This study presents a prediction-based approach to determine thresholds for a medication alert in a computerized physician order entry. Traditional static thresholds can sometimes lead to physician's alert fatigue or overlook potentially excessive medication even if the doses are belowthe configured threshold. To address this problem, we applied a random forest algorithm to develop a prediction model for medication doses, and applied a boxplot to determine the thresholds based on the prediction results. An evaluation of the eight drugs most frequently causing alerts in our hospital showed that the performances of the prediction were high, except for two drugs. It was also found that using the thresholds based on the predictions would reduce the alerts to a half of those when using the static thresholds. Notably, some cases were detected only by the prediction thresholds. The significance of the thresholds should be discussed in terms of the trade-offs between gains and losses; however, our approach, which relies on physicians' collective experiences, has practical advantages.
IOS Press, Inc.
6751 Tepper Drive
Clifton, VA 20124
Tel.: +1 703 830 6300
Fax: +1 703 830 2300 firstname.lastname@example.org
(Corporate matters and books only) IOS Press c/o Accucoms US, Inc.
For North America Sales and Customer Service
West Point Commons
Lansdale PA 19446
Tel.: +1 866 855 8967
Fax: +1 215 660 5042 email@example.com