JMIR Publications

ABSTRACT

Background:

Research on the integration of artificial intelligence (AI) into community based primary health care (CBPHC) has highlighted several advantages and disadvantages in practice, including facilitating diagnosis and disease management in different fields, as well as doubts concerning implementation of AI in health care. Despite, the potential benefits and risks, there is no comprehensive knowledge synthesis that clearly identifies and evaluates AI systems that have been tested and/or implemented in CBPHC.

Objective:

To identify and evaluate the published papers which tested and/or implemented artificial intelligence (AI) in community based primary health care (CBPHC). Materials: Systematic scoping review informed by Joanna Briggs Institute scoping review framework.

Methods:

An information specialist performed a comprehensive search from date of inception until February 2020, on seven academic databases: Cochrane Library, MEDLINE, EMBASE, Web of Science, CINAHL, ScienceDirect, and IEEE Xplore. The population included all who provide and receive care in CBPHC. The intervention included AI interventions, implemented or tested. The outcomes of interest were related to patients, health care providers and CBPHC systems. Two authors independently screened the papers, extracted data using a validated extraction form and the third author validated all. Risk of bias was assessed using the prediction model risk of bias assessment tool (PROBAST).

Results:

We retrieved 22,113 documents. After the removal of duplicates, 16,870 documents were screened, and 90 studies met our inclusion criteria. Machine learning (45%), natural language processing (26%), and expert systems (18%) were the most highly studied in CBPHC. The AI systems were primarily implemented for diagnosis/detection/surveillance purposes. Highest accuracy was reported for neural networks, considering the given database for the given clinical task. The risk of bias was the least in participants category (4%) and the highest in outcome category (43%). Discussion: AI systems are not widely tested and implemented in CBPHC, and those which were tested/implemented poorly reported their datasets, AI models, and outcomes.

Conclusions:

Further studies are needed to efficiently guide development and implementation of AI interventions in CBPHC setting.