To assess the impact of problem lists' accuracy in electronic health records on clinical decision-making: protocol for a randomized controlled trial

Abstract

Background Problem-oriented electronic health record (EHR) systems, with complete, coded and up-to-date (i.e. accurate) problem lists, aid healthcare providers to track patient’s health status and make better decisions, as problem lists can provide useful summaries of important health issues. Unfortunately, problem lists are still incomplete and out-of-date (i.e. inaccurate). That is, among others, because providers are under time pressure, prefer writing free-text notes and are often unwilling to update problem lists, unless receiving returns for their effort. This study aims to assess the impact of accuracy of problem lists in EHRs on clinical decision-making. Methods In a laboratory setting we will perform a crossover randomized controlled trial in which we will recruit individual Dutch healthcare providers on-site at Amsterdam University Medical Centers. Participants will be presented with two records of two patients (A and B): one with an accurate and one with an inaccurate problem list, created in a training environment of the EPIC EHR in agreement with clinical experts. Randomization determines which record will have the accurate problem list. Participants are informed that EHR-usage is investigated and do not know which record has an accurate or inaccurate problem list. Participants will provide a motivated (Yes/No) answer on whether prescribing medication X and medication Y is appropriate. Medication Y cannot be prescribed in both patient records, due to contraindicated diagnosis (A) and diagnosis-related medical history (B). Medication X serves as a control question, and is contraindicated based on an allergy, which is equally documented in the records. The primary outcome measure is the correctness of the motivation for the correct answer of medication Y. Secondary outcome measures are: correctness of X (and Y) with the right motivation(s); total time to answer X and Y where the motivation for Y (and X) is correct. Time stamps are registered between opening the question and confirming two Yes/No answers. Proportion of retrieved correct answers and time to answer(s) retrieval will be compared by Chi-square, McNemar tests and Log Rank survival analyses. Alternative analytical models will be applied if necessary. Discussion If accurate problem lists lead to faster and better decision-making resulting in better patient outcomes, one may be motivated to develop future policies for this area and healthcare providers may be persuaded to use and update problem lists, also leading to improved data quality and opportunities for reuse.