Health Services Research & Development

Veterans Crisis Line Badge
Go to the ORD website
Go to the QUERI website

2007 HSR&D National Meeting Abstract

Printable View

National Meeting 2007

3061 — Developing Indicators of Adverse Drug Events via Non-Linear Analysis of Administrative Data

Nebeker JR (TREP-Salt Lake City) , Yarnold PR (VA Chicago), Soltysik RC (VA Chicago), Sauer BC (TREP-Salt Lake City), Xu W (TREP-Salt Lake City)

Because of uniform availability, hospital administrative data are appealing for surveillance of adverse drug events (ADEs). Expert-generated surveillance rules that rely on presence of International Classification of Diseases, 9th Revision Clinical Modification (ICD-9-CM) codes have limited accuracy. Rules based on non-linear associations among all types of available administrative data may be more accurate. By applying hierarchically optimal classification tree analysis (HOCTA) to administrative data, we derived and validated surveillance rules for 3 pre-specified ADE groups: narcotic-related reactions, bleeding/anticoagulation problems, and delirium/psychosis.

Retrospective cohort design. Subjects: A random sample of 3,987 admissions drawn from all 41 Utah acute-care hospitals in 2001 and 2003. Measures: Professional nurse reviewers identified ADEs using implicit chart review. Pharmacists assigned Medical Dictionary for Regulatory Activities codes to ADE descriptions for identification of clinical groups of events. Hospitals provided patient demographic, admission, and diagnostic data to the Utah Department of Health.

The base rate of drug-induced bleeding/anticoagulation problems was 0.8% and that of delirium/psychosis was 1.0%. For these ADEs, we derived models with good discrimination: area under the receiver operator characteristic = 0.83 and 0.80, respectively. An accurate model could not be derived for narcotic-related reactions. Poisoning and adverse event codes designed for the 3 types of reactions had low sensitivities and, when forced in, did not improve model accuracy. Ancillary, non-ICD-9 data such as age and length of stay were the most important variables for the models.

HOCTA is a promising method for rapidly developing surveillance rules for administrative data. Data other than ICD-9 codes are useful predictor variables.

Risk-adjusted surveillance rules developed with HOCTA may provide tools to estimate rates of ADEs and track performance in selected patient-safety areas. All available administrative data should be considered when constructing patient safety indicators.