Materials and Methods

Outbreak information—According to the USDA, the 2004 outbreak of VS started on May 18, 2004, and ended on January 14, 2005.⁸ During that period, 490 livestock premises were investigated by state departments of agriculture (or equivalent agencies) and the USDA in Texas (93/490 investigated premises [19.0%]), New Mexico (98/490 [20.0%]), and Colorado (299/490 [61.0%]). Investigations were initiated on the basis of suspicious clinical signs reported by veterinarians (practitioners) or other animal caretakers. At the end of the outbreak, there were 294 premises (15/294 [5.1%] in Texas, 79/294 [26.9%] in New Mexico, and 200/294 [68.0%] in Colorado) confirmed with VS, including positive results for 403 horses, 1 donkey, 1 mule, 63 cattle, 1 alpaca, and 1 llama. All animals had positive results for the New Jersey virus serotype. The remaining 196 investigated premises (78/196 [39.8%] in Texas, 20/196 [10.2%] in New Mexico, and 98/196 [50.0%] in Colorado) were confirmed negative for VS.

Sample population—A list of all premises investigated for VS between May and September 2004 was obtained from the states and USDA in the fall of 2004. The list contained contact information for the premises, the animal species suspected of having VS, test results, and final status of the premises (positive or negative for VS). All livestock premises investigated in Texas (81/369 [22.0%]), New Mexico (94/369 [25.4%]), and Colorado (194/369 [52.6%]) in that specific time period during the outbreak were included in the study.

The definition used by the USDA to classify an animal as the index case for VS infection was clinical signs compatible with VS and either a positive result for the VI test or a 4-fold increase in antibody titers for the CF or SN tests on paired sera obtained at least 7 days apart. Subsequent cases were defined as animals with compatible clinical signs of VS and at least one of the following: a positive result for the VI test; a positive result on an antigen-capture ELISA, IgM ELISA, or CF test; or a 4-fold increase in antibody titers for the CF or SN tests in paired sera obtained at least 7 days apart. However, most of the subsequent VS cases were diagnosed on the basis of clinical signs compatible with VS and a positive result for the VI or CF test. All tests were performed at the National Veterinary Services Laboratories at Ames, Iowa, or the National Veterinary Services Laboratories Foreign Animal Disease Diagnostic Laboratory at Plum Island, New York, in accordance with their protocol that was based on published methods.²³ Premises with at least 1 animal fulfilling these criteria were considered positive for VS virus and classified by the states and USDA as VS case premises. Premises where investigated animals had negative results for testing in accordance with the aforementioned protocol were considered negative for VS virus infection. For purposes of the study reported here, negative premises were classified as control premises. In addition, premises at which the primary animal suspected of having VS was a horse were defined as equine premises. In equine case premises, the first horse identified as positive for VS by use of the aforementioned definitions was classified as a case horse. The first horse in an alphabetic list of names of horses at each equine control premises was classified as a control horse. On the basis of these definitions, there were 215 (17/215 [7.9%] in Texas, 77/215 [35.8%] in New Mexico, and 120/215 [55.8%] in Colorado) case premises and 154 (63/154 [40.9%] in Texas, 15/154 [9.7%] in New Mexico, and 75/154 [48.7%] in Colorado) control premises. Among these, there were 187 (13/187 [7.0%] in Texas, 77/187 [41.1%] in New Mexico, and 97/187 [51.9%] in Colorado) equine case premises and 107 (37/107 [34.6%] in Texas, 11/107 [10.3%] in New Mexico, and 59/107 [55.1%] in Colorado) equine control premises.

Data collection—A questionnaire used in another VS study⁶ was modified and adapted for use in the study reported here. The questionnaire comprised 2 sections, which requested information about the livestock premises and specific horses (when applicable). Information collected regarding premises included the primary animal species with clinical signs compatible with VS; area of the premises; main purpose of the premises (eg, breeding operation or dairy farm); use and type of chemical (eg, sprays or pour-on products) and nonchemical (eg, insect traps or fly strips) measures used to control insect populations on animals or the premises; predominant type of vegetation coverage on the premises; whether there was a water body (and the type of water body) on the premises; visitation and type of visitors (eg, veterinarians, farriers, and trainers) on the premises; animal movement to and from the premises; sharing of housing, feed, or water by multiple animal species; and prior vesicular or ulcerative disease on the premises. Information collected on specific horses included sex; age; breed; use and type of chemical (eg, sprays or wipes) and nonchemical (eg, fly masks or sheaths) measures used to prevent insect bites or harassment; amount of time spent in shelters, drylots, or pasture; type, quality, and origin of hay or forage; main source of water; and method of water delivery. Most questions referred to the 2-week period prior to the VS investigation, except those related to premises characteristics (such as area and purpose of the premises, vegetation coverage, water body and type of water body, and prior VS on the premises) or horse characteristics (such as sex, age, and breed).

Questionnaires were administered to the owner or person in charge of the management of the premises or care of the livestock during telephone interviews conducted between September and December 2004. Twenty-four second-year veterinary students participating in the honors credit section of the Preventive Veterinary Medicine class at the College of Veterinary Medicine and Biomedical Sciences at Colorado State University conducted the interviews. Each student was provided with a telephone card to perform the calls and a standardized script to introduce the study. The number of premises contacted was allocated equally among the participating students. Prior to the start of data collection, training sessions were conducted with the students to review the questionnaire, address concerns regarding content and clarity of the questions, and standardize data collection and recording procedures. Weekly meetings between students and the study coordinator (PCD) were conducted until the end of the study to assess the project status and discuss problems related to use of the questionnaire. At least 3 attempts were made to contact someone at each premises. An effective contact was defined as an established telephone connection with 1 or more valid answers entered in the questionnaire.

In this study, the interviewers were aware of the disease status of the premises. They needed to know the disease status of the premises to obtain information for case and control horses. Methods used in the study were approved by the CSU Human Research Committee.

Data analysis—Data were entered into a computer database.^a Data were validated, and separate analyses were conducted at the premises and animal (horse) level. Descriptive statistics for the various variables were calculated on the basis of the total number of valid, nonmissing responses for each variable. For categoric variables, categories were merged when they contained a small number of observations or when it made biological sense. For example, because insects reproduce in various types of water environments, the types of water body on a premises were merged, and results were compared with those for premises reporting no water body. Stratified analysis and the χ² test were used for variable selection and assessment of associations among variables and between variables and VS status. The OR was used as the measure of strength of association between variables and VS status of the premises or animal. Odds ratios that differed from 1 with a liberal χ² test (P ≤0.1) indicated a protective (OR < 1) or risk (OR > 1) effect of variables on disease. Changes in ORs by a third variable indicated a confounding effect when stratum-specific ORs were similar but crude and the Mantel-Haenzel (adjusted) OR differed by > 10%. When stratum-specific ORs differed from each other and the value for the Breslow-Day χ² test for interaction was P ≤0.1, then an interaction effect was considered. Variables that had protective, risk, confounding, or interaction effects in the stratified analysis and that were suspected to be associated with disease (on the basis of historical information) were included in logistic regression models. Results were interpreted on the basis of adjusted ORs and 95% CIs estimated from the logistic regression models. Associations were interpreted on the basis of the magnitude and precision of the OR estimates and their biological plausibility. Further evaluation of confounding effects was performed by adding and deleting variables from each model, including variables initially left out of the models, and assessing changes in OR estimates, as described previously. The final models included the variables selected in the stratified analysis and any other variable that had a confounding effect during the modeling process. The methods for data analysis, variable selection, and model building used in the study were based on published guidelines.²⁴ When an interaction term was included in the model, 95% CIs were calculated by use of the variance and covariance values for the logistic regression coefficients, as described elsewhere.²⁵ The Hosmer-Lemeshow test was used to assess the goodness-of-fit of the models to the data. Values of P > 0.05 in this test indicated a good fit. Adjusted ORs for the variables not considered associated with disease were estimated by use of logistic regression models. All analyses were performed by use of commercially available statistical software programs.^b,c

Results

Premises-level data—People at 138 premises (18/138 [13.0%] in Texas, 41/138 [29.7%] in New Mexico, and 79/138 [57.2%] in Colorado) were effectively contacted during the survey. Of these, 100 (9/100 [9.0%] in Texas, 37/100 [37.0%] in New Mexico, and 54/100 [54.0%] in Colorado) were case premises and 38 (9/38 [23.7%] in Texas, 4/38 [10.5%] in New Mexico, and 25/38 [65.8%] in Colorado) were control premises. Equids (including 1 donkey) were the primary species investigated on the majority (118/134 [88.1%]) of these premises, followed by cattle (11/134 [8.2%]) and other species (sheep, goat, and llama, 5/134 [3.7%]). The median premises area was 4.9 hectares (12 acres), with most (114/134 [85.1%]) consisting of premises ≤ 40.8 hectares (≤ 100 acres). The reported main purpose of the premises was residence (36/136 [26.5%]), followed by use for pleasure (35/136 [25.7%]), breeding operation (17/136 [12.5%]), boarding facility (9/136 [6.6%]), cattle ranch (8/136 [5.9%]), and other (31/136 [22.8%]).

Most (102/137 [74.5%]) of the premises used measures to control insect populations in the 2 weeks prior to the investigation. Most (88/96 [91.7%]) used chemical-based methods, although some (35/76 [46.0%]) used nonchemical methods. The most frequently used methods of chemical and nonchemical insect control measures reported were sprays (51/64 [79.7%]) and insect traps or strips (16/30 [53.3%]), respectively. Most (101/137 [73.7%]) premises had grassland or pasture as the predominant type of vegetation coverage. The predominant type of vegetation coverage for the remaining premises (36/137 [26.3%]) included scrub and brush rangeland (19/36 [52.8%]), hardwood or evergreen forest (12/36 [33.3%]), cropland (2/36 [5.6%]), and other (3/36 [8.3%]). A water body on the premises was reported for most (87/125 [69.6%]) of the premises. These included irrigation canals (35/87 [40.2%]), rivers or streams (23/87 [26.4%]), lakes or ponds (17/87 [19.5%]), and other (such as drainage, swamps, or city water; 12/87 [13.8%]). Visitors on the premises in the 2 weeks prior to the investigation were reported for some (40/135 [29.6%]) of the premises and included veterinarians (17/40 [42.5%]), farriers (30/40 [75.0%]), and other (such as breeders, livestock haulers, or trainers; 13/40 [32.5%]). Movement of animals from the premises was most frequently reported (29/134 [21.6%]), followed by movement of animals to and from premises (11/134 [8.2%]) and movement of animals onto premises (7/134 [5.2%]). For approximately a fourth (37/134 [27.6%]) of the premises, multiple species of animals sharing housing, feed, or water in the 2 weeks prior to the investigation was reported. Prior vesicular or ulcerative disease was reported for 9 of 134 (6.7%) premises.

Variables selected for use in the stratified analysis and entered in the logistic regression model for estimation of adjusted ORs included use of measures to control insect populations in the 2 weeks prior to the VS investigation, predominant type of vegetation coverage on the premises, and whether there was a water body on the premises. The ORs, logistic regression coefficients, and 95% CIs were determined (Table 1). There was a good fit of the model to the data (P = 0.77). On premises predominantly covered by grassland or pasture, the odds of being a case premises when measures to control insect populations were used in the 2 weeks prior to the investigation were 0.08 that of being a case premises when measures to control insect populations were not used in that same period. On premises covered by other types of vegetation, the odds of being a case premise were similar for those that did and did not use measures to control insect populations in the 2 weeks prior to the investigation. Coverage by grassland or pasture increased the odds of being a case premises by 11 times for premises that did not use insect control measures in the 2 weeks prior to the investigation. For premises that used measures to control insect populations in the 2 weeks prior to the investigation, the odds of being a case premises were similar between those covered with grassland or pasture and those covered with other types of vegetation. A body of water on the premises increased the odds of being a case premises by approximately 2 times. These associations were not confounded or modified by premises area (ie, number of hectares [acres]). The odds of being a case premises were similar for premises that did and did not receive visitors in the 2 weeks prior to the investigation (OR, 0.9; 95% CI, 0.3 to 2.3); that did and did not move animals to and from the premises in the 2 weeks prior to the investigation (OR, 0.6; 95% CI, 0.3 to 1.6); and where multiple animal species did and did not share housing, feed, or water in the 2 weeks prior to the investigation (OR, 1.3; 95% CI, 0.5 to 3.4). The odds of being a case premises were also similar for premises that did and did not have a history of vesicular or ulcerative disease (OR, 0.8; 95% CI, 0.1 to 4.8).

Table 1—

Results for variables included in premisesand animal (horse)-level logistic regression models to evaluate factors associated with VS.

View Table

Horse-level data—Information was obtained on 86 (7/86 [8.1%] in Texas, 35/86 [40.6%] in New Mexico, and 44/86 [51.2%] in Colorado) case horses and 32 (6/32 [18.8%] in Texas, 4/32 [12.5%] in New Mexico, and 22/32 [68.8%] in Colorado) control horses within respective equine premises. Slightly more horses were males (66/118 [55.9%]) than females (52/118 [44.1%]). Mean age of horses was 13 years (range, 1 to 28 years). Horses were Quarter Horse or Quarter Horse–crossbred horses (48/118 [40.7%]), followed by Paint or Paintcrossbred horses (15/118 [12.7%]), Thoroughbred or Thoroughbred-crossbred horses (10/118 [8.5%]), and other (45/118 [38.1%]). There were no significant (P = 0.6) differences in the frequency distribution for sex, age, and breed between case and control horses.

Measures to prevent insect bites or harassment by insects were applied to 81/118 (68.6%) horses. Use of chemical and nonchemical measures was reported for 79/81 (97.5%) and 31/80 (38.8%) horses, respectively. The most frequently reported methods of chemical and nonchemical control measures were sprays (62/73 [84.9%]) and fly masks (9/13 [69.2%]), respectively. More than half (65/117 [55.6%]) of the horses spent time in some type of shelter. For horses that used shelters, the median time spent in shelters was 13.5 h/d (range, 1 to 24 h/d). Horses that spent time in shelters or drylots were significantly (P = 0.009) less likely to have spent time on pasture. The amount of time spent in shelters, drylots, or pasture was spread over all 7 days of the week for > 80% of the horses. Horses that spent time in shelters or drylots were significantly (P = 0.01) more likely to be fed hay or forage. For horses fed hay, the main type was alfalfa or an alfalfa-grass mix (54/88 [61.4%]). Most of the hay consumed was produced in the same state as the horse's residence (73/88 [82.9%]). When adjusted on the basis of spending time in shelters, there was no significant (P = 0.17) difference in the frequency distribution of hay feeding between case and control horses.

Wells were the most frequently reported source of water for horses (56/118 [47.5%]), followed by city or municipal sources (42/118 [35.6%]) and natural sources, including rivers, creeks, lakes, ponds, or springs (20/118 [16.9%]). Water was most frequently delivered to horses in tanks (83/118 [70.3%]), followed by provision by use of automatic waterers (15/118 [12.7%]), directly from the source (12/118 [10.2%]), and in a bucket (8/118 [6.8%]). Horses that spent time in shelters or drylots were significantly (P = 0.005) less likely to drink directly from the source. There were no significant (P = 0.2) differences in the frequency distribution for water source and delivery method between case and control horses.

Variables selected in the stratified analysis and entered in the logistic regression model for estimation of adjusted ORs included use of measures to prevent insect bites or harassment by insects in the 2 weeks prior to the investigation and a horse spending time in shelters in the 2 weeks prior to the investigation. The ORs, logistic regression coefficients, and 95% CIs were determined (Table 1). There was a good fit of the model to the data (P = 0.83). The odds of being a case horse when measures to prevent insect bites were used in the 2 weeks prior to the investigation were 0.2 that of being a case horse when measures to prevent insect bites were not used in the same period. The odds of being a case horse for horses that spent time in shelters were 0.4 that of being a case horse for horses that did not spend time in shelters.

Discussion

Results of the study reported here provided additional epidemiologic evidence to support the role of insect vectors in transmission of VS virus. At the premises level, use of measures to control insect populations had a protective effect against disease caused by VS virus, specifically when grassland or pasture was the predominant type of vegetation coverage. Concurrently, type of vegetation coverage when measures to control insects were not used and the fact there was a body of water on the premises were found to be risk factors for infection with VS and increased the odds of disease. Variations in the effects of insect control measures by type of vegetation coverage, and vice versa, reflected an interaction between these 2 factors. This interaction could be explained by differences in type and effectiveness of insect control measures used by premises with differences in vegetation coverage or variations in density and prevalence of various insect populations in different environments. In the study, the proportion of premises that used chemical and nonchemical control measures was similar between the 2 categories of vegetation coverage. However, a more specific analysis by type of chemical and nonchemical measures was not possible because of scarce data. Variations in density and prevalence of various types of insects were not assessed as part of the study. Most of the premises enrolled in the study had a predominant coverage of grassland or pasture. Therefore, the sample size in the category of vegetation coverage other than grassland or pasture was relatively small, which reduced the precision of the estimated OR for use of measures to control insect populations in that group.

At the horse level, measures to prevent insect bites and spending time in a shelter decreased the odds of being a case horse. The protective effect of measures to prevent insect bites appeared to be greater in horses that spent time in a shelter, and conversely, the protective effect of a shelter appeared to be evident only when measures to prevent insect bites were used. These potential variations in the effects of measures to prevent insect bites and spending time in shelters could be attributed to variations in type and frequency of use of measures to prevent insect bites and factors that affect frequency and duration of contact between insects and horses (such as type of shelter) and number of hours and period of the day spent in a shelter. In the study reported here, horses on which measures were applied to prevent insect bites typically spent 5 hours more in a shelter, compared with that for horses on which such measures were not applied (data not shown). Because of loss in precision of OR estimates and lack of statistical significance, the potential interaction between these 2 variables was not reported in the results.

In the study reported here, visitors on the premises, movement of animals to the premises, and multiple animal species sharing housing, feed, or water in the 2 weeks prior to the investigation were not found to be risk factors for development of VS. Visits by veterinarians had an apparent protective effect against the disease. However, because multiple visitors were common, the effects of veterinarian visits alone could not be clearly assessed. Movement of animals to the premises was not frequent, which made it difficult to assess its effect. Nevertheless, movement of animals has been considered responsible for spread of VS in prior outbreaks, and direct contact is 1 form of transmission of VS virus.^4,20 Therefore, findings of a lack of association between movement of animals and sharing of housing, feed, or water should not discourage animal caretakers, livestock owners, and authorities from implementing basic biosecurity measures to prevent spread of the disease. Similar to results of another study,⁹ a history of VS or ulcerative disease on the premises was reported for a low percentage of premises in the study reported here and was not found to be associated with VS. Whether this finding reflects a true lack of association or a negative response bias is not known.

Results of the study reported here corroborate and complement findings of another study.⁹ In that study,⁹ a reported increase in biting insect populations, lack of access to shelter, access to pasture, and proximity to running water were found to increase the risk of infection with VS. In addition, a potential interaction between increased insect populations, access to shelter, and proximity to running water was suggested.⁹ However, associations between infection and use of insect control measures and infection and type of vegetation coverage were not detected. In the study reported here, spending time in shelters was inversely correlated with time spent in pastures. Hence, to prevent a correlation between predictor variables (collinearity), only shelter was included in the logistic regression analysis. Time spent in a pasture increased the risk of infection with VS, similar to the results described elsewhere.⁹

Results reported here were interpreted on the basis of the magnitude and precision of the ORs and biological plausibility of the associations in light of prior research findings. The 95% CIs for 2 factors (premises covered by grassland or pasture that did not use measures to control insect populations and whether a horse spent time in a shelter) slightly crossed a value of 1. The rationale to interpret and discuss these as factors associated with infection with VS included the fact that these variables had significant results in the univariate-stratified analysis. The point estimates for the ORs differed considerably from 1, whereas the CIs were only slightly < or > 1. The sample sizes in some categories were small, which reduced statistical power. For instance, there were only 6 premises with vegetation coverage other than grassland or pasture that did not use measures to control insect populations, and only 8 control horses did not spend time in a shelter. In addition, these associations were plausible based on the biological aspects of the disease and prior research findings. Access to shelter has been associated with protection from infection with VS,⁹ avoiding exposure to insect vectors has been recommended for prevention of other vector-borne diseases (bluetongue and West Nile virus) in domestic animals and humans,^26,27 and type of vegetation coverage has been associated with variations in insect populations and vector-borne diseases.^19,28,29

Hematophagous insect species in which virus replication and transmission to susceptible hosts have been verified include black flies (Simulium spp), phlebotomine sand flies (Lutzyomia spp), and biting midges (Culicoides spp).^11,12,16,18 Virus has also been isolated from nonhematophagous insects including house flies (Musca domestica) and eye gnats (Hippelates spp), which suggests that mechanical transmission is possible.¹⁴ In addition, grasshoppers (Melanoplus sanguinipes) have been able in experimental conditions to acquire, replicate, and transmit the virus to susceptible hosts that ingested infected specimens.¹⁷ The role of each potential vector in virus transmission during outbreaks remains to be determined. Knowledge of the insect species most important in virus transmission during outbreaks, in addition to information on the reproductive and feeding processes of those insects, can direct the establishment of specific and effective insect control measures to prevent and control spread of VS. Additional studies on the composition and distribution of insect populations in outbreak zones and the prevalence and efficacy of transmission of the virus in various insect populations may assist in determining the insect species on which to focus control efforts.

In the study reported here, the case and control premises effectively contacted represented 34% and 19%, respectively, of the case and control premises investigated by the states and USDA during the entirety of the 2004 outbreak and 47% and 25%, respectively, of the case and control premises investigated by the states and USDA between May and September 2004. The higher percentage of effective contacts for case premises may have reflected a greater willingness of owners or operators of case premises to respond to the survey. Nonparticipation bias is possible when a pattern of exposure differs among participants (premises effectively contacted) and nonparticipants (premises available for contact but not effectively contacted and premises investigated after September 2004 and not available for contact). On the basis of the state distribution for participant (13%, 30%, and 57% for Texas, New Mexico, and Colorado, respectively) and nonparticipant (21%, 16%, and 63% for Texas, New Mexico, and Colorado, respectively) premises, it is possible there was some degree of selection bias. Texas and New Mexico are underrepresented and overrepresented, respectively, among the participant premises. It is not possible to assess the magnitude and direction (underestimation or overestimation) of effect of nonparticipation bias in the OR estimates reported here. In addition, whether the distribution for the 3 states is a reasonable proxy to assess nonparticipation bias for other factors is not known. However, the results reported here are consistent with other research findings and the biological aspects of VS, as discussed earlier.

Case (false-positive) and control (false-negative) premises and horses may have been misclassified. Variables that influence the degree of premises misclassification include sensitivity and specificity for the diagnostic test, number of animals tested for each premises, cutoff number of animals with positive test results necessary to classify a premises as positive, and true prevalence of infection for a premises.³⁰ Population estimates of sensitivity and specificity of the most commonly used tests were not found in the literature. However, another study²³ in which investigators compared results of CF tests, competitive ELISA, IgM ELISA, and SN tests in experimentally infected animals revealed detection of exposure by all methods. The low cutoff number of animals (≥ 1 VS-positive animal) used to classify a premises as positive could have increased the likelihood of false-positive premises. On the other hand, the low number of animals tested for each premises could have led to false-negative premises. Typically, only suspected animals (1 to 3) were tested on investigated premises. The prevalence of infection for premises was not determined. However, a study³¹ on equine sentinel premises in Colorado revealed within-herd seroprevalence varied between 0% and 100%, which suggests a wide range of possible true prevalence values. In the study reported here, all management information obtained for the survey referred to the 2-week period prior to the stateUSDA investigation. It is possible that there were management changes prior to the investigation but after the virus had been introduced onto the premises because of knowledge of the ongoing outbreak. This could have led to spurious associations between management factors and VS. The results reported here are biologically plausible and consistent with prior research results; hence, this was likely not an important source of bias.

In case-control studies, ORs are estimated as an approximation of the relative risk because measures of disease incidence cannot be estimated in such study designs.³² However, for this approximation to hold true, it is necessary to assume that the disease is rare in the target population.³² Whether this assumption holds true for the study reported here depends on the definition of the target population. When the target population is considered as all of the premises investigated for VS during the outbreak, the rare disease assumption may not hold true because most of the investigated premises were, in fact, VS-confirmed cases. In such situations, the ORs typically overestimate effects³²; hence, the protective effects of insect control measures and shelter or the risk effects for type of vegetation coverage and a body of water on the premises would be less pronounced. The term odds (instead of risk) of being a case premises or case horse was used in the study reported here so that it would not be implicit that the rare disease assumption holds true.

Finally, in the process of fitting the logistic models to the data, only records with complete information were used. Hence, there was a potential for loss of precision attributable to sample size reduction and introduction of bias should the pattern of exposure differ between records included and excluded from the model. In the study reported here, 89% of the observations at the premises level were used for model fitting. However, there were no major differences in the ORs estimated by use of the stratified analysis and logistic regression methods, except that estimates from the stratified analysis were slightly more precise in some instances because of larger sample sizes in specific strata. The advantage of the use of logistic regression in this situation was that the estimated ORs were adjusted for > 1 variable at a time, which reduced potential residual confounding effects. The OR estimates at the premises level should be considered conservative. At the horse level, 117 of 118 (99%) observations were used in the logistic regression model. Hence, changes in precision or introduction of bias during modeling were not a concern. On the basis of the goodness-of-fit statistic, both logistic regression models fit the data.

On the basis of the study reported here, the combination of use of insect control measures and providing a shelter for horses may assist in the prevention and control of VS. It is probable that a combination of insect control measures that includes prevention of insectanimal contact as well as control of insect populations (such as treatment of sites that harbor insects) would be most effective. In addition, shelters that are most effective for preventing the access of insects to the animals would probably be more successful in the prevention of the disease. Owners of premises covered with grassland or pasture or that have a body of water on the premises should be aware of their increased risk for VS.