JMIR Publications

ABSTRACT

Background:

Electronic health records (EHRs) and practice management systems represent the digital foundation of outpatient care worldwide. Integrating artificial intelligence (AI) and machine learning (ML)-supported medical and administrative processes (AMP) into these platforms offers a profound opportunity to streamline clinical workflows and standardize patient care. However, existing research remains highly fragmented and focused on isolated tools. This creates a critical research gap, as there is currently no comprehensive mapping of how these diverse medical and administrative workflows collectively integrate and perform within the real-world clinical environment of ambulatory care.

Objective:

This study aims to address this research gap by comprehensively mapping the global status quo of natively EHR-integrated AMP in ambulatory care. Specifically, we explore the types of digital processes currently available and how physicians and medical assistants interact with these technologies in clinical environments.

Methods:

We conducted a systematic review of PubMed and Scopus databases (searching from inception to March 24, 2026) following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. We targeted empirical studies evaluating outpatient clinical and administrative processes driven by AI, ML, or AI-enabled decision-support algorithms natively embedded within EHR or practice management systems. To ensure literature saturation, backward citation searching of included studies was conducted (yielding no additional eligible studies). Study selection, data extraction into a standardized matrix, and methodological quality assessment using Joanna Briggs Institute (JBI) critical appraisal checklists (tailored to RCTs, quasi-experimental, cohort, cross-sectional, and qualitative studies) were conducted by one reviewer and independently verified by a second reviewer.

Results:

A total of 39 empirical studies were included. Methodological quality assessed via JBI tools was moderate to high, with common limitations being a lack of participant and personnel blinding (system-inherent) and selection bias due to reliance on voluntary early adopters. Our synthesis identified 5 primary categories of natively EHR-integrated AMP: point-of-care clinical decision support systems (CDSS) (16/39, 41%), active background screening and triage (9/39, 23%), ambient intelligence for clinical documentation (8/39, 21%), administrative processes and patient communication (4/39, 10%), and remote patient monitoring (RPM) (2/39, 5%). While ambient AI demonstrated success in reducing documentation effort and burnout, point-of-care CDSS was heavily hindered by alert fatigue (with override rates up to 96%), and active background screening was constrained by algorithmic opacity (“black-box” characteristics) and clinician distrust.

Conclusions:

Rather than achieving full automation, AMP technologies currently function as assistive tools within primary care. Real-world utilization requires a strict human-in-the-loop verification process, where medical professionals actively oversee both AI- and ML-driven outputs. Successful integration depends on addressing critical workflow barriers, particularly minimizing alert fatigue and enhancing algorithmic transparency to foster clinician trust. Clinical Trial: PROSPERO [CRD420261327094]