Skip to Content
SBMI logo

Comprehensive analysis of drug alert data with high override and backout rate

Author: Heidi Jones, BS

Primary Advisor: Jingchun Sun, PhD

Committee Members: Hua Xu, PhD, Dean F. Sittig. PhD

Masters thesis, The University of Texas Health Science Center School of Health Information Sciences at Houston.

Abstract:

Background: Medication prescription errors are both numerous and costly, but can be alleviated with the help of computerized physician order entry (CPOE ) and clinical decision support (CDS) technology. To provide the best improvement in quality, CDS interventions should be minimized to only generate alerts that are accurate and relevant. This is a difficult task because of poor physician agreement on what content is valuable. A systematic analysis of alert responses among different healthcare settings may provide a quantifiable perspective to aid in CDS content modification.

Materials and Methods: An SPSS crosstabs analysis was used to compare alert rates among facilities. Anatomical Therapeutic Chemical (ATC) classification codes were assigned to all drugs using information from the Kyoto Encyclopedia of Genes and Genomes (KEGG) and DrugBank databases. ATC similarity was assessed with a probabilistic model.

Results: Some variability in override rates exist among facilities, but all have high override rates at least greater than 75% with an overall override rate of 95%. When particular high frequency pairs are compared, only small variability occurs among facilities. Most alert generating pairs occur infrequently and do not provide enough data to draw conclusions., but 7% of override drug pairs are overridden >100 times. ATC similarity scores indicate that our CDS rules signal drug pairs with higher similarity scores than DrugBank interaction pairs or randomly compiled pairs.

Discussion:Override rates are high at all facilities, but have small amounts of variability among facilities. The most interesting alerts that may warrant further evaluation for possible exclusion are those that are frequently overridden. The ATC similarity score may be a useful tool to help make quantifiable CDS content decisions.

Conclusions: CDS content should be reviewed and improved regularly. However, content modification is difficult due to conflicting opinions. A systematic analysis of drug pair overrides and backouts along with ATC similarity evaluation help make quantifiable choices in CDS content improvement.