Purpose: Errors in the entry of body weight can affect accuracy of medications prescribed in the electronic health record system. Simple decision support (like the identification of extreme weights for age) are useful but have low predictive values. We deployed a web-services method to extract patients' body weight patterns to boost the rate of detection and mitigation of weights used in medication ordering at our children's hospital. Conventional decision support (weight < 3rd or > 97th percentile and for weights far from the expected WHO/CDC weight) remained in place throughout the study. Methods: Using an existing web-services framework, we arranged to have admitted patients' weights sent to a rules engine that was capable of fitting these data points to predictive curves using the R statistical package. For each patient, the system would calculate a probability that the current weight was accurate, and set a vulnerability flag when it probability was low. it then “listened” for a medication order, and, upon detection of a new medication order, sent a message to the hospital pharmacist message pool for use by the pharmacist(s) involved in medication order validation. A pharmacist (SS) compiled the results of these messages and recorded whether the algorithm correctly identified an error and what action the pharmacist took to mitigate any errors. Results: Examination of simple decision-support alerts (to detect a 50% variation between the past 2 weights) aimed at the nursing MAR workflow over a trial period of 17 days resulted in a positive predictive value of 2.6%. Desiring a less noisy method, we trialed and implemented the statistical web-services method to reduce false positives. Over 130 days of study, 123 messages were sent on 65 patients. 47 (65%) of the patient identified had a true weight error as judged by a clinician (SAS) who reviewed each chart along with the growth chart. The positive predictive value was lowest in the < 1 year old group (53%) and highest for 6-12 year olds (100%). In 32 (68%) of the 47 true-positive patients, the pharmacist recorded an intervention in the EHR, suggesting an alert salience rate of at least that percentage. The only mis-dosed medications were for those medications that were given immediately upon weighing the child (i.e., antipyretics in the ED). Other, more conventional alerts fired in each of those cases. Conclusion: A more sophisticated approach to determining the appropriateness of a weight based on statistical patterns of growth performed better than a simple alert based on absolute weight variation in the context of a system where conventional weight decision support was also active. More work is required to increase the positive predictive value, especially for infants, whose robust weight gain can simulate weight entry errors.
Weight Entry Error Detection: A Web Service for Real-time Statistical Analysis
- Views Icon Views
- Share Icon Share
- Search Site
Stephen A. Spooner, Sarah Shields, Judith W. Dexheimer, Cecilia M. Mahdi, Philip Hagedorn, Thomas Minich; Weight Entry Error Detection: A Web Service for Real-time Statistical Analysis. Pediatrics January 2018; 141 (1_MeetingAbstract): 21. 10.1542/peds.141.1MA1.21
Download citation file: