Data
The analyses conducted in this study rest principally on data collected from ED visits in the United States and New York. The national-level data are derived from the Web-based Injury Statistics Query and Reporting System (WISQARS) (Centers for Disease Control and Prevention, 2019). WISQARS is an online, interactive database which provides national estimates of both fatal and nonfatal injuries. The present study utilizes the nonfatal injury data which comes from the National Electronic Injury Surveillance System–All Injury Program (NEISS-AIP), sponsored by the U.S. Consumer Product Safety Commission and the CDC’s National Center on Injury Prevention and Control. The NEISS-AIP is based on a sample of 66 hospitals randomly selected from all hospitals in the United States which have a 24-hour ED and a minimum of six beds. The sample is stratified by hospital size measured in terms of the number of ED visits each year. The nonfatal injury data provide estimates of injuries treated in EDs by cause of injury (e.g., dog bites), race/ethnicity, gender, and disposition of the patient after being released from the ED.
In addition to the WISQARS database, this study examines individual-level patient records from New York. These patient records include a large number of demographic, diagnostic, and treatment variables. The patient records also include more detailed information concerning the racial and ethnic characteristics of patients than is contained in the national data sets. Importantly, the New York patient records include geographic identifiers such as the county or the 5-digit zip code in which the patient resides.
The data for New York come from the Statewide Planning and Research Cooperative System (SPARCS), which is under the auspices of the New York State Department of Health (2019). SPARCS assembles data on outpatient, inpatient, and ambulatory surgery patients treated in all hospitals in New York state.
This study also draws upon data gathered by New York city’s Department of Health and Mental Hygiene (DOHMH) (2019). The database consists of dog bites which are reported via online, fax, or phone to the city’s DOHMH Animal Bite Unit. Each record in the database provides information on: (1) the date of the bite, (2) the breed, age, and gender or the dog, (3) whether the dog was spayed or neutered, and (4) the zip code and borough of the person who was bitten. Altogether, there were 10,280 records spanning the years from 2015 to 2017.
Variables
Injury Code. For both the national and state data sets, identification of patients who were treated for a dog bite was based on two separate injury codes. The International Classification of Diseases, Ninth Revision (ICD-9) External Cause of Injury code (E-code) E906.0 – Dog Bite – was utilized for the years prior to 2015. Both the ICD-9 E-code E906.0 and the ICD-10 E-code W54.0XXA – Bitten by dog (initial encounter) – were utilized for the year 2015. Just the ICD-10 E-code W54.0XXA was used for the years 2016 - 2018.
Sociodemographic Characteristics. Both the WISQUARS and SPARCS data sets furnished information about the age and gender of patients. The SPARCS data sets also included two separate variables about the race and ethnicity of patients. A typology was created from these two variables with the following five values: “white, non-Hispanic,” “black, non-Hispanic,” “Asian, non-Hispanic,” “other, non-Hispanic,” and “Hispanic.” Importantly, the SPARCS database included the patient’s county of residence and his/her 5-digit zip code.
Statistical analyses
To measure the combined effects of year, background characteristics (i.e., gender, age, race/ethnicity), and geographic location on the incidence of dog bites, we conducted a negative binomial regression analysis using the patient records from New York. A negative binomial regression analysis was performed instead of a Poisson regression due to overdispersion of the data.
The population-based counts of both the number of outpatients and inpatients who were bitten by a dog served as the dependent variable in this analysis. The predictor variables comprised the year, geographic location, and the demographic characteristics of the patients. Year was measured as an interval-level variable ranging in values from 1 (corresponding to the year 2005) to 14 (corresponding to the year 2018). To capture possible curvilinear effects of year on the incidence of dog bites, a multiplicative term created by squaring the year variable was also incorporated into the analysis. Geographic location was a dichotomous variable with a value of 1 indicating New York City and a value of 0 indicating New York State omitting New York City. Gender was also a dichotomous variable with a value of 1 indicating male and a value of 0 indicating female. The age variable consisted of 7 categories: under 5, 5 to 9, 10 to 14, 15 to 19, 20 to 44, 45 to 64, and 65 and older. The racial-ethnic background of patients was made up of 5 groups as mentioned above: non-Hispanic white, non-Hispanic black, non-Hispanic Asian, non-Hispanic other, and Hispanic.
Since it can be assumed that the risk of being bitten by a dog varies by population sizes, an offset variable was introduced into the analysis. The offset variable was created in two steps. First, population counts were tallied for each combination of year, geographic location, gender, age group, and racial-ethnic category. So, for example, one count might comprise non-Hispanic Asian females between the ages of 10 to 14 living in New York City in 2014. Altogether, this step yielded 1960 different counts. Next, natural log transformations were carried out on each of these counts.
To measure the demographic correlates of the rate of dog bite injuries at the county level in New York state (N =62), a three-step process was undertaken. First, the number of both outpatients and inpatients were combined for each county for the year 2018 (the most recent year for which data are available). Second, these figures were divided by the population of each county to obtain an injury rate. Finally, the rates were correlated with an array of socio-demographic variables at the county derived from the American Community Survey 2014-2018 (5-Year Estimates) (U.S. Census Bureau, 2018). These variables consisted of the following: (1) population density per square mile, (2) the racial-ethnic composition of the county, (3) median family income, (4) per capita income, (5) percent of families with income below the poverty level, (6) percent of the population 25 and over with a B.A. degree or more, (7) percent of the population with no health insurance, and (8) percent of the insured population with public health insurance.
A similar procedure was conducted to examine the socio-demographic correlates associated with dog bite injuries at the neighborhood level in New York city. For this analysis, the number of outpatients and inpatients were combined for each 5 digit zip code in New York city (N = 179). Next these figures were aggregated up to the United Health Fund (UHF) level (N =42) and divided by the population of each UHF district to obtain an injury rate. These rates were then correlated with the same set of socio-demographic variables described above calculated for each UHF district.
Spatial analysis
To determine the geographic distribution of patients injured by dog bites at the neighborhood level in New York city, a thematically shaded map of the injury rate by the United Health Fund (UHF) district in which the patient resided was created. A Global Moran’s I was computed to assess whether the spatial distribution of the residences of the patients was geographically clustered or dispersed.