Complexities and approaches for deriving longitudinal daily morphine milligram equivalents using electronic health record prescription data

Abstract

Objective To describe challenges and solutions for calculating longitudinal daily opioid dose in morphine milligram equivalents from electronic health record prescriptions for a clinical trial of voluntary opioid reduction in patients with chronic non-cancer pain.Materials and Methods Researchers obtained opioid prescriptions for 525 participants from the National Patient-Centered Clinical Research Network datamart at three health systems. Daily opioid dose was calculated using dose conversions and summing across prescriptions after applying assumptions, reviewing suspect prescribing patterns, and removing spurious prescriptions.Results Out of 16 071 extracted prescriptions, 1207 (8%) were unusable, and 14 864 (92%) were analyzed.Discussion Numerous challenges were identified related to incomplete data, inaccurate refill dates, and overlapping or duplicate prescriptions.Conclusion Using electronic prescription data to calculate daily doses of opioid consumption is challenging and requires significant cleaning prior to use in research. This paper recommends steps to review and clean electronic opioid prescription data.This manuscript explores challenges encountered with opioid prescription data from electronic health records (EHRs) and offers solutions to derive opioid doses into morphine milligram equivalents (MME) across multiple prescriptions for extended periods of time for research purposes. The issues described here were identified in a randomized controlled trial evaluating two interventions on voluntary opioid reduction among chronic non-cancer pain patients, followed for up to 18 months. Prescription-level issues were primarily composed of missing data elements required for the derivation of daily MME. More complex challenges were identified when evaluating temporal patterns across multiple prescriptions over the study period, such as monthly prescriptions for long-term medications being prescribed too early and apparent duplicate prescriptions. The novel systematic data cleaning steps and assumptions described in this paper, which are not yet sufficiently described in existing literature, were developed in collaboration with a subject matter expert and prescribing physicians. Implementing these data cleaning steps is important to ensuring that EHR opioid prescription data are an accurate reflection of patients' opioid use over time, which in turn is critical for research of long-term opioid therapy.