| Abstract
|
:
|
A meta-analysis was conducted on data from three program arms of the U.S. National Antibiotic Resistance Monitoring System (NARMS). Data representing over 37,000 non-Typhi Salmonella isolates were extracted from 15 final reports published by USDA, CDC, and FDA from program years 1996-2003. Three antibiotics representing older drug classes; tetracycline, ampicillin, and sulfamethoxazole; were selected for analyses of resistance levels among chicken, turkey, swine, and cattle isolates in comparison to human diagnostic sets. Resistance analyses applied ANOVA, post hoc comparison of means, effect sizes, and trend graphics among animal subcategory and human diagnostic sets. Trend Chi-square analyses were conducted on all sets. Associations among the top 15 Salmonella serotype sets were analyzed using Spearman rank order and a two-component predictive model. The results of the study showed significant differences in mean resistance between human and several animal subcategories. Mean effect sizes of human versus diagnostic and inspected slaughter animal sets were d = -0.83 and -0.32, respectively. No statistically significant resistance trends were detected among inspected slaughter or human isolates for the three antimicrobials. Increasing trends were found among some diagnostic (ill and/or treated) animal subcategories, in conjunction with slightly declining resistance in human sets. Serotype analyses indicated strong and statistically significant yearly correlation within each respective program, but weaker associations between programs. Prediction Indices (PI%) for serotype data were: CDC, 92.9 (95% C.I. 79.3,100); USDA, 71.1 (55.8, 86.6); and USDA-CDC, 29.6 (9.9, 49.3). The 2002 USDA-FDA serotype sets had PI% of 17.2; USDA-CDC, 14.1; and FDA-CDC, 3.4. In conclusion, resistance to three older antibiotics in classes approved for production animals, showed level or declining trends among sets representing inspected animal or diagnostic human sources. Serotype analyses indicated generally low association levels between programs. Meta-analysis and statistical methods, along with predictive models may be useful tools for integrating resistance, trend, and serotype data in risk evaluations.
|