- Research
- Open Access
Factors influencing U.S. canine heartworm (Dirofilaria immitis) prevalence
- Dongmei Wang^{1},
- Dwight D Bowman^{2},
- Heidi E Brown^{3},
- Laura C Harrington^{4},
- Phillip E Kaufman^{5},
- Tanja McKay^{6},
- Charles Thomas Nelson^{7},
- Julia L Sharp^{1}Email author and
- Robert Lund^{1}
https://doi.org/10.1186/1756-3305-7-264
© Wang et al.; licensee BioMed Central Ltd. 2014
- Received: 7 November 2013
- Accepted: 12 May 2014
- Published: 6 June 2014
Abstract
Background
This paper examines the individual factors that influence prevalence rates of canine heartworm in the contiguous United States. A data set provided by the Companion Animal Parasite Council, which contains county-by-county results of over nine million heartworm tests conducted during 2011 and 2012, is analyzed for predictive structure. The goal is to identify the factors that are important in predicting high canine heartworm prevalence rates.
Methods
The factors considered in this study are those envisioned to impact whether a dog is likely to have heartworm. The factors include climate conditions (annual temperature, precipitation, and relative humidity), socio-economic conditions (population density, household income), local topography (surface water and forestation coverage, elevation), and vector presence (several mosquito species). A baseline heartworm prevalence map is constructed using estimated proportions of positive tests in each county of the United States. A smoothing algorithm is employed to remove localized small-scale variation and highlight large-scale structures of the prevalence rates. Logistic regression is used to identify significant factors for predicting heartworm prevalence.
Results
All of the examined factors have power in predicting heartworm prevalence, including median household income, annual temperature, county elevation, and presence of the mosquitoes Aedes trivittatus, Aedes sierrensis and Culex quinquefasciatus. Interactions among factors also exist.
Conclusions
The factors identified are significant in predicting heartworm prevalence. The factor list is likely incomplete due to data deficiencies. For example, coyotes and feral dogs are known reservoirs of heartworm infection. Unfortunately, no complete data of their populations were available. The regression model considered is currently being explored to forecast future values of heartworm prevalence.
Keywords
- Canine heartworm
- Dirofilaria immitis
- Head-banging smoothing
- Mosquito vectors
- Prevalence rates
Background
The Companion Animal Parasite Council (CAPC) has compiled a data set of over nine million heartworm antigen tests performed on dogs in the United States during 2011 and 2012[1]. These data are test results taken from dogs that visited veterinary clinics throughout the United States. From this data, our goal is to quantify the environmental, socio-economic, and vector factors that influence canine heartworm prevalence rates. Brown et al.[2] describe the data and list factors posited to influence canine heartworm prevalence rates (these are discussed fully below). This paper quantitatively assesses which of the factors are important (or unimportant) and in what direction the factors impact (increase or decrease) heartworm prevalence rates.
The data will be used to estimate the probability that a dog entering a veterinary clinic will test positive for heartworm. Although prevalence rates increase if a dog is from an area with a high heartworm transmission rate, the raw data do not describe transmission, but rather the risk that a dog’s infection is detected if it enters a clinic within the United States and is tested for heartworm for any reason whatsoever. The most common reasons for testing a dog are: assessing the negative status of a dog before it begins heartworm prevention, annual testing of dogs on preventive prophylaxis to verify that the dog has been protected, and assessing whether or not a dog with clinical signs suggestive of heartworm disease is indeed infected. In many areas of the United States, dogs are kept on prophylaxis year round; however, in some areas, veterinarians utilize heartworm prevention seasonally. Here, annual testing verifies whether or not infection occurred during the period when preventives were not taken. To assess true transmission rates, it would be more appropriate to follow dogs or other canines, specifically coyotes, that are not receiving any form of prophylaxis[3, 4]— these are not the canines studied here. According to the unpublished abstract of Pulaski et al. for the 58th Annual Meeting of the American Association of Veterinary Parasitologists and the unpublished presentation of Blagburn et al. at The Triennial Symposium of the American Heartworm Society in 2013, in parts of the United States, resistance to heartworm preventatives has been recognized. Thus, future data may help detect preventive failure.
A spatial logistic regression model will be fitted to the CAPC county-by-county heartworm test results and related to factor measurements. Logistic regression methods are used in lieu of ordinary regression techniques because prevalence probabilities, which must lie in the interval [0,1], are being modeled. A significant technical challenge involves the large number of counties reporting a small number of tests (often this count is zero). Small sample sizes from isolated counties can adversely impact results if not properly handled. Therefore, methods are developed that account for sample size issues. The head-banging algorithm, a method for smoothing the county-by-county prevalence rates, will be used to extract general spatial structure in the prevalence estimates; this procedure is adept at dealing with outlying observations and boundary (edge) features.
Our results are useful in a variety of contexts. First and foremost, predicting heartworm prevalence rates alerts the pet owner to high-risk areas. This will be evident from the baseline risk maps constructed in Section "Construction of the baseline heartworm prevalence map". Second, pinpointing the factors accompanying high heartworm prevalence rates provides an opportunity to target those factors in mosquito and heartworm control programs. Third, our results provide a quantitative analysis of canine heartworm across the entire country, allowing us to confirm that cases do occur in the Western United States. Finally, the fitted regression model can be used to forecast future prevalence levels of heartworm or its response to climate change.
Methods
The data and factors
This section describes the data provided by CAPC and lists the factors considered by our analysis.
Other data aspects are worth illuminating. First, heartworm infections are not detectable by almost all testing methods until about 6 or 7 months after the dogs have become infected; this is how long it takes for either microfilariae or antigen to appear in the blood after infection. Thus, many of the detected infections likely commenced the year prior to a positive test result. Because no travel information exists, it cannot be known where a dog acquires infection. However, it is suspected that the majority of infected dogs were infected close to home. Second, it is not known if a dog has been tested more than once. Dogs may be tested more than once annually to verify the need for treatment, to verify successful treatment, or annually tested after the infection is first identified.
Heartworm factors considered for inclusion in the study
Factors | Data available period | Scale | Source | |
---|---|---|---|---|
Climate factors | Annual temperature | 2011 and 2012 | Division | National Climate Data Center (NCDC) |
Annual precipitation | 2011 and 2012 | Division | NCDC | |
Annual relative humidity | 2011 and 2012 | Station | NCDC | |
Geographic factors | Elevation | 2012 | County | |
Percentage forest coverage | 2007 | County | United States Department of Agriculture (USDA) | |
Percentage surface water coverage | 2010 | County | U.S. Census Bureau | |
Societal factors | Population density | 2010 | County | U.S. Census Bureau |
Median household income | 2011 | County | U.S. Census Bureau | |
Mosquito species | Aedes aegypti | 2008 | County | Moore, CG.[15] |
Aedes albopictus | 2012 | County | Hynes NA[16] | |
Aedes canadensis | 2004 | County | RF Darsie, Jr. and RA Ward[17] | |
Aedes sierrensis | 2004 | County | RF Darsie, Jr. and RA Ward | |
Aedes trivittatus | 2004 | County | RF Darsie, Jr. and RA Ward | |
Anopheles punctipennis | 2004 | County | RF Darsie, Jr. and RA Ward | |
Anopheles quadrimaculatus | 2004 | County | RF Darsie, Jr. and RA Ward | |
Culex quinquefasciatus | 2004 | County | RF Darsie, Jr. and RA Ward |
Construction of the baseline heartworm prevalence map
We first construct a baseline heartworm prevalence map. This is done on an annual basis as there is too little data to consider seasonal effects. As will become apparent, this analysis is a necessary precursor to assess factor importance — informative factors should be able to reproduce the structure of our baseline prevalence map. For the years 2011 and 2012, all data were combined into a single sample.
For a county s, let p(s) denote the probability that a single dog tests heartworm positive. For notation, n(s) is the number of tests in county s and k(s) is the number of positive tests at county s. For example, if county s has 3 positive tests out of 100 during 2011 and eight positive tests out of 200 during 2012, then k(s) = 11, n(s) = 300, and$\widehat{p}(s)=11/300$ (a hat over a quantity indicates it is an estimate). Figure1 displays county-by-county values of$\widehat{p}(s)$. This figure indicates that heartworm is most problematic in the Lower Mississippi Valley. The role of factors in explaining the prevalence rates will be discussed in Section "Factor quantification". No factors are involved in the calculation of$\widehat{p}(s)$.
The estimated standard error of$\widehat{p}(s)$ is the square root of (1). We weight the values of$\widehat{p}(s)$ inversely proportional to this standard error. Before doing this, an adjustment was made to the values of$\widehat{p}(s)$ for small sample sizes.
This estimator has desirable sampling properties and cannot be zero or unity[18]. Of course, for large k(s) and n(s),$\widehat{p}(s)$ and${\widehat{p}}_{W}(s)$ are approximately equivalent.
The raw values of${\widehat{p}}_{W}(s)$ greatly vary across counties. Even counties in close proximity to one another often have highly different prevalence estimates. This said, some spatial structure clearly exists in Figure1. Our next goal is to extract and explore this structure. To accomplish this, the weighted head-banging spatial smoothing algorithm was applied to the county-by-county values of${\widehat{p}}_{W}(s)$. This procedure serves to remove localized small-scale variations due to random chance, illuminating large-scale structures that are actual features of the prevalence rates.
While we will not delve into the details of weighted head-banging procedures, the technique is a median-based algorithm proposed by Tukey and Tukey[19] for smoothing spatial data. One needs to input the longitude and latitude of the centroid of each county, the county value to be smoothed, and the corresponding weights. Mungiole et al.[20] discuss the algorithm in detail. Head-banging takes its name from a child’s game, where the child presses their face against pins protruding from a board that are of various lengths. The result leaves an impression of the child’s face, smoothing the lengths of adjacent nails but leaving the general structure of the face’s impression. Head-banging techniques are very effective for down-weighting or removing noisy ‘spikes’ while preserving edge structures. A spike is an isolated observation that lacks confirmation from nearby data. Because of different testing practices from county to county, many spikes exist in the heartworm prevalence estimates. An edge occurs where data changes significantly in pattern — perhaps due to a mountain range. Edges are informative as they often demarcate distinct data regions.
Figure2 shows our smoothed prevalence rates based on the weighted head-banging procedure. The larger the triple parameter is, the smoother (less rough) the resulting map will be. We have intentionally left the graphic slightly under-smoothed. This is because it is easy to visually smooth variabilities away with the eye, but impossible to recover true fluctuations that are erroneously smoothed away. Thirty triples were used to produce this graphic.
Figure2 has interesting implications. First and foremost, heartworm is most prevalent in the Lower Mississippi Valley. While the northern latitudes show less activity, places where the prevalence rates were relatively higher do exist. Michigan, Vermont, and Northwest Washington, for example, show greater heartworm disease prevalence than some of the other states at the same latitude. The Northern Rockies perhaps show the least heartworm disease prevalence. While many inferences can be made from Figure2, we caution the reader not to over-interpret minutia. The map is constructed from only two years of data and there are variations in the results that may be spurious. For example, two very close locations — say Baton Rouge and New Orleans, Louisiana — might be shaded different colors on the map, but should not be expected to have radically different prevalence rates. As additional years of data are collected, we expect our baseline to become more accurate. Another issue involves dispersal of the disease: there is no a priori reason to think that prevalence rates are static in time. With only two years of observations, time trends will be difficult to discern and are not explored herein.
Factor quantification
This section examines the significance of the individual factors presented in Table1 in predicting heartworm prevalence. A logistic regression model was created using data from 2011 and 2012. Logistic regression methods (as opposed to ordinary regression methods)[22] are specifically designed for cases involving a binary outcome that can be summarized by a probability, and hence limited to take values in the interval [0,1]. Our goal is to reproduce the structure in Figure2.
Clarifying terms, ln denotes natural logarithm and logit(x) = ln(x)- ln(1 - x). Notice that e^{g(X(s))}/(1 + e^{g(X(s))}) ∈ [0,1] for any value of g(X(s)). This guarantees that all predicted prevalence rates lie between zero and unity. The overall location parameter, β_{0}, is common to all counties while β_{1},…,β_{8}, are regression coefficients for the eight non-mosquito factors, and γ_{1},…,γ_{8}, are regression coefficients for the eight mosquito species. The notation 1_{ i }(s),1 ≤ i ≤ 8, are zero-one indicators: 1_{ i }(s) is taken as unity if the i th mosquito type is present in county s and zero otherwise.
Results and discussion
Clarifying details, in the head-banging procedure, the inverse of the standard deviation of${\widehat{p}}_{\text{Logistic}}(s)$ is used as the weight. For any factor, if both 2011 and 2012 data are available (such as temperature and precipitation) then the average of the observations from the two years is used in equation 5. Otherwise, the most recent measurement is used.
Significance of factors**
Effect | Estimate | Standard error | Exp(β) | 95% Wald CI |
---|---|---|---|---|
Intercept | -8.0610 | 0.0819 | ||
Temperature | 0.0720 | 0.0011 | 1.075 | (1.072, 1.077) |
Median household income | -0.0227 | 0.0003 | 0.978 | (0.977, 0.978) |
Population density | -0.0175 | 0.0004 | 0.983 | (0.982, 0.984) |
Precipitation | 0.0982 | 0.0049 | 1.103 | (1.093, 1.114) |
Elevation | -0.0605 | 0.0030 | 0.941 | (0.936, 0.947) |
Relative humidity | -0.0147 | 0.0008 | 0.985 | (0.984, 0.987) |
Forest coverage | 1.2493 | 0.1156 | 3.448 | (2.781, 4.375) |
Surface water coverage | 0.1054 | 0.0265 | 1.111 | (1.055, 1.170) |
Aedes trivittatus | 0.9987 | 0.0153 | 2.715 | (2.634, 2.797) |
Culex quinquefasciatus | 0.4063 | 0.0152 | 1.501 | (1.457, 1.547) |
Aedes sierrensis | 0.6947 | 0.0293 | 2.003 | (1.892, 2.121) |
Anopheles punctipennis | 0.4132 | 0.0182 | 1.512 | (1.459, 1.567) |
Anopheles quadrimaculatus | -0.1663 | 0.0185 | 0.847 | (0.817, 0.878) |
Aedes aegypti | -0.1232 | 0.0112 | 0.884 | (0.865, 0.904) |
Aedes canadensis | 0.1481 | 0.0143 | 1.160 | (1.128, 1.192) |
Aedes albopictus | -0.0894 | 0.0112 | 0.915 | (0.895, 0.935) |
The model results in a fairly good fit: McFadden’s pseudo R^{2} = 0.62[24]. All predictors were significant predictors of heartworm prevalence rates at level 0.0001. Higher temperatures are associated with higher prevalence – heartworm is generally more prevalent in warmer temperatures. Higher median household incomes are associated with lower prevalence rates. Logically, higher income pet owners can more easily afford heartworm preventives. Not surprisingly, heartworm prevalence decreases with increasing population density and elevation. It is surprising to see that prevalence declines with higher humidities. One explanation for this may lie with the high correlations among elevation, relative humidity, and temperature. This will be explored further below when a model that allows the factors to interact is examined.
Presence of Culex quinquefasciatus, Aedes sierrensis, Anopheles punctipennis, Anopheles quadrimaculatus, and Aedes canadensis are associated with higher heartworm prevalence. The other three mosquito species are estimated to decrease prevalence rates. The reader should not a priori believe that presence of any mosquito species acts to increase prevalences in the model: the 16 factors are being "judged" in tandem.
It is perhaps remarkable that every factor considered is judged to influence prevalence rates with a significance level of 0.0001. It leaves an unsettling feeling that other important factors may have been omitted. We reiterate that this study is exploratory and will hopefully be improved in the future. We should also mention the usual regression caveat: (3) presupposes a linear relationship on the factors that may not be realistic at all factor levels. This drawback is not germane to this study; linear regression models, of course, serve as rudimentary guidance.
Significance factors and interactions*
Effect | Estimate | Standard error |
---|---|---|
Intercept | -8.4312 | 0.0993 |
Median household income | -0.0267 | 0.0004 |
Temperature | 0.0655 | 0.0015 |
Elevation | 0.5609 | 0.0171 |
Population density | -0.0444 | 0.0015 |
Temperature*Elevation | -0.0087 | 0.0003 |
Temperature*Forest coverage | 0.4840 | 0.0195 |
Temperature*Surface water coverage | -0.0345 | 0.0034 |
Elevation*Relative humidity | -0.0041 | 0.0002 |
Forest coverage | -36.5371 | 2.1127 |
Population density*Median household | 0.0005 | <0.0001 |
income | ||
Forest coverage*Surface water coverage | -18.1276 | 1.5896 |
Precipitation | 0.3049 | 0.0278 |
Precipitation*Surface water coverage | -0.3247 | 0.0299 |
Precipitation*Elevation | 0.0196 | 0.0020 |
Precipitation*Relative humidity | -0.0032 | 0.0004 |
Elevation*Surface water coverage | 0.2472 | 0.0342 |
Elevation*Woodland coverage | 0.7325 | 0.1019 |
Forest coverage*Relative humidity | 0.0903 | 0.0234 |
Surface water coverage*Relative humidity | 0.0281 | 0.0075 |
Surface water coverage | 1.1773 | 0.5507 |
Aedes trivittatus | 1.0682 | 0.0158 |
Aedes sierrensis | 1.1311 | 0.0313 |
Culex quinquefasciatus | 0.5483 | 0.0163 |
Anopheles punctipennis | 0.3948 | 0.0191 |
Aedes canadensis | 0.1567 | 0.0151 |
Aedes albopictus | -0.0976 | 0.0116 |
Aedes aegypti | -0.0879 | 0.0126 |
Anopheles quadrimaculatus | -0.1084 | 0.0163 |
Including interactions increased McFadden‘s pseudo R^{2} to 0.65. Table3 summarizes the results of this procedure. In this table, Exp(β) and their confidence intervals were not included. This is because when factor interactions are included, the value of Exp(β) depends not only on the coefficient of this factor, but also on all the other factors that interact with the factor. All listed factors are significant with a significance level of 0.0001, except the surface water coverage*relative humidity interaction (p-value of 0.0002) and surface water coverage (p-value of 0.0327). Differences from the Table2 results exist, but are not radical. Relative humidity, for example, is not itself significant but interacts with several other factors. All other individual factors remain significant predictors. The parameter estimates of the main effects are slightly different from those in Table2 since interactions are now included. For example, the parameter estimates of both median household income and population density are smaller than those in Table2 and their interaction is significant. The significant interaction indicates that the changes of heartworm prevalence as median household income changes are conditional on the value of population density, and vice versa.
Conclusions
Our factor list in Table1 is likely incomplete; however, it should serve as a starting point. Several important factors that were discussed in Brown et al.[2] could not be included in the analysis due to data deficiencies. For example, coyotes and feral dogs are known reservoirs of heartworm infection. Unfortunately, no complete data of their populations were available. Additionally, mosquito vector abundance data were unavailable. Should information regarding coyote counts and vector abundance be consistently collected in the future and incorporated in the methods discussed herein, the model would improve in power and practical implications.
Some of the factors could be refined. For example, it is frequently posited that heartworm degree units for temperature likely influence prevalence rates[26]. Such units are typically measured from 14 degree Celsius (57 degrees Fahrenheit), below which it is difficult to have high heartworm transmission. We have not looked at temperature departures above 14 degree Celsius because our data are annual. Also, subtracting 14 from the annual temperatures would reparametrize the value of β_{0}, but would not change the overall regression fit.
The fitted regression model is currently being explored to forecast future values of heartworm prevalence. Indeed, many of the predictive factors vary with time (temperature and precipitation, for example). From a forecast of these factors — say a year in advance — and our fitted logistic regression model, predictions of prevalence rates can be made. This can inform the pet owner and practitioner in advance of a potentially bad heartworm season. The results can also be used to assess prevalence rate changes due to climate change. For example, if annual temperatures are expected to increase by one degree F, one could add one degree to the temperature in the fitted logistic regression model to predict the change.
Declarations
Acknowledgements
The authors would like to thank the Companion Animal Parasite Council for providing the data and supporting this study financially.
Authors’ Affiliations
References
- CAPC parasite prevalence maps heartworm. [http://www.capcvet.org/parasite-prevalence-maps],
- Brown HE, Harrington LC, Kaufman PE, McKay T, Bowman DD, Nelson CT, Wang D, Lund R: Key factors influencing canine heartworm, Dirofilaria immitis, in the United States. Parasites Vectors. 2012, 5: 245-10.1186/1756-3305-5-245.PubMed CentralView ArticlePubMedGoogle Scholar
- Sacks BN, Woodward DL, Colwell AE: A long-term study of non-native-heartworm transmission among coyotes in a Mediterranean ecosystem. Oikos. 2003, 102: 478-490. 10.1034/j.1600-0706.2003.12590.x.View ArticleGoogle Scholar
- Levy JK, Lappin MR, Glaser AL, Birkenheuer AJ, Anderson TC, Edinboro CH: Prevalence of infectious disease in cats and dogs rescued following Hurricane Katrina. J Am Vet Med Assoc. 2011, 238: 311-317. 10.2460/javma.238.3.311.View ArticlePubMedGoogle Scholar
- ANTECH Diagnostics. [http://www.antechdiagnostics.com/Main/Home.aspx],
- IDEXX Laboratories. [http://www.idexx.com/view/xhtml/en_us/corporate/home.jsf],
- National climatic data center. [http://www7.ncdc.noaa.gov/CDO/cdoselect.cmd?datasetabbv=GSOD&countryabbv=&georegionabbv],
- National climatic data center. [http://www.ncdc.noaa.gov/monitoring-references/maps/us-climate-divisions.php],
- Paroscientific, Inc. [http://www.paroscientific.com/dewpoint.htm],
- Peakbagger.com. [http://peakbagger.com],
- United States department of agriculture. [http://quickstats.nass.usda.gov/],
- Census bureau. [http://www.census.gov/support/USACdataDownloads.html\#LND],
- Census bureau. [http://factfinder2.census.gov/faces/nav/jsf/pages/searchresults.xhtml?refresh=t],
- Bureau of labor statistics, U.S. Department of labor. [http://www.bls.gov/cpi],
- Moore CG: Exotic mosquitoes in the USA: updates and trends. 27th Biennial State Public Health Vector Control Conference. March 25–26, 2008, Fort Collins, Colorado, [http://www.cdc.gov/ncidod/dvbid/westnile/conf/27thbiennialVectorControl/index.htm],Google Scholar
- Hynes NA: Dengue: a reemerging concern for travelers. Clevel Clin J Med. 2012, 79 (7): 474-482. 10.3949/ccjm.79a.11048.View ArticleGoogle Scholar
- Darsie RF, Ward RA: Identification and Geographical Distribution of the Mosquitoes of North America, North of Mexico. 2004:384, Gainseville: University Press of FloridaGoogle Scholar
- Wilson EB: Probable inference, the law of succession, and statistical inference. J Am Stat Assoc. 1927, 22: 209-212. 10.1080/01621459.1927.10502953.View ArticleGoogle Scholar
- Tukey PA, Tukey JW: Graphical Display of Data Sets in 3 or More Dimensions, Interpreting Multivariate Data. 1981, New York: WileyGoogle Scholar
- Mungiole M, Pickle LW, Simonson KH: Application of a weighted head-banging algorithm to mortality data maps. Stat Med. 1999, 18: 3201-3209. 10.1002/(SICI)1097-0258(19991215)18:23<3201::AID-SIM310>3.0.CO;2-U.View ArticlePubMedGoogle Scholar
- Hansen KM: Head-banging: robust smoothing in the plane. IEEE Trans Geosci Remote Sensing. 1991, 29: 369-378. 10.1109/36.79427.View ArticleGoogle Scholar
- Pohlman JT, Leitner DW: A comparison of ordinary least squares and logistic regression. Ohio J Sci. 2003, 103 (n5): 118-125.Google Scholar
- Czepiel SA: Maximum likelihood estimation of logistic regression models: theory and implementation. [http://czep.net/contact.html],
- McFadden D: Conditional logit analysis of qualitative choice behavior. Frontiers In Econometrics. Edited by: Zarembka P. 1973, New York: Academic Press, 105-142.Google Scholar
- Hosmer DW, Lemeshow S: Model-building strategies and methods for logistic regression. Applied Logistic Regression, 2nd Edition. 2000, New Jersey: John Wiley & Sons, Inc.View ArticleGoogle Scholar
- Shearer P: Literature review - heartworm disease. Banfield Appl Res & Knowl Team. 2011, 1-16. [http://www.banfield.com/getmedia/e456eec3-77f2-46f1-b302-87a17a8fba0a/5906a327-4e82-49f7-9432-e7da7fc568a9-pdf0],Google Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate. Please note that comments may be removed without notice if they are flagged by another user or do not comply with our community guidelines.