely temporal analysis estimated 119 000 snake bites per year [10]. The underestimation from a purely spatial model is likely to be due to the lack of adjustment for seasonal variations and recall bias. The overestimation from a purely temporal model is likely to be due to lack of adjustment for the spatial variation and subnational level seasonal variations. This highlights the importance of spatiotemporal patterns when estimating disease burdens in countries.

We used harmonic mathematical functions (sine and cosine terms) to represent the seasonally varying weather (rainfall, temperature and humidity), agricultural practices and other seasonally varying phenomena in Sri Lanka [10]. In the present analysis, these harmonic functions alone were able to explain the temporal variations in snakebites and envenoming bites. None of the weather indices improved the model fits. However, other studies have indicated associations between snakebite incidence and weather indices such as rainfall [37–39]. A possible explanation for this apparent discrepancy is that harmonic functions in our models act as proxies for seasonal variation in weather-related and any other unmeasured seasonally varying correlates of snakebite incidence.

Our spatial explanatory variables (elevation from sea level, population density, percentage of agricultural workers in a cluster, long-term averaged rainfall and temperature) showed complex patterns of association with snakebite and envenoming incidences in different parts of the country. In general, incidence dropped when the population density increased, as snake habitats are less likely to coexist in population-dense areas [4]. Percentage of agricultural workers living in communities showed positive association with bite incidence, reflecting the close interaction between human and environmental factors [27]. Long-term averages of rainfall and temperature showed inverse associations with bites. Snakes are ectothermic animals and are less likely to inhabit locations subject to extreme weather conditions [29]. However, direct interpretation of the adjusted effects of variables require care. The regression coefficients in these models represent the effect of changing the values of a single variable fixing the values of all other variables. In the current context, our explanatory variables cannot typically be controlled in this way. Hence, our objective is to explain the joint effect of all the explanatory variables rather than their individual effects.

The NSS of Sri Lanka reported snakebite events experienced by family members of households in the preceding year from the date of the interview. It is possible that these reported bite counts are affected by recall bias [40]. We used an exponentially decaying function of recall duration to estimate recall bias in the fitted models and adjusted for this when constructing our estimates of snakebite incidence. The results for different provinces indicated a mix of under-reporting or over-reporting of bites. Under-reporting was estimated in provinces with low bite incidences and over-reporting in provinces with high bite incidences. A possible explanation is that in areas with high incidences, where bites are a common feature of day-to-day life, inhabitants are more likely to report a false positive memory of a bite of a previous month and vice versa [41].

Decision-making requires information of known precision. The PCMs identify the areas that can be confidently classified as high risk and low-risk, and areas where risk status is highly uncertain. Red patches on the PCMs indicate the hotspots and blue patches the cold spots, relative to the monthly mean incidences of snakebites and envenoming bites across all the provinces. These PCMs enable decision-makers to identify priority areas for establishment of specialized snakebite management centres and allocate resources, including snakebite antivenom, at operational level. Further, identification of high-risk geographical areas and seasons of snakebite are important to formulate management plans for sustainable use of natural resources. This can be achieved by environmental education processes and reduction of accidental snakebites will benefit the country including the tourism industry.

In this study, we adopted a point process modelling framework instead of the conventional regression modelling approach in order to make best use of the individual level data provided by the NSS. Times of individual events at given locations were modelled using a multivariate Poisson process. The NSS was conducted for 11 months and the survey sampled different numbers of individuals at each cluster. Our methodology allowed us to estimate snakebites and envenoming bites per person incorporating spatial and temporal variables while accounting for recall bias and varying sample sizes at different clusters. The same methods could be applied for spatiotemporal modelling of other diseases for which the times and places of individual events are recorded.

There are limitations in this study. Although NSS had high spatial coverage capturing snakebite events occurring in all the provinces and districts of Sri Lanka, its temporal coverage was limited to a 13-month sampling window at each sampled location. Consequently, we were only able to disentangle seasonal trends from long-term trends by allowing for possible recall bias. Our spatiotemporal models were built from previously published separate spatial and temporal models, and the explanatory variables considered were limited to the variables used in those models. Envenoming bites are a subset of overall bites. The smaller number and fewer data-points available to fit the point process model resulted in higher standard errors and hence wider confidence intervals for envenomings than for overall snakebites. Our spatiotemporal models were constructed at the provincial level and parameter estimates were optimized for each province as the survey was designed to estimate bite incidences at provincial level. This resulted in artificial discontinuities in our fitted incidence maps at province boundaries, which are particularly apparent along the North-Western province. This can be overcome by designing future surveys at sub-province level or at district level to estimate the burden.

The present study aimed at evaluating spatiotemporal patterns of snakebite and envenoming bite in the country at the time of the survey. However, the climate change is known to affect the weather patterns of the country and hence the distribution of snakes [10]. Therefore, we intend to investigate the spatiotemporal patterns of snakebites in the changing climate, in a future study.

In conclusion, this study is the first demonstration of spatiotemporally interacting variations in snakebites and envenomings in a country. The findings can provide insights into healthcare decision-making at local level, taking into account seasonal variations, to prevent and manage snakebites. Sri Lanka could benefit by establishing a notification system for snakebites. This would provide additional documented information and augment future studies. These methods could also be applied for spatiotemporal modelling of other diseases for which the times and places of individual events are recorded.

Supporting information

Acknowledgments

The authors wish to acknowledge the assistance given by the Director General of Health Services, Provincial Directors of Health Services of the nine provinces, District Secretaries and Regional Directors of Health Services of the 25 administrative districts of Sri Lanka, all the Divisional Secretaries and the respective Grama Niladharis of the sampled Grama Niladhari Divisions, all the research managers and research assistants for implementation of the study. All the study participants are also acknowledged.