Epidemiology of Rodent Bites and Prediction of Rat Infestation in New York City

Transcription

1 American Journal of Epidemiology Vol. 148, No. 1 Printed in U.S.A. Epidemiology of Rodent Bites and Prediction of Rat Infestation in New York City James E. Childs, 1 Sara L. McLafferty, 2 Ramses Sadek, 1 Gayle L Miller, 1 Ali S. Khan, 1 E. Randy DuPree, 3 Ranjan Advani, 3 James N. Mills, 1 and Gregory E. Glass 4 The authors examined the epidemiology of rodent bites occurring in New York City from 1986 through 1994 to identify factors contributing to increased probability of rodent bite and rat infestation. City blocks on which a rodent bite case had been reported (n = 415) and three control blocks per bite block, matched by borough and randomly selected, were compared according to demographic characteristics obtained from US Census data. Environmental variables were defined using a geographic information system to extract distances to areas potentially providing food or refuge for rats, such as parks. Borough-specific models of bite risk were generated by logistic regression using data collected from 1991 to 1994; risk values were then generated for all city blocks. Field surveys for signs of rat infestation conducted on 31 randomly selected blocks indicated a significant association between degree of infestation and predicted risk. Spatial analyses comparing neighboring blocks showed that blocks with bite cases were significantly clustered. The models based on data from previous years correctly predicted 72 percent of 53 block addresses of rodent bite cases from 1995 as being locations of high or intermediate risk. A combination of geographic and epidemiologic analyses could help investigators identify the spatial occurrence of rat infestation over a large area and might help to focus control activities. Am J Epidemiol 1998; 148: bites and stings; geography; information systems; rats; risk assessment; rodent control; urban health In cities across the United States, introduced rodents of the family Muridae (the Norway rat (Rattus norvegicus), the black rat (Rattus rattus), and the house mouse (Mus musculus)) are locally abundant and the sources of considerable public health and economic concern. Numerous zoonotic infections can be acquired through contact with these rodents or their ectoparasites (1). Within the past decade, reports of human disease associated with a rat-borne Hantavirus have been made from Baltimore, Maryland (2), and diseases such as leptospirosis (3) and murine typhus (4) are ongoing concerns in many US cities. House mice and Norway rats are common throughout the Received for publication September 2, 1997, and in final form January 16, Division of Viral and Rickettsial Diseases, National Center for Infectious Diseases, Centers for Disease Control and Prevention, Atlanta, GA. 2 Department of Geography, Hunter College, New York, NY. 3 New York City Department of Health, New York, NY. 4 Department of Molecular Microbiology and Immunology, Johns Hopkins University School of Public Health, Baltimore, MD. Reprint requests to Dr. James E. Childs, National Center for Infectious Diseases, Centers for Disease Control and Prevention, 1600 Clifton Road NE, Mailstop G13, Atlanta, GA This paper was prepared under the auspices of the US Government and is therefore not subject to copyright. United States and are reported from all states, while black rats are largely limited to coastal regions, primarily in the warmer South (5). Contact with rodents can result in bites. In 1970 and 1971, the Center for Disease Control conducted a national animal-bite surveillance program in 15 health jurisdictions; rodents accounted for 4.3 percent of the 196,117 recorded bites (6). However, the epidemiology of rodent bites has received little attention, and the information available is negligible compared with data concerning dog and cat bites (7-10). Cases of rat bite sporadically catch public attention when the victims are infants, but most accounts are limited to case reports of rat-bite fever, a rare but potentially fatal infection primarily caused by Streptobacillus moniliformis in the United States (11). In one of the few epidemiologic studies of rat bites, investigators examined 50 consecutive bite patients appearing in a Los Angeles, California, emergency room (12). The majority of patients were under 15 years of age, and most of the bites were not severe; only one patient developed a bacterial infection that required antibiotic treatment. In contrast to bites from carnivores, bites from rodents (except the larger species such as beavers or woodchucks) are not considered to carry a high risk for rabies virus transmission 78

2 Rodent Bites and Rat Infestation in New York City 79 (13), and victims are rarely prescribed postexposure treatment (14). The distribution of rodent populations within urban centers has only rarely been determined directly by trapping or by examining urban neighborhoods for signs of infestation (15). Mice are caught almost exclusively within residences, while rats are found within dwellings and in alleys, sewers, and other locations (16, 17). Indirect survey methods, such as mapping of reports of rodent sightings in Baltimore (17) or of the home addresses of Detroit, Michigan, children found to be seropositive for Leptospira icterohaemorrhagiae (18), have permitted inferences about the urban conditions that support extensive rodent infestation, particularly rat infestation. Although less affluent inner-city neighborhoods are typically the areas identified as having the most severe problems, no studies have systematically defined factors that might predict rat infestation and potentially the risk of rodent-associated problems. In this investigation, we examined the epidemiology of rodent bites reported from New York City and used these data and the environmental and social characteristics of the home blocks of bite patients to generate predictive models for rodent bite. After control blocks were selected and the risk of a block's having a resident bite case was modeled, a New York City-wide map of rodent bite risk was produced. We then tested the predictions of statistical and spatial modeling by comparing block addresses of additional rodent bite victims from a subsequent year of surveillance with the addresses' predicted risk classification, and by environmental sampling of randomly selected blocks for evidence of rat activity. MATERIALS AND METHODS Rodent bite data Animal bites to humans in the five boroughs of New York City (Brooklyn, Bronx, Manhattan, Queens, and Staten Island) are reportable to the New York City Department of Health. Information on the descriptive epidemiology of rodent bite was derived from surveillance reports submitted to the health department from January 1986 through September A subsample of reports from January 1991-September 1994 was examined for statistical modeling and for use in the geographic information system analyses. The analysis was limited to the previous 4 years to reduce the influence of environmental and neighborhood change within the city. Of the initial 829 reports of rodent bite, 315 cases were excluded because they occurred in a work-related setting (mostly university laboratories; n = 20) or had incorrect addresses {n = 295). This left 514 bite patients whose addresses could be mapped to 415 distinct blocks using Arclnfo software (Environmental Systems Research Institute, Redlands, California) and the Bureau of the Census' enhanced TIGER files for New York City (19). Blocks were defined as the geographic units for study, which permitted the use of 1990 US Census data on demographic determinants at either the block level or the census tract level. Demographic variables Block statistics on total population, race (percentage black, Asian, or white), percentage Hispanic, and numbers of people younger than 18 years and older than 65 years were extracted from 1990 Census files for determination of the demographic characteristics of each block. Information on median household income was available at the census tract level, and that value was assigned to all blocks within each tract. For assessment of crowding and the potential for environmental conditions which provide harborage for rodents (e.g., vacant buildings), the total number of housing units in each block and the total number of units occupied were included. A density or crowding variable was computed by dividing the total block population by the number of occupied housing units. Environmental variables Environmental variables were assessed by determining the block's location relative to sites potentially serving as rodent refuge or food resources. Arclnfo was used to query existing databases and to compute the distance (in meters) from the block center to the geographic feature. Attention was focused on parks, subways, railroad lines, and highways, because these areas can provide suitable habitat or exposed earth which can be used as burrow sites by Norway rats. Data files on sanitation records of street cleanliness were not available, but the distance to the nearest waste transfer station was measured. Analyses For analysis of the effects of demographic and environmental factors on rodent bite risk, each block from which a bite had been reported was compared with approximately three control blocks randomly selected from all of the remaining blocks in each borough. The final database consisted of 1,660 blocks: 528 in Brooklyn (132 bite blocks and 396 control blocks), 448 in Manhattan (112 and 336 blocks), 372 in the Bronx (93 and 279), 272 in Queens (68 and 204), and 40 in Staten Island (10 and 30). For each

3 80 Childs et al. control block, a suite of variables identical to that of the corresponding bite block was derived. Preliminary analyses included univariate comparisons of variables by means of nonparametric statistical tests performed independently for each borough. The spatial aggregation of blocks with rat bites, through the 10th nearest neighboring block, was tested by comparing the distances (in meters) from the centers of bite and control blocks using the method of Cuzick and Edwards (20), with Simes correction for multiple comparisons (21). Logistic regression analyses for the New York City data set and for each borough were used to generate equations for prediction of the risk of a block's containing a rat-bite case. Data on most demographic and environmental variables were dichotomized at the median value before being entered into the model. Blocks were assigned a value of 1 if their value exceeded the median value for the entire borough and a 0 otherwise. When continuous variables fitted the model better, they were retained. In each analysis, logistic regression was performed for all subsets to determine the two best subsets of different numbers of variables (22). Subsets were examined for their fit to the data (Hosmer-Lemeshow goodness-of-fit statistic (C)), for sensitivity (number of bite blocks correctly classified as bite blocks/[number of bite blocks correctly classified as bite blocks plus number of bite blocks incorrectly classified as nonbite blocks]), for specificity (number of nonbite blocks correctly classified as nonbite blocks/fnumber of nonbite blocks correctly classified as nonbite blocks plus number of nonbite blocks incorrectly classified as bite blocks]), and for overall correct classification. The contribution of each combination of variables was assessed by selectively removing variables and evaluating the data using partial F tests. In some analyses, variables that were not statistically significant were retained because they contributed to increased model sensitivity. Final logistic regression models were selected to maximize sensitivity and overall correct classification, and 2-5 variables without interaction terms were retained. For each borough, the logistic regression equation was used with the Arclnfo data layers to generate a rodent bite risk score for every block with a nonzero residential population. Statistical analyses were performed using SAS (23) or SPSS (24). Specific tests are indicated below, and all p values are based on twotailed tests of significance. Field assessments Field surveys were conducted in autumn 1996 to test the accuracy of the borough-specific predictions of rodent bite classification. A list of 31 randomly selected blocks in Manhattan and Brooklyn (13 predicted to be low risk, 13 predicted to be high risk, and five predicted to be intermediate risk) was provided to a survey team that was blinded to their predicted value. Surveys were performed using a standardized environmental survey form (25) designed to assess rat infestation. The housing characteristics of each block were assessed by determining the proportions of addresses that were residential, commercial (businesses), mixed (residential and commercial), and food-oriented (groceries or restaurants). Proportions of vacant buildings and empty lots were determined. The proportion of properties without proper refuse containers or with accessible garbage, pet food, and/or animal feces provided an indicator of food resources for rats, and the presence of abandoned vehicles, lumber piles, appliances, and other large items was recorded as a measure of potential harborage for rats. The proportion of houses with physical evidence of infestation or other indicators of rat activity was also recorded. All proportions were transformed by taking the arcsin of the square root, and one-way analysis of variance was used, with predicted risk classification as the categorical variable. Analytical results with two-tailed p values less than were considered statistically significant. Prediction of future block sites of rat-bite cases Addresses of rodent bite cases reported to the New York City Department of Health in 1995 were analyzed by the methods described above. Rodent bite risk values generated by the borough-specific logistic regression analysis were computed for each geocoded block for ascertainment of how accurately sites had been predicted. RESULTS Demographic characteristics The 514 evaluated cases of rodent bite occurred throughout the five boroughs of New York City (table 1 and figure 1). The bite cases were evenly distributed by sex, and the median age of case-patients was 22 years (range, <1 year to 93 years). Most bites occurred at the extremities of the body, with approximately half occurring while the victim was asleep (table 1). The biting animal was most commonly (81 percent) identified as a rat. Most bites were treated by simple wound washing, and most (98 percent) patients were immediately released from the treatment facility. Clustering of bite cases within each borough was apparent, and the aggregation of 415 blocks with bite

4 Rodent Bites and Rat Infestation in New York City 81 TABLE 1. Characteristics of 514 New York City residents reported to have received a rodent bite during the period Characteristic Year Borough Brooklyn Bronx Manhattan Queens Staten Island Sex Male Female Bite site Finger/hand Toe/foot Leg Head/face/neck Arm Body/trunk Biting animal Rat Mouse Other Situation Asleep Awake Treatment Medical None Surgical Disposition Released Admitted No.* % * Numbers do not total 514 in some categories because of missing values. cases, through the 10th nearest neighboring block, was statistically significant (all z scores > 6.0; all p values < ; Simes correction: p < ). Bites were most common in northern Brooklyn, the southern Bronx, and the southern and northern extremes of Manhattan (figure 1). Overall, 371 (89.4 percent) of the 415 blocks contained a single bite case, 34 (8.2 percent) contained two cases, nine (2.2 percent) contained three, and one contained four. The number of blocks reporting more than one bite case was 13 for the Bronx (14 percent of all bite blocks), eight for Brooklyn (6.1 percent), 22 for Manhattan (19.6 percent), zero for Queens, and one for Staten Island (10 percent). Characterization of block of residence Most people who had been bitten by rodents lived on blocks with significantly lower median incomes than those of control blocks (table 2). In addition to containing more residents per block (four of five boroughs), case blocks also had a higher percentage of residents younger than 18 years of age and a lower percentage of residents older than 65 years of age. The mean number of housing units per block and the mean percentage of rental units per block were greater for case blocks in each borough (table 2). In each borough, the percentage of the population characterized as black or Hispanic was higher on the case blocks. Blocks with more than one bite case showed trends identical to those of blocks with single bite cases; only in Manhattan were significant differences found between blocks with different numbers of reported bite cases. Blocks with multiple bites in Manhattan had lower median incomes, lower percentages of Asian and white residents, higher percentages of black residents, and higher percentages of residents under 18 years of age (Wilcoxon rank sum test: all p's < 0.05) in comparison with single-bite blocks or control blocks. These differences were similar to those between all bite blocks and control blocks, and therefore blocks with any number of bite cases were pooled for future analyses. Distances to environmental features varied substantially by borough, but some patterns in their association with rodent bite blocks were discernable (table 2). Distances to subways (five of five boroughs), waste stations (four of five), railroads (four of five), and parks (five of five) were shorter for case blocks. No pattern was apparent in distance to the nearest highway. Logistic regression modeling of rodent bite blocks The best subset logistic regression models for New York City and for each borough contained 2-5 variables (table 3). Variables included in the models of more than one borough or of New York City overall were median income, total population of block, distance to subway, percentage Hispanic, and percentage younger than 18 years. Seven other variables were used in individual borough models (table 3). The fit of each model varied considerably, and individual borough models provided better fits to the data than did the overall New York City model (table 4). Four of the five borough models provided good fits to the data, as indicated by Hosmer-Lemeshow goodness-of-fit statistics (table 4). In each case, these models provided a sensitivity greater than 60 percent

5 82 Childs et al. FIGURE 1. Distribution of 415 blocks in the five boroughs of New York City on which a person reporting a rodent bite lived, The boundaries of each borough are defined by thicker lines. A single dot is displayed even for blocks from which more than one bite was reported. Data on rodent bite, a reportable condition in New York City, were analyzed for , but not all addresses could be coded and mapped. (range, 61.8 percent to 70.0 percent) and a specificity near or above 70 percent (range, 69.8 percent to 86.7 percent). The citywide and Brooklyn models provided the poorest fits to the data, with sensitivities below 60 percent, although both the overall percentage of blocks correctly characterized and the percent specificities were over 70 percent (table 4). Mapping risk of rodent bite Overall, 28,478 (85 percent) of the 33,468 blocks in the five boroughs had nonzero populations, and a rodent bite risk score could be obtained for each borough using the logistic regression models (table 5). Maps were drawn by stratifying risk probabilities at the median value (0.041) and the 72nd percentile (0.7) (figure 2 (Staten Island not shown)); this formed three classes of blocks: high risk (22 percent of blocks), intermediate risk (28 percent), and low risk (50 percent) blocks (table 5). The second cutoff point was selected after exploratory analyses revealed a cluster of values between 0.7 and 0.9. Risk scores were not generated for blocks without human residents. Field assessment High risk blocks were only marginally more likely to be infested than were low and intermediate risk blocks, but the degree of infestation was significantly greater. Among the 31 blocks surveyed, 10 of 13 predicted to be high risk had evidence of active rat infestation in one or more residences, compared with eight of 13 predicted to be low risk and three of five predicted to be intermediate risk. However, in blocks predicted to be high risk, the average proportion of addresses infested was 27.3 percent as compared with 8.3 percent for intermediate risk blocks and 9.1 percent for low risk blocks. An initial one-way analysis of variance indicated no difference between intermediate risk and low risk blocks, and when these blocks were combined, a significant difference was evident in the proportion of addresses infested on high risk blocks

6 Rodent Bites and Rat Infestation in New York City 83 TABLE 2. Comparison of demographic data (mean values) from 415 city blocks on which rodent bite patients resided and 1,245 randomly selected control blocks from each borough, New York City, t Characteristic Total population ot block %<18 years of age % >65 years of age Median annual income (dollars)* Race/ethnlclty (%) White Black Hispanic Asian No. of housing units/block % of rental units Distance (m) to a geographic feature Highway Waste station Subway Railroad Park Bite 519"* 23*** 8 26, "* 3 186"* 82«* 500 5, , Brooklyn No bite "* 32,966"* 53*" " 4, "* 2, Bite 873" 19*" 12 31, "* 35«* 7 396* 85*" 2,601" 2, , Bronx No bite ,757'" 64*** ,136 2, " 161 Bite 759*** 32«" 9 19, «53«* 2 259"* 90"* 4,911 5, Manhattan No bite "* 32,880"* 49** ,184" 6,008" 758"* 1,124"* 261 Staten Island Bite , ,331 9,294 6, No bite ,861" 89* ,552* 11,539* 8,997 1,203* 370 * pi 0.05; " p <. 0.01; * p i (Wlfcoxon rank sum test). t Asterisks indicating significant differences are placed beside the category with the higher value. t Income data were available by census tract only, and the tract value was given to each Mock within the tract. Race (white, black, Asian) and ethnicity (Hispanic) sum to more than 100% because these are not mutually exclusive categories. Bite 584*** 23* 12 33, * 25"* 15* 218"* 82"* 563 6, , Queens No bite " 43,259*" 58* ,743** 2,167*" 1, compared with other blocks (F = 4.28, df = 1, 29; p = 0.04). High risk blocks contained a higher proportion of vacant addresses (4.4 percent) than did low and intermediate risk blocks (0.9 percent) (F = 3.88, df = 1, 21; p = 0.06). No other significant differences were found between high risk blocks and other blocks with regard to types of housing, proportion of residences lacking garbage cans, or proportion of residences with refuse or potential rat harborage sites. Prediction of bite blocks Of the 78 cases of rodent bite reported in New York City in 1995, 53 could be geocoded and assigned risk scores for rodent bite based on borough-specific models. Thirty (56.6 percent) of the blocks on which a case resided were classified as high risk, eight (15.1 percent) were classified as intermediate risk, and 15 (28.3 percent) were classified as low risk. This distribution was significantly different from that expected on the basis of the proportion of New York City blocks in each risk classification (^ = 58.5, df = 2; p < 0.001). The percentage of blocks classified as being of medium or high bite risk in 1995 was 83.3 percent in Brooklyn (n = 24), 77.8 percent in the Bronx (n = 9), 45.5 percent in Manhattan (n = 11), and 66.7 percent in Queens (n = 9). DISCUSSION This study examined the epidemiology of rodent bites in New York City, with the goal of generating predictive models for the risk of encountering rodents, especially rats, at the city block level, using the demographic, social, and physical attributes of blocks. Blocks were selected as the study units, rather than the exact addresses of the rodent bite cases, for several reasons. Practically, census data for many demographic characteristics were available for the block or tract level. More importantly, rat infestations are typically a community problem and are not usually restricted to single residences (15, 17, 26, 27). The significant clustering of bite blocks throughout New York City corroborates this conclusion. The accuracy of the models and maps was examined by two methods, one involving physical inspection of randomly selected blocks of different predicted risk levels and the second assessing model predictions based on rodent bite data generated from an additional year of surveillance. Both approaches indicated that the models provide information that can help in distinguishing areas of rat infestation and quantifying the potential risk of rodent bite over large geographic areas. However, the evaluation procedures also demonstrated that rat infestation is a continuous rather than a categorical factor. The fact that some blocks were misclassified with regard to the occurrence of rodent bite or rat infestation was expected. Rodent bite is a rare event, occurring infrequently even in locations where rats and mice are common. A questionnaire survey of 1,363 persons residing in Baltimore (17) indicated that although rats and mice were frequently seen on streets and in alleys (64 percent of respondents reported seeing rats and 32 percent reported seeing mice), they were infrequently

7 84 Childs et al. a 2. 2 O X o I o I os I i 1 a S a is i I 1.a co 0 2 o *; r-: in! oi o oj CNI * * co ci r^ d r^ lo CO (O tb CO 8 S) q o r-: ci w I I I I I I I I & I I I II I I I I I I «I I I S I a 2 2 J # a T^ T^ 4 i- J> 0) i-m o> ^ a> 1 odd?- 0) I I I ^(O I I I I CO ' <D ' ' ' ' 0) 00 ' ' ' ' T- T- O O cq p o) r^ c\i ^ 10 o LO CO O IT) O CO 00 CNJ f* O r^ O T&88 I I 9 I I I I I I T T 7 i3- o _- o o_ Q ^ o reported within residences (6 percent for rat infestation, 49 percent for mice) and were rarely associated with bites (1.2 percent of respondents reported experiencing any rodent bite during their lifetime). Furthermore, rodent bites are underreported. Previous studies in New York City have indicated that only 41 percent of animal bites treated in emergency rooms are reported to health authorities (28). Even bites inflicted by dogs are frequently not reported: In Pennsylvania, a study of children aged 4-18 years found that the rate of dog bite was 36 times greater than the rate reported to health authorities (29). In addition to underreporting, we were unable to geocode a high percentage of addresses, primarily because inaccurate addresses were given or recorded at the treatment centers. These limitations would result in some bite blocks' being classified as nonbite blocks and would reduce precision, biasing our model results toward the null (30). However, the subsequent assessments of the models and the evidence of clustering among bite blocks provided confidence that characteristics relevant to rat infestation and risk of rodent bite were captured. Another concern was that the field assessment measured only rat infestation, but the statistical models used to classify blocks were based on reported rat or mouse bites. A previous study in another urban location found a similar spatial distribution of rat and mouse infestations (17), which suggests that infestations are influenced by similar conditions. Because 81 percent of our bite patients reported rats as the source (table 1) and because juvenile rats can be misidentified as mice, our models were heavily weighted toward identification of potentially rat-infested blocks. The variables associated with bite blocks were consistent with the ecology of urban rats (31) and their distribution within urban centers (15, 18). These variables mostly consisted of a set of characteristics generally associated with areas of lower socioeconomic status, including high percentages of ethnic minorities (except Asians), young populations with a lower proportion of individuals over age 65 years, more residential units per block, with rental units comprising a high proportion, and lower median incomes. A study conducted in Baltimore in the 1940s found that the degree of rat infestation was most highly associated with blocks containing a high proportion of residences in need of repair (15), while a survey conducted in the 1980s found an association between rat exposure and lower income (17). The associations of the environmental variables with case blocks were less consistent and more difficult to interpret. Case blocks tended to be closer to

8 Rodent Bites and Rat Infestation in New York City 85 TABLE 4. Summary statistics for final regression models used to predict blocks with cases of rodent bite, New York City, Borough Adjusted FP HosmerLemeshow statistic All boroughs Brooklyn Bronx Manhattan Queens Staten Island P value correct* sensitivity* specificity* * Calculated at a probability of being a block with a rodent bite case of subways, railroads, and parks, all of which are a potential source of exposed ground in which Norway rats can burrow. Parks and subways can also be sources of food through human refuse. Zoologic parks have been associated with major foci of rat infestations in other cities (32). However, living in proximity to noisy transportation systems may also be a characteristic of less affluent neighborhoods. Mean distance to waste stations also tended to be shorter for bite blocks, although it was not clear whether this was related to potential food sources or other neighborhood characteristics. Still, the significance of environmental variables in the logistic regression models controlling for socioeconomic variation suggests some independent effect on the incidence of rat bites. The restriction of our sample base to bite patients Risk Probability tm o H No Data Scale 1:250,000 FIGURE 2. Risk of a block's being a site where a rodent bite could have occurred, based on logistic regression modeling of bite risk for each of five boroughs (Staten Island not shown), New York City, Risk was stratified into three classifications, with half of the blocks being low risk and 22% and 28% being intermediate and high risk, respectively. Am J Epidemiol Vol. 148, No. 1, 1998

9 86 Childs et al. TABLE 5. Classification of all blocks in New York City into areas of varying rodent bite risk using common cutoff values, New York City, * Borough H '9 h Bite risk flhvti^ o >0-7 > p7o,7* Brooklyn 3,290 (43)t 0 Bronx 2,118(60) 33 Manhattan 849 (34) 839 (34) Queens 8 6,508 (58) Staten Island (19) Low (p<004) population 4,287(57) 1,223 1,404(40) 1, (32) 564 4,715(42) 1,727 2,963(81) 441 All boroughs 6,265(22) 8,062(28) 14,150(50) 4,989 * This scheme was used to select blocks for field surveys of rat infestation for assessment of the scoring indices in terms of their usefulness in predicting locations of rat abundance. t Numbers in parentheses, percentage of area in that risk category. with geocoded addresses raises questions concerning how representative of rodent bite victims our sample was. However, few data on rodent bites with which to compare our results are available. In a case series from Los Angeles (12), the mean age of treated patients was 10.8 years, approximately half of our figure, and fewer bites to the lower extremities were reported (~14 percent vs. >28 percent). These differences may reflect the epidemiology of rat bite in various geographic locations or different methods of case selection. In southern areas of the United States, especially in port cities, the most common rat may be the black or roof rat, R. rattus (33). The black rat is highly arboreal, and its ecology is sufficiently different from that of R. norvegicus that the epidemiology of bites inflicted by this species may be distinct. The coupling of geographic information system applications and epidemiologic modeling is a relatively new procedure, and the methodology is still evolving rapidly. In a similar study, investigators used casecontrol methods based on residential units to model environmental parameters associated with Lyme disease (34). In that study, residences of Lyme disease patients in Baltimore County, Maryland, were matched with randomly selected addresses. A risk map and model correctly classified 85.8 percent of the addresses of Lyme disease patients identified during a subsequent year of study. These complementary approaches identifying factors associated with a disease or an event and subsequently mapping where the factors occur are most useful when limited resources are available for assessment of risk over a large geographic area. The traditional field survey approach used in this study to characterize rat infestations required three field crews of two or three persons working for a 2-week period yet only surveyed a minute fraction of New York City. Some of the most useful applications of this type have been studies of zoonotic diseases, a category which can include animal bites (35). The potential use of these methods for assessing human risk for rat-borne diseases such as leptospirosis and murine typhus requires additional exploration. The potential application of these methods in helping to focus control efforts is apparent, although the means of implementing control may be lacking. Many cities have rodent control programs that are understaffed, and efforts are typically reactive rather than proactive. Responding to public complaints of rodent infestation does not ensure that efforts are directed toward locations with the greatest problem. In addition, the fact that boroughspecific models provided substantially better fits to the data indicates that generalization of these models to other cities would not necessarily be useful. The refinement of methods explored in this study would hold promise in terms of helping to direct control efforts through a process that is less costly than field surveys and less prone to misdirection by differences in the degree of tolerance of rodent infestation among communities. REFERENCES 1. Gratz NR. Rodents and human disease: a global appreciation. In: Prakash I, ed. Rodent pest management. Boca Raton, FL: CRC Press, 1988: Glass GE, Watson AJ, LeDuc JW, et al. Domestic cases of hemorrhagic fever with renal syndrome in the United States. Nephron 1994;68: Vinetz JM, Glass GE, Flexner CE, et al. Sporadic urban leptospirosis. Ann Intern Med 1996;125: Dumler JS, Taylor JP, Walker DH. Clinical and laboratory features of murine typhus in south Texas, 1980 through JAMA 1991;266: Brown RZ. Biological factors in domestic rodent control training guide. Atlanta, GA: US Communicable Disease Center, (Public Health Service publication no. 773). 6. Moore RM Jr, Zehmer RB, Moulthrop JI, et al. Surveillance of animal-bite cases in the United States, Arch Environ Health 1977;32: Beck AM. The epidemiology and prevention of animal bites. Semin Vet Med Surg (Small Anim) 1991;6: Gershman KA, Sacks JJ, Wright JC. Which dogs bite? A case-control study of risk factors. Pediatrics 1994;93: Mathews JR, Lattal KA. A behavioral analysis of dog bites to children. J Dev Behav Pediatr 1994; 15: Wright JC. Reported cat bites in Dallas: characteristics of the cats, the victims, and the attack events. Public Health Rep 1990;105: Holmes B, Pickett MJ, Hollis DG. Unusual gram-negative bacteria, including Capnocytophaga, Eikenella, Pasteurella, and Streptobacillus. In: Murray PR, Baron EJ, Pfaller MA, et al, eds. Manual of clinical microbiology. 6th ed. Washington, DC: ASM Press, 1995: Ordog GJ, Balasubramanium S, Wasserberger J. Rat bites: fifty cases. Ann Emerg Med 1985; 14: Childs JE, Colby L, Krebs JW, et al. Surveillance and spatiotemporal associations of rabies in rodents and lagomorphs in the United States, J Wildl Dis 1997;33:20-7.

10 Rodent Bites and Rat Infestation in New York City Rabies prevention United States, Recommendations of the Immunization Practices Advisory Committee (ACIP). MMWR Morb Mortal Wkly Rep 1991 ;40(RR-3): Davis DE, Fales WT. The distribution of rats in Baltimore, Maryland. Am J Hyg 1949;49: Childs JE, Glass GE, Korch GW, et al. Lymphocytic choriomeningitis virus infection and house mouse (Mus musculus) distribution in urban Baltimore. Am J Trop Med Hyg 1992; 47: Childs JE, Glass GE, LeDuc JW. Rodent sightings and contacts in an inner-city population of Baltimore, Maryland, U.S.A. Bull Soc Vector Ecol 1991;16: Demers RY, Thiermann A, Demers P, et al. Exposure to Leptospira icterohaemorrhagiae in inner-city and suburban children: a serologic comparison. J Fam Pract 1983;17: Marx R. The TIGER system: automating the geographic structure of the US Census. Governm Publ Rev 1986;13: Cuzick J, Edwards R. Spatial clustering for inhomogeneous populations. J R Stat Soc Ser B 1990;52: Simes RJ. An improved Bonferroni procedure for multiple tests of significance. Biometrika 1986;73: Hosmer DW Jr, Lemeshow S. Applied logistic regression. New York, NY: John Wiley and Sons, Inc, SAS Institute, Inc. SAS/STAT software: changes and enhancements through release Cary, NC: SAS Institute, Inc, Norusis MJ. SPSS for Windows: base system user's guide, release 6.0. Chicago, IL: SPSS, Inc, Davis H, Casta A, Schatz G. Urban rat surveys. Washington, DC: US Department of Health, Education, and Welfare, (DHEW publication no. (CDC) ). 26. Kaukeinen D. Rodent control in practice: householders, pest control operators and municipal authorities. In: Buckle AP, Smith RH, eds. Rodent pests and their control. Cambridge, England: CAB International, 1994: Margulis HL. Rat fields, neighborhood sanitation, and rat complaints in Newark, New Jersey. Geogr Rev 1977;67: Beck AM. The epidemiology of animal bite. Compend Continu Educ Pract Vet 1981;3: Beck AM, Jones BJ. Unreported dog bites in children. Public Health Rep 1985;100: Kleinbaum DG, Kupper LL, Morgenstern H. Epidemiologic research: principles and quantitative methods. New York, NY: Van Nostrand Reinhold Company, Glass GE, Childs JE, Korch GW, et al. Comparative ecology and social interactions of Norway rat (Rattus norvegicus) populations in Baltimore, Maryland. Occas Pap Mus Nat Hist Univ Kans 1989;130:l Farhang-Azad A, Southwick CH. Population ecology of Norway rats in the Baltimore Zoo and Druid Hill Park, Baltimore, Maryland. Ann Zool 1979;15:l Lund M. Commensal rodents. In: Buckle AP, Smith RH, eds. Rodent pests and their control. Cambridge, England: CAB International, 1994: Glass GE, Schwartz BS, Morgan JM III, et al. Environmental risk factors for Lyme disease identified with geographic information systems. Am J Public Health 1995;85: Mott KE, Nuttall I, Desjeux P, et al. New geographical approaches to control of some parasitic zoonoses. Bull World Health Organ 1995;73: