Skip to main content


Figure 3 | BMC Health Services Research

Figure 3

From: Chronology of prescribing error during the hospital stay and prediction of pharmacist's alerts overriding: a prospective analysis

Figure 3

CART tree for predicting alert's overriding among new alerts. Numbers in ellipses and rectangles report the number of observed alerts. Classifying variables are indicated in the ellipses as questions and splitting rules are printed at the branches. In the terminal nodes (rectangles), the number and the proportions of correctly predicted outcomes by the tree (respectively incorrectly predicted) are presented as "correct decision" (respectively "incorrect decision"). The ward was the first discriminating variable, i.e. most influential reasons for alert's overriding. For the vascular medicine and geriatrics wards (23 alerts), alert's overriding was dependent from the first level of the 'Anatomical Therapeutic Chemical classification '. Among the alerts due to 'Alimentary tract and metabolism, Systemic hormonal preparations, excluding sex hormones and insulins, Musculo-skeletal system, Nervous system, Respiratory system, Sensory organs and Various' errors (n = 14), it correctly classified 100% of the alerts with an overriding on the next day. Among the errors belonging to the categories 'Blood and blood forming organs, Cardiovascular system, Anti-infectives for systemic use' (n = 9), it correctly classified 5 alerts (56%) with a non-overriding on the next day and missed 4 alerts' overriding. For the other wards (internal medicine 1, clinical immunology, internal medicine 2, diabetes care, and nephrology), the next differentiating factor was type of errors. For new alerts with 'Inappropriate choice of drug and/or drug dose, Wrong unit, Wrong route, Drug omission or Duplicate order' errors, the classification and regression tree predicted a non-overriding of the alert whereas for 'Drug-drug interaction' error, it anticipated an overriding on the next day.

Back to article page