Identifying outliers of antibiotic usage in prevalence studies on nosocomial infections

Abstract

Objective: To investigate whether the correlation between patients' antibiotic treatment (yes/no) and patients' infections (yes/no) in each hospital department, described by Pearson's correlation coefficient (p) for binary data as a measure for adequate use of antibiotics, is an appropriate quality indicator.

Design: Comparison of the results of repeated prevalence studies in different hospitals with the data of a national prevalence study, comparing the hospital (p) and reference (pNIDEP[Nosokomiale Infektionen in Deutschland: Erfassung und Pravention]) correlation coefficients for "use of antibiotics/ presence of infections."

Setting: The data of 5,377 surgical patients were separated from the total data of a national prevalence study in 72 representative hospitals to create a reference correlation coefficient (rhoNIDEP) with a reference range. Nine additional prevalence studies, involving a total of 4,984 patients, were repeatedly performed in the surgical departments of 8 other hospitals during a 12-month period, whereby the correlation coefficients rho(n) for every prevalence investigation were determined.

Results: In the national prevalence study, 15.3% of the surgical patients received antibiotics on the study day. Surgical patients had a 3.8% prevalence of nosocomial infections and a 7.0% prevalence of community-acquired infections. Pearson's correlation coefficient rhoNIDEP for correlation between patients' binary data use of antibiotics and presence of infection was 0.62. To compare the correlation coefficient of each department with the appropriate reference range, the coefficients of the single departments were plotted against the number of patients; in these plots, three lines indicated the value rhoNIDEP and the upper and lower reference ranges, depending on the number of patients. Seven of eight surgical departments investigated during the repeated prevalence studies were found to be within the reference range, near the reference value, in the majority of prevalence studies; only one of the departments was identified as an outlier as regards antibiotic use.

Conclusion: The correlation between patients' antibiotic treatment (yes/no) and patients' infections (yes/no) in hospitals or departments, as described by Pearson's correlation coefficient p for binary data with a definitive reference range depending on the number of patients, is useful for quality management in identifying the overall necessity for evaluating the indications for antibiotic use in one's own hospital.