Background and Importance Medication discrepancies occur in a majority of hospital admissions to the emergency department (ED) and are a major source of avoidable harm. Obtaining an accurate medication history is essential to identify drug-related problems early on. However, the medication reconciliation process is prone to many errors and is furthermore labour-intensive. As a result, many patients simply do not receive a complete medication reconciliation, mostly owing to limited recourses. We hence need an approach which enables us to identify those patients at the ED who are at increased risk for clinically relevant discrepancies.

Aim and Objectives To develop and prospectively validate a prediction model to identify patients who are at risk for at least one clinically relevant medication discrepancy at the time of ED presentation.

Material and Methods A prospective study was carried out at the ED. Medication histories were obtained and clinically relevant discrepancies were identified, using an internally validated scoring tool. Two distinct datasets were created. A first dataset (n=824) was used to build and internally validate a prediction model. We used multivariable logistic regression with backward stepwise selection to select the final model. A second dataset (n= 350) was used to prospectively validate the prediction model. The predictive performance of this model was assessed by measuring calibration, discrimination and classification.

Results The final model contained nine predictors that can easily be obtained upon ED admission, including age, origin before admission (home/nursing home) and number of drugs. Prospective validation showed excellent calibration with a slope of 1.09 and an intercept of 0.18. Discrimination was moderate with a c-index of 0.66. Using a probability threshold of 0.4, the sensitivity, specificity, positive predictive value, negative predictive value and alert rate was 41%, 81%, 56%, 70% and 27%, respectively.

Conclusion and Relevance The presence of at least one clinically relevant medication discrepancy can be predicted by our model with moderate performance. Using our prediction model is more efficient than performing medication reconciliation at random and can guide the rational use of limited resources at the ED. Depending on the available resources, different probability thresholds can be applied to increase either the specificity or the sensitivity of the prediction model.

Conflict of Interest No conflict of interest