Hotspots of canine leptospirosis in the United States of America

Accepted Manuscript Title: Hotspots of canine leptospirosis in the United States of America Author: Allison M. White, Carlos Zambrana-Torrelio, Toph Allen, Melinda K. Rostal, Andrea K Wright, Eileen C. Ball, Peter Daszak, William B. Karesh PII: DOI: Reference:

S1090-0233(17)30059-X http://dx.doi.org/doi: 10.1016/j.tvjl.2017.02.009 YTVJL 4964

To appear in:

The Veterinary Journal

Accepted date:

28-2-2017

Please cite this article as: Allison M. White, Carlos Zambrana-Torrelio, Toph Allen, Melinda K. Rostal, Andrea K Wright, Eileen C. Ball, Peter Daszak, William B. Karesh, Hotspots of canine leptospirosis in the United States of America, The Veterinary Journal (2017), http://dx.doi.org/doi: 10.1016/j.tvjl.2017.02.009. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

1 2 3 4 5 6 7 8 9 10 11 12 13 14

Hotspots of canine leptospirosis in the United States of America Allison M. White a, Carlos Zambrana-Torrelio a,*, Toph Allen a, Melinda K. Rostal a, Andrea K Wright b, Eileen C. Ball b, Peter Daszak a, William B. Karesh a a b

EcoHealth Alliance, 460 West 34th Street, 17th Floor, New York, NY 10001, USA Zoetis, 100 Campus Drive, Florham Park, NJ 07932, USA

* Corresponding author: Tel.: +1 212 3804460. E-mail address: [email protected] (C. Zambrana-Torrelio).

1

Page 1 of 25

15 16

Highlights 

17 18

leptospirosis in the USA. 

19 20 21

Machine learning methods were used to predict the spatial occurrence of canine

Positive canine leptospirosis tests were more likely to occur in the Midwest, Eastern and Southwestern USA.



Deciduous forest and precipitation appear to influence the likelihood of occurrence of canine leptospirosis in the USA.

22 23 24

Abstract Leptospirosis is a widespread zoonotic disease that causes hepatic and renal

25

disease in dogs and human beings. The incidence of leptospirosis in dogs in the USA

26

appears to be increasing. This study used 14 years of canine leptospirosis testing data

27

across 3109 counties in the USA to analyze environmental and socio-economic correlates

28

with rates of infection and to produce a map of locations of increased risk for canine

29

leptospirosis. Boosted regression trees were used to identify the probability of a dog

30

testing positive for leptospirosis based on microscopic agglutination test (MAT) results,

31

and environmental and socio-economic data. The Midwest, East and Southwest were

32

more likely to yield positive tests for leptospirosis, although specific counties in

33

Appalachia had some of the highest predicted probabilities. Location (suburban areas or

34

areas with deciduous forest) and climate (precipitation and temperature) were predictors

35

for positive MAT results for leptospirosis, although the precise direction and strength of

36

the effects was difficult to interpret. Wide geographic variation in predicted risk was

2

Page 2 of 25

37

identified. This risk mapping approach may provide opportunities for improved

38

diagnosis, control and prevention of leptospirosis in dogs.

39 40

Keywords: Canine; Leptospirosis; Zoonosis; Risk; Topography; Climate

3

Page 3 of 25

41

Introduction

42

Leptospirosis is a common and widespread zoonotic disease, with reservoirs in

43

domestic and wild animals (Waitkins, 1985; Bharti et al., 2003; Nelson and Couto, 2003;

44

Heymann, 2008; Costa et al., 2015). The disease is caused by spirochaetal bacteria

45

belonging to the genus Leptospira, which infect a range of mammals, including humans,

46

livestock (e.g. cattle pigs and goats) and companion animals (e.g. dogs and horses)

47

(Nelson and Couto, 2003; Heymann, 2008)1. Infection is typically transmitted through

48

direct contact of oral or nasal mucosa, or broken skin, with contaminated urine or water,

49

and dogs are at risk of infection from drinking contaminated water (Nelson and Couto,

50

2003; Heymann, 2008). Leptospirosis can cause severe clinical disease in dogs, including

51

acute hepatic and/or renal failure. It can also produce a chronic carrier status, presenting

52

as idiopathic polyuria/polydipsia, which may not be preceded by severe hepatic or renal

53

disease (Ward, 2002b; Nelson and Couto, 2003).

54 55

Canine leptospirosis has been reported in the USA for more than 100 years

56

(Bolin, 1996) and the prevalence of leptospirosis is reported to be increasing; the rate

57

increased by 1.2 cases/100,000 dogs/year from 1983 to 1998 (Ward et al., 2002). The

58

national proportion of positive microscopic agglutination test (MAT) results increased

59

from 8.7% to 12% from 2002 to 2004 (Glickman et al., 2006; Moore et al., 2006).

60

Clusters of cases of canine leptospirosis have been detected in Texas, California and the

61

upper Midwest, suggesting that, whilst leptospirosis is ubiquitous across the USA, some

62

areas are disproportionately affected (Ward, 2002a; Gautam et al., 2010; Hennebelle et

63

al., 2013). 1

See: CDC, 2014. Leptospirosis. http://www.cdc.gov/leptospirosis/2015 (accessed 10 May 2016).

4

Page 4 of 25

64 65

Environmental and socio-economic factors are thought to explain the distribution

66

and transmission of leptospirosis (Gubler et al., 2001; Sehgal, 2006; Reis et al., 2008).

67

Flooding has been linked to outbreaks of leptospirosis (Reis et al., 2008). The risk of

68

leptospirosis is also related to land cover (e.g. evergreen forests, percentage of wetlands

69

and public open spaces) and proximity to forests (Tangkanakul et al., 2000; Ward et al.,

70

2004; Ahern et al., 2005; Ghneim et al., 2007; Alton et al., 2009; Raghavan et al., 2011;

71

Raghavan et al., 2012a and b, 2013). To date, there has been no systematic analysis of the

72

distribution of leptospirosis in the USA and risk factors associated with the occurrence of

73

the disease.

74 75

In the current study, we used novel statistical approaches to analyze large sets of

76

leptospirosis test data. We tested for the correlation of positive results with a wide range

77

of hypothesized drivers and used the results to identify ‘hotspots’ of leptospirosis (areas

78

of relative higher probability for leptospirosis cases) and the factors that are likely

79

associated with them. Based on the ecology and epidemiology of leptospirosis, we

80

hypothesized that abiotic factors, such as precipitation and temperature, as well as

81

anthropic activities, measured as change in land use cover, can be used to detect areas of

82

higher prevalence (i.e. hotspots) of canine leptospirosis in the USA.

83 84

Materials and methods

85

Literature review

5

Page 5 of 25

86

Literature related to the prevalence of canine leptospirosis and associated factors

87

was reviewed to identify variables for selection and to assist model building. PubMed and

88

ISI Web of Knowledge data bases were searched for the terms ‘(leptospirosis OR

89

Leptospira) AND (dog OR dogs OR canine OR Canis familiaris)’. Articles used were

90

limited to those published from 2000 onwards that were published in English, French or

91

Spanish that focused on North America (Canada, USA and Mexico); most data were from

92

post-2000, although some publications included earlier data, extending as far back as

93

1970.

94 95 96

Data sources and explanatory variables We obtained serological canine leptospirosis MAT results in the USA through an

97

agreement with IDEXX Laboratories. A total of 87,355 tests were available in the

98

proprietary database; however, vaccine history and whether samples were submitted for

99

paired MATs were unknown. All MAT results for tests conducted from 2000 to 2014 at

100

IDEXX Laboratories were included. The sensitivity of MAT has been estimated to be 50-

101

67% at 1:400 dilution and 22-67% at 1:800 dilution, while the specificity ranges from 69-

102

93% at 1:400 dilution and 69-100% for 1:800 dilution in dogs with confirmed

103

leptospirosis, and dogs with clinical and laboratory indicators for leptospirosis, but where

104

ultimately leptospirosis was ruled out (Miller et al., 2008). MATs performed at the same

105

laboratories on samples from specific pathogen free dogs were 100% specific at both

106

1:400 and 1:800 dilutions (Miller et al., 2008). Although the sensitivity and specificity of

107

both titers were similar, a cut-off of ≥1:800 dilution was selected for this study to limit

108

the effect of unknown vaccination status; this was also the cut-off value used by IDEXX

6

Page 6 of 25

109

for diagnostic purposes. All serovars were included in our analysis without distinction.

110

All data was anonymized and geo-referenced at USA zip code level.

111 112

To control the effect of dog population size on the number of samples submitted

113

for testing, we estimated the dog population at county level. We weighted total state level

114

dog population data (American Veterinary Medical Association, 2012) by county level

115

human population data2, assuming the proportion of dog owners is uniform within a

116

state’s counties.

117 118

Variables for the spatial analyses were compiled in four main categories: (1)

119

climate; (2) land cover type; (3) ecology; and (4) socio-economic status. Results from the

120

literature review were used to provide information for selection of variables. Climatic

121

variables included precipitation and temperature across the USA over a 30 year period

122

(1981-2010) (Hijmans et al., 2005)3. Land cover type and percentage cover were

123

collected from the 2011 National Land Cover Database (Jin et al., 2013).

124 125

The mean number of rodent species per county distributed across the USA was

126

used as an ecological indicator of host diversity under the assumption that higher

127

diversity increases the likelihood of transmission of leptospirosis, in the absence of viable

128

data on the number of rodents in a region (which is also likely to fluctuate widely over

129

time).

130 See: US Census Bureau, 2010. 2010 American Community Survey 1-Year Estimates, Washington, DC. https://factfinder.census.gov (accessed 10 May 2016). 3 See: PRISM Climate Group. http://prism.oregonstate.edu (accessed 10 May 2016). 2

7

Page 7 of 25

131

Median household income and percentage of residents with a Bachelor’s degree

132

or higher were used as socio-economic variables4. All variables were aggregated at

133

county level using US Census 2010 county boundaries5. Since the number of requested

134

tests varied highly among counties, we determined the percentage of positive tests by

135

county and used this proportion for all our analyses.

136 137

Modeling

138

Predictive models were constructed using boosted regression trees (BRTs)

139

(Leathwick et al., 2006) with internal cross-validation, previously used to model species

140

distributions (Leathwick et al., 2006; De'ath, 2007; Pittman et al., 2009) and disease

141

ranges (Hay et al., 2013). Boosted regression trees fit an ensemble of regression trees to

142

data in a stepwise fashion, up-weighting poorly predicted data points at each step and

143

monitoring predictive accuracy to prevent over-fitting. Compared to traditional statistical

144

techniques, they predict patterns in complex data sets accurately and are robust for use on

145

data sets with many interacting variables or non-linear relationships; no P values,

146

regression coefficients or confidence intervals are reported (Cutler et al., 2007; Elith et

147

al., 2008).

148 149

The BRT used 31 county level variables for climate and precipitation, dog

150

ownership, landscape composition and rodent diversity. Data analysis was performed

151

using R (Jorge et al., 2015) and the R program dismo (R package, version 1.0-12). The

See: US Census Bureau, 2010. 2010 American Community Survey 1-Year Estimates, Washington, DC. https://factfinder.census.gov (accessed 10 May 2016). 5 See: US Census Bureau, 2014. 2014 TIGER/Line Shapefiles. https://www.census.gov/geo/mapsdata/data/tiger-line.html (accessed 10 May 2016). 4

8

Page 8 of 25

152

predicted probability per county was mapped using ArcGIS version 10.2 (Environmental

153

Systems Research Institute).

154 155

Results

156

Literature review

157

Five hundred articles were identified in the initial search, 474 of which were

158

excluded on the basis of geography and relevance (e.g. randomized control trials, vaccine

159

studies, laboratory diagnostic development, case studies and studies examining acute

160

renal failure). We identified 26 peer reviewed publications with detailed testing data (see

161

Appendix: Supplementary material). The sample size in these articles ranged from 32

162

(Martin et al., 2014) to over 1 million dogs (Ward et al., 2002). We identified a range of

163

climatic, land-cover and socio-economic factors associated with canine leptospirosis

164

(Tangkanakul et al., 2000; Ward et al., 2004; Ahern et al., 2005; Ghneim et al., 2007;

165

Alton et al., 2009; Raghavan et al., 2011; Raghavan et al., 2012a and b, 2013) and

166

included these variables in the model when data was available (Table 1; see Appendix:

167

Supplementary material).

168 169 170

Descriptive statistics A total of 12,317/87,355 (14.1%) MAT tests performed by IDEXX in the USA

171

from 2000 to 2014 were positive for leptospirosis. At least one MAT result was available

172

for 1260/3109 (40.5%) counties in the contiguous USA (Fig. 1) from which 746/1260

173

(59.2%) counties had one or more positive tests. The number of samples submitted for

9

Page 9 of 25

174

MAT from a county ranged from 1 to 3761 (mean 62) and the maximum number of

175

positive tests in one county was 670 (Cook County, Illinois).

176 177 178

Modeling Values from predictive modeling were interpreted as the probability that a

179

diagnostic test in a given county would be positive; for example, a probability of 0.14

180

indicates that there is a 14% chance that any given MAT will be positive. The Midwest,

181

East and Southwest were more likely to yield positive tests for leptospirosis, although

182

specific counties in Appalachia had some of the highest predicted probabilities (Table 1).

183

The highest predicted probability observed was 0.37 in Webster County, West Virginia,

184

whilst the lowest was 0.02 in Harney County, Oregon. The overall median predicted

185

probability was 0.12. Eighty-four (2.7%) counties had predicted probabilities >0.2. The

186

final model fitted 1200 trees; the cross-validation accuracy was 0.58, which is better than

187

random chance, but closer to random chance than perfect accuracy. Hotspot maps of the

188

relative risk of leptospirosis were produced from the BRT model for MAT results (Figs. 2

189

and 3). These show the probability by county that a given test will be positive for

190

leptospirosis. Since the BRT model analyses complex interactions and non-linear

191

relationships, the directionality and precise relationship between any individual variable

192

and the outcome is too idiosyncratic and context dependent to be meaningful on its own;

193

for example, there appears to be a slightly decreased probability of a positive test at

194

higher ranges of precipitation in the coldest quarter of the year, conditional on other

195

variables, and there appears to be a modest increase as the proportion of a county covered

196

by low intensity developed land approaches 0.3. Instead, the overall predictions of a

10

Page 10 of 25

197

model for a county should be considered. Variables important to the model’s overall

198

output included precipitation, temperature, deciduous forested land, and low density

199

developed land (e.g. residential areas with houses built on large lots or areas with 20-49%

200

of surface areas covered with impervious material) (Table 2).

201 202 203

Discussion This analysis has produced the first comprehensive predictive risk map for canine

204

leptospirosis in the contiguous USA and statistical support for some previously

205

hypothesized risk factors. The variation in canine leptospirosis risk in specific counties

206

and regions of the USA appears to be mainly influenced by environmental and land use

207

factors. Our model suggests that landscape and environmental factors, specifically low

208

density developed land (e.g. residential areas with houses built on large lots or areas with

209

20-49% of surface areas covered with impervious material), deciduous forested land,

210

precipitation and temperature, are useful in predicting MAT test results. However, as

211

stated above, the model’s predictions for individual variables are too idiosyncratic and

212

context dependent to be broadly applicable or simply summarized.

213 214

Whilst boosted regression trees allow for the building of complex models with

215

multiple variables, one of the major drawbacks of this approach is that the individual

216

effects of any one specific variable are not easily interpretable. Given the complexity of

217

our model and the number of variables included (e.g. multiple land cover variables), we

218

could identify important variables, but could not describe the precise relationship

219

between one variable and the risk of leptospirosis. Despite the complexity, the analysis

11

Page 11 of 25

220

allowed us to indicate how much influence a particular variable might have on the model

221

prediction (Table 2).

222 223

The mechanistic relationship between different vegetation types and the risk of

224

leptospirosis is unclear and may be a reflection that rural regions provide more

225

opportunity for rodent reservoirs to transmit leptospirosis to dogs. Deciduous forest

226

cover, which contributed most to the predictive power of the model, has not been

227

proposed previously as a risk factor for canine leptospirosis. It is likely that deciduous

228

forest is a proxy for other conditions favorable for leptospirosis transmission, such as

229

specific precipitation regimes, higher rodent density and specific dog behavior.

230 231

One drawback of our model is that it does not account for flooding; therefore, it is

232

possible that variables representing precipitation could explain some of the leptospirosis

233

risk attributed to flooding in the literature (Ahern et al., 2005; Raghavan et al., 2012a).

234

Our finding of the risk for leptospirosis in spacious residential areas (lower density

235

developed land) aligns with the findings of previous studies (Ward et al., 2004; Ghneim

236

et al., 2007; Raghavan et al., 2011), whilst only one study has cited rural areas as a risk

237

for leptospirosis (Alton et al., 2009). Lower density developed land provides a

238

combination of impervious surfaces that can concentrate rain water run-off and acts as a

239

good habitat for reservoir animals, such as rodents and raccoons, hence simulating the

240

effects of ‘flooding’ without requiring a specific level of annual rainfall.

241

12

Page 12 of 25

242

Many of the counties with the highest overall predicted risk are in Appalachia (i.e.

243

West Virginia, Eastern Kentucky and Western Virginia). The Appalachian Mountains

244

receive high annual precipitation6 and contain predominantly deciduous forests. Clusters

245

of leptospirosis cases have been identified in Illinois and Michigan in other studies

246

(Ward, 2002a; Gautam et al., 2010) and coincide with the results of our modeling.

247

Significant spatial clusters were observed throughout the USA, including in Texas,

248

California, the greater Chicago area and some regions of the upper Midwest (Ward,

249

2002a; Gautam et al., 2010; Hennebelle et al., 2013). This increased risk may be

250

explained by the proximity to the Great Lakes and the moderately high annual

251

precipitation in these areas.

252 253

In addition, we identified several counties in Kansas and Nebraska at higher risk

254

for leptospirosis, despite the absence of available testing data for many counties in those

255

states. The presence of leptospirosis in these states is supported by the literature.

256

Raghavan et al. (2011, 2012a and b) examined specific risk factors (hydrologic,

257

environmental and socio-economic factors) for canine leptospirosis in Kansas and

258

Nebraska. In the study of Harkin et al. (2003), 41/500 (8.2%) dogs were shedding

259

leptospires.

260 261

Our analytical approach has some important limitations. Firstly, due to a lack of

262

available testing data, some areas of the contiguous USA (i.e. the upper Midwest and

263

parts of the Southeast) were poorly represented in our testing data. Whilst the use of

264

predictive methods allows us to make predictions even when data are lacking, the 6

See: PRISM Climate Group. http://prism.oregonstate.edu (accessed 10 May 2016).

13

Page 13 of 25

265

inclusion of data from these areas would increase predictive accuracy. Additional

266

analyses using testing data from these areas would provide a means to refine and

267

externally validate our results.

268 269

Secondly, we used a titer threshold ≥1:800 to select positive tests for our model.

270

This cut-off threshold is expected to identify more exposures, whilst minimizing the risk

271

of including positive titers due to vaccination. When vaccinated dogs were monitored for

272

1 year post-vaccination, some dogs developed titers ≥1:800 by weeks 7-15, whereas by

273

weeks 29-52 none of the dogs had titers ≥1:400 and only a small percentage had titers

274

≥1:100 (Martin et al., 2014). Our literature review indicated that MAT titer thresholds for

275

considering a test to be positive ranged from ≥ 1:100 to ≥ 1:3,200. The 2010 American

276

College of Veterinary Internal Medicine (ACVIM) Small Animal Consensus Statement

277

on Leptospirosis (Sykes et al., 2011) states that there is a lack of consensus as to which

278

titer level should be used to determine that a test is negative for leptospirosis and also that

279

single positive titers must be considered with clinical signs and a paired titer, since even a

280

titer ≥1:800 does not confirm a diagnosis of leptospirosis. We followed the

281

recommendation of the diagnostic laboratory (positive titer ≥1:800), which fits with our

282

desire to exclude animals vaccinated within the past year and to have enough confidence

283

in the likelihood of exposure to proceed with the epidemiological analysis.

284 285

Given the relatively low vaccination rates for canine leptospirosis in the USA

286

(<4% of dogs in 2011) (American Veterinary Medical Association, 2012), we do not

287

expect this to have a major impact on the results (Klaasen et al., 2003). Although vaccine

14

Page 14 of 25

288

history and whether samples were submitted for paired MATs were unknown, given the

289

overall size of the data set, this was considered unlikely to have an impact on results.

290 291

Data were aggregated at county level due to data availability and to estimate

292

leptospirosis risk in an easily interpretable manner to aid veterinarians in describing risks

293

to owners. County lines are often arbitrary and may have changed during the course of

294

our study. Using an arbitrary unit of area (county) may have augmented some of the

295

observed effects. This would be likely to have the greatest impact in the Western USA,

296

where the counties are large.

297 298

Our analysis provided some counterintuitive results. Several counties with high-

299

predicted values were found in areas where clusters of leptospirosis had not previously

300

been identified. In some of these counties, the available data did not have a high

301

proportion of positive tests, indicating differences between the predicted values and

302

existing leptospirosis clusters (Figs. 2 and 3) (Gautam et al., 2010; Hennebelle et al.,

303

2013). Counties with low risk sometimes adjoin counties at high risk; for example, in

304

Virginia, small counties with high risk are sometimes entirely surrounded by larger

305

counties with lower risk. This may be due to smaller counties being population centers or

306

due to artefacts in the data. These specific counties should be examined further to

307

determine whether the low level of reported leptospirosis is due to lack of data, lack of

308

testing or better vaccination coverage.

309

15

Page 15 of 25

310

Areas identified by our model as higher risk areas differed from the overall

311

distribution of proportion of available positive tests throughout the USA (Figs. 1 and 2),

312

indicating that this method provides utility compared to analyzing historical testing data

313

alone. Identifying the county level risk for leptospirosis can contribute to determining

314

where to implement prevention and control measures, testing and vaccination.

315 316

Leptospirosis is a disease of increasing concern for both people and dogs.

317

Identified canine leptospirosis incidences in the USA have ranged from 0.04% in a study

318

of hospital prevalence from 1970-1998 across the USA, to as high as 29% in a study

319

examining tests submitted to the veterinary diagnostic lab in Illinois from 1996-2001.

320

This variation makes it important to identify risk areas for this disease (Ward et al., 2002;

321

Boutilier et al., 2003). The US Centers for Disease Control and Prevention recently

322

reinstated leptospirosis in human beings as a nationally notifiable disease; in 2014, there

323

were 21 human cases of leptospirosis (Centers for Disease Control, 2015). Dogs play an

324

important role as potential indicators of areas with high endemicity for leptospirosis and,

325

although infrequent, zoonotic transmission of leptospirosis from dogs to human beings

326

can occur. Thus, recognizing and preventing canine leptospirosis has implications for

327

human health as well as dogs.

328 329 330

Conclusions Our model can be used to characterize the risk of canine leptospirosis across the

331

USA and can be applied to improve delivery of veterinary services, including diagnostics

332

and vaccination, and to identify areas for increased research and surveillance efforts.

16

Page 16 of 25

333

Canine leptospirosis remains an important disease, widely distributed in the USA, with

334

varying levels of disease risk. Recognizing and identifying the areas most at risk will help

335

to provide improved veterinary care and control of this disease.

336 337 338

Conflict of interest statement Funds for this project were provided by Zoetis, which currently markets a vaccine

339

against canine leptospirosis. Andrea Wright and Eileen Ball are employees of Zoetis.

340

None of the other authors of this paper have a financial or personal relationship with

341

other people or organizations that could inappropriately influence or bias the content of

342

the paper.

343 344 345 346

Acknowledgements Funding for this project was provided by Zoetis. The authors would like to thank IDEXX Laboratories for providing the data for this project.

347 348 349 350

Appendix. Supplementary material Supplementary data associated with this article can be found, in the online version, at doi: …

351 352

References

353 354 355 356 357 358

Ahern, M., Kovats, R.S., Wilkinson, P., Few, R., Matthies, F., 2005. Global health impacts of floods: Epidemiologic evidence. Epidemiologic Reviews 27, 36-46. Alton, G.D., Berke, O., Reid-Smith, R., Ojkic, D., Prescott, J.F., 2009. Increase in seroprevalence of canine leptospirosis and its risk factors, Ontario 1998-2006. Canadian Journal of Veterinary Research 73, 167-175.

17

Page 17 of 25

359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404

American Veterinary Medical Association (AVMA), 2012. US Pet Ownership and Demographics Sourcebook 2012. AVMA, Schaumberg, IL, USA, pp. 21-23. Bharti, A.R., Nally, J.E., Ricaldi, J.N., Matthias, M.A., Diaz, M.M., Lovett, M.A., Levett, P.N., Gilman, R.H., Willig, M.R., Gotuzzo, E., et al., 2003. Leptospirosis: A zoonotic disease of global importance. The Lancet Infectious Diseases 3, 757771. Bolin, C.A., 1996. Diagnosis of leptospirosis: A reemerging disease of companion animals. Seminars in Veterinary Medicine and Surgery (Small Animal) 11, 166171. Boutilier, P., Carr, A., Schulman, R.L., 2003. Leptospirosis in dogs: A serologic survey and case series 1996 to 2001. Veterinary Therapeutics: Research in Applied Veterinary Medicine 4, 387-396. Centers for Disease Control, 2015. Notifiable diseases and mortality tables. Morbidity and Mortality Weekly Report (MMWR) 63, ND-719-ND-732. Costa, F., Hagan, J.E., Calcagno, J., Kane, M., Torgerson, P., Martinez-Silveira, M.S., Stein, C., Abela-Ridder, B., Ko, A.I., 2015. Global morbidity and mortality of leptospirosis: A systematic review. PLoS Neglected Tropical Diseases 9, e0003898. Cutler, D.R., Edwards, T.C., Beard, K.H., Cutler, A., Hess, K.T., Gibson, J., Lawler, J.J., 2007. Random forests for classification in ecology. Ecology 88, 2783-2792. De'ath, G., 2007. Boosted trees for ecological modeling and prediction. Ecology 88, 243251. Elith, J., Leathwick, J.R., Hastie, T., 2008. A working guide to boosted regression trees. Journal of Animal Ecology 77, 802-813. Gautam, R., Guptill, L.F., Wu, C.C., Potter, A., Moore, G.E., 2010. Spatial and spatiotemporal clustering of overall and serovar-specific Leptospira microscopic agglutination test (MAT) seropositivity among dogs in the United States from 2000 through 2007. Preventive Veterinary Medicine 96, 122-131. Ghneim, G.S., Viers, J.H., Chomel, B.B., Kass, P.H., Descollonges, D.A., Johnson, M.L., 2007. Use of a case-control study and geographic information systems to determine environmental and demographic risk factors for canine leptospirosis. Veterinary Research 38, 37-50. Glickman, L.T., Moore, G.E., Glickman, N.W., Caldanaro, R.J., Aucoin, D., Lewis, H.B., 2006. Purdue University-Banfield National Companion Animal Surveillance

18

Page 18 of 25

405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450

Program for emerging and zoonotic diseases. Vector Borne and Zoonotic Diseases 6, 14-23. Gubler, D.J., Reiter, P., Ebi, K.L., Yap, W., Nasci, R., Patz, J.A., 2001. Climate variability and change in the United States: Potential impacts on vector- and rodent-borne diseases. Environmental Health Perspectives 109 (Suppl. 2), 223233. Harkin, K.R., Roshto, Y.M., Sullivan, J.T., Purvis, T.J., Chengappa, M.M., 2003. Comparison of polymerase chain reaction assay, bacteriologic culture, and serologic testing in assessment of prevalence of urinary shedding of leptospires in dogs. Journal of the American Veterinary Medical Association 222, 12301233. Hay, S.I., George, D.B., Moyes, C.L., Brownstein, J.S., 2013. Big data opportunities for global infectious disease surveillance. PLoS Medicine 10, e1001413. Hennebelle, J.H., Sykes, J.E., Carpenter, T.E., Foley, J., 2013. Spatial and temporal patterns of Leptospira infection in dogs from northern California: 67 cases (2001-2010). Journal of the American Veterinary Medical Association 242, 941947. Heymann, D., 2008. Control of Communicable Diseases Manual, 19th Edn. American Public Health Association, Washington, DC, USA. Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G., Jarvis, A., 2005. Very high resolution interpolated climate surfaces for global land areas. International Journal of Climatology 25, 1965-1978. Jin, S., Yang, L., Danielson, P., Homer, C., Fry, J., Xian, G., 2013. A comprehensive change detection method for updating the National Land Cover Database to circa 2011. Remote Sensing of Environment 132, 159-175. Jorge, S., Monte, L.G., De Oliveira, N.R., Collares, T.F., Roloff, B.C., Gomes, C.K., Hartwig, D.D., Dellagostin, O.A., Hartleben, C.P., 2015. Phenotypic and molecular characterization of Leptospira interrogans isolated from Canis familiaris in Southern Brazil. Current Microbiology 71, 496-500. Klaasen, H.L., Molkenboer, M.J., Vrijenhoek, M.P., Kaashoek, M.J., 2003. Duration of immunity in dogs vaccinated against leptospirosis with a bivalent inactivated vaccine. Veterinary Microbiology 95, 121-132. Leathwick, J.R., Elith, J., Francis, M.P., Hastie, T., Taylor, P., 2006. Variation in demersal fish species richness in the oceans surrounding New Zealand: An analysis using boosted regression trees. Marine Ecology Progress Series 321, 267-281.

19

Page 19 of 25

451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495

Martin, L.E., Wiggans, K.T., Wennogle, S.A., Curtis, K., Chandrashekar, R., Lappin, M.R., 2014. Vaccine-associated Leptospira antibodies in client-owned dogs. Journal of Veterinary Internal Medicine 28, 789-792. Miller, M., Annis, K., Lappin, M., Gill, M., Lunn, K., 2008. Sensitivity and specificity of the microscopic agglutination test for the diagnosis of leptospirosis in dogs. Journal of Veterinary Internal Medicine 22, 787-788 (Abstract 287). Moore, G.E., Guptill, L.F., Glickman, N.W., Caldanaro, R.J., Aucoin, D., Glickman, L.T., 2006. Canine leptospirosis, United States, 2002-2004. Emerging Infectious Diseases 12, 501-503. Nelson, R.W., Couto, C.G., 2003. Small Animal Medicine 3rd Edn. Mosby, St Louis, Missouri, USA. Pittman, S.J., Costa, B.M., Battista, T.A., 2009. Using LiDAR bathymetry and boosted regression trees to predict the diversity and abundance of fish and corals. Journal of Coastal Research, 27-38. Raghavan, R., Brenner, K., Higgins, J., Van der Merwe, D., Harkin, K.R., 2011. Evaluations of land cover risk factors for canine leptospirosis: 94 cases (20022009). Preventive Veterinary Medicine 101, 241-249. Raghavan, R.K., Brenner, K.M., Harrington, J.A., Jr., Higgins, J.J., Harkin, K.R., 2013. Spatial scale effects in environmental risk-factor modelling for diseases. Geospatial Health 7, 169-182. Raghavan, R.K., Brenner, K.M., Higgins, J.J., Hutchinson, J.M., Harkin, K.R., 2012a. Evaluations of hydrologic risk factors for canine leptospirosis: 94 cases (20022009). Preventive Veterinary Medicine 107, 105-109. Raghavan, R.K., Brenner, K.M., Higgins, J.J., Shawn Hutchinson, J.M., Harkin, K.R., 2012b. Neighborhood-level socioeconomic and urban land use risk factors of canine leptospirosis: 94 cases (2002-2009). Preventive Veterinary Medicine 106, 324-331. Reis, R.B., Ribeiro, G.S., Felzemburgh, R.D.M., Santana, F.S., Mohr, S., Melendez, A.X.T.O., Queiroz, A., Santos, A.C., Ravines, R.R., Tassinari, W.S., et al., 2008. Impact of environment and social gradient on Leptospira infection in urban slums. PLoS Neglected Tropical Diseases 2, e228. Sehgal, S.C., 2006. Epidemiological patterns of leptospirosis. Indian Journal of Medical Microbiology 24, 310-311.

20

Page 20 of 25

496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521

Sykes, J.E., Hartmann, K., Lunn, K.F., Moore, G.E., Stoddard, R.A., Goldstein, R.E., 2011. 2010 ACVIM Small Animal Consensus Statement on Leptospirosis: Diagnosis, epidemiology, treatment, and prevention. Journal of Veterinary Internal Medicine 25, 1-13. Tangkanakul, W., Tharmaphornpil, P., Plikaytis, B.D., Bragg, S., Poonsuksombat, D., Choomkasien, P., Kingnate, D., Ashford, D.A., 2000. Risk factors associated with leptospirosis in northeastern Thailand, 1998. American Journal of Tropical Medicine and Hygiene 63, 204-208. Waitkins, S.A., 1985. From the PHLS. Update on leptospirosis. British Medical Journal (Clinical Research Edition) 290, 1502-1503. Ward, M.P., 2002a. Clustering of reported cases of leptospirosis among dogs in the United States and Canada. Preventive Veterinary Medicine 56, 215-226. Ward, M.P., 2002b. Seasonality of canine leptospirosis in the United States and Canada and its association with rainfall. Preventive Veterinary Medicine 56, 203-213. Ward, M.P., Glickman, L.T., Guptill, L.E., 2002. Prevalence of and risk factors for leptospirosis among dogs in the United States and Canada: 677 cases (19701998). Journal of the American Veterinary Medical Association 220, 53-58. Ward, M.P., Guptill, L.F., Wu, C.C., 2004. Evaluation of environmental risk factors for leptospirosis in dogs: 36 cases (1997-2002). Journal of the American Veterinary Medical Association 225, 72-77.

21

Page 21 of 25

522 523

Fig. 1. Percent of microscopic agglutination tests (MAT) results positive by county in the

524

USA in 2000-2014. Darker colors indicate a greater proportion of positive tests. Counties

525

with no data indicate that no MATs were submitted to the reference laboratory during the

526

study period.

527 528

Fig. 2. Predicted probability of a positive microscopic agglutination test (MAT) result for

529

canine leptospirosis in the continental USA. Predicted probabilities range from 0.023 to

530

0.371, indicating that approximately 1/3 dogs tested is expected to be positive for

531

leptospirosis. Scale is green to red where green indicates lower probability and red

532

indicates higher probability.

533 534

Fig. 3. Insets from Figs. 1 and 2, showing the counties for Maine (above) and Idaho

535

(below). The left side shows the percent of positive microscopic agglutination test (MAT)

536

results and the right side shows the predicted probability of a positive MAT result.

537

22

Page 22 of 25

538

Table 1 Counties with the highest predicted probabilities for canine leptospirosis in the USA. Rank

County

State

Predicted probability

Dogs expected to test positive

1

Webster County

West Virginia

0.370

1/2.7

2

Marion County

Indiana

0.34

1/2.9

3

Harrisonburg City

Virginia

0.33

1/3.0

4

Staunton City

Virginia

0.33

1/3.0

5

Waynesboro City

Virginia

0.31

1/3.2

6

Adair County

Kentucky

0.31

1/3.2

7

Covington City

Virginia

0.30

1/3.3

8

Curry County

Oregon

0.30

1/3.3

9

Nicholas County

West Virginia

0.30

1/3.3

10

Bedford City

Virginia

0.29

1/3.4

539

23

Page 23 of 25

540

541

Table 2 Relative influence of the five most important variables on the boosted regression tree model. Percentage relative influence a

Variable

Group

Deciduous forest

Land cover

10.9

Precipitation (coldest quarter mean)

Bioclimate

9.0

Scrubland and shrub land

Land cover

6.30

Developed (low intensity)

Land cover

5.7

Temperature (mean)

Bioclimate

5.0

a

Greater relative influence indicates greater contribution to model results.

542

24

Page 24 of 25

543

25

Page 25 of 25