Logo image
Back
2894. Metrics of Antimicrobial Use Within Inpatient Settings: Impacts of Statistical Methods and Case-Mix Adjustments
Abstract   Open access

2894. Metrics of Antimicrobial Use Within Inpatient Settings: Impacts of Statistical Methods and Case-Mix Adjustments

Michihiko Goto, Rajeshwari Nair, Bruce Alexander, Brice Beck, Christopher Richards, Eli N Perencevich and Daniel J Livorsi
Open forum infectious diseases, Vol.6(Supplement_2), pp.S81-S81
10/23/2019
DOI: 10.1093/ofid/ofz359.172
PMCID: PMC6809196
url
https://doi.org/10.1093/ofid/ofz359.172View
Published (Version of record) Open Access

Abstract

Abstract Background The necessary data elements and optimal statistical methods for benchmarking hospital-level antimicrobial use are still being debated. We aimed to describe the relative influence of case-mix adjustment and different statistical methods when ranking hospitals on antimicrobial use (AU) within inpatient settings. Methods Using administrative data from the Veterans Health Administration (VHA) system in October 2016, we calculated total antimicrobial days of therapy (DOT) and days present according to the National Healthcare Safety Network (NHSN) protocol. Patient-level demographics, comorbidities, and recent procedures were used for case-mix adjustments. We compared hospital rankings across 4 different methods: (A) crude antimicrobial DOT per 1,000 days present, aggregated at the hospital-level; (B) observed/expected (O/E) AU ratio with risk adjustment for ward-level variables (analogous to NHSN’s Standardized Antimicrobial Administration Ratio); (C) O/E AU ratio with risk adjustment for ward-/patient-level variables; (D) predicted/expected (P/E) AU ratio with risk adjustment for ward-/patient-level variables, based on a multilevel model accounting for clustering effects at hospital- and ward-levels. Results The cohort included 165,949 DOTs and 318,321 days present at 122 acute care hospitals within VHA. Crude DOTs per 1,000 days present ranged from 153.6 to 900.8 (Figure A), and ward-level risk adjustment only modestly changed rankings (Figure B). When adjusted for ward- and patient-level variables (including demographics, 14 comorbidities and 22 procedures), rankings changed substantially (Figure C). Risk-adjustment by a multilevel model changed rankings even further, while shrinking variabilities (Figure D). Ten hospitals in the lowest and 11 hospitals in the highest quartiles by O/E risk adjustment for only ward-level variables were classified to different quartiles on P/E risk adjustment. Conclusion We observed that the selection of variables and statistical methods for case-mix adjustment had a substantial impact on hospital rankings for antimicrobial use within inpatient settings. Careful consideration of methodologies is warranted when providing benchmarking metrics for hospitals. Disclosures All Authors: No reported Disclosures.

Details

Metrics

32 Record Views
Logo image