JMIR Publications

ABSTRACT

Background:

Best practice standards, in the form of clinical practice guideline (CPG) and clinical pathway (CP), aim to standardize care and improve outcomes. However, variation in clinical practice exists, and not all deviations are inappropriate. Measuring adherence to best practice standards remains a challenge due to limitations in representation methods and data fidelity.

Objective:

This scoping review aims to survey and synthesize existing literature on the computable representation of guideline recommendations and explores methods used to detect and quantify deviations from best practice.

Methods:

We followed the Arksey and O’Malley framework and PRISMA-ScR guidelines. Five databases (Ovid Medline, EMBASE, IEEE Xplore, Web of Science, and Scopus) were searched in April 2025. Studies were included if they used EMR-derived data to measure adherence to CPGs or CPs and provided detail on data representation or deviation detection. Titles, abstracts, and full texts were screened using Covidence. Data were extracted on guideline representation, clinical context, data sources, adherence metrics, and modelling techniques. A narrative synthesis was conducted to identify themes.

Results:

Twenty-four studies were included. Most originated from the US (29%) and were published as conference proceedings (56%). Cardiovascular conditions were the most common focus (54%). Data sources included HL7 messages, structured EMR data, event logs, and Fast Healthcare Interoperability Resources (FHIR) -transformed data. Best practice standards were formalized using Business Process Model and Notation (BPMN), ontologies, FHIR, or hybrid approaches. While some studies used structured modelling, others defined pathways as linear sequences of steps. The most common method for adherence measurement was rule-based alignment of patient data with guideline components. Several studies applied weighted scoring to distinguish clinically significant deviations. Process mining was used in a subset to detect sequence and timing variations. However, most models lacked contextual sensitivity and did not incorporate patient-specific factors, such as comorbidities or clinician intent.

Conclusions:

Despite promising advances, challenges persist in representing best practice standards in computer-interpretable formats and measuring adherence in a clinically meaningful way. Models often prioritize technical alignment over clinical relevance and are limited by data quality and standardization. FHIR shows potential for interoperability and real-time monitoring, but further research is needed to enable context-aware, scalable adherence evaluation. Future efforts should focus on standardized modelling, integration with clinical workflows, and real-time decision support to support adaptive and patient-centered care.