Fecal Shedding of Salmonella in Horses Admitted to a Veterinary Teaching Hospital

Fecal samples were collected from 724 horses admitted to, and 232 horses hospitalized at, a veterinary teaching hospital between October 2000 and June 2001, and cultured for Salmonella. Salmonella was isolated on 24 occasions from 12 horses. S. Newport was the most commonly isolated serotype. The estimated prevalence (95% confidence interval CI]) of Salmonella shedding on admission was 0.5% (0.1, 2.0). The estimated incidence rate of Salmonella shedding during hospitalization was 4.3% (95% CI, 2.2-8.1). The highest incidence of Salmonella shedding during hospitalization occurred during June 2001, but no significant (P > .5) temporal clusters of horses shedding Salmonella were detected.Horses shed Salmonella in their feces at a low rate. Horses shedding Salmonella occurred only sporadically and no obvious serotype pattern was observed. The occurrence of fecal shedding probably reflected the prevalence of inapparent Salmonella infection in the hospital source population, providing a baseline against which future potential outbreaks can be identified.

Introduction

Infection of horses by Salmonella organisms is a serious health issue. It is particularly troubling when outbreaks occur in hospitalized patients because these outbreaks can result in substantial economic losses and have a major impact on the welfare of patients.1] Establishments with a high-density of horses, including veterinary teaching hospitals (VTHs) and private veterinary clinics, 1 and 2] are most vulnerable to outbreaks of disease attributable to Salmonella infection. Host susceptibility and environmental persistence of Salmonella are also factors contributing to outbreaks.Estimates of the prevalence of Salmonella-shedding horses admitted to veterinary hospitals have generally been made under outbreak conditions.1] For example, between 1971 and 1982, 245 hospitalized horses (1.7%) at the University of California were found to shed Salmonella.3 and 4] Three major outbreaks occurred during the study period, with no apparent periodicity. Between 1996 and 1999, 35 hospitalized horses (5.5%) at the Michigan State University were found to shed Salmonella.5] One major outbreak occurred during the study period. Only one national survey of nonhospitalized horses in the United States for Salmonella infection has been undertaken: the prevalence of fecal shedding of Salmonella was estimated to be 0.8%, and the farm prevalence of shedding was 1.8%.6]Many factors have been associated with the risk of Salmonella isolation from hospitalized horses, including diarrhea, fever, change in diet, large colon impaction, colic, withholding feed, feeding bran mash, antibiotic treatment, intubation with nasogastric tubes, and average daily ambient temperature.7, 8, 9 and 10] Many of theses factors are thought to operate primarily through the effect of stress, increasing the susceptibility of horses to infection. Also, if a horse is infected by Salmonella but not shedding the organisms in its feces, the presence of stressors may reactivate fecal shedding. Most studies that have been conducted on risk factors for Salmonella shedding in horses have included horses with clinical salmonellosis, with or without inclusion of horses inapparently infected by Salmonella.7, 8 and 10] Risk factors for Salmonella fecal shedding versus clinical salmonellosis have not been clearly delineated.An outbreak (epidemic) of disease can be defined as “an occurrence of disease in excess of its anticipated frequency.”11] To more effectively identify future outbreaks of Salmonella infection in hospitalized horses, it is necessary to have accurate estimates of the prevalence of endemic fecal shedding of Salmonella in horses admitted to VTHs, and the incidence of fecal shedding during hospitalization. The aims of this study were to estimate the prevalence of fecal shedding in horses admitted to a VTH, to estimate the incidence of fecal shedding during hospitalization, and to describe the seasonal distribution of fecal Salmonella-shedding prevalence and incidence.

Materials and Methods

Study Design

Fecal samples were collected from horses admitted to the Purdue University VTH between October 12, 2000 and June 30, 2001. Horses admitted as inpatients were sampled at least on the day of admission, the day after admission, the day of discharge, and once or more in between. All horses admitted to the VTH during the study period were eligible to be sampled. In the case of mares accompanying sick foals, samples were also collected from the mare. Fecal samples were collected generally from freshly voided fecal material in stalls. Samples were stored at 4°C for up to 24 hours before processing.

Data Collection

For all horses included in the study, date of examination (outpatients) or date of hospitalization (inpatients) was recorded. For inpatients, date of discharge or date of death was also recorded. Horse characteristics were recorded as part of each horse's medical record, and included date of birth, sex (mare, stallion, gelding), and specific breed. The outcome of each admission (discharged alive, died, euthanized) and whether a necropsy was performed were also recorded. The number of samples collected per horse was recorded in a laboratory-reporting system, but specific date of collection of each sample (except for the first and last samples collected) was not routinely recorded.

Bacteriologic Cultures

All fecal samples were cultured for Salmonella species using standard techniques. Specimens were streaked onto brilliant green (BG) and xylose-lysine-tergitol (XLT-4) plates, and approximately 10 g of fecal material was put into 100 mL of tetrathionate Hajna broth. BG plates were incubated at 35° to 37°C for 18 to 24 hours and XLT-4 plates were incubated for 24 to 48 hours. Tetrathionate broth was incubated at 35° to 37°C for 24 to 48 hours, and then streaked to BG and XLT-4 plates. These plates were incubated as previously described. Suspect colonies on plates were subcultured and further identified by the Vitek GNI system. All Salmonella isolates were speciated and serotyped (National Veterinary Services Laboratory, Ames, Ia).

Data Analysis

The total number of horses examined (admissions), the total number of examinations (including admissions), the total number of horses hospitalized during the study period, and the number of samples collected per horse were calculated (Excel 2000, Microsoft Corp, Redmond, Wash) from recorded information. The frequency distributions of admissions (1-6) per horse, sex, breed, and patient outcome were calculated based on owner/horse identity and hospital record number, and the frequency distribution of number of samples collected per horse (nil to 8) was calculated from laboratory records and owner/horse identification and laboratory submission number. Length of hospitalization (days) was calculated from recorded date of hospitalization and date of discharge information, and was summarized by median and mean lengths of hospitalization and 95% confidence intervals (CIs), based on the Wilcoxon test (Minitab for Windows, Minitab Inc, State College, Penn) and the normal distribution (Statistix for Windows, Analytical Software, Tallahassee, Fla), respectively. Normality of the distributions of lengths of hospitalization and age were tested using the normal probability plot and Wilks-Shapiro statistic (Statistix).To estimate the incidence of Salmonella shedding, only hospitalized horses that were sampled on at least three occasions were included, because the sensitivity of culture is suboptimal and repeated attempts to culture Salmonella are necessary to increase the sensitivity of this technique.12] The total number of days at-risk of shedding Salmonella was calculated as the sum of lengths of hospitalization. The incidence density rate (true incidence) of Salmonella shedding was calculated as

(No. horses detected shedding Salmonella)/(total No. horse-days at-risk)

and expressed as the number of horses shedding Salmonella per 1000 horse-days at-risk. For example, if 10, 10, and 30 horses were hospitalized for periods of 3, 6, and 9 days each, horse-days at-risk would be (10.3)+(10.6)+(30.9) = 360. If one case of Salmonella-shedding was detected in this group, the estimated incidence density rate would be one per (1000/360) 1000 horse-days AT-RISK = 2.8 per 1000 horse-days at-risk. Horses that were culture-positive on the first sampling were not considered incident cases of Salmonella-shedding, and therefore were not included in the numerator of the incidence density rate. Horses shedding Salmonella after hospitalization were assumed to have commenced shedding at the midpoint of their hospitalization, and therefore contributed half their length of hospitalization to the rate denominator. A 95% CI for the estimated incidence density rate was calculated assuming shedding Salmonella to be Poisson distributed.13] Incidence-density rates were also calculated for each month of the study, the contribution of each horse to each month's horse-days at-risk being calculated on the basis of date of hospitalization and date of discharge information. Cumulative incidence (risk) rate (%) was calculated as described for the incidence density rate, except that the denominator used was the number of horses at-risk of Salmonella-shedding when hospitalized. For cumulative incidence rate, a 95% CI was calculated based on the binomial distribution.14]The prevalence of Salmonella-shedding in horses admitted to the VTH was calculated as

(No. horses detected shedding Salmonella by culturewhen admitted)/(total No. horses admitted and sampled)

Ninety-five percent CIs for prevalence estimates were calculations based on the binomial distribution.14]The temporal clustering of horses shedding Salmonella was investigated using the scan statistic.15] For horses shedding Salmonella after hospitalization, the midpoint of their length of hospitalization was used as the date of first occurrence of Salmonella-shedding. The population at-risk used in these analyses was the monthly total number of horse-days at-risk. The occurrence of horses shedding Salmonella was assumed to be Poisson distributed, so the expected number of horses shedding Salmonella in any given time period was proportional to the incidence of Salmonella-shedding during the entire study period. The study period was scanned for clusters of horses shedding Salmonella using a scanning window of as much as 50% (130 days) of the time period (SatScan, Bethesda, Md).

Results

Between October 12, 2000 and June 30, 2001, 724 horses were admitted to the VTH. Six hundred and thirty-two horses (87.3%) were admitted only once during the study period; 69 (9.5%), 15 (2.1%), 2 (0.3%), 5 (0.7%), and 1 (0.1%) horses were admitted on 2, 3, 4, 5, and 6 separate occasions, respectively. Length of hospitalization was not recorded for 2 horses. Three-hundred and sixty (42.3%) of the 854 admissions performed did not result in hospitalization of the horse. The distribution of length of hospitalization of all horses hospitalized during the study is shown in Figure 1. The median and mean lengths of hospitalization (95% CI) were 3.0 (2.5, 3.5) and 4.3 (3.8, 4.7) days, respectively. The minimum and maximum lengths of hospitalization were 1 and 57 days. The distribution of lengths of stay was nonnormally distributed (Wilks-Shapiro statistic, 0.7317).