Machine Learning and NLP Improve Medication Reconciliation, Patient Safety

By Joe Farr, RN, and David Sellars

Three-quarters of hospital executives reported worrying that their patients’ medication history data is incomplete and/or inaccurate, according to a 2017 survey by the College of Healthcare Information Management Executives. The news was disappointing, coming after years of investment in improving the medication reconciliation process. Many hospital leaders likely wondered what more could be done to ensure accurate medication data was available to boost patient safety and care outcomes.

That was also a concern for leaders at King’s Daughters Medical Center (KDMC), a 99-bed community hospital in Brookhaven, Mississippi, 60 miles south of the state capitol in Jackson. At the time, virtually all patient records at KDMC had incomplete medication histories, requiring manual—and therefore error-prone—data entry in the EHR by nurses at the point of care. In late 2018, KDMC sought to improve patient safety and streamline the medication reconciliation process by automating EHR transcription of critical medication data.

The solution KDMC came up with has contributed to increased patient safety and better health outcomes, as well as higher nurse productivity.

The root of the problem: Free-text sigs

Medication history discrepancies typically occur during patient triage/intake, when data from an outside EHR is transferred into the hospital’s resident EHR. One of the most difficult problems in the medication reconciliation process is that the distinct nomenclature of different EHR systems often results in missing or indiscernible “sigs”—the important shorthand prescribing instructions that mean the difference between a patient receiving 1.0 mg and 10 mg of a medication, or “qhs” being interpreted as “every hour” instead of “nightly at bedtime.” An estimated 66% of data from the nation’s largest medication history database is missing essential sig information. That leads to staff and clinicians spending hours on the computer or phone conferring with outside providers and pharmacies to gather the missing sig data, in the interest of preventing adverse drug events.

Further complicating the issue, current prescription routing technologies provide free-text sig information for dosing instructions rather than the discrete text fields that are more easily translatable to an EHR. This poses a challenge for pharmacists, who must manually sift through the sig data in order to create prescriptions, as well as for clinical staff responsible for medication reconciliation during triage/intake. At KDMC, nurses had to manually transcribe sigs from the medication history into the EHR’s current visit list.

The solution: Automated sigs

KDMC’s leadership decided that the best way to improve the safety of medication reconciliation was to automate the transcription of sig data into the EHR. Doing that meant finding a way to convert free-text sig data into programmable data yielding discrete sig components within a patient’s medication history. The hospital hoped this would reduce medication reconciliation clicks and keystrokes; ensure a more accurate patient medication history; and help fill important medication history gaps.

For assistance, KDMC turned to DrFirst, its e-prescribing partner, for implementation of an AI-powered solution that uses natural language processing and machine learning to process and validate results, thereby codifying sigs into each facility’s standard terminology (e.g., “by mouth” versus “oral” or “PO”). The automated process operates entirely in the background, without clinician intervention, and employs statistical validation and clinical analysis to produce real-time sig translation. In addition, the solution helps staff resolve gaps by supplying alternative drug IDs for best-case drug matching and details for incomplete or uncommon sigs. Multiple safety checks enable disqualification of transactions that are deemed clinically invalid. The solution is programmed to prefer no data to wrong data, so if it determines that a sig data point poses a safety risk, it will withhold the data rather than transcribe it.

Challenges along the way

The automated transcription solution required less than two hours to implement, and it integrated seamlessly with the health system’s inpatient EHR. KDMC’s EHR vendor needed to update its code to accommodate the solution, a task it accomplished quickly.

The hospital was concerned that nurses responsible for medication reconciliation might revert to manual sig translation because the new process was unfamiliar. To prepare end users for the new tool, KDMC created a two-minute educational video to ensure fast adoption free of technological barriers.

Outcomes

Implementation of the automated transcription solution at KDMC reduced the number of incomplete or error-filled patient medication records, which in turn minimized pharmacy call-backs, reduced workflow disruptions, and eased patient treatment delays. Implementation also significantly reduced the average number of computer “clicks” required for medication reconciliation, resulting in additional time and dollar savings.

After implementing the tool, medication reconciliation required a total of 45,000 fewer “clicks” per month, based on an analysis of data from May and June 2019 compared with pre-implementation data. The resulting time savings of 34 hours per month for clinicians (408 hours per year) translates into about $11,000 in recaptured nursing productivity over a 12-month period, based on 19,390 annual patient visits and an average of five medications per patient.

More significantly, the solution appears to have contributed to increased patient safety and improved health outcomes. In the seven months following its implementation, KDMC’s overall 30-day readmission rate fell by 11.3%, from 6.2% pre-implementation to 5.5% post-implementation. The hospital is currently reviewing the data to determine the exact impact of the solution on 30-day readmission rates. The hypothesis is that improved accuracy of medication dosage accounts for a significant portion of the decrease in readmissions, possibly due to a decline in post-discharge adverse drug reactions.

Anecdotally, KDMC’s nurses have expressed satisfaction with the tool, which has streamlined patient triage and saves nursing time. In light of these results, healthcare organizations are advised to find ways to automate the transcription of sig data with tools that enhance patient safety and outcomes while saving time.

Joe Farr is clinical applications coordinator at KDMC. David Sellars is principal of product innovations at DrFirst, a leading provider of e-prescribing, price transparency, and medication management solutions.