The onset of the COVID-19 pandemic in 2020 resulted in the rapid expansion of virtual consultations in primary care [1]. The shift from a primarily face-to-face (F2F) model of health care provision led to approximately 70% [2] and 65% [3] of primary care contacts happening virtually in the United Kingdom and United States, respectively.

Virtual consultations can be defined as real-time communication between patients and clinicians through telephone or videoconferencing [4]. It is argued that not only have virtual consultations helped minimize COVID-19 transmission, but they may also improve the efficiency of and access to care [5]. This may be particularly important in rural areas with geographical disparities in service provision [6] and resource-constrained settings with workforce shortages [4]. In addition, virtual care may improve patient-provider communication and engagement [7] and better facilitate multiperson involvement in care [8].

However, concerns have been raised over the speed at which virtual consultations were implemented, with both patients and clinicians reporting lack of confidence in the underlying technology and poorer clinical decision-making as key issues. [9]. Patients have expressed concerns regarding confidentiality, privacy, and the safety of their data [7,8]. Virtual care limits clinicians’ capacity to conduct physical examinations [10] and increases reliance on patients’ abilities to articulate their symptoms [5], posing potential safety risks. Moreover, those with limited access to technology or with lower digital literacy may be at risk of “digital exclusion” [11].

Previous reviews have investigated the impact of virtual consultations on the effectiveness [3,12-15] and efficiency [16,17] of care, often focusing on specific clinical conditions or taking broad definitions of remote services. However, there is a notable lack of systematic reviews assessing the impacts on safety, patient-centeredness, timeliness, and equity, with much of the existing literature limited to scoping or rapid reviews of the evidence [18-21]. Furthermore, although some systematic reviews investigated the impact of virtual care on aspects of quality exclusively in primary care settings [13,16,17,22], others examined heterogeneous evidence, including data on interventions delivered in secondary care facilities or by specialists [3,15]. There is, therefore, a need to comprehensively evaluate the impact on all aspects of care quality specifically in primary care settings.

The aim of this review was to systematically evaluate the impact of virtual consultations on the quality of primary care. We chose to use the Institute of Medicine (IOM)’s theoretical framework to map the impact across 6 domains of quality, including efficiency, effectiveness, safety, patient-centeredness, timeliness, and equity [23]. This widely accepted model for describing care quality has been previously used in other systematic reviews investigating the effects of digital health interventions on the quality of care [24,25]. When evaluating the impact of virtual consultations on quality, it is essential that all quality domains are assessed, as what is beneficial in one domain may be detrimental in another. Therefore, the IOM’s quality of care framework provides a comprehensive model for evaluating the impact of virtual consultations across all aspects of care quality.

Six databases (MEDLINE, Embase, HMIC, PsycINFO, CINAHL, and Cochrane) were searched on June 20, 2022. The search included studies published between January 2017 and June 2022, as the last 5 years have seen most of the shift toward virtual care. Concepts covered in the search strategy primarily included (1) virtual consultations, (2) primary care, and (3) the IOM’s 6 domains of care. The concepts of virtual consultations and primary care were intentionally kept broad to address variations in language, and search terms for the domains of quality were adapted from a previous review [24].

A detailed breakdown of the exact combination of free text and Medical Subject Headings terms used can be found in Multimedia Appendix 1. The reference lists of relevant systematic reviews and gray literature sources were also screened.

Studies were included if they focused on adult patients accessing primary care services (including care homes), involved telephone or videoconference consultations delivered by health care professionals, compared outcomes with F2F consultations, and reported outcomes that fit under any of the IOM’s quality of care domains. Detailed descriptions of the inclusion and exclusion criteria are provided in Textboxes 1 and 2, respectively. Studies focusing on specific health conditions were not excluded to characterize the general use of virtual consultations in primary care.

Duplicates were automatically excluded using Covidence web-based screening software (Veritas Health Innovation [27]). During the subsequent title, abstract, and full-text screening processes, the decision to exclude a study was made through consensus between the first and second independent reviewers. Iterative meetings were held to discuss any discrepancies that arose. Only when consensus was not able to be reached, a third reviewer, the most senior member of the research team, made the final decision. Cohen κ was used to measure the intercoder agreement in title and abstract screening and full-text screening (0.22 and 0.65, respectively). Disagreements were resolved through discussion with a third investigator. Data extraction was conducted using a standardized data extraction form (Multimedia Appendix 2). Effect sizes such as mean differences, odds ratios, and risk ratios were extracted. Where available, the rates of intervention adherence, follow-up, and survey response were also extracted.

Risk of bias was assessed using the Mixed Methods Appraisal Tool (version 18) [28] by 2 independent reviewers (Multimedia Appendix 3 [29-58]). Disagreements were resolved through consensus with a third reviewer. A study was considered high risk if it scored “yes” in ≤2 dimensions, moderate risk if it scored “yes” in 3 dimensions, and low risk if it scored “yes” in 4 or all dimensions.

A narrative synthesis of the selected studies was performed. The overall characteristics of the included studies are summarized. Pertinent findings across the studies were mapped onto the IOM framework to comprehensively illustrate the evidence for all the quality of care domains. For each domain, relevant outcome data were compiled into summary tables to facilitate the comparison of results across the studies (Multimedia Appendix 4 [29,30,32-34,36,37,41-45,50,51,53,56], Multimedia Appendix 5 [31,36,42,54,55,57,58], Multimedia Appendix 6 [29,45,46,50,52], Multimedia Appendix 7 [39,40,57], and Multimedia Appendix 8 [35,38,43,47-49]). The key findings for the 6 IOM care quality domains are presented.

Our initial searches retrieved a total of 6272 records (Figure 1). After title, abstract, and full-text screening, 30 (0.48%) papers met the inclusion criteria. No relevant records were found from searching gray literature or from the reference lists of relevant articles.

The 30 included studies (Table 1) comprised a total of 5,469,333 participants. Sample sizes ranged from 28 [29] to 1,490,734 [30] participants, and publication year ranged from 2017 to 2022. The 30 studies included 14 (47%) retrospective cohorts [30-43], 6 (20%) cross-sectional studies [44-49], 4 (13%) quasi-experimental studies [29,50-52], 3 (10%) randomized controlled trials (RCTs) [53-55], 2 (7%) cohort studies [56,57], and 1 (3%) cluster RCT [58]. Most studies were conducted in the United States (20/30, 67%) [30-34,36-44,46-49,53,58], whereas the rest were conducted in Australia (2/30, 7%) [35,45], Canada (2/30, 7%) [50,52], Kenya (1/30, 3%) [54], the United Kingdom (1/30, 3%) [51], Japan (1/30, 3%) [55], Singapore (1/30, 3%) [29], New Zealand (1/30, 3%) [56], and Sweden (1/30, 3%) [57].

Studies considered patients with a range of health conditions (ie, mental illnesses [36,40,42,53]; urgent and nonemergent conditions [33,44]; overweight and obesity [31,58]; lower back pain [32]; alcohol use disorders [54]; nicotine dependence [55]; hard-to-heal ulcers [57]; and acute, nonurgent conditions [37]). The remaining studies (17/30, 57%) considered primary care users in general. All consultations were delivered in primary care settings. Of the 30 included studies, 3 (10%) studies specified occurring in rural locations [48,54,58], and 1 (3%) study was conducted in an urban, socioeconomically deprived area [51].

Of the 30 included studies, 19 (63%) studies had a moderate risk [29,31-33,37-39,41-43,48-53,56-58], 3 (10%) had a low risk [35,40,55], and 8 (27%) had a high risk of bias [30,34,36,44-47,54] (Figure 2). For quantitative descriptive studies, the main sources of bias were poor sample representativeness and risk of nonresponse bias (ie, potential lower engagement from those with lower digital literacy) [45,46,50]. Main sources of bias for RCTs included issues with blinding [54,55,58], low or unclear adherence to the intervention [53,54,58], lack of details on randomization [54], and differences between groups at baseline [53]. Nonrandomized studies had globally a high risk of bias stemming from the uncertain accuracy of the measurements of exposure and outcome [30-35,37,38,41,44,47-49], unaccounted confounders [30,32,33,43,44,47,48], the overrepresentation of certain subgroups [52], and selection bias [39,41,49].

Of the 30 included studies, 4 (13%) investigated only telephone consultations [51,52,54,58], and 10 (33%) assessed consultations delivered only via videoconference using a range of platforms (ie, Skype [Skype Technologies], Zoom [Zoom Video Communications, Inc], video call, or bespoke telehealth portals) [29,31-33,44-46,53,55,57]. The remaining 16 (53%) of the 30 studies considered consultations via both telephone and videoconference as interventions. In almost all studies, the consultations were the first in an episode of care. Other studies consisted follow-up consultations, including behavioral therapy [42,53,58], motivational interviewing [54], and smoking cessation counseling [55].

Our findings suggest that virtual consultations are equally effective to or more effective than F2F care for the management of conditions, including mental illness, excessive smoking, and alcohol consumption. Evidence for the impact on clinical safety is extremely limited. Only 5 studies investigated patient-centeredness, and of these, 4 (80%) found positive impacts on some aspects of patient-centeredness; however, a negative impact was noted on patients’ perceived autonomy support (ie, the degree to which people perceive those in positions of authority to be autonomy supportive). Virtual consultations may reduce waiting times, lower patient costs, and reduce rates of follow-up in secondary and tertiary care settings. However, there is evidence that virtual consultations may increase the need for additional general practitioner visits compared with F2F consultations. Evidence regarding equity was considerably mixed. Overall, it appears that virtual care is more likely to be used by younger, female patients, with disparities among other subgroups depending on contextual factors.

This study provides insight into the recent changes in the delivery of primary care and the impact of virtual consultations on care quality. The review uses the IOM’s comprehensive quality framework and maps findings to this model, lending a structured approach to the evaluation of the impact of virtual consultations.

This review did not consider pediatric populations, care delivered by non–health care professionals (eg, community health workers), or outcomes related to medication prescriptions. Many of the included studies considered patients who were part of specific subpopulations, such as older veterans [53] or young military servicemen [29], which may limit the generalizability of the findings.

Eligible studies were restricted to those published within the last 5 years; although this time frame was considered reasonable in line with the changing health care landscape, this might have potentially led to some earlier papers being missed. The decision to conduct the literature search only up to June 2022 may have excluded some more recently studies published in the past year. However, this should not substantially impact our review’s findings, given that it is largely in line with the median time of approximately 8 months as reported in other literature reviews [25].

It is important to note that this study was neither specifically designed to evaluate changes after the COVID-19 pandemic nor optimized to detect those differences. In addition, the studies that covered the COVID-19 pandemic were conducted when the pandemic was in its early stages and, therefore, cannot provide additional information on how the pandemic accelerated adoption, or its impact, after the COVID-19 pandemic.

All but one of the included studies were from high-income countries, and most were from the United States, highlighting the lack of research from low- and middle-income countries. Therefore, many of the findings of this review will have limited relevance outside of the United States and certainly will lack generalizability in low-income regions with dissimilar health system financing structures or technological infrastructure. The lack of studies evaluating safety aspects, such as medication safety incidents, demonstrates the gap in knowledge previously highlighted by Gleeson et al [20]. Similarly, current evidence has common limitations of bias introduced through the lack of adjustment for confounders and selection bias toward the inclusion of patients with greater digital literacy or engagement with health care.

Finally, the apparent lack of studies investigating the effectiveness of virtual care for a wider range of morbidities in primary care highlights a key area for further research.

Although evidence of improved efficiency is likely to keep driving the implementation of virtual care, it is critical to ensure that the transition to new service delivery models does not pose additional patient harm. However, even within this domain, the results were sometimes broad and unspecific, making it difficult for readers to contextualize the reported findings. Future research could benefit from a more focused approach when defining the contexts in which evaluations were taking place to aid in the generalizability of the insight gained. The apparent lack of studies investigating the safety of virtual consultations highlights a concerning gap in the literature, and future evaluations should focus on the evaluation of diagnostic error and medication safety in this context. Furthermore, health technology assessments investigating the impact on patient-centeredness should capitalize on the use of validated patient-reported measures whenever possible to allow a rigorous comparative approach.

Policy efforts to support improvements in data collection in primary care (ie, consultation type, duration, and quality outcomes) will be critical to developing a strong evidence base capitalizing on real-world data. The mixed findings on the impact on equity highlight the need for investigations at the local level, which will be vital to develop context-specific strategies tailored to community health needs and characteristics. Data collection should adopt an intersectional approach, considering a breadth of patient characteristics, to inform the design of locally appropriate interventions and ensure equitable access to care. Importantly, virtual consultation interventions and access schemes must incorporate participatory approaches in their research and design, encouraging input from marginalized voices and including community knowledge, values, and preferences in decision-making processes.