West Nile virus (WNV) is a global emerging infectious disease. Initially characterized in Uganda in 1937, WNV first appeared in North America in 1999 in New York, USA, with subsequent spread to Canada in 2001 [1]. The lifecycle of this flavivirus is sustained in a mosquito-avian enzootic cycle, with spillover to humans and other mammals [2]. The majority of human infections are asymptomatic, although approximately 20% of infections cause febrile illness and 1% of infections lead to neuroinvasive disease [1, 3].

In Ontario, Canada, WNV incidence peaks in May to October each year corresponding to the period of mosquito and virus activity. High incidence years occur infrequently, with epidemics in 2002 and 2012 resulting in high WNV incidence rates, of 3.5 and 2.0 WNV cases per 100,000 population, respectively [4]. Importantly, the sporadic nature of the disease poses a challenge for planning and sustaining public health surveillance and intervention strategies to prevent human infection.

Environmental factors, such as climate and land cover, have an important influence on WNV transmission through their effects on mosquito population dynamics and ecology. Higher winter temperatures often correspond with increased mosquito abundance, mosquito biting rates, viral replication and rates of transmission [5–9], while grasslands, wetlands and urban cover have all been associated with WNV activity [10–12]. Notably, the relative importance of different environmental factors can vary greatly across large geographic areas due to differences in primary vector species. In western Canada, Culex tarsalis is the predominant vector species, while in eastern Canada, Culex pipiens/restuans dominates [13]. These two species have different lifecycles and habitat preferences, which in turn affect the nature of environmental risk factors for disease transmission.

In addition to the environment, individual and population characteristics, such as age, sex, and behavioural risk factors can also influence the incidence of WNV in a given area. Human behaviour, and hence contact with mosquitoes, can vary greatly within and between population groups [14]. Variance in population structure between areas, such as the male/female ratio of a region [15] and the number of senior households in a given area [10] have been identified as predictors of WNV incidence. Importantly, the spatial dependence of factors that contribute to WNV incidence requires consideration of geographically-specific risk factors.

In Ontario, a limited number of studies have investigated local predictors of high WNV incidence seasons, i.e. years with elevated numbers of reported WNV cases during the May to October transmission season. This study aimed to determine PHU-level predictors of annual WNV incidence specific to Ontario, recognizing that early identification of WNV risk can assist PHUs in tailoring their surveillance and response efforts. Herein we used a mixed modelling approach combining climate, land cover, population and surveillance data to ascertain environmental and population predictors of human WNV incidence in southern Ontario.

To address the knowledge gap on population level predictors of human WNV incidence in Ontario, we applied a multivariable mixed modelling approach that incorporated key environmental and population factors at the PHU level. Monthly climate indices, particularly February, March and April minimum temperature and February precipitation, and the annual percentage of WNV-positive mosquito pools in a given PHU, were significantly predictive of human WNV incidence across PHUs of the southern portion of Ontario. When considering only ‘early season’ variables, which effectively excludes mosquito surveillance data, January and February mean minimum temperatures were of primary importance, highlighting the utility of climate data in predicting WNV risk in the upcoming transmission season.

Our findings are similar to previous studies, which also found higher winter temperatures to be significantly predictive of increased rates of WNV transmission in an upcoming season. For example, Wimberly et al. found that winter temperature variables had the greatest influence on West Nile virus human infection rate in the United States in 2014; December and January temperatures in particular were most significant [21]. Similarly, Manore et al., found mean minimum January temperature to be a highly significant predictor of human WNV [9]. Our analysis indicates that January and February temperatures may be relevant predictors for WNV incidence in the southern portion of Ontario. Temperatures during the winter months have a considerable impact on the ability of WNV to survive into the spring, and in colder years effective overwintering of mosquitoes is lessened [22].

Maximum temperatures also had a significant influence on WNV activity. It was found that lower April and higher August mean maximum temperatures were significantly related to human WNV incidence. The significance of the April mean maximum temperature may be related to virus amplification in the avian host, which is believed to occur in the early spring [8]. Warmer temperatures during this period may be unfavourable since spring temperatures that are too warm might result in faster melting of snow which could dilute nutrients in standing water or flush Culex breeding sites, impeding larval proliferation [23]. The causal association between warmer August maximum temperatures and human WNV infection is expected, as warmer summer temperatures increase mosquito abundance and biting rates and decrease viral amplification time [24].

In addition to temperature associations, precipitation was found to contribute significantly to increased human WNV risk, albeit more subtly. We found that lower February and March mean precipitation was associated with higher WNV incidence in a given year. While this is in contrast to some studies in the United States that found increased March precipitation to be associated with outbreaks, these results may be attributable to the variable influence of precipitation across the study region and differences in primary vector species [21]. Further to this, it has been proposed that early spring drought may concentrate vectors and hosts around pools of water, and allow for low populations of vector predators [25].

Finally, we found that the annual percentage of WNV-positive mosquito pools was significantly predictive of human WNV incidence, confirming our expectation that higher mosquito infection rates should result in more transmission events. This has been previously remarked by Brownstein et al., who noted that mosquito surveillance data is the most sensitive marker for human risk with positive mosquito pools accounting for 38% of human risk [26], and suggested that this data should be included in any surveillance system. The association between WNV-positive mosquito pools and human WNV case counts has also been noted by Rochlin et al., who found a strong association between human risk and proximity to a single WNV-positive mosquito pool [10], and Liu et al., who found that presence of a WNV-positive pool in the last 30 days was significantly predictive of risk for human infection [14]. The results of our study indicate that a 1% increase in annual WNV-positive mosquito pools would result in a 7% increase in the annual human WNV incidence rate.

This study identifies several predictors of human WNV incidence in southern Ontario that can be of practical use for public health. Importantly, readily available data on climate indices may be used by public health officials for predicting more severe WNV seasons at the PHU level. Limitations include the fact that data were aggregated at the PHU level, which may have masked smaller-scale variations in WNV incidence and predictor variables (including land cover and population structure) and hence associations. Avian data was not included in the analyses due to paucity of data. Since birds are the main reservoir hosts of WNV, bird dynamics may substantially influence seasonal risk, and previous studies have found links between bird community composition and WNV incidence [27]. Future studies may seek to include this type of data to produce a more comprehensive model. In addition, PHU-level census data availability was restricted to 2011, which may have limited the importance of population structure variables in our analysis. Finally, while the proposed multivariable model was successful in identifying key predictors of human WNV incidence in southern Ontario PHUs, some overdispersion was noted particularly for the most southern PHUs, which could affect the reliability of model estimates for these areas.

Despite these limitations, the variables identified as significant in our model may be useful for public health planning. In practical terms, this could entail monitoring of monthly average temperature and precipitation trends in an area, and comparing the data to normal or historical values. Identification of higher than average minimum winter temperatures and lower than average spring precipitation at the PHU level could serve as indicators of elevated risk of WNV in an upcoming transmission season. These early year estimates of virus activity could then be used to inform decisions regarding the quantity of resources that could be put towards in-season risk measures based on mosquito surveillance. This is important, since small scale weather events such as sudden heavy rains or short cold periods, which affect mosquito survival, can have a large impact on WNV risk [8]. While early season predictors may be useful, our results support the ongoing monitoring of climate and entomological indices to more accurately predict WNV risk at the local scale.

Overall, although WNV does not consistently pose a public health risk to the majority of Ontarians, our results indicate that by using measures of risk that are detectable early in the year it may be possible to estimate the level of WNV activity for the upcoming season, allowing PHUs to tailor appropriate preventive strategies and decrease risk to public health.