Factors associated with West Nile virus disease fatalities in horses

Abstract

In 2003, the occurrence and location of horses with clinical signs of West Nile virus infection were identified in the southern portion of Saskatchewan with the help of veterinarians, owners, and the regional laboratory. A total of 133 clinical cases were reported between July 30 and September 19, 2003; however, postseason surveillance suggests that the number of cases was underestimated. The case fatality rate was 43.8% (95% CI 35.2, 52.4). Factors associated with fatality in clinical cases included sex, week of onset of clinical signs, and coat color. Reported clinical cases clustered within regional health authority districts, suggesting regional differences in geographic factors, potentially including climate and mosquito control, that could contribute to the risk of disease. However, most of the variation in the risk of fatality in clinical cases is explained at the individual level rather than the Regional Health Authority level, which suggests the outcome of clinical disease is primarily determined by characteristics of, or management factors affecting, the individual horse.

Introduction

West Nile virus (WNV) is an arbovirus (arthropod-borne virus) that affects the nervous system of humans, horses, and birds, causing mild to severe illness and sometimes death (1). It was first reported in Africa in 1937 (2) and has been identified sporadically in European countries from the early 1960s to present time (3–5). Since being diagnosed in New York state in 1999, WNV has spread across the North American continent (6–9).

In Canada, WNV infection was first seen in birds in southern Ontario in 2001 (10). Since 2001, throughout Canada, the date and location of birds found dead that tested positive for WNV were systematically recorded by the Canadian Cooperative Wildlife Health Centre. In 2002, the first equine cases (those showing clinical signs consistent with WNV infection) were identified in Ontario, Manitoba (MB), and Saskatchewan (SK) (10,11). In SK, there was only a passive system to monitor the occurrence of clinical disease in horses, which, unfortunately, resulted in the numbers reported varying, depending on the source of information.

The virus is amplified in a natural cycle between birds and mosquitoes and incidentally infects humans and horses. In western Canada, Culex restuans and Culex tarsalis are the major mosquito species responsible for transmission in the amplification cycle. Culex tarsalis is considered the most important vector for transmission of WNV to humans and horses (12). Horses can become infected when bitten by a mosquito carrying the virus, but they do not contribute to the spread or amplifcation of the virus in the natural cycle (13). Once bitten, a horse may eliminate the virus uneventfully or show clinical signs including fever, depression, muscle tremors, weakness, lack of coordination, inability to rise, and paralysis (14). The incubation period of infection to manifestation of clinical signs is 5–15 d (15). Most reports suggest that around 25% to 45% of those horses that show clinical signs die or require euthanasia (4–8,14).

Factors significantly associated with fatality from WNV clinical disease include nonvaccinated status for WNV; age; clinical signs, such as inability to rise; early season onset of clinical disease; sex; and breed (8,16). Horses given either 1 or 2 doses of vaccine (even if it was not given according to the manufacturer’s recommendations) seemed less likely to die than unvaccinated horses (8,16). In 1 of these studies, death was more likely if clinical signs occurred first from March 1 to August 19, the animal was female, the animal was over 18 y of age, or was unable to rise at any time during clinical illness (8).

Given the pattern of WNV spread across the continent, the number of horses in SK showing clinical disease was expected to be higher in 2003 than in 2002. The 1st objective of this study was to document the date of onset, geographic location, and outcome of clinical cases of WNV infection in horses in SK in 2003. The primary objective was to identify risk factors that could be associated with fatality in infected horses, including management, mosquito control, vaccination, individual horse characteristics, and environmental conditions.

Materials and methods

Study population and protocol

Veterinarians in private practice, horse owners, and the staff of the regional diagnostic laboratory [Prairie Diagnostic Services (PDS), Saskatoon, SK] provided information on clinically affected horses in SK during the summer and early fall of 2003. In June 2003, all large and mixed animal practices in the southern portion of SK were contacted by facsimile and asked 1) to send any serum samples collected from horses where WNV infection was suspected to PDS in Saskatoon for immunoglobulin (Ig)M enzyme-linked immunosorbant assay (ELISA), and 2) to allow the laboratory to notify the research team of any positive result from the samples submitted. Of the 79 veterinary clinics contacted, 51 responded to the facsimiles signed in June 2003. Twelve more practices were contacted by PDS at the time a submitted sample was declared positive, giving a total cooperation rate of 80% (63/79).

For the purposes of this study, the definition of a ‘clinical case’ was the presence of 1 or more of the classic signs of WNV clinical disease (ataxia, recumbency, paresis, or paralysis) or death; and a positive IgM ELISA. Once a ‘clinical case’ was identified, the veterinarian was contacted to obtain consent to contact the owner. If consent was given, owners were telephoned 2 to 4 wk after the onset of clinical signs and asked for additional information. The telephone survey included questions about location, characteristics of the affected horse, management of that horse, vaccination status, other resident equids (horses, ponies, mules, donkeys), and mosquito control measures applied at both the farm and the individual animal level.

In October, after the mosquito season, large or mixed animal veterinary clinics in each Regional Health Authority (RHA) were contacted to estimate the number of untested probable cases of WNV infection per region (postseason survey). A random numbers procedure was used to rank all of the veterinary clinics in each RHA. Clinics from each RHA were then contacted by facsimile in random order until at least 1/3 had responded. The participating veterinarians were asked to estimate the number of horses that were known or reported to have shown clinical signs consistent with WNV but were not tested for IgM antibodies during the 2003 season.

Horse data

The legal land location of the clinical case at the estimated time of exposure (up to 2 wk before the 1st clinical signs) was requested for all cases where follow-up was obtained. Legal land location included quarter, section, township, range, and meridian. In one instance, a global positioning system (GPS) unit was used to obtain the latitude and longtitude. For those cases in which the owner was unsure or unwilling to give the location, directions from the nearest town or township/range data were collected. This information was cross-referenced with an electronic map of SK. By using commercial geographical information systems (GIS) software (ArcView GIS version 3.2 and ArcGIS version 8.0; Environmental Systems Research Institute, Redlands, California, USA), spatial location was displayed as a point by using the centroid-of-the-polygon for the land location. Confidentiality was maintained in all publications by summarizing location data at the RHA level.

Statistics Canada’s 2001 Census of Agriculture estimated the number of horses and ponies (horse data) in SK to be approximately 71 300 and provided horse data by Census Consolidated Subdivisions (CCS). The CCS has similar boundaries to rural municipalities (RM), allowing estimation of the horse population by RHA (aggregations of RMs) and the mapping of cases per 1000 horse population by RHA. Data provided by the post-season survey were aggregated and mapped by RHA to compare and assess the accuracy of surveillance, using only confirmed cases of WNV for surveillance in 2003.

The animals were classified by sex, color, age, and breed type, but not by specific breed. Breed type was divided into light horses, draft horses and crosses, ponies, and “other”. The category of “other” was composed of 3 miniature horses and 1 mule. Sex was reported in 3 categories: mare, gelding, and stallion. Color was divided into 3 categories: dark (chestnut, sorrel, brown, black, or bay), light (gray, white, palomino, buckskin, or dun), and multicolored (roan, paint, pinto, and appaloosa). Age was classified into 3 categories: young (≤ 5 y), adults (6–18 y), and geriatrics (> 18 y). Age was categorized because the odds of outcome was not thought to increase on a linear scale and so that the results were comparable with those of other studies (8).

Day of onset of clinical signs was aggregated into week of onset with Sunday as the start of the week. To allow for easier interpretation, week of onset was further categorized into weeks of the year, 31–33 (beginning of August), weeks 34–35 (epidemic peak, end of August), and weeks 36–38 (September).

Housing was classified as housed in an enclosed barn, access to a simple 3-sided shelter with corral or pasture, or no access to a shelter in open pasture with or without trees. The method of mosquito control on farm was recorded as smudges (smoky fire used to ward off mosquitoes), water removal, sealed barn, use of fans or special lighting in a barn, use of general insecticides, or no mosquito control attempted. The use of insecticides sprays on individual horses was recorded as yes (any amount or frequency) or no.

Mosquito data

Mosquito data were obtained from the 2003 Saskatchewan Health mosquito trapping program. Center for Disease Control (CDC) traps were used for the collection of live mosquitoes for virus testing, and New Jersey Light traps (NJLT) were used for the collection of dead mosquitoes for the determination of mosquito species. Culex tarsalis and C. restuans data were provided on a weekly basis from June to September as the average number of mosquitoes trapped per night. Variables of interest were (a) the highest weekly average for C. tarsalis from the whole season (number of C. tarsalis), (b) the highest weekly average for C. restuans from the whole season (number of C. restuans), (c) whether WNV was found in pooled mosquitoes of each species (positive pool), and (d) the weekly average for C. tarsalis from the likely week of exposure for each case. For each variable of interest, the results of the nearest mosquito CDC and NJLT trap were linked to each horse case.

Environment data

Daily average temperature and precipitation data for all climate stations in SK were obtained from Environment Canada. These values were used to create a ‘season average’ for temperature and ‘season total’ for precipitation. These variables were calculated for each climate station, using daily values for June, July, and August 2003. The season average temperature was determined by dividing the sum of the daily temperatures by the number of days; the season total for precipitation was the sum of the daily amounts of precipitations. The variables from the nearest climate station were linked to each horse case.

Cumulative growing degree days (GDD) are the sum of the positive daily differences between 16°C (the threshold temperature for activity of C. tarsalis and virus transmission) and the mean temperature for that day (17). ‘Season GDD’ was calculated for each climate station by using temperature data from May 1 to September 30. Cases were assigned data from the nearest climate station with complete daily temperature and precipitation information.

Statistical analysis

The association between each risk factor and the final health outcome (fatality) was examined by using generalized linear mixed models with a binomial distribution and logit link function (18). The calculations were performed by using penalized quasi-likelihood estimates and the 2nd order of the Taylor series expansion (MLwiN version 2.0; Centre for Multilevel Modeling, Institute of Education, London, UK). The strength of the association between outcome and exposure was reported as odds ratio (OR) with 95% confidence intervals and P values. A 2-level hierarchical (RHA, Horse) logistic regression model was used to account for geographic clustering of observations and because the method of surveillance used clinics within RHAs (18).

The linearity assumption was assessed for all continuous risk factors considered in the modeling process. Exposures were reanalyzed after being classified into quartiles and the linearity assumption was examined by looking for an increasing (or decreasing) series of coefficients. The “effect” of exposure in each higher category, as determined by the odds ratio, was compared with the level of the outcome in the baseline or referrent exposure category to identify patterns consistent with a linear, monotonic, or threshold effect.

All exposure variables where the association with the odds of fatality was significant at P < 0.25 were considered in developing the final multivariable model. Nonexposure risk factors were defined as confounders, if removing or adding the factor changed the effect estimate for the exposure by more than 10%, in which case, the factor was retained in the final model. Manual backwards elimination of variables was used to achieve a final model containing only the statistically significant exposure variables and any nonexposure risk factors that were either significant or acted as important confounders. Biologically reasonable interactions were assessed between significant risk factors (P < 0.05) in the final model.

An approximation method based on latent variables was used to estimate the intraclass (intraregion) correlation coefficient [ρ = σ²_r/(σ²_r+π²/3)] to examine the clustering of clinical cases of WNV within an RHA (18).

Results

Onset date and location

The date of onset of clinical signs for the 1st WNV IgM-positive clinical case was July 30, 2003. The last recorded date of onset of clinical signs in a horse was September 19, 2003. Of the 130 horses for which a questionnaire was completed out of the total 133 clinical cases, 57 died or were euthanized because of complications associated with clinical disease (43.8% case fatality rate, 95% CI 35.2, 52.4) and the remainder recovered completely. Case fatality rates differed by RHA (Table 1). The epidemic curve appeared to peak 1 wk earlier for horses that died from WNV clinical disease than for horses that survived, suggesting that there could be a difference in risk of fatality associated with time of onset (Figure 1). This observation was explored in subsequent analyses.

Table 1.

Postseason survey results and case fatality rates by Regional Health Authority (RHA)

RHA	Surveyed clinics (total number of clinics in the RHA)	Total probablesa	Estimated probablesb	Ratio (estimated probables: confirmed)	Confirmed cases ( )c	Case fatality rates (%)d
Sun Country	4 (11)	20	55	5.5:1	10	50
Five Hills	3 (10)	58	193	16:1	12	75
Cypress	3 (5)	36	60	3:1	17 (1)	47
Regina Qu’Appelle	4 (10)	40	100	5.6:1	18	39
Sunrise	2 (6)	21	63	4.8:1	13	23
Saskatoon	5 (8)	7	11	0.5:1	22	41
Heartland	3 (6)	8	16	1.6:1	10	30
Kelsey Trail	1 (5)	2	10	2:1	4 (1)	75
Prince Albert Parkland	1 (4)	12	48	6:1	8	63
Prairie North	2 (7)	21	74	4.4:1	16 (1)	31
Totals	28 (72)	225	630	4.8:1	130 (3)	44

^aFrom surveyed clinics

^bBased on proportion of surveyed clinics

^cNumbers in parentheses are cases with no further follow-up obtained

^dCase fatality rates are calculated on confirmed cases only

Figure 1.

Number of clinical cases of West Nile virus infection in horses by week of onset of clinical signs in 2003. The bars represent the total number of cases by week. The lines represent the number of cases by health outcome per week. Weeks begin on a Sunday and end on a Saturday (Week 31 runs from July 27 to August 2. Weeks 31–35 correspond to the end of the month of July and the whole month of August 2003, and weeks 36–39 correspond to the month of September 2003).

For all 133 confirmed clinical cases, location was reported at the RHA level, based on the location of the owner (n = 130) or of the submitting clinic (n = 3). For the 130 clinical cases with further follow-up, exact legal land location was obtained for 117 cases (90%), exact GPS coordinates for 1 (1%), and partial legal land location (township and range) for 12 (9%).

Clinical cases of WNV infection were reported in all 10 RHAs in the southern portion of the province (Figure 2). When the raw data were adjusted for potential regional differences in reporting, based on the results of the postseason survey, the RHA with the highest number of clinical cases per 1000 horses was in the south central portion of the study area (Figure 3). Postseason results differed by RHA (Table 1), with the overall number of estimated untested probable clinical cases exceeding that of the tested clinical cases by approximately 4.8 to 1.

Figure 2.

Confirmed positive IgM clinical cases of West Nile virus infection per 1000 horses at risk reported by Statistics Canada in each Regional Health Authority (RHA).

Figure 3.

Combined number of diagnosed positive IgM clinical cases and undiagnosed probable clinical cases per 1000 horses at risk reported by Statistics Canada in each Regional Health Authority.

Description of affected horses

Most clinical cases occurred in horses ≤ 10 y; the fatality rate for this age group was 46% (38/83). The oldest horse that survived was 25 y old, while the oldest horse that died was 32 y old. Most horses that died (as opposed to being euthanized) did so during the month of August (11/12 or 92%). The fatality rate was 36% (21/59) for mares, 45% (27/60) for geldings, and 82% (9/11) for stallions. Dark colored horses were most commonly affected (89/124), but they had a lower fatality rate (36%) than did those in other color classifications. Most of the clinical cases were light horse breeds and most were used for pleasure riding (Table 2).

Table 2.

Summary of individual risk factors by survival outcome and case fatality rates (n = 130)

			Health outcome
Variable		Cases (n)	Dead	Alive	Case fatality rate	Odds ratioa	95% CI
Week of onset	31–33	27	17	10	63%	4.6	1.7, 12.5
	34–35	53	26	27	49%	2.5	1.1, 5.6
	36–38	50	14	36	28%	ref b
Age	< 5 years	44	22	22	50%	ref b
	6–18 years	69	28	41	41%	0.7	0.3, 1.5
	> 18 years	17	7	10	41%	0.7	0.2, 2.2
Gender	Gelding	60	27	33	40%	1.4	0.7, 3.0
	Mare	59	21	38	36%	ref b
	Stallion	11	9	2	82%	8.5	1.6, 43.5
Breed	Pony	6	2	4	33%	0.6	0.1, 3.6
	Light	106	48	58	45%	ref b
	Draft	11	4	7	36%	0.9	0.2, 3.3
	Cross	3	2	1	67%	2.6	0.2, 29.8
	Other	4	1	3	25%	0.4	0.04, 4.0
Color	Dark	89	32	57	36%	ref b
	Light	19	14	5	74%	4.4	1.5, 13.1
	Multi-color	16	8	8	50%	1.9	0.6, 5.4
Individual shelter	No shelter	86	42	44	49%	2.7	1.2, 6.1
	Simple shelter	41	29	12	71%	ref b
Individual	No	108	51	57	47%	2.3	0.7, 7.4
insecticide use	Yes	19	5	14	26%	ref b
Vaccine	No vaccine	121	53	68	44%	1.0	0.3, 3.9
	Vaccinatedc	9	4	5	44%	ref b
Primary use	Pleasure	108	49	59	45%	ref b
	Breeding	13	6	7	46%	1.6	0.5, 4.9
	Farm work	5	2	3	40%	1.3	0.3, 5.4
	Competition	4	0	4	0%	d
Herd size	10 or less	92	53	39	58%	ref b
	11–25	21	10	11	48%	1.1	0.4, 2.5
	25 or more	13	8	5	62%	1.1	0.5, 2.7
Farm mosquito	None	97	42	55	43%	0.9	0.4, 2.1
control	Used at least one	29	13	16	45%	ref b

^aUnivariable analysis adjusted for spatial location (RHA) by inclusion of a single random effect

^bref = reference category

^cVaccinated refers to fully vaccinated horses with the standard 2 vaccinations. Those horses with only 1 vaccination are considered nonvaccinates

^dNot enough data entries to converge

Of the 130 clinical cases with follow-up, only 9 were reported to have been fully vaccinated according to the manufacturer’s recommendations (2 doses given 3 to 6 wk apart, at least 2 to 3 wk before the peak mosquito season), while 3 were vaccinated only once (Table 2). Initial vaccination dates for the 9 fully vaccinated horses were February (n = 2) and April (n = 7). The vaccination dates for the 3 horses given only 1 dose were May, August, and September. Of the 9 fully vaccinated horses, 5 recovered, while 4 died or were euthanized (44% fatality rate). None of the 3 horses that were vaccinated once died or were euthanized.

Five of the 130 (3.8%) horses had been taken more than 15 km (10 miles) from their home location in the 3 wk before onset of clinical signs. Of these, 2 stayed within the same RHA, 2 travelled to a different RHA in SK, and 1 travelled to BC. In 2003, BC had no evidence of WNV infection. Five horses had a change of residence 2–3 d before onset of clinical signs. The location at which they were most likely exposed (given an incubation period of 5–15 d) was, therefore, recorded as the location of the previous residence. Most of the clinical cases were in the same RHA as the veterinary clinic that submitted the blood sample (94%).

No horses were housed in a barn for any period of time and most horses (68%) did not have access to shelter. Nineteen of the horses had been sprayed periodically with insecticide during the mosquito season. None of the horses were blanketed for any period of time. The majority of clinical cases were kept on farms that did not use any on-farm mosquito control methods (97/126). For horses on farms that did use at least 1 method, the primary choice was smudges (18/29, 62%). Seven of the 8 horses on farms that used water removal survived and 9 of the 18 horses on farms that used smudges survived (Table 2).

Herd size ranged from a single horse to 300 horses. The majority of horses were in small herds of 10 or fewer horses (73%). On most of the farms, all horses were kept under the same conditions; however, at 1 farm, the horses were housed in a barn, with the exception of the case horse. There were 5 farms that had 2 clinical cases and 5 farms that had at least 1 undiagnosed neurologic case, in addition to the tested case.

Description of environmental and mosquito data for the reported cases

The mean ‘season average temperature’ was 18.7°C and the mean ‘season total precipitation’ was 123 mm (Table 3); both of these values differed by climate station and by region. The mean ‘season GDD’ was 357. There was no obvious or statistically significant difference between these values for those horses that died and those that survived. Average temperature and total precipitation for the week prior to the estimated exposure period were also calculated, but, again, there was no apparent or statistically significant difference between values for those horses that died and those that survived (Table 3). The median distance between horse cases and the nearest climate station was 21 km (min, max = 2, 49 km).

Table 3.

Summary of individual environmental data (continuous variables) by health outcome. These variables were calculated by using only those climate stations or mosquito traps for which the data was matched up with the nearest case

			Temperature (°C)
Variable	Health outcome	Cases (n)	Min	Max	Mean	Overall Mean (°C)	Odds ratiob (95% CI)
Season average temperaturea (°C/day)	Dead	57	16.7	20.3	18.7	18.7	1.0 (0.7, 1.5)
	Alive	73	16.4	20.3	18.6
Season total precipitationa (mm)	Dead	57	39.2	206	118	123	1.0 (1.0, 1.0)
	Alive	73	39.2	215	128
Season GDDa (degree days)	Dead	57	199	490	350	357	1.0 (1.0, 1.0)
	Alive	73	178	507	355
Exposure week average temperature (°C/day)	Dead	57	14.8	26.5	21.3	20.9	1.1 (1.0, 1.2)
	Alive	73	14	26.5	20.6
Exposure week total precipitation (mm)	Dead	57	0	44.6	7.65	7.30	1.0 (0.9, 1.0)
	Alive	73	0	39.4	6.85
Exposure week	Dead	57	0	4.80	0.38	0.5	0.8 (0.5, 1.3)
C. tarsalis (number per night)	Alive	73	0	4.40	0.51
Number of C. tarsalis (number per night)	Dead	57	0	10.40	1.67	1.9	0.9 (0.8, 1.1)
	Alive	73	0	15.00	1.98
Number of C. restuans (number per night)	Dead	57	0	6.00	0.41	0.3	1.4 (0.8, 2.4)
	Alive	73	0	2.60	0.25

^aIn this table Growing degree days (GDD) is the cummulative growing degree days for the whole season (May to September), season average temperature is calculated with daily values from June to August divided by the total number of days and season total precipitation is the cumulative daily data from June to August

^bUnivariable analysis adjusted for spatial location (RHA) by inclusion of a single random effect

Variables summarizing mosquito data were similar for those horses that died or survived. The highest numbers of C. tarsalis and C. restuans over the whole season were 1.9 and 0.3, respectively. The number of C. tarsalis mosquitoes during the estimated exposure week of each case ranged from 0 to 4.8 (mean, 0.5) (Table 3). Culex tarsalis numbers were higher for those horses that survived, while C. restuans numbers were higher for those horses that died; however, the comparisons were not statistically significant.

A WNV-positive C. tarsalis and C. restuans mosquito pool was found in the nearest CDC trap for 54% (70/130) and 39% (51/130) of clinical cases, respectively (Table 4). The median distance between horse cases and the nearest mosquito trap was 29 km (min, max = 4, 92 km).

Table 4.

Summary of individual environmental data (categorical variables) by survival outcome

			Health outcome
Variable		Cases (n)	Dead	Alive	Odds ratioa (95% CI)
Positive pool	no	60	28	32	ref c
Culex tarsalisb	yes	70	29	41	0.8 (0.4, 1.7)
Positive pool	no	79	35	44	ref c
Culex restuansb	yes	51	22	29	1.0 (0.5, 2.0)

^aUnivariable analysis adjusted for spatial location (Regional Health Authority) by inclusion of a single random effect

^bA positive pool of mosquitoes (collection of species of mosquito designated) at the nearest to the trap location of the case horse

^cref = reference category

Association between individual horse risk factors, environmental variables, and the odds of fatality in clinical cases of WNV

Variables where there was a statistically significant association with fatality in the initial univariable mixed models that adjusted for clustering by RHA included week of onset, coat color, gender, and availability of individual shelter. Horses affected in August had 3.1 times (OR 95% CI 1.4, 6.8) greater odds of dying than those affected in September. The month of August was further broken down into 2 categories, weeks 31–33 and weeks 34–35. The odds of dying in both weeks 31–33 and weeks 34–35 (August) were higher than the odds of dying in weeks 36–38 (September) (Table 5).

Table 5.

Risk factors associated with survival outcomes in horses with clinical infection with West Nile virus (WNV). Final mixed multivariable model^a

Variable	Category	Cases (n)	Odds ratio	95% CI	P
Week of Onset	Week 31–33	27	6.3	1.9, 20.4	0.003
	Week 34–35	53	2.4	0.9, 6.2	0.08
	Week 36–38	50	ref b
Gender	Mare	59	ref b
	Gelding	60	1.2	0.5, 2.9	0.70
	Stallion	11	14.7	2.5, 87.4	0.005
Coat Color	Dark	89	ref b
	Light	19	6.4	1.9, 21.6	0.004
	Multicolor	16	1.9	0.6, 6.5	0.31

^aAdjusted for Regional Health Authority (RHA) by the inclusion of a single random effect

^bref = reference category

The odds of dying was greater for light colored horses than for dark colored horses, while multicolored horses were at similar risk to dark colored horses (Table 5). The odds of dying for males (stallions and geldings combined) was 1.9 times greater (95% CI 0.9, 3.8) than the odds of dying for females (Table 5). Specifically, the odds of dying for stallions was 8.5 times greater (95% CI 1.6, 43.5) greater than the odds of dying for females (Table 5). The odds of dying was greater for those with no shelter than for those with a simple shelter.

The final mixed model included week of onset, gender, and coat color (Table 5). Shelter did not remain statistically significant in the presence of the other variables and was removed from the final model. The final model was a mixed model with a random intercept to account for geographic clustering with RHA. The variance accounted for at the region level (σ²_r) 0.35 (standard error, 0.36) or approximately 9.5% of the overall variance in the model.

Discussion

West Nile virus infection and the development of associated clinical disease are influenced by many factors, including environment, mosquito populations, and individual susceptibility (3–5,9). The onset dates of WNV clinical disease reflect the time necessary for development of climatic conditions required for mosquito reproduction, natural amplification of the virus, and shift in mosquito feeding patterns to allow transmission of the virus to horses (12). The location of the cases corresponded to areas where mosquito populations capable of transmitting WNV and susceptible horse populations coexist.

Distribution of clinical cases

Travel history was essential in determining where horses were potentially exposed to WNV. However, as few horses had travel histories, exposure could be assumed to have occurred near their place of residence, even without complete travel history. Clinical cases and the submitting clinic were within the same RHA. Thus, on the basis of this study, we could assume WNV activity in the locality of either the submitting clinic or of the individual horse, depending on the degree of available location accuracy.

Most clinical cases were recorded in the 4 most southerly RHAs, while the fewest occurred in the most northerly RHAs of the study area. This distribution corresponds mainly to differences in climate and mosquito populations; however, veterinarians and horse owners located in the Saskatoon RHA might have been more inclined to pursue veterinary services and diagnosis, as this is the location of the PDS. No clinical cases were diagnosed from the northern portion of the province (outside the study area). This was probably because habitat and climate do not favor the principal vector, Culex tarsalis (12).

The postseason survey of veterinary clinics suggested that there were numerous cases of neurologic symptoms in horses for which owners did not seek a diagnosis. Diagnosis by clinical signs alone could be mistaken for other neurological conditions in horses, such as rabies, wobbler syndrome, equine herpesvirus encephalomyelopathy (EHV1), and western equine encephalomyelitis (14); however, all of these are considered relatively unlikely in this region. Thus, typical neurologic symptoms in an area known to have active WNV are reasonably predictive of WNV clinical disease presentation.

A mixed model was used to account for regional distribution of clinical cases, because WNV has been shown to cluster by geographic area in other studies, including 1 recent study that identified statistically significant clusters of both horse and human cases of WNV in 2003 (9,19,20). In the current study, RHA was chosen as the unit to represent geographic location in the mixed model, because it was a reasonable geographic proxy for the veterinary clinics that were the main source of our information. Individual veterinary clinics were not used in the analysis, as most of these only sampled 1 or 2 cases. The boundaries of the RHAs often reflect patterns of travel from rural areas to larger centers for health services, and based on the maps, appear to reflect a reasonable pattern of access to, and use of, veterinary services. In addition, case fatality rates differed by RHA, with the highest case fatality rate and highest postseason survey ratio in the Five Hills RHA. Decisions on the form and timing of public education campaigns related to WNV and mosquito control programs are often made at the level of the RHA or by rural municipalities located within the RHA.

Risk of fatality for clinical cases

The only significant predictors of risk of fatality in the final model were week of onset of clinical signs, sex, and color, while controlling for region. A small amount of variation (9.5%) was accounted for by geographic region in the final mixed model (18). This could reflect variability of the RHA mosquito programs and other inherent differences between the regions. Differences in clinic policy could have influenced the variation, but, overall, most clinics followed similar guidelines for diagnosis or treatment. Therefore, programs to control the outcome of clinical disease should be directed toward the individual horse. Unfortunately, the variables found to be significant in this study are difficult to manage for disease control purposes.

The case fatality rate was similar to rates from other studies conducted in North America (6–8). The increased survival rate of horses with clinical signs as the season progressed was similar to findings in the western United Stated (US). Salazar et al (8) found that, in 2003, horses were 1.7 times more likely to die if they showed clinical signs early (before August 19) in the season. The increased risk of death in August could be a result of earlier recognition of clinical cases as the season progressed. Owners may have checked their horses for clinical signs more often when they heard of clinical cases in nearby areas, sought veterinary assistance sooner (subtler signs), and begun treament earlier. This assumes that horses have a steady progression of clinical signs and that owners of horses that died merely missed the onset of mild signs. However, in this study, many owners reported seeing nothing unusual with their horses during the day before the onset of moderate to severe clinical disease (recumbency, paresis, or paralysis of limbs). An alternate theory is that the incubation period for different doses of virus corresponding to different time periods results in the difference in immune response and the varying presentations and outcomes of clinical disease. Further clarification of this finding is warranted.

Stallions were more likely to die as a result of WNV clinical disease than either mares or geldings. In some mammals, it has been noted that while testosterone boosts sexual characteristics, it also simultaneously impairs immune system functioning (21). Therefore, stallions may have a poorer immune response than either females or geldings. To compare our results to those of other studies from the US, we recategorized the data as males (both stallions and geldings) and females (mares) for reanalysis. In both the other studies, the univariable analysis suggested that females were more likely to die, but this result was not statistically significant (8,16). In our analysis, no statistically significant difference existed between mares and geldings, but rather between stallions and mares or geldings. Therefore the percentage of males that were intact stallions could influence this association. This was reported as 8% for 1 study (8), compared with 15% for our study.

Prevention education programs for humans list wearing light colored clothing as a method to decrease the possibility of WNV infection (22). When seeking out potential explanations, no clear evidence for the reasoning behind the use of light colors as prevention of infection could be found. In this study, we looked at risk of outcome once infection and clinical disease had occurred. We found the light colored horses had greater odds of dying from WNV clinical disease than did dark colored horses. A possible explanation could be that the immune response is linked to color genetically. However, no references to justify this were found. Further information is needed to determine the role of coat color in the outcome of WNV clinical disease.

Other factors assessed in this study, particularly age and vaccination, were not significantly associated with the risk of fatality in clinical cases. In other studies, older affected horses have been more likely to die (8,16). In this study, the case fatality rate was highest in horses ≤ 5 y; however, this was not statistically significant. In this study, only 7% of horses reported to have clinical signs in 2003 were vaccinated, so there was little power to assess vaccine effectiveness in prevention of fatality. Other studies have shown that the vaccine has a protective effect against fatality in clinically affected horses (8,16).

The environmental variables (temperature, precipitation, GDD, and mosquitoes) examined also were not associated with individual horse risk of fatality. Calculation of these variables involved averaging 3 to 5 mo of data, which may not represent the exposure period of each clinically affected horse or the developmental period of the mosquitoes prior to exposure. However, when temperature and precipitation were calculated by using only data from the week before the estimated exposure date of each clinical horse, the results were still not statistically significant. With so much variation in environmental variables by local macro- and micro-climates, the use of the nearest climate station may not have represented the conditions at the actual location of the clinically affected horse. More likely, however, is the conclusion that environmental conditions would not play a role in an individual horse’s health outcome once it had become infected with WNV, rather it would be associated with the risk of infection only.

The role of various other prevention strategies were also explored. Mosquito reduction and avoidance strategies were not widely used by the owners’ of clinically affected horses. None of the mosquito reduction methods were significant predictors of fatality. Housing horses in a barn at night has been suggested as a means of preventing infection and disease. In this study, no horses with WNV clinical disease were housed in a barn at any time. When variables sex, coat color, and onset of clinical signs were included in the analysis, the protective effect of simple shelter was not significant. Even with these findings, avoidance of mosquitoes is still the best method of prevention of clinical disease, with death as an possible outcome. In addition, owners should be aware of the timing and risk of WNV infection in their geographic area, so that they can use all methods available to them in the prevention of infection and in having a favorable outcome of clinical cases.