use of a case-control study and geographic information systems to

Vet. Res. 38 (2007) 37–50 37c© INRA, EDP Sciences, 2007DOI: 10.1051/vetres:2006043

Original article

Use of a case-control study and geographicinformation systems to determine environmental and

demographic risk factors for canine leptospirosis

George S. Ga,b*, Joshua H. Vc, Bruno B. Cd,Philip H. Kd, Daphne A. Da, Michael L. Je

a Center for Companion Animal Health, School of Veterinary Medicine, University of California,Davis, CA 95616, USA

b RTI International, 3040 Cornwallis Rd, PO Box 12194, Research Triangle Park, NC, 27709-2194, USAc Department of Environmental Science and Policy, University of California, Davis, CA 95616, USA

d Department of Population Health and Reproduction, School of Veterinary Medicine,University of California, Davis, CA 95616, USA

e John Muir Institute of the Environment, University of California, Davis, CA 95616, USA

(Received 1 April 2006; accepted 19 July 2006)

Abstract – Leptospirosis is increasingly diagnosed as a re-emerging canine disease in the USA.Our objectives were to describe potential risk factors for canine leptospirosis infections in northernCalifornia, through the use of a case-control study, and to perform a spatial analysis to investigatewhich aspects of the landscape and land use patterns are important in the transmission of leptospiro-sis. Forty-three cases and 59 controls were enrolled into the study. Serological results showed that17 (39.5%) of the 43 dog cases were infected with serovar pomona. Cases were 7.86 times morelikely to have been walked in a rural environment rather than an urban environment. Cases also hadeight times higher odds of swimming in outdoor water, and approximately 12 times higher oddsof drinking from outdoor water in the two weeks preceding illness. At smaller distances from thedogs’ homes (radius ≤ 0.5 km) hydrographic density was positively correlated with cases, while atlarger distances (radius ≥ 5 km) there was a positive relationship between leptospirosis cases andpercent of wetlands or public open space. Intervention measures for the prevention of canine lep-tospirosis should include reducing access to potentially infectious bodies of water that are close tocanine homes, and to large areas of wetlands and public open space in the general vicinity. We haveshown that a spatial analysis in conjunction with traditional epidemiological analysis is a powerfulcombination in identifying risk factors for infectious diseases.

leptospirosis / dogs / GIS / California / disease geography

1. INTRODUCTION

Leptospirosis is a widely distributed dis-ease of humans, domestic animals, andwildlife caused by bacteria of the genusLeptospira. There are currently over 200

* Corresponding author: [email protected]

pathogenic serovars of this organism, andit has a worldwide distribution [8]. Lep-tospirosis has long been known as a dis-ease of dogs, with Leptospira interrogansserovars canicola and icterohaemorrha-giae as the primary agents historically in-volved in canine infections. From the early

Article available at http://www.edpsciences.org/vetres or http://dx.doi.org/10.1051/vetres:2006043Article available at http://www.edpsciences.org/vetres or http://dx.doi.org/10.1051/vetres:2006043

http://www.edpsciences.org/vetres

http://dx.doi.org/10.1051/vetres:2006043

38 G.S. Ghneim et al.

1970’s to the early 1990’s there was a sharpdecline in the incidence of canine lep-tospirosis throughout the USA [16]. Thischange was attributed to the commercialleptospirosis vaccine that was commonlyused at that time, since that vaccine pro-tected dogs against serovars canicola andicterohaemorrhagiae [16].

But the last decade has seen a rever-sal of this trend, and now leptospirosis isbeing increasingly diagnosed, and docu-mented as a re-emerging canine diseasenationwide in the USA [1,3,22,23,25]. Co-inciding with the increase in incidence wasa change of the infecting serovars. In recentstudies dogs were primarily infected withthree serovars: bratislava, grippotyphosa,and pomona, which were not included inhistoric canine vaccines. It should also benoted that most of the diagnoses weremade based on serology, primarily the mi-croscopic agglutination test (MAT), andthat cross-reactivity among serovars is acommon occurrence [3].

We know that leptospirosis is intri-cately tied to fresh water, and human orcanine outbreaks have usually followedheavy rainfall or flooding [7, 21]. But itis still unclear what led to the recent in-crease in incidence of canine leptospirosisand more specifically the change in infect-ing serovars. Leptospira bacteria are shedin the urine of an infected individual andsurvive well only in a moist and pH neu-tral environment [7]. Infection is spreadwhen another animal’s mucous membranesor disrupted epithelium come into contactwith contaminated water [7]. While ro-dents are generally considered the primaryreservoir for leptospirosis, cattle and smallmammals such as raccoons, opossums, andskunks are often implicated as other im-portant reservoirs [16, 17]. The diagnos-tic challenges associated with this diseasemake monitoring incidence difficult andonly with time consuming and costly ac-tive surveillance can we accurately assessthe incidence of leptospirosis. Dogs are

good sentinels for detecting the presenceof Leptospira in the environment and arekey factors to understanding the ecology ofthis disease. Because of their importance aspets, many dogs benefit from aggressive di-agnostics and treatment which has enabledclinicians and researchers to gain insightinto the current status of leptospirosis inmany geographic areas.

Our primary objective was to describethe potential risk factors associated withcanine leptospirosis infections in northernCalifornia, through the use of a retrospec-tive case-control study of dogs that werepresented at the Veterinary Medical Teach-ing Hospital (VMTH) at the School ofVeterinary Medicine, University of Cali-fornia, Davis, USA.

Our secondary objective was to performa spatial analysis to investigate which as-pects of the landscape and land use patternsare important in the transmission of lep-tospirosis. We used a Geographical Infor-mation System (GIS) as a tool to performour spatial analyses. Recent applications ofthese spatial analytic techniques to vectorborne diseases, such as Lyme disease [12],and arboviral diseases [15] have demon-strated the emerging utility of this tech-nique.

2. MATERIALS AND METHODS

2.1. Case-control study

2.1.1. Animals

We performed a search in the VMTHdatabase for any dog that was serologicallytested for leptospirosis from January 1st1998 to December 31st 2000. The MATwas the primary diagnostic test used at theVMTH. Not all canines included in thestudy had paired serum titers performed.Therefore, a case definition was developedthat was applied to individuals that weresampled at least once. Dogs that had a

Risk factors for canine leptospirosis 39

single titer of 1:800 or greater without in-dication of vaccination for leptospirosis inthe three months prior to illness were pre-sumptively considered to be infected withLeptospira [3], and were enrolled as cases.All dogs that had been vaccinated in thethree months before being examined at theVMTH were excluded, to prevent any mis-classification. It was necessary to excludethese recently vaccinated dogs because wecould not evaluate which vaccines the dogshad received from their regular veterinar-ian. Similarly, any dogs with titers of 1:400or 1:200 (i.e. suspects) were excluded toensure that the results were valid and toavoid misclassification. Since serologicaltiters in response to leptospirosis infectionscan decrease rapidly, it is unclear if dogswith a titer of 1:200 or 1:400 were truly in-fected or just had a residual titer from pastvaccinations [3]. Dogs that had been ex-amined at the VMTH and had leptospirosisserology performed with resulting titers of1:100 or less were enrolled as controls. Thecontrols were generally dogs with acute re-nal disease, as the primary clinical sign inthe most recent cases of leptospirosis wasgenerally acute renal disease [1, 17, 25].Due to the small sample size, we selectedall the controls that were available, result-ing in a 1 to 1.4 ratio of cases to controls,and we could not perform any matching.The serovar with the highest titer was re-ported as the serovar most likely to havecaused the infection [4].

2.1.2. Demographics

All cases and controls were examinedfor any difference in demographic at-tributes that included county of residence,breed category, gender, and age. For dogswith two residences, we used the primaryresidence for county of residence. Countieswere designated as either coastal or inland.Coastal counties were those defined as anycounty that bordered the ocean, bay, or

delta water. Breeds were classified into theseven breed categories, as defined by theAmerican Kennel Club Inc. (AKC), withan eighth category designated for all mixedbreed dogs. These seven AKC categoriesare sporting dogs, terriers, working dogs,non-sporting dogs, herding dogs, houndsand toy dogs. Four gender categories weredefined as intact male, neutered male, in-tact female, and neutered female. We de-fined age groups as less than one year, oneto three years, four to seven years, andeight years or older.

2.1.3. Questionnaire

We developed a questionnaire that ad-dressed various environmental issues thatwere potential risk factors for leptospiro-sis infection. The questionnaire includedquestions about 21 potential risk factors.The questions dealt with all aspects of adog’s environment that could lead to ex-posure to Leptospira organisms, e.g. ex-posure to water, being off leash, walkingin an urban environment, and exposure towildlife. Vaccine history was excluded be-cause few owners knew if the Leptospiraportion of the vaccine was included intheir dog’s most recent annual vaccination.We also included a question relating towhether the owners saw wildlife speciessuch as rodents, raccoons, skunks, or opos-sums in the dog’s environment. The ques-tionnaire was administered to the ownersof both cases and controls and referred tothe two week period before the patientswere seen at the VMTH.

2.1.4. Data analysis

We evaluated differences in exposure todemographic and environmental risk fac-tors in cases versus controls using chi-square tests of homogeneity and comput-ing odds ratios (OR) with 95% confidence


intervals (95% CI) and P-values. Thosevariables whose associated P-values were< 0.05 were further analyzed using an un-conditional logistic regression model. Foreach category of variables, all variablelevels were compared to the first level(baseline) of the variable. Adjusted ORand 95% CI were estimated using EGRETStatistical Software ( c© 2001 Cytel Soft-ware Corp., Cambridge, MA, USA). Lev-els within variables for which there wereno values were removed.

2.2. GIS study

2.2.1. Map design

Each case and control dog was iden-tified for location by using Topo USA3.0 ( c© DeLorme, Yarmouth, ME, USA)address-matching software to determinea general latitude and longitude. We ex-cluded cases and controls for which wecould not obtain a home address. We alsoexcluded three geographic outliers whichhad home addresses in Nevada (1) and thesouthern portion of California (2), since itwas not reasonable to include these dogsin the same geographic analysis; further-more, for dogs with more than one res-idence, we used the primary residence.A point coverage of case and control lo-cations was generated and projected inArcGIS 8.1 (Environmental Systems Re-search Institute, Redlands, CA, USA) toAlbers Equal Area (Fig. 1), to match pre-existing geospatial data. Initial results fromthe case-control portion of the study ledus to focus on a suite of landscape scalefactors to be analyzed within the GIS.These included hydrographic features (e.g.streams and rivers), forms of land use (e.g.urban, forest, agriculture, and wetland),and accessibility to open space.

We examined cases and controls asa nominal response variable against sev-eral landscape factors at varying scales

within the GIS to (a) determine which fac-tors are likely better predictors for coarsescreening of disease carriers and (b) to de-termine at which scale these factors arelikely more influential. We addressed thesecond objective by buffering each pointusing fixed radii of 0.5 km (78.5 ha),2 km (1256.6 ha), 5 km (7854 ha), and10 km (31415.9 ha). These circular areas,which we termed Levels of Spatial Influ-ence (LSI), contain landscape features indifferent proportions (Fig. 2) allowing usto detect appropriate scales of influence byvarious landscape predictors. The four dif-ferent radii represent proxies for differentmodes of exposure; LSI 1 to LSI 4 en-compass a range of dog-owner behaviors,from off-leash near-home access, to neigh-borhood walks, to local park visitation, andto regional park access. For each LSI, wecalculated landscape predictors from threeseparate sources of data: stream features,where we calculated hydrographic density(unit length/unit area LSI); percent man-aged open space (total area of managedopen space/unit area LSI); and percent landuse category (total area of land use/unitarea LSI) using urban, forest, agriculture,and wetland categories.

Our stream features are defined by theNational Hydrographic Dataset (NHD)1;which included streams, rivers, lake edges,and other similar edge feature types. Al-though our water features (e.g. ponds,streams) are effectively static, preva-lent dynamic hydrological processes (e.g.floods) are inherently spatially coincidentwith the presence of pre-existing water fea-tures.

We focused on public open space, man-aged for multiple uses such as recreation

1 USGS 2003, Standards for National Hy-drography Dataset: National Mapping ProgramTechnical Instructions, 1999, US Environmen-tal Protection Agency, US Department of theInterior, US Geological Survey National Map-ping Division, http://nhd.usgs.gov/.


Figure 1. Geographical representation of canine leptospirosis cases and controls in northern Cali-fornia identified during the three year period, 1/1/1998 to 12/31/2000. (A color version of this figureis available at www.edpsciences.org/vetres.)

as a landscape factor. Open space, as de-fined for our study, and developed byothers [2] includes the following: city, re-gional, state, and national parks; land trustsand sanctuaries; wilderness areas and na-ture reserves; recreation areas and beaches;as well as parcels designated as conserva-tion easements. While not all such openspaces guarantee access to dogs, they doserve as preserved habitats for wildlife.We also focused on land uses, as definedand mapped by the National Land CoverDatabase (NLCD 2001) [10]. We isolatedthe following land uses: urban areas, char-acterized by > 30% constructed materi-als (e.g. asphalt, buildings, etc.); forestedlands, characterized > 25% by tree cover;agriculture, defined as > 75% land plantedor cultivated for food, feed, or fiber (note:

we excluded urban grasses); and wetlands,which were identified using the saturatedsoil criteria defined by Cowardin [5].

2.2.2. Statistical analyses

We examined the overall geographictrend for any spatial biases by perform-ing a bivariate logistic regression analysisof cases and controls as a function of lati-tude and longitude. We also examined therelationship between mean annual precip-itation and disease status, by examininglong term rainfall data [6]. We then per-formed a series of bivariate logistic regres-sion analyses to examine factor relation-ships between case/control responses andhydrographic density, open space avail-ability, and categorical land-uses for each


Figure 2. Example of a case/control with the four LSI, and types of managed open spaces. (A colorversion of this figure is available at www.edpsciences.org/vetres.)

LSI. We constructed multivariate mod-els for each LSI using respective land-scape factor values, selecting forwardstep-wise predictors (P ≤ 0.10) to de-termine relative scale of influence usingReceiver Operating Characteristic (ROC)area-under-the-curve (AUC) for compar-ing model performance [18]. We thencreated a single, multi-scale multivariatenominal logistic regression model to seeif similar landscape factors predominatedacross scales and across the study area.We used JMP (version 5.0) for all statis-tical analyses (SAS Institute, Cary, NC,USA) and examined our results for poten-tial non-linear relationships by performingsimilar regressions with square and squareroot transformations of the independent

variables. We also assessed linearity ofour continuous predictor variables to avoidmodel misspecification by using a Box-Tidwell transformation test with interac-tion terms in our full model. Lastly, in ourfull landscape model, we report likelihoodratio (LR) χ2 values of significance forthe continuous independent factors, sincethe Wald statistic is prone to Type II er-rors [13].

3. RESULTS

3.1. Case-control study

3.1.1. Animals

We enrolled 43 cases and 59 controlsinto the study, since 28 individuals were


Table I. Leptospira interrogans serovar(s) in-fections diagnosed by MAT.

Serovar(s) with Number Diagnosed

highest titer (N = 43)

pomona 17 (39.5%)

canicola 4 (9.3%)

bratislava 2 (4.7%)

grippotyphosa 2 (4.7%)

icterohaemorrhagiae 2 (4.7%)

hardjo 1 (2.3%)

pomona/ 11 (25.6%)

bratislava

grippotyphosa/ 1 (2.3%)

canicola

pomona/bratislava/ 2 (4.7%)

icterohaemorrhagiae

pomona/bratislava/ 1 (2.3%)

grippotyphosa/canicola

defined as suspects and thus not included inthe study. Questionnaires were completedfor only 30 cases (70%) and 36 con-trols (61%), due to a lack of current con-tact information or refusal. The serologi-cal results for the 43 cases demonstratedthat 17 (39.5%) dogs were infected withserovar pomona, four (9.3%) were infectedwith serovar canicola, two dogs (4.7%)were infected with serovar bratislava, twodogs (4.7%) were infected with serovargrippotyphosa, two dogs (4.7%) were in-fected with serovar icterohaemorrhagiae,one dog (2.3%) was infected with serovarhardjo, and 15 dogs (34.9%) had equaltiters to more than one serovar (Tab. I).

3.1.2. Unconditional multiple logisticregression model

For the demographic risk factors, onlyage was statistically significant (Tab. II).Cases had significantly higher odds of be-ing less than a year old or of being eight

years or older when compared to dogs be-tween one and three years of age. Theseresults indicate that cases were approxi-mately 16.5 times more likely to be juve-nile dogs and approximately 11 times morelikely to be older dogs than dogs betweenone and three years of age. The compar-isons of intact males to all females, andcoastal versus inland counties were border-line significant (Tab. II).

The analysis of the questionnaire re-sponses indicated five significant environ-mental risk factors (Tab. III). Cases had11.15 higher odds of living only in oneresidence rather than sharing their timebetween two residences. Cases were 7.86times more likely to have been walked ina rural environment rather than an urbanenvironment. Our examination of watersources or non-specific swimming activ-ities did not lead to any significant riskfactors. However, we determined that dogsthat drank outdoor water or swam in out-door water in the two weeks before presen-tation at the VMTH were at risk. Cases hadeight times higher odds of swimming inoutdoor water, and approximately 12 timeshigher odds of drinking from outdoor wa-ter in the two weeks preceding illness. Thelast significant risk factor was wild animalexposure, with cases having approximatelyfive times higher odds of having a knownexposure to wild animals (e.g., rodents,skunks, opossums, raccoons, or coyotes).Exposure to different categories of wild an-imals was not significant.

3.2. GIS study

Geographic location was useable for36 (84%) cases and 50 (85%) con-trols (Fig. 1). Examination for geographictrends indicated no spatial correlation inlatitude (P = 0.78). Longitude was sig-nificant (P = 0.035), but accounted forminimal variation in our observations (r2 =

0.038). Such longitudinal trends are of


Table II. Case-control study of leptospirosis in dogs, VMTH, Jan 1, 1998–Dec 31 2000: Demo-graphic risk factors, and logistic regression model results.

Risk factor Variables Cases Controls Total Odds ratio 95% Confidenceinterval

County Inland 12 28 40 1 NACoastal 28 29 57 2.25 (0.96, 5.29)

Breed group Mixed 11 8 19 1 NAToy 0 3 3 0.00 (0.00, ∞)

Hound 6 3 9 1.45 (0.28, 7.64)Herding 8 12 20 0.48 (0.14, 1.74)

Non-sporting 0 10 10 0.00 (0.00, ∞)Working 6 7 13 0.62 (0.15, 2.58)Terrier 1 4 5 0.18 (0.017, 1.95)

Sporting 11 12 23 0.67 (0.20, 2.27)

Gender Intact females 2 7 9 1 NANeutered females 15 24 39 2.19 (0.40, 11.96)

Intact Males 10 6 16 5.83 (0.90, 37.82)Neutered males 16 22 38 2.55 (0.47, 13.91)

All females 17 31 48 1 NAIntact males 10 6 16 3.04 (0.94, 9.81)

Age 1–3 years 1 11 12 1 NA< 1 year 3 2 5 16.50 (1.09, 250.20)4–7 years 16 24 40 7.33 (0.86, 62.50)≥ 8 years 22 22 44 11.00 (1.31, 92.63)

possible concern because precipitation ispositively correlated with longitude – dueto the proximity of the Pacific Ocean to ourstudy area (r = 0.36; n = 86 case/controls)and to the state of California (r = 0.67; n =1000 random sites). However, there was nodetectable difference in mean annual pre-cipitation (1961–1990 annual average) be-tween cases and controls (P = 0.33). Whenobservations greater than –120.10 ˚W inlongitude were excluded (n = 5 con-trols), the longitudinal trend was no longerpresent (P = 0.38). Furthermore, theseeasterly controls were significantly furtherfrom the geographic center of all obser-vations than the other data (P < 0.0001;mean difference = 164.9 km) and were ex-cluded from all subsequent analyses.

We analyzed cases as a positive re-sponse using logistic regression. Our initialscreening results showed that log-distanceto the nearest hydrographic feature waspositively related to infection status (P =0.0006, pseudo-r2 = 0.11), indicatingthat the probability of infection increasedthe closer the dwelling address was to ahydrographic feature. However, we alsowanted to determine which other factors,and at which scale of proximity, were act-ing to promote infection. To address thisquestion, we constructed multivariate lo-gistic regression models at each of our fourlevels of spatial influence using a suite oflandscape factors.

We used forward stepwise nominal lo-gistic regression to find the best model at


Table III. Case-control study of leptospirosis in dogs, VMTH, Jan 1 1998–Dec 31 2000: Environ-mental risk factors with significant logistic regression model results.

Risk factor Variables Cases Controls Total Odds 95% confidenceratio interval

2nd residence Yes 1 10 11 1.00 NANo 29 26 55 11.15 (1.34, 93.18)

Where walked Urban 2 10 12 1.00 NARural 11 7 18 7.86 (1.31, 47.05)

Remote (wild) 14 15 29 4.67 (0.87, 25.14)

Swam 2 weeks before illness No 18 32 50 1.00 NAYes 9 2 11 8.00 (1.56, 41.14)

Drank outdoor water 2 weeks before illness No 9 28 37 1.00 NAYes 16 4 20 12.44 (3.30, 46.97)

Wild animal exposureNo 2 10 12 1.00 NAYes 28 26 54 5.39 (1.08, 26.92)

each scale. At LSI 1 (0.5 km), the bestmodel consisted of a single factor in hy-drographic density (P = 0.0017; ROCAUC = 0.71), which was positively cor-related with infection. The best model forLSI 2 (2 km) included urban cover andagriculture (P = 0.05; ROC AUC = 0.63);these factors were negatively correlatedwith infection. For LSI 3 (5 km) the bestmodel contained solely percent agriculture(P = 0.08; ROC AUC = 0.57), whichwas again negatively correlated with in-fection. Lastly, at LSI 4 (10 km), the bestmodel contained percent open space, per-cent wetlands, and percent agriculture (P =0.02; ROC AUC = 0.67); only agriculturehad a negative parameter estimate, whileopen space and wetlands were positivelycorrelated with infection. From this ap-proach, we concluded that localized effectsof LSI 1 constructed the best single scalemodel based on ROC values, albeit weakly.

To gain a more synoptic view of theeffect of landscape parameters on ob-served infections, we constructed a com-bined scale model with factors from mul-tiple LSI. Again using forward stepwise

nominal logistic regression (P < 0.10),we found that the best combined model(P < 0.0001; ROC AUC = 0.85) includedfive factors from four scales. At LSI 1,hydrographic density was positively cor-related with infection (LR χ2 = 18.94;P < 0.0001; OR = 375.15; 95% CI 19.88–13 303), whereas agriculture was nega-tively associated with infection at LSI 2(LR χ2 = 5.77; P = 0.0163; OR = 0.0006;95% CI < 0.0001–0.37). At LSI 3, per-cent wetland was a positive factor (LRχ2 = 8.21; P = 0.0042; OR 37.35; 95% CI3.03–1 221). At LSI 4, percent open spacefactored positive (LR χ2 = 11.19; P =0.0008; OR = 6189.29; 95% CI 23.36–> 1 000 000), while percent agriculture wasnegative in predicting cases (LR χ2 = 4.15;P = 0.0416; OR 0.12; 95% CI 0.0088–0.93).

4. DISCUSSION

Adin and Cowgill [1] described 36 casesof canine leptospirosis from California thatwere seen at the VMTH between 1990and 1998, with 30 (83%) of these cases


occurring in the last three years (1996–1998). We identified 43 cases in a three-year period (1/1/1998 to 12/31/2000), sug-gesting an increase in the incidence ofcanine leptospirosis since 1996. We can-not completely rule out diagnostic bias, butthe VMTH has performed active dog lep-tospirosis surveillance since at least 19902.However, recent national diagnostic dataalso indicate that there is an increase inincidence of canine leptospirosis in theUSA [14]. The most frequent serovar iden-tified in dogs from northern California waspomona (39.5%), similar to the 44% pre-viously reported [1]. In the MidwesternUSA, serovar grippotyphosa was the mainserovar identified in cases of leptospiro-sis in dogs (72.1% in a study from Illinoisand 59% in a study from Indiana) [4, 23],whereas it was rather uncommon in north-ern California (< 5% of the cases). Sucha difference in serovar prevalence is mostlikely due to the presence of different reser-voir species. While there has not beenany recent work in California on identi-fying wildlife reservoirs of leptospirosis,four of 30 skunks tested in Connecticuthad serologic evidence of infection withserovar grippotyphosa [17]. While none ofthe skunks in this study had evidence ofinfection with serovar pomona, previousresearch has shown that skunks could alsobe reservoirs of serovar pomona [17, 20].Richardson and Gauthier [17] also foundthat 11 of 31 raccoons had serologic ev-idence of infection with serovar ictero-haemorrhagiae. One factor that may ex-plain the preponderance of serovar pomonain California could be the cattle industry.In one recent study from Texas, 263 out of1193 (22.3%) cattle had at least a titer of1:100 against serovar pomona [19]. Withthe large number of cattle present in Cal-ifornia, it is likely that serovar pomonais common in the environment, and may

2 Cowgill L., personal communication.

be carried by skunks, raccoons, and othersmall mammals.

Our results also included four cases ofleptospirosis caused by serovar canicolaand two cases caused by serovar ictero-haemorrhagiae. These results are impor-tant because leptospirosis due to these twoserovars has rarely been diagnosed in thelast 10 years [1, 16, 25]. In recent years,portions of the general public have de-clined using the leptospirosis vaccine intheir dogs due to a perceived risk of vac-cine reaction and a lesser risk of infec-tion with leptospirosis. The increased in-cidence of infection with serovars canicolaand icterohaemorrhagiae may indicate thatthe lack of leptospirosis vaccination in thegeneral dog population has influenced theincidence of these serovars as well. How-ever, it is difficult to assess the significanceof these cases since there were only sixcases reported and our study was not de-signed to look at changes in incidence overtime. It is important for pet owners andveterinarians to realize that a lack of com-pliance in using the Leptospira portion ofthe standard canine vaccine may lead toincreased incidence of leptospirosis infec-tions with serovars icterohaemorrhagiaeand canicola.

Several recent studies have examinedthe demographic characteristics of Lep-tospira infected dogs, but the findings areambiguous. Rentko et al. [16] did not findany age or breed predilection, but foundmore infected males than females. Harkinand Gartrell [9] found more infected fe-males (n = 10) than infected males (n =7), and almost half of their cases wereeither German shepherds or German shep-herd mixed-breed dogs (n = 8). Wardet al. [22] found that sexually intact maleshad a greater risk than sexually intact fe-males, and also found that herding dogs,working dogs, and hounds had a greaterrisk than companion dogs. Age was alsosignificantly associated with risk of lep-tospirosis, since dogs that were four to nine


years old had significantly greater oddsof contracting leptospirosis than dogs lessthan one year old [22]. Our logistic re-gression analysis found that there were nosignificant breed differences. This findingmay have been due to the small numbersof cases and controls. We found a dif-ferent age group at risk than reported byWard et al. [22]. The clinical significanceof our results could be questionable be-cause of our small sample size, but maybe due to the behavior and naïve immunestatus of younger animals and the weak-ened immune status of older animals. Oneexplanation for this difference is the geo-graphical origin of the cases and regionaldifferences as it relates to sources of infec-tion.

The county of residence (coastal versusinland counties) was marginally signifi-cant. While coastal waters are not directlyimplicated in the disease, this factor cor-relates with coastal regions having higherrainfall and more standing water, which fa-cilitate increased Leptospira survival [7].Previous studies have shown that canineleptospirosis is seasonal and that there isa significant correlation between cases ofleptospirosis and rainfall [1, 21]. In theSan Francisco bay area, annual rainfall waspositively correlated with the incidence ofcanine leptospirosis [1]. However, our re-sults did not show a positive correlationwith rainfall, possibly due to the small tem-poral and spatial scales of our study.

Although many environmental variablesare important in the ecology of leptospiro-sis, our goal was to specifically identifysome of these risk factors and to furtherquantify their importance. The uncondi-tional logistic regression model identifiedfive significant risk factors. It is unclearwhy dogs in one residence versus those oftwo residences have such greater odds ofbeing infected with leptospirosis. It is pos-sible that dogs in only one residence hadmore consistent exposure and therefore agreater chance of infection. We included

this question to look at the possibility thatdogs with two residences may spend alarge amount of time at a lake or mountainhome but could not draw a conclusion fromthe data.

Rural environments are thought to posea higher risk of infection because theseenvironments tend to have larger num-bers of livestock, rodents, and small mam-mals, which are typical reservoirs for Lep-tospira [17, 24]. Rural environments alsotend to have more standing water, suchas irrigation canals and reservoirs. Wa-ter plays a large role in the ecology ofLeptospira and is generally necessary forthe occurrence of any epidemic. We couldnot identify any specific water source as ahigher risk factor, but the primary risk fac-tors for infection with leptospirosis wereswimming or drinking from an outdoorwater source, with respective odds ratios ofeight and 12.44. Indirect exposure to wildanimals, as reported by owners, also poseda significant risk factor for infection in thecase-control study.

Based on our results as well as the workof Ward et al. [24], we believe that habitatswhere there are a large number of interac-tions between canines and wildlife pose asignificant risk for acquiring leptospirosis.Ward et al. [24] also used GIS to iden-tify environmental risk factors and reportedthat being located within 1000 m of an areathat was urbanized between 1990 and 2000was significant for increased risk of devel-oping leptospirosis. While our study didnot consistently find urbanization to be asignificant factor of influence, we did findthat surrogates for wildlife reservoir expo-sure (e.g. open space and wetlands) wereimportant factors.

The relationship between disease sta-tus and the distance to the nearest hy-drographic feature was apparent from ourstudy. Hydrographic density is measuredas unit length of a water feature per unitarea of the LSI; thus it indirectly measurespotential dog-to-water contact within that


LSI. At close proximity, the important fac-tor probably was exposure to leptospiresand not the amount of habitat availablefor supporting leptospires. One of the in-teresting features of this analysis is thesmall amount of variation in case outcomeexplained by the distance to the nearesthydrographic feature, despite the highlysignificant relationship that existed. A pos-sible explanation for the small amount ofvariation determined by geographic vari-ables is the individual behavior of the dogsthemselves. Different breeds of dogs, andeven individuals within the same breed,can show marked differences in their at-traction to water, and therefore their risk ofexposure to leptospirosis.

Our results show that the best logisticregression model for a single LSI is themodel for LSI 1, where hydrographic den-sity is positively correlated with infection.In the logistic regression model for LSI 2urban cover and agriculture are both nega-tively correlated with infection. Our defini-tion of agriculture is an area where > 75%of the land is planted. So both agriculturalland and urban cover would have almost nostanding water, would not provide a goodhabitat for most small mammals, and aretherefore not ideal for the transmission ofleptospirosis. For LSI 3 agriculture is againnegatively correlated with infection.

The second best logistic regressionmodel for a single LSI is the model forLSI 4, where open space and wetlands arepositively correlated with infection, andagriculture is again negatively correlatedwith infection. A likely reason that publicopen space becomes significant at LSI 4(radius = 10 km) is that there is a min-imum number of reservoir hosts neededfor an infectious disease to maintain itselfin a specific environment, termed CriticalCommunity Size [11]. Large tracts of pub-lic open space are more likely to containlarge numbers of raccoons, opossums, andskunks, and lead to higher risk of exposureto leptospirosis. Large tracts of open space

with higher amounts of wetlands are idealhabitats for the presence of leptospires andtheir reservoirs.

When we looked at combined scalemodels we found that the best combinedmodel is one where hydrographic den-sity is positively correlated in LSI 1, per-cent agriculture is negatively correlated inLSI 2, percent wetlands is positively cor-related in LSI 3, and percent open spaceis positively correlated while percent agri-culture is negatively correlated in LSI 4,reinforcing the findings of the individualLSI models.

One aspect of retrospective case-controlstudies that could affect the results of ourstudy is recall bias. The general concernwith these situations is that owners of sickdogs will recall the past better than theindividuals associated with control dogs.There are many proposed reasons for thisbias but the primary reasons are believed tobe an extensive search for the cause of dis-ease, and better recall associated with casesdue to the stress of the situation. While wecannot completely rule out recall bias it isunlikely to affect our results because ourcontrols were all leptospirosis suspects atone point, so the same search for expla-nations could have taken place. Anotherpossible bias of our study was our selec-tion of controls, which were dogs that weresuspected of having leptospirosis but wereseronegative. Generally these are animalsthat are referred to the VMTH with acuterenal failure. There are many causes ofacute renal failure in dogs, but the majorityof cases of acute renal failure in dogs arecaused by toxicities. While we cannot ruleout any bias with our selection of controls,causes of toxicities are generally iatrogenicand randomly distributed in space with re-lation to the natural environment.

One of the limitations of this study isthat we were not able to identify and countpotential wildlife reservoirs of leptospiro-sis, and directly quantify the levels of riskin different habitats. Another limitation of


this study was that it was not possible toisolate the infecting Leptospira serovarsand in this way specifically define whichrisk factors apply to each of the serovars.Future research should focus on isolatingLeptospira organisms from water sourcesand on identifying the specific reservoirspecies involved in canine leptospirosis in-fections, with emphasis placed on skunksin California. The GIS study’s results vali-date the findings of the case-control studyand underline the usefulness of GIS inelucidating environmental risk factors forinfectious diseases. While it was importantto demonstrate the correlation between theresults of the case-control study and theGIS analysis, it is now clear that GIS anal-ysis of possible risk factors can be usedalone. Overall, we have demonstrated thatGIS analysis in conjunction with more tra-ditional epidemiological analysis can be apowerful tool in identifying risk factors forinfectious diseases.

ACKNOWLEDGEMENTS

This study was funded by generous supportfrom The Center for Companion Animal Healththrough a special grant provided by Mary AnnCharles. I would like to thank the staff at TheCenter for Companion Animal Health, RickieKasten, Amy Poland, and Dr Janet Foley.

REFERENCES

[1] Adin C.A., Cowgill L.D., Treatment andoutcome of dogs with leptospirosis: 36cases (1990–1998), J. Am. Vet. Med. Assoc.(2000) 216:3371–3375.

[2] Beardsley K., Stoms D., Compiling a DigitalMap of Areas Managed for Biodiversity inCalifornia, Natural Areas J. (1993) 13:177–190.

[3] Bolin C.A., Diagnosis of leptospirosis: areemerging disease of companion animals,Sem. Vet. Med. Surg. (Small Anim.) (1996)11:166–171.

[4] Boutilier P., Carr A., Schulman R.L.,Leptospirosis in dogs: a serologic survey andcase series 1996 to 2001, Vet. Ther. (2003)4:178–187.

[5] Cowardin L.M., Wetland Classification inUnited-States, J. For. (1978) 76:666–668.

[6] Daly C., Neilson R.P., Phillips D.L., A sta-tistical topographic model for mapping cli-matological precipitation over mountainousterrain, J. Appl. Meteorol. (1994) 33:140–158.

[7] Faine S., Leptospira and leptospirosis, CRCPress, Boca Raton, 1994.

[8] Farr R.W., Leptospirosis, Clin. Infect. Dis.(1995) 21:1–20.

[9] Harkin K.R., Gartrell C.L., Canine lep-tospirosis in New Jersey and Michigan: 17cases (1990-1995), J. Am. Anim. Hosp.Assoc. (2002) 56:203–213.

[10] Homer C., Huang C.Q., Yang L.M., WylieB., Coan. M., Development of a 2001National Land-Cover Database for theUnited States, Photogramm. Eng. RemoteSensing (2004) 70:829–840.

[11] Kermack W.O., McKendrick A.G., A con-tribution to the mathematical theory of epi-demics, Proc. R. Soc. Lond. (1927) 115:700–721.

[12] Maupin G.O., Fish D., Zultowski J., CamposE.G., Piesman J., Landscape ecology ofLyme disease in a residential area ofWestchester County, New York, Am. J.Epidemiol. (1991) 133:1105–1113.

[13] Menard S., Applied Logistic RegressionAnalysis, Sage University Press, ThousandOaks, 1995.

[14] Moore G.E., Guptill L.F., Glickman N.W.,Caldanaro R.J., Aucoin D., Glickman L.T.,Canine leptospirosis, United States, 2002–2004, Emerg. Infect. Dis. (2006) 12:501–503.

[15] Reisen W.K., Lothrop H.D., Presser S.B.,Hardy J.L., Gordon E.W., Landscapeecology of arboviruses in southeasternCalifornia: temporal and spatial patternsof enzootic activity in Imperial Valley,1991–1994, J. Med. Entomol. (1997)34:179–188.

[16] Rentko V.T., Clark N., Ross L.A., SchellingS., Canine leptospirosis: A retrospectivestudy of 17 cases, J. Vet. Intern. Med. (1992)6:235–244.


[17] Richardson D.J., Gauthier J.L., A serosurveyof leptospirosis in Connecticut peridomesticwildlife, Vector Borne Zoonotic Dis. (2003)3:187–193.

[18] Swets J.A., Pickett R.M., Evaluation of di-agnostic systems: methods from signal de-tection theory, Academic Press, New York,1982.

[19] Talpada M.D., Garvey N., Sprowls R.,Eugster A.K., Vinetz J.M., Prevalence ofleptospiral infection in Texas cattle: impli-cations for transmission to humans, VectorBorne Zoonotic Dis. (2003) 3:141–147.

[20] Verts B.J., The biology of the striped skunk,University of Illinois Press, Baltimore,Maryland, 1964.

[21] Ward M.P., Seasonality of canine lep-tospirosis in the United States and

Canada and its association with rainfall,Prev. Vet. Med. (2002) 56:203–213.

[22] Ward M.P., Glickman L.T.G., Guptill L.,Prevalence of and risk factors for leptospiro-sis among dogs in the United States andCanada, J. Am. Vet. Med. Assoc. (2002)220:53–58.

[23] Ward M.P., Guptill L.F., Prahl A., WuC.C., Serovar-specific prevalence and riskfactors for leptospirosis among dogs: 90cases (1997–2002), J. Am. Vet. Med. Assoc.(2004) 224:1958–1963.

[24] Ward M.P., Guptill L.F., Wu C.C., Evaluationof environmental risk factors for leptospiro-sis in dogs: 36 cases (1997–2002), J. Am.Vet. Med. Assoc. (2004) 225:72–77.

[25] Wohl J.S., Canine leptospirosis, Comp.Small Anim. (1996) 18:1215–1224.

To access this journal online:www.edpsciences.org/vetres

use of a case-control study and geographic information systems to

Documents