AI-Based Clinical Decision Support Systems in Primary Care: Systematic Review of Clinical Implementation.
Florian Naye, Steven Ouellet, Farzaneh Yousefi et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review assessed the effectiveness and safety of AI-based clinical decision support systems (AI-CDSSs) used by health professionals in primary care, compared with usual care or non-AI decision support. It included 14 records covering 11 experimental or quasi-experimental studies. Benefits were more consistent for workflow and process outcomes in structured tasks, including screening, referral, and alert optimization; patient-level effects were limited and inconsistent. Safety evidence was insufficient: only 2 studies prespecified safety outcomes. All experimental studies had high risk of bias, while quasi-experimental studies had serious to critical risk of bias.
Key findings
- The review included 14 records on 11 studies: 4 experimental and 7 quasi-experimental.
- Workflow-related benefits were more consistent in structured, protocolized tasks such as screening, referral, and alert optimization.
- Patient-level effects were limited and inconsistent, particularly for heterogeneous conditions or preference-sensitive decisions.
- Only 2 studies prespecified safety outcomes; reported safety information mainly concerned technical malfunctions or self-reported side effects.
Why it matters
Primary care AI-CDSSs may improve processes in well-defined decision contexts, but evidence of patient benefit and safety is not yet dependable. The review highlights the need for robust, comparable evaluations that measure clinical outcomes, safety, and how clinicians interact with AI outputs before reliable implementation can be assessed.
What to keep in mind
Evidence was heterogeneous, follow-up was short, and risk of bias was predominantly high or critical. Studies came exclusively from high-income countries, and none reported clinician interaction measures such as agreement with or override of AI outputs.