Visceral leishmaniasis (VL) is a tropical disease of public health importance in India. Caused by Leishmania donovani parasites and transmitted by the sand fly Phlebotomus argentipes, the disease is anthroponotic in India and endemic in four States: Bihar, Jharkhand, West Bengal and Uttar Pradesh ( Addy & Nandy, 1992). It is estimated that 90% of VL cases in India originate from Bihar ( Singh et al., 2010). In 2005, India with Bangladesh and Nepal, set a target to eliminate VL with incidence at less than one case per 10,000 population by 2015, which was subsequently readjusted to 2017. This is in line with the London declaration on Neglected Tropical Diseases (NTDs), which set the aim to eliminate VL as a public health problem by 2020.

The seasonal and interannual scale climate trends seen in vector-transmitted diseases have been widely reported ( Connor et al., 1999). In India, malaria prediction was first conducted using climatic and socioeconomic data in the 1910s by Captain S. R. Christophers from the British Army ( Gill, 1938). This system was implemented until the 1940s, when malaria was not seen as disease of public health concern within the Indian subcontinent ( Gill, 1938). Despite the burden of VL disease in India, little work has been done to associate climatic indicators with case incidence. Historical trends have previously shown that there is a resurgence of VL every 15 years post control: whilst widely accepted, there is limited understanding about the cause of this pattern ( Malaviya et al., 2011; Muniaraj, 2014).

Napier first proposed that VL in India was cyclical in nature in 1946, before the suspected epidemics of 1977 and 1991. Visceral leishmaniasis epidemics in Assam (1875 and 1950) however were thought to be caused by the influenza pandemic of 1918–1919, malaria, famine and earthquakes ( McCombie Young, 1924; Napier, 1943; Rogers, 1908). Findings from McCombie Young (1924) in Assam, supported this and after the introduction of antimonial drugs, a dramatic decrease in case mortality from 90% to 10% was observed ( Dye, 1992; McCombie Young, 1924; Muniaraj, 2014). More recently, Muniaraj attempted to elucidate the cause behind the 10–15 year cycle, attributing the VL case trend to vector control practices and the need for appropriate therapeutic strategies to reduce VL mortality ( Muniaraj, 2014). However, Dye has referred to this cyclical trend of L. donovani VL as “part of the folk wisdom of tropical medicine”, suggesting that post-kala-azar dermal leishmaniasis (PKDL) patients serve as a parasite reservoir: chronically infective and available to sand flies ( Dye, 1992; Rogers, 1908). Studies by Addy and Nandy have suggested that disease transmission during the suspected epidemic period of 1977 in Bihar, was most likely passed on person to person from active VL cases ( Addy & Nandy, 1992).

Associations between climatic indicators, vector density and VL incidence in India have previously been detected when analysing district level data ( Bern et al., 2010; Dhimal et al., 2015; Malaviya et al., 2011; Tiwary et al., 2013). Generalised associations for Bihar as a State, currently do not exist. Humidity, temperature, rainfall, soil temperature and moisture are all widely accepted factors that influence the development of Ph. argentipes ( Bhunia et al., 2010). There are two seasonal peaks of Ph. argentipes in Bihar, one between March to June and another in October to November ( Dinesh et al., 2001; Picado et al., 2010). Associations have also been made between VL incidence and air temperature, relative humidity and annual rainfall in the Gangetic plain, which includes regions within Bihar ( Bhunia et al., 2010). Malaviya et al. noted seasonality of VL case trends, with peaks in March to April and a minor secondary peak in July ( Malaviya et al., 2011). In addition, the onset of disease symptoms were recorded, in descending order, from April to June, June to September and October to December ( Perry et al., 2013). This paper investigates the impact of annual climatic and disease trends on the cyclic nature of VL burden in Bihar.

For full details on data sources accessed for this research study, see supplementary Table 1.

Associations between climatic indicators and vector-borne diseases have been established within many diseases, including VL ( Chen et al., 2012; Dhimal et al., 2015; Malaviya et al., 2011; Perry et al., 2013; Thomson et al., 1999; Thomson et al., 2005; Tiwary et al., 2013). Predictors of Leishmaniasis spp. transmission cycles and sensitivity to meteorological and climatic variables is known to vary spatially; dependent on a range of factors including species composition, host competence, contact rates, vector competence, sensitivity to weather and other environmental stressors ( Chaves et al., 2008; Lewnard et al., 2014). These factors however also remain essential for developing early warning systems to prevent epidemics. Due to the lack of spatial and temporal open access granular case data, there are a limited number of studies investigating the interaction of climatic variables and VL in Bihar. Previous studies have predominantly focused on district level climatic trends identifying various monthly or annual temperature, relative humidity and rainfall measurements as the key variables affecting VL transmission ( Malaviya et al., 2011; Napier, 1926; Sivaramakrishnaiah & Ramanatham, 1967; Tiwary et al., 2013). This is the first study to look at historical data spanning 43 years and include climatic variables during specific month periods estimated to impact transmission, namely seasonal sand fly abundance and monsoon period ( Picado et al., 2010).

In this study, four statistically significant negative associations were detected between Average Temperature and VL incidence ( RR= 0.682, p=0.0186 ( Annual) and RR=0.796, p=0.0494 ( Pre-Sand fly Peak)) and Maximum Temperature and VL incidence ( RR=0.667, p=0.003 (Annual) and RR= 0.814, p=0.0501 (near significant - Pre-Sand fly Peak), using GOB data ( Table 2). However after considering the FD correction for multiple testing, only Maximum Temperature and VL incidence retained a significant association (0.0360). Overall this shows that risk is reduced with each unit increase of temperature. This could be due to human behavioural patterns, such as more people sleeping outside in warmer temperatures, unfavourable temperatures for optimal sand fly emergence or other trends associated with the vector and transmission dynamics. This is the first time that an association with temperature has been suggested to VL incidence for the Pre-Sand fly Peak time period. Annual humidity and VL incidence when fitted to climate anomalies (1977–2013) and adjusted for a decreasing temporal trend in incidence, also showed some association (p=0.0398) however only when corrections for multiple testing were not considered. This suggests that Annual Humidity could also play a similar key role in VL transmission to Annual Maximum Temperature: influencing human sleeping behaviour.

Historical research indicated that the factors favouring the development of an epidemic would include: altitude below 609.60m above sea level, greater than 1270mm annual rainfall, greater than 70% mean humidity, presence of alluvial soil, maximum temperature below 37.78°C, with diurnal variation less than -6.666°C, abundant vegetation and subsoil water, and a rural environment ( Napier, 1926; Sivaramakrishnaiah & Ramanatham, 1967). Bihar is estimated to be 52.73m above sea level, and maximum temperature was below 37.78°C across all time periods ( Table 2), however rainfall during years of high VL incidence was considerably lower than previously suggested ( Table 4). Other criteria such as “abundant vegetation and subsoil water” cannot be quantified to form a comparable variable within this study. Given the changes in global climate, these historical guidelines on optimal conditions for a VL epidemic may no longer be suitable.

More recent research conducted by Malaviya et al. ( Malaviya et al., 2011) showed associations between monthly VL incidence, rainfall and specific humidity, but contrary to this study, found no clear association between incidence and average or maximum temperature. A study conducted in the Gangetic plain, which includes VL endemic West Bengal and Bangladesh, suggested optimal meteorological factors for VL incidence included air temperature of 25.0–27.5°C, relative humidity of 66–75% and annual rainfall between 1000 and 1600 mm ( Bhunia et al., 2010). Due to a lack of available open access data, only annual rainfall data can be compared to findings by Bhunia et al. ( Bhunia et al., 2010): rainfall during the periods of high VL incidence fell below the suggested optimal range (1992: 834.667mm, 1991: 670.250mm).

The term epidemic is generally referred to as a sudden increase in the number of cases, beyond what is normally expected within the population of a specific area ( CDC, 2016). The underlying assumption when identifying an epidemic is that the data collection methods adopted are comparable in accuracy and precision. When including GOB&I data, the highest VL incidences were in 1977 and 1991, and historically they have been referred to as “epidemics” ( Bora, 1999; Thakur, 1984b; Thakur, 2007; Thakur et al., 2008). Data forming the basis for these suspected epidemics were collected through active case detection surveys. Such surveys typically detect higher levels of incidence than those recorded through normal passive reporting channels and highlight the need for better case reporting mechanisms. Comparisons between GOB and GOB&I data suggest that case numbers recorded through standard reporting channels in 1977 and 1991 were under-reported (81.41% and 76.15% respectively). Under-reporting of VL in Bihar has been documented since 1977, with most recent reports up to 2006 ( Bora et al., 1994; Singh et al., 2010; Thakur, 1984b). In 2010, it was reported that a “substantial proportion” of patients were visiting private laboratories for diagnosis, which could explain underreporting documented for VL ( Hasker et al., 2010). The quality of passive case reporting may have fluctuated greatly over the years, so a single correction factor cannot be applied across all years of data without further evidence. While the question remains if these are true epidemics, both surveys did result in the implementation of IRS programmes to control the disease.

When considering the current data set, given that the peaks originate from independent surveys, they may simply be artefacts highlighting the true burden of disease. As reported by Bora et al., case data recorded by GOB typically includes cases recorded through government health facilities only as private practitioners treating for VL fail to report cases they have treated to the health authorities ( Bora et al., 1994). This was also the issue faced by Dye and Wolpert when modelling the Assam epidemics of 1875 and 1950; where poor quality data limited the findings for the cause behind the second epidemic ( Dye & Wolpert, 1988). Given the discrepancies in data and sources, further evidence is required to confirm the presence of a VL disease cycle.

The anthroponotic nature of the disease in India, suggests records of PKDL could provide additional evidence to explain the peaks seen in 1977 and 1991 ( Addy & Nandy, 1992). Open access case data sources used within this analysis, provided the number of deaths and cases only, restricting further exploration using this variable.

Historically VL vector monitoring within Bihar has been sporadic and limited by lack of spatial and temporal granularity ( Dinesh et al., 2001; Malaviya et al., 2011; Picado et al., 2010; Tiwary et al., 2013). This has prevented the robust multi-annual evaluation of P. argentipes abundance peaks, physiology status and transmission effectiveness. For effective retrospective analysis to understand disease incidence, understanding the vector is crucial, furthermore climatic parameters can be further refined based on vector behaviour.

Time periods used within this study were identified as periods of high sand fly activity or adverse weather. The months included within the Sand Fly Peak time period (March - June and October - November) have been shown previously to coincide with months of high case numbers ( Dinesh et al., 2001). This relationship needs further exploration, through spatial models, if monthly granular data were available, to further understand disease trends and outbreak prediction.

The available published data describes Bihar’s primary intervention against VL, IRS, as a dichotomous variable, limiting the understanding of its impact on the natural disease patterns. The VL incidence between IRS and non-IRS years is very similar when considering the overall dataset and interestingly increases (3.590) during years of IRS in comparison to years of non-IRS (3.000), suggesting that associations are potentially being masked by the lack of detail about the intervention. Other IRS programme indicators, such as spray quality, would also supplement understanding the success of IRS round, allowing for a robust statistical model to be developed ( Coleman et al., 2015).

Since the start of the VL control efforts strategies adopted for case finding have changed, with the most concerted efforts ongoing considering the current elimination target. Accredited Social Health Activists (ASHAs) today play a key role in promoting healthcare activities such as immunisation, reproductive and child health and detection of VL cases ( Das et al., 2016). Research has shown this incentive based approach for identifying VL patients led to an improved referral rate, translating into early detection of cases and a long-term drop in case numbers ( Das et al., 2016).

The Indian VL elimination programme have taken significant steps to improve case data quality over recent years: increasing VL awareness, providing incentives to local health workers (ASHAs) and implementing district level tracking systems. For the VL elimination programme this means that to sustain the gains made a system such as reactive IRS as seen in other diseases must be established, or, the levels of VL will increase again as observed in Figure 1 ( Coleman et al., 2008). Using the current WHO standards for initiating alerts in India, as in Brazil, after months when the incidence has been double the monthly average, is likely to be a sub-optimal criterion with limited infrastructure and funding to ensure rapid response: particularly after the elimination target has been achieved and the disease is no longer considered a public health concern ( Coleman et al., 2008; Lewnard et al., 2014). A comprehensive understanding of the causative factors for disease trends is required to develop a suitable outbreak detection and response system within the VL Indian context.