Confusions between drug names that look and sound alike (e.g., Keppra(r) and Kaletra(r), Indocid(r) and Endocet(r)) continue to occur frequently, and each confusion poses a threat to patient safety.2-5 Our long term objective is to design, build, test and continuously improve tools that minimize the harm caused by drug name confusion errors. For a patient to be harmed, an error must occur and it must go undetected until it reaches the patient. Harm is minimized either by preventing the error from occurring in the first place or by rapidly detecting the error so its adverse effects can be mitigated. Both prevention and mitigation efforts have been hindered by the lack of valid, reliable and efficient methods for assessing name confusion error rates. The gold standard for measuring medication error rates is direct observation of the prescribing-dispensing- administering process. This method is valid and reliable but is too time consuming and expensive to be widely used. As a result, many error reduction interventions have been designed, but few have been tested, and their effectiveness is, for the most part, unknown. Similarly, efforts to mitigate the effects of wrong drug errors are virtually non-existent because there has been no accurate and efficient way to detect such errors after they occur. The key to improving both prevention and mitigation of harm is the development of scalable, efficient, valid and reliable methods for detecting these drug name confusion errors. Our short-term goal is to develop and validate an algorithm for detecting drug name confusion errors by analyzing suspicious patterns in real-world prescription drug databases (in our case, integrated electronic medical records from the US Veterans Health Administration). We plan to test the following three hypotheses: 1. Computerized measures of drug name confusability can be used to identify wrong-drug errors in real-world prescription drug databases. 2. The number of errors detected will increase as the predicted probability of confusion increases. 3. The classification performance of the error detection algorithm (i.e., its accuracy, sensitivity and specificity) can be enhanced by applying machine learning techniques and by incorporating additional information from the electronic medical record (e.g., time between refills, diagnosis, lab values, demographics, etc.) To test these hypotheses, we propose studies with the following specific aims: 1. To design and implement an algorithm for the detection of suspicious patterns in prescription drug databases. 2. To test and validate this algorithm using real-world prescription data from the US Veterans Health Administration. 3. To use machine learning techniques to optimize and further validate the performance of the error detection algorithm, incorporating additional information from the electronic medical record. Health care professionals often confuse drug names that look and sound alike. Wrong drug errors occur in hospitals and in community pharmacies and can cause serious harm to patients. Our project seeks to improve patient safety by developing and testing new techniques for detecting wrong drug errors in integrated electronic medical records. ? ? ? ?
