Accuracy of Administrative Data in Comparison to Administered Antimicrobial Doses for Hospitalized Children

Background: The Centers for Disease Control and Prevention have adopted days of antimicrobial therapy as a metric for healthcare facilities to benchmark antimicrobial use.  Hospital billing data are often used as a proxy for medication administration, which is a convenient method to study antimicrobial prescribing practices using large administrative datasets. We sought to evaluate the accuracy of institutional billing data compared to medication administration data.

Methods: This study compared medication billing data from the Pediatric Health Information System (PHIS) with aggregate electronic medical record (eMAR) data from 5 freestanding children’s hospitals between January 1st, 2010 through June 30th, 2013. A de-identified aggregate dataset of eMAR days of therapy for targeted systemic antimicrobials was generated from each institution.  Billing data for these same patients were extracted from the PHIS database.  Each site also provided information on its current pharmacy billing practice (bill upon dispense versus bill upon administration) and if the practice had changed during the study period.  Each monthly institutional data point was described as either administration or dispense-based billing.

Results: A total of 897,187 days of therapy were included in the datasets from the five participating institutions.  Only 2 of 5 institutions utilized dispense-based pharmacy billing during the study period; both converted to administration-based billing prior to July 1, 2012.  The derived linear regression models (Figure 1) indicated that PHIS data may better predict eMAR charge data when compared to an ideal reference line, where PHIS data perfectly match eMAR data. The linear model for administration data also had a better fit than that for dispense (r²=0.98 and r²=0.46).

Conclusion: In these hospitals, antimicrobial charge data derived from bill upon administration accurately reflects the source eMAR data.  Some caution may be needed when using data from hospitals that bill upon dispense.  As these are aggregate data, they may not reflect variation among different medications and routes of administration.

Figure 1: Correlation of Administrative versus Medication Administration Record Data