We selected a scoping review as the best method for assessing the research landscape of AI and health equity in primary care because it offers a way to systematically identify key research gaps, opportunities, evidence, and concepts in this understudied space. This type of review differs from systematic reviews and meta-analyses in that it does not narrow the parameters of the review to a specific quality assessment. Instead, it is a systematic approach to examine the landscape of a research field using broad questions to examine both empirical and conceptual aspects [16,17]. This is particularly important in the fields of health equity, primary care, and AI, where much of the literature is currently focused on specific outcomes or aspects of care [18-21]. Equity considerations extend across multiple outcomes and therefore require a scoping review to draw overall conclusions. Our protocol, developed on the basis of seminal work by Arksey and O’Malley [16], includes six stages: (1) identification of the research question; (2) identification of relevant studies; (3) study selection; (4) data extraction; (5) collation, summarization, and reporting of the results; and (6) consultation of knowledge users. We followed the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses for Scoping Reviews) checklist [22] and registered our protocol with the Open Science Framework (digital object identifier: 10.17605/OSF.IO/WGSB3). To identify articles of interest, we conducted a search update on the basis of a previous study by Kueper et al [15], who conducted a systematic scoping review of AI and primary care research in May 2020.

A committee of medical professionals at different levels (medical students and attending physicians) with multiple domain expertise (AI, primary care, and fairness in machine learning) and training in recognition of health care disparities led the scope of this study. We used the methodology of Arksey and O’Malley [16] and Levac et al [23] to guide the discussions for determining the research questions we sought to investigate. We considered vulnerable populations on the basis of the PROGRESS (place of residence, race/ethnicity/culture/language, occupation, gender/sex, religion, education, socioeconomic status, and social capital) criteria [8], which include the following variables to ascertain vulnerabilities: place of residence; race, ethnicity, and culture; occupation; gender; religion; education; socioeconomic status; and social capital. We identified three key domains for assessment: representation of vulnerable populations in the underlying data set relative to the intended target population, as assessed, for example, by subgroup prevalence; author reporting of the types of bias outlined by Suresh et al [9]; and whether these studies attempt to mitigate these pre-existing biases in their systems upstream of, during, or downstream of model development (Table 1).

We built a preliminary data framework in accordance with the suggestions of Daudt et al [28] to align data extraction with the initial research question (Table 2). One category we extracted is measuring compliance with existing AI ethics guidelines developed for the European Commission. This category was chosen after examining multiple other AI ethical guidelines, including those of the House of Lords [29] and IEEE [30]. The European Commission's guidelines were chosen because of the comprehensible key requirements, orientation toward conceptual, higher-level evaluation (rather than technical specificities), and wide adoption across the AI community [31-33]. Two authors (JW and SS) will independently extract data from the first 10 included studies and meet to determine whether the framework is specific enough for consistency and the data are sufficient for research questions outlined initially. During this process and in prior stages, it is likely that additional categories and adjustments will be made to our data extraction framework, at which time we shall consult with the research team to guide decisions on how to appropriately modify the framework. Once the reviewers reach a consensus using the data extraction framework, it will be circulated among the research team and consultation team for final comments and suggestions. Following this, additional reviewers may be brought on, in which case they are expected to match the data extracted from these first 10 included studies in order to take part in the data collection (Cohen κ>0.8). For any disputes on data extraction, a third reviewer will be involved in settling the discussion, and appropriate adjustments to the data extraction framework will be made.

Our analysis will involve both a descriptive numerical summary and an interpretive synthesis. While our approach in stage 5 will be an iterative process, we will use this section to first provide descriptive tables, frequency tables, and visual representation of the results. Further synthesis will be performed to identify current obstacles, gaps, and opportunities in the literature.

Our scoping review will include consultation with other AI researchers in academia, nonprofit, and industry to enhance the perspective, applicability, and purpose of our study and ultimately offer more practical recommendations. We will engage with stakeholders at three timepoints: (1) prior to the submission of this protocol, (2) during the finalization of the data collection framework, and (3) at the end of the study during the collation, summarization, and reporting of the results.

To our knowledge, this will be the first scoping review that applies an equity lens to the existing literature on AI in primary care. Primary care has a large potential to reduce costs and improve quality of life, especially for underserved populations [37]. Many experts have lauded AI’s potential to affect primary care [14] and issues in vulnerable patient care management. By understanding AI’s current place in primary care through the lens of health care equity, researchers can develop AI interventions that address the field’s existing gaps and opportunities.

After completing this scoping review, we will write a briefing paper to address the implications of the findings in a narrative. We will also develop a manuscript and PRISMA-ScR checklist to submit for publication.

Our scoping review will not incorporate a peer review process for our search strategy despite being recommended in Peer Review of Electronic Search Strategies [38]. This is typically conducted for systematic reviews rather than for scoping reviews and is not feasible with the time and resource constraints we have to achieve this review [23]. Additionally, we do not engage with community members or underserved populations themselves for consultation or feedback. We believe this is important for any study related to health equity as it improves the quality and applicability of studies for the populations they hope to serve [23,39]. However, identifying and consulting with these groups has been difficult and costly to incorporate into the protocol, which has been a recurring problem in this field of research. Instead, we rely on expert stakeholders to guide our critical appraisal of the existing literature. Considering the design of this study, we also will not conduct a rigorous assessment of the included articles beyond an inequity lens [16]. Additionally, scoping reviews do not provide a clear understanding of the efficacy of current interventions in practice as systematic reviews do, which is offset by the benefit of providing breadth from a large number of studies [16]. We also limit our work to English language articles, and no proprietary research is captured in this review.

AI has immense potential to improve the patient-physician relationship by augmenting physician capabilities. Primary care is an especially viable area for the integration of AI, given its early entry point, broad scope of vulnerable populations, the heavy toll these socioeconomic factors have on patient care, and the need to address these factors to manage disease more effectively. However, algorithms are susceptible to performance disparities across different subgroups, which may further reinforce pre-existing health inequities if not rigorously assessed before deployment. With this scoping review protocol, we aim to provide a process to assess the state of AI in primary care for vulnerable populations.