As a means to mitigate the risk of viral transmission for both patients and clinicians during the COVID-19 pandemic, many health systems have rapidly converted ≥70% of their outpatient visits to telehealth via phone or video delivery [1-5]. To support this shift, the Centers for Medicare and Medicaid Services in the United States issued an emergency ruling to decrease regulatory requirements for telehealth and created payment parity between in-person care and telehealth delivered via phone or video [6]. Increased telehealth use during the COVID-19 pandemic provided health systems, technology companies, and health care providers experience with telehealth at scale and raised the possibility that telehealth could become a standard option in the postpandemic period. However, concerns remain that care delivered via telehealth is potentially low in quality of care, is difficult to incorporate into workflows, and can exacerbate health disparities [7-10]. Specifically, evidence is needed regarding the efficacy of telehealth as a replacement for in-person care when treating patients.

Extensive literature supports telehealth as a supplement or adjunct to in-person care for the management of chronic conditions [11] such as congestive heart failure (CHF) and type 2 diabetes mellitus (T2DM) [12-14]. These 2 highly prevalent chronic diseases are among the most common and costly conditions affecting approximately 13.4% [15] and 10.5% [16] of all adults in the US, respectively. In addition, CHF and T2DM typically require physical assessment to establish disease status and assess the presence and extent of exacerbations. However, the effects of telehealth as a replacement for in-person health care delivery for CHF, T2DM, and other chronic illnesses remain uncertain [10,17,18] Before the COVID-19 pandemic, many patients with chronic medical conditions, such as CHF, chronic obstructive pulmonary disease (COPD), and T2DM, uniformly received in-person evaluation. During the pandemic, these patients often received telehealth to unknown effect. Although telehealth can increase accessibility to health care by lowering barriers to access [19-21], few studies exist to support the use of telehealth as an equivalent and equitable replacement for in-person care, and the potential adverse effects have not been well defined [18]. Assuming that telehealth can readily replace in-person care may be inappropriate, given the scarcity of evidence examining telehealth applied in this way.

A first step to address the question of equivalence of synchronous (real time) telehealth via phone or video as a replacement for in-person care for chronic diseases is a review focused specifically on evidence from the comparative literature. If there is moderate to strong evidence that telehealth is equivalent to in-person care for patients with chronic conditions, its promise should be developed more fully and incorporated as a standard option for delivering longitudinal care. Early during the COVID-19 pandemic, there was the first complete replacement of telehealth with in-person care [1-5]. However, since then, we have started to see the routine substitution of telehealth for in-person care visits across many specialties and contexts. This substitution (meaning only video) is not usually for all care, but rather can often be a replacement for part of in-person care (some phone visits replaced by video). In addition, currently, there are multiple commercial health care providers who provide only telehealth (Teledoc and CallonDoc). It is within this context that we formulated the questions for this review.

We conducted a systematic review to summarize and report the use of telehealth as a replacement or substitute for in-person care in the context of chronic management of CHF, COPD, and T2DM. The questions guiding this review were the following:

This systematic review was conducted as part of a Veterans Health Administration (VHA)–funded report [22] in response to a topic proposed by the VHA Office of Rural Health. For this review, similar to completed previous reviews and to meet the goals of the VHA as a learning health care system [23], (1) the partners from the Office of Rural Health were not involved in conducting the review, but informed topic and question development and provided contextual relevance for the study; (2) the partners from the Office of Rural Health were not involved in approving the final write-up of the report; and (3) a technical expert panel guided the conduct of the review and discussion of the findings. We developed and followed an a priori protocol for this review, and there were no significant deviations after registration (PROSPERO [International Prospective Register of Systematic Reviews] registration number CRD42021239756) [24]. Each step was pilot-tested to train and calibrate the study investigators. We adhered to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [25].

We developed an analytic framework [26] (Figure 1) that outlined the population, outcomes, mediation effect of the care visit modality, moderation effect of patient characteristics, and any adverse effects. First, we identified clinical activities (medication management, symptom monitoring, and physical examination) for longitudinal follow-up and the ability to complete them via phone, video, or either of these for CHF, COPD, and T2DM. We determined the relevant aspects that should be abstracted from the eligible literature to obtain critical evidence about conducting a telehealth visit in any clinic setting. Then, with this foundation, we determined that the telehealth modality (eg, telephone, video, and in person) mediates the relationship between the clinical visit and prespecified clinical-level and system-level outcomes. The telehealth interventions matched with our operationalized definition of telehealth and included important contextual elements such as delivery mode (telephone, video, and in person), dose (duration and frequency of contact), and clinical context of care provision. In addition, we specified that care delivered via telehealth should be for clinical activities provided by the prescribing clinician such as for evaluation, diagnosis, or medication prescription and not for the provision of self-management education or other support provided adjunctively by a clinical team member other than the prescribing clinician (eg, nurse care manager), because such interventions have been previously evaluated [11].

In collaboration with an expert medical librarian, we conducted a primary literature search from inception to February 7, 2021, in 2 databases (MEDLINE [via Ovid] and Embase [via Elsevier]). We used database-specific subject headings and keywords to search for relevant titles and abstracts (Multimedia Appendix 1). The search strategies were peer-reviewed by a second expert medical librarian before execution using the Peer Review of Electronic Search Strategies Checklist [27]. In addition, we manually searched previous systematic reviews conducted on this or a related topic for potential inclusion.

Studies identified through our primary search were classified independently by 2 investigators from the study team for relevance to the questions based on the title and abstract from our a priori established eligibility criteria. Study eligibility criteria were organized by population, intervention, comparator, outcome, timing, and setting elements and other criteria such as study design, language, and publication type (Table 1). All studies classified for inclusion by at least one investigator were reviewed at the full-text level. The studies designated for exclusion by one investigator at the title and abstract level were screened by a second investigator. If both investigators agreed on exclusion, the study was excluded. Full-text review included 2 independent reviewers. Conflicts were resolved via discussion. All articles that met the eligibility criteria at the full-text level were included for data abstraction. All results were tracked in an electronic database (EndNote [Clarivate Analytics] for referencing and DistillerSR [Evidence Partners Inc] for data abstraction).

Data from published reports were abstracted into a customized DistillerSR database by one reviewer and overread by a second reviewer. Disagreements were resolved by consensus or by obtaining a third reviewer’s opinion. Data elements included descriptors to assess applicability, quality elements, intervention details, and outcomes including adverse events. Key characteristics that were abstracted included participant descriptors (eg, race and ethnicity, gender, age, and rural status), intervention characteristics (eg, clinician type and telehealth modality), comparator, and outcomes (eg, glycosylated hemoglobin [HbA_1c] level, hospital admission, emergency department visits, and New York Heart Association functional classification). We abstracted all outcomes that were used to evaluate telehealth, but prioritized outcomes identified a priori in collaboration with our partners from the Office of Rural Health and technical expert panel for analysis. Multiple reports from a single study were treated as a single data point, prioritizing results based on the most complete and appropriately analyzed data. When critical data were missing or unclear in the published reports, we requested supplemental data from the study authors. We emailed the authors of 1.6% (2/129) of the studies to obtain additional information and did not receive a reply from any of them. When we did not have sufficient information, we left the field blank.

The investigators who participated in data extraction also completed the quality assessment. Disagreements were resolved by consensus between the 2 investigators or, when needed, by arbitration by a third investigator. For randomized, nonrandomized, and controlled before-after studies, we used the criteria from the Cochrane Effective Practice and Organization of Care (EPOC) risk of bias (ROB) tool [28]. We assigned a summary ROB score (low, unclear, or high) to individual studies. Among the investigators, no ROB disagreements occurred owing to missing results in a synthesis.

The certainty of evidence for each question was assessed using the approach described by Grading of Recommendations Assessment, Development, and Evaluation [29]. We limited the Grading of Recommendations Assessment, Development, and Evaluation ratings to the questions that had at least two included studies. In brief, this approach requires the assessment of four domains: ROB, consistency, directness, and precision. Additional domains to be used when appropriate are coherence, dose-response association, impact of plausible residual confounders, strength of association (magnitude of effect), and publication bias. We considered these domains qualitatively and assigned a summary rating as high, moderate, or low strength of evidence after discussion by a subteam of 5 investigators. In some cases, high, moderate, or low ratings were impossible or imprudent to be provided. In these situations, a grade of insufficient was assigned.

The research questions guided our subgroup analysis. Prespecified potential effect modifiers included study design characteristics (eg, allocation concealment), disease context (CHF, COPD, or T2DM), and intervention type (eg, telehealth modality). Regarding patient-level characteristics of interest (race and ethnicity, gender, age, and rural status), we looked for analyses conducted within the primary literature that sought to identify effect modifications (eg, subgroup analyses and regression model explanatory variables). Manuscripts included in this review did not specify descriptions of gender or sex. For consistency, we use gender throughout the Results and Discussion sections because the interventions examined are more relevant to self-identity and not specific to one’s biology at birth. However, we realize that this terminology may not reflect patients who would not have self-identified as such.

We summarized the primary literature using relevant data abstracted from the eligible studies. Summary tables describe the key characteristics of the primary studies: study design, patient demographics, and details of the intervention and comparator. Owing to conceptual heterogeneity related to the structure, purpose, and delivery of telehealth visits, we did not conduct a meta-analysis, but rather described findings narratively, focusing on identifying patterns in the efficacy and safety of the interventions across conditions and outcome categories.

Continuous outcomes were summarized using the mean patient-level difference (follow-up minus baseline) when the outcome was reported using the same scale. For studies that did not directly report the mean and SD of patient differences, we used the difference in means between the follow-up and baseline. For 20% (1/5) of the studies [30], we computed the SD of the difference based on the reported P value for the difference between the 2 arms, assuming the same correlation between follow-up and baseline in each arm. When studies reported only medians and ranges, we translated them into means and SDs [31], and if a study reported only baseline SD, we assumed the same SD at follow-up. Finally, in the absence of other information, we assumed a conservative 0.5 correlation between the follow-up and baseline measures.

Given the limited amount of existing literature we identified that addressed our questions, we sought to assess the pool of ongoing studies that would add relevant findings in the near future. To conduct such a scan of the literature on the horizon, we applied our previously developed search terms to the Cochrane Central Register of Controlled Trials. Notably, we did not apply the same rigor to this process as for our primary search process. At least one reviewer screened the studies identified through the horizon scan at the title and abstract level, and all the included studies were verified by a second reviewer.

The search identified 11,245 studies from the 2 databases (Figure 2). After deduplication, 77.03% (8662/11,245) of the articles underwent the screening process. In total, 0.06% (5/8662) of the studies met the inclusion criteria. Of those 5 studies, 4 (80%) focused on diabetes and 1 (20%) focused on CHF. The details of the included studies are provided in Table 2. We have provided the details of study characteristics (Multimedia Appendix 2), intervention characteristics (Multimedia Appendix 3), all outcomes reported in the included studies (Multimedia Appendix 4), and excluded studies and the reason for exclusion (Multimedia Appendix 5). Common reasons for excluding studies by intervention included telehealth that supplemented rather than replaced in-person care, telehealth interventions delivered by nonprescribing clinicians, and telehealth delivered asynchronously only. In the following sections, we describe the results by chronic disease (CHF, COPD, and T2DM). The certainty of evidence for the included studies is presented in (Table 3).

The 20% (1/5) of the studies of CHF, by Hansen et al [33], did not report on adverse events. The 40% (2/5) of studies of T2DM reported adverse events [30,32]. Jeong et al [32] described four groups of adverse events: (1) general events, (2) diabetes-related events, (3) serious events, and (4) biochemical events. Adverse events were noted in the control (n=33 or 29.20%, in-person appointments at 8, 16, and 24 weeks), telemonitoring (n=30 or 26.55%, in-person appointments at 8, 16, and 24 weeks, with remote monitoring of blood glucose data), and telemedicine (n=23 or 20.54%, video visits at 8 and 16 weeks, in-person visits at 24 weeks) arms. Diabetes-related events were noted in the control (n=7 or 6.19%), telemonitoring (n=7 or 6.19%), and telemedicine (n=3 or 2.68%) arms. Serious reported adverse events were noted in the control (n=2 or 1.7%), telemonitoring (n=2 or 1.70%), and telemedicine (n=1 or 0.90%) arms, and it included angina pectoris, rotator cuff syndrome, malignant hepatic neoplasm, skin ulcer, and hematuria [32]. Biochemical parameters for serum alanine aminotransferase (ALT), aspartate aminotransferase, and creatinine levels were measured, and samples were obtained at baseline and 24 weeks [32]. Comparing the relative percentage of patients with worsened laboratory values, ALT was the only parameter that showed significant worsening between the telemedicine and telemonitoring groups. Specifically, none of the participants in the telemonitoring arm and 7 participants in the telemedicine arm (6.7%; P=.01) experienced worsening of ALT values. Klingeman et al [30] described two types of adverse events: (1) severe hypoglycemia and (2) foot ulcers. Severe hypoglycemia was noted in the experimental (n=1 or 3.3%) arm, but not in the control (n=0 or 0%) arm. Foot ulcer was noted in the experimental (n=1 or 3.3%) and control (n=3 or 10%) arms.

The 20% (1/5) of the studies of CHF [33] that met our inclusion criteria was rated as having high ROB owing to low numbers of patients enrolled, unclear method for patient randomization, and poor description of both patient dropout and how primary outcomes were assessed. Among the 80% (4/5) of randomized T2DM studies, the ROB (Figure 4) for patient-reported outcomes was judged to be low for 1 (25%) study, unclear for 1 (25%) study, and high for 1 (25%) study and 1 (25%) study did not report this type of outcome [30,32,34,35]. For objective outcomes, ROB was judged to be low for 50% (2/4) of the studies [32,34] and high for 50% (2/4) of the studies [30,35]. Patterns that led to judgments of low ROB (Figure 5) included (1) noting randomization of study participants, (2) collecting objective outcome data, and (3) generally limited expected impact of bias from patient knowledge of the treatment arm. Patterns that led to high ROB included (1) missing or unclear data on randomization methods, data collection, and analysis; (2) unblinded treatment arm; (3) absence of predetermined intervention assessment patterns in the protocol; (4) unclear primary outcomes; and (5) missing or unclear reporting of patient-reported outcomes.

This search identified 1787 unique studies. We found only 0.17% (3/1787) of studies [36-38] in our horizon scan that reported on studies without published results (Multimedia Appendix 6) that may potentially meet the inclusion criteria of our systematic review. All of these studies (3/3, 100%) are randomized controlled trials that were designed before the COVID-19 pandemic. Of the 3 studies, 2 (67%) studies focus on T2DM [36,37], whereas the remaining 1 (33%) study is on CHF [38]. Although the noninferiority study [37] will not meet our inclusion criteria as it is conducted in Brazil (a non–Organisation for Economic Co-operation and Development [OECD] country) and the findings may not be applicable to the US population or setting, we mention it here given the low number of studies that otherwise met our inclusion criteria. The other T2DM study [36] is specifically focused on reducing emergency diabetes care for older (aged >50 years) African Americans. The CHF study by Komkov et al [38] has very limited detail available. Although using these registries to identify trials has limitations and there are likely other relevant studies, it appears that there are few trial-based studies currently in the pipeline to inform our questions in this review.

We found scant evidence examining chronic disease management delivered through synchronous telehealth compared with in-person delivery for T2DM (4/5, 80%), COPD (0/5, 0%), and CHF (1/5, 20%). This suggests that there is little evidence to help guide practice on when to use telehealth instead of traditional in-person visits while managing these chronic diseases. Our review sought to include studies that used telehealth to replace all or part of in-person care. In other words, some specific in-person visits in the intervention arm were replaced by telehealth visits, whereas the comparator arm maintained all visits as in-person. Note that we consider this to be different from using telehealth as a supplement or add-on to the usual in-person care. However, we did not find any studies that only partially substituted in-person visits. We did not attempt to include studies that used telehealth as an add-on to existing in-person care, as there are already existing high-quality, peer-reviewed publications on this question [11-13]. However, despite the paucity of evidence, telehealth modalities such as video or telephone have increasingly been used to replace in-person clinic visits for managing chronic conditions, particularly during the COVID-19 pandemic [1-5]. Understanding the benefits and risks associated with shifting in-person care to telehealth is critical in shaping how health systems deliver care going forward. Although in-person visits have since increased as more has become known about COVID-19 transmission and prevention practices, telehealth continues to play a much larger role in outpatient care than before the pandemic [39,40].

Evidence indicates that telehealth can be used effectively as an adjunctive or supplemental approach to in-person care. A recent review by Albritton et al [18] examined the impact of video teleconferencing visits on prevention and management of chronic illness. Results from that review indicated that video teleconferencing resulted in similar clinical effectiveness as in-person care for certain diseases [18]. The results from our review differ in indicating clinical effectiveness of telehealth from those of Albritton et al [18] owing to several differences in the review type (systematic vs rapid review), date limitations, search strategies, databases searched, and operationalization of telehealth. Our approach to identify relevant telehealth papers was broad and more comprehensive, which resulted in a large number of articles to review. Of the 7 papers included in the review by Albritton et al [18], only 1 (14%) was not captured in our search. Additional previous reviews have examined various ways of using telehealth modalities in the context of these conditions of interest, but none of them have focused on replacing in-person care with telehealth visits [11]. Although we found only 20% (1/5) of the studies on telehealth for chronic management of heart failure as a substitute for in-person care, previous reviews report mixed results for the impact of other supplemental types of telehealth on heart failure outcomes [41-43]. Several recent analyses on the impact of telehealth in T2DM indicated that health outcomes did not worsen because of switching to telehealth compared with those in-person clinic care [14,40,44,45]. However, there is evidence that telehealth as an adjunctive strategy to typical in-person care can be associated with a decrease in HbA_1c level in patients with both type 1 and type 2 diabetes [46-49].

The successful incorporation of telehealth into health care delivery relies upon the fit between the telehealth modality, care delivery context, and disease management approaches [9,19,50]. Presumably, not all areas of health care delivery lend themselves equally well to telehealth, but management of certain chronic diseases (CHF, COPD, and T2DM) may provide good opportunity to replace routine in-person care with telehealth. In our review, we sought to address a critical evidence gap by examining the comparative literature on telehealth as a replacement for in-person care in chronic disease management. Interestingly, our findings came from studies that were conducted in specialty settings, and aspects of the studied telehealth interventions were often incompletely described. However, much of the long-term management of these chronic conditions occurs within the context of primary care settings. As primary care settings likely have different pressures and challenges with telehealth modalities, given the need to address multiple comorbidities during the same visit, the results from our review may not be directly applicable. Thus, we recommend future reviews to examine and provide evidence-based guidance about the effect of telehealth interventions to deliver high-quality care using the right modality for the right patients with the right clinical condition at the right time.

A way to determine the effect telehealth is to use noninferiority analytic approaches when hypotheses focus on whether telehealth delivered care is equally effective to in-person care. Our eligibility criteria focused on randomized controlled trials and did not include observational study designs. Randomized controlled trials are the gold standard; however, conducting these trials is time-consuming and resource-intensive. Importantly, findings from randomized controlled trials take years to affect clinical practice, if they are implemented at all. Randomized controlled trials should not be expected to fill all the research gaps in the implementation and adoption of telehealth for chronic disease management. Thus, given the paucity of randomized controlled trials, we strongly recommend that future reviews focused on telehealth include what are likely to be rich and robust, but potentially biased; observational; and alternatively designed studies that emerge during and after the COVID-19 pandemic.

Overall, there are 5 key areas in which future studies on this topic can fill the existing gaps and improve the approach. First, and perhaps most critical, telehealth interventions should be thoroughly described to maximize reproducibility and generalizability in other clinical contexts. Guidance exists on mobile and web-based interventions, which may provide indirect suggestions about key characteristics for telehealth intervention description. Second, there is a need to evaluate how best to integrate telehealth as a replacement for in-person care. Furthermore, there is a need to assess which clinical settings are best suited to the telehealth environment (eg, primary care vs specialty care settings). Approaches to integrating telehealth can be expected to vary across settings with different workflow patterns, clinical resources, and competing clinical demands, which emphasizes the need for solid evidence. Third, outcomes varied across the included studies, and some important outcomes were not addressed by any study (eg, impact on clinical workflow, patient satisfaction with telehealth experience, and subsequent use). Fourth, investigators should be encouraged to consider a priori subgroup evaluations or make individual patient-level data available, so that future reviews can identify patient-level characteristics associated with better outcomes with telehealth. Finally, future studies should also actively solicit and report patient perspectives and feedback on telehealth interventions to better inform intervention design. Such information can guide clinics and health care systems to offer optimal patient-centered telehealth delivery and support efforts to ensure equitable benefits and access to telehealth.

Our review benefited from being protocol-driven, leveraging input from an expert panel consisting of clinicians and telehealth researchers, identifying disease-specific clinical outcomes, using an analytic framework to guide the understanding of telehealth modalities, and using a detailed approach to categorize and define telehealth components in chronic disease self-management. In addition, our review was based on a clear definition and use of telehealth. Notably, we acknowledge that individual patient characteristics (eg, race and ethnicity, gender, age, and rural status) may moderate the relationship between the modality in which the clinical visit occurs and any clinical-level and system-level outcomes.

Despite these strengths, our approach had some limitations. First, we included only the studies that met the EPOC criteria in this review; however, observational studies may have findings relevant to the provision of synchronous telehealth for chronic illness management. However, we do not believe that this limitation largely affected our findings. Second, we focused this review on 3 of the most prevalent chronic diseases, but there may be appropriately designed studies that targeted other conditions that we did not include. Third, we only included studies conducted in OECD countries, and thus, we may have missed relevant studies conducted in other countries. Fourth, given the small number of studies that we identified, statistical methods to detect publication bias were not conducted. Although it is possible that individual health systems or clinics have conducted quality improvement studies evaluating differences in experiences between synchronous and in-person care—especially during the COVID-19 pandemic—we suspect it to be unlikely that studies meeting EPOC criteria on this intervention have not been published, given the recent emphasis on the role of telehealth. Fifth, we identified few studies overall, and most studies had <100 patients and were assessed as having unclear or high ROB. Intervention core components, intervention fidelity, and impact of intervention on clinical workflow were not reported in any study. In addition, the interactions between clinicians and patients during telehealth episodes were not adequately or explicitly described, and most of our outcomes of interest were not consistently reported across the studies. These omissions limited the interpretation and replication of the evaluated interventions. Sixth, the included telehealth interventions used different telehealth modalities (email, phone, and video) with different hardware, delivered via different numbers of clinical interactions between patients and clinicians, over a wide range of intervention durations, and within different health care systems, which inherently make comparison between them challenging. Finally, the studies included in our review did not specify how they used or defined gender (man, woman, or nonbinary) or sex (male, female, or intersex) in their publications. Information on gender and sex is important to be captured and described for telehealth studies and research. Future studies should consider including observational studies; studies on additional, highly prevalent chronic diseases; studies conducted in non-OECD countries; and studies that do not meet the EPOC criteria especially, as those conducted since the onset of the COVID-19 pandemic may provide useful information.

The COVID-19 pandemic precipitated a rapid shift from in-person to telehealth delivery, without a clear understanding about the impacts of telehealth on important health outcomes. Previous studies have found that telehealth modalities can improve health outcomes through the supplementation of in-person management of certain chronic diseases, particularly with approaches such as remote monitoring and patient education. However, we found that, currently, there is very little evidence on the use of telehealth as a replacement for in-person care for several chronic conditions and that the studies in this area remain insufficient and methodologically inconsistent. In conclusion, our review builds on this existing body of literature by evaluating the comparative literature on the effectiveness of telehealth visits delivered as a substitute for in-person visits for chronic disease management and provides recommendations for future studies in this area.