Antimicrobial de-escalation of treatment for healthcare-associated pneumonia within the Veterans Healthcare Administration

Abstract

Objectives: The objective of this study was to measure quantitatively antimicrobial de-escalation utilizing electronic medication administration data based on the spectrum of activity for antimicrobial therapy (i.e. spectrum score) to identify variables associated with de-escalation in a nationwide healthcare system.

Methods: A retrospective cohort study of patients hospitalized for healthcare-associated pneumonia was conducted in Veterans Affairs Medical Centers (n = 119). Patients hospitalized for healthcare-associated pneumonia on acute-care wards between 5 and 14 days who received antimicrobials for ≥ 3 days during calendar years 2008-11 were evaluated. The spectrum score method was applied at the patient level to measure de-escalation on day 4 of hospitalization. De-escalation was expressed in aggregate and facility-level proportions. Logistic regression was used to assess variables associated with de-escalation. ORs with 95% CIs were reported.

Results: Among 9319 patients, the de-escalation proportion was 28.3% (95% CI 27.4-29.2), which varied 6-fold across facilities [median (IQR) facility-level de-escalation proportion 29.1% (95% CI 21.7-35.6)]. Variables associated with de-escalation included initial broad-spectrum therapy (OR 1.5, 95% CI 1.4-1.5 for each 10% increase in spectrum), collection of respiratory tract cultures (OR 1.1, 95% CI 1.0-1.2) and care in higher complexity facilities (OR 1.3, 95% CI 1.1-1.6). Respiratory tract cultures were collected from 35.3% (95% CI 32.7-37.7) of patients.

Conclusions: De-escalation of antimicrobial therapy was limited and varied substantially across facilities. De-escalation was associated with respiratory tract culture collection and treatment in a high complexity-level facility.

Figure 1.

Study flow diagram for HCAP cohort.

Figure 2.

Predicted probability of de-escalation by facility for patients hospitalized with HCAP. The predicted probability of de-escalation (%) by facility is based on a mixed-effects logistic regression model for a typical patient from the cohort. Values for covariates used to fit the model included: admitted 1 January 2010 (study midpoint) to a general medical/surgical unit at a complexity level 1 VA facility, not admitted from a skilled nursing facility, median age (73 years), male, median number of days hospitalized in the last 90 days (6 days), no recent outpatient intravenous therapy, no chronic wound care, median spectrum score on day 2 of inpatient stay (44.50), blood culture(s) obtained and respiratory tract culture not obtained. Regression analyses were based on 9283 admissions at 118 VA facilities with complete data. Facilities are ordered left to right from highest to lowest predicted probability of de-escalation. The broken line represents a facility with median predicted probability of de-escalation (31.6%).