PS2-06: Feasibility of Determining Medication Adherence from Electronic Health Record Medication Orders

Abstract

Background and Aims: Medication adherence is essential to optimizing outcomes for many chronic diseases. Typically, calculation of adherence is based on dispensings identified from pharmacy claims databases. However, this procedure excludes prescription orders that are never dispensed and potentially results in biased adherence estimates. To explore extracting newly-ordered medications from the Electronic Health Record (EHR), we conducted a project to:

1. Determine the process needed to identify ordered medications from the EHR,

2. Apply the process to identify a study cohort, and

3. Validate the process with medical record review.

Methods: The study team prepared an inclusive list of study drugs, identified EHR tables and fields that contained medication order data, explored methods that facilitated inclusion of only the final medication order (when applicable), and explored optimal sequencing of inclusion criteria. These four steps were then used to identify a patient cohort newly-prescribed an antihyperlipidemic agent, and record review was used iteratively to validate each process step.

Results: Multiple challenges were encountered and solutions identified. First, drug orders in the EHR were based on an unfamiliar drug categorization scheme. Second, Clarity tables, fields and linkages required exploration to discern useful data elements. For example, the “stop reason” field was not useful in identifying cancelled orders and the “order date” was occasionally overridden by the most recent refill of that order. Other factors complicating identifying new orders included human error, reversed orders, sending orders to incorrect dispensing locations, and medication reconciliation after nursing home discharge. We also found that by changing the order in which we applied the selection criteria resulted in a “cleaner” data set. For example, applying diagnosis criteria earlier in the process allowed us to remove prevalent users and their refill history.

Conclusion: Identifying medications ordered in the EHR, particularly medications ordered but not dispensed, was a process fraught with multiple challenges. We believe, however, that we have developed an iterative process that yields a reliable medication order algorithm. This is an important first step in achieving less biased estimates of medication adherence in chronic disease populations.