Text Parsing-Based Identification of Patients with Poor Glaucoma Medication Adherence in the Electronic Health Record.

PURPOSE

To assess the feasibility of automated text parsing screening of physician notes in the electronic health record (EHR) to identify glaucoma patients with poor medication compliance.

DESIGN

Cross-sectional study.

METHODS

An automated EHR extraction identified a cohort of patients at the University of Michigan with a diagnosis of glaucoma, ≥40 years old, taking ≥1 glaucoma medication, and having no cognitive impairment. Self-reported medication adherence was assessed with 2 validated instruments: the Chang scale and the Morisky medication adherence scale. In tandem, a text parsing tool that abstracted data from the EHR was used to search for combinations of the following words in patient visit notes: "not," "non," "n't," "no," or "poor" accompanied by "adherence," "adherent, "adhering," "compliance," "compliant," or "complying." The proportion of patients with self-reported poor adherence was compared between the EHR extraction and text parsing identification using a Fisher exact test.

RESULTS

Among 736 participants, 20.0% (n = 147) self-reported poor adherence and 6.1% (n = 45) had EHR documentation of poor adherence (P < .0001). Using text parsing as a pre-screening tool, 22 of the 45 patients (48.9%) with non-adherence identified by text parsing also self-reported poor medication adherence compared to the 20.0% by self-report overall (P < .0001).

CONCLUSIONS

Text parsing physician notes to identify patients' noncompliance to their medications identified a larger proportion of patients who then self-reported poor medication adherence than an automated EHR pull alone but was limited by the small number of patients identified. Optimizing the documentation of medication adherence would maximize the utility of this automated approach to identify medication noncompliance.