In phase 2, MEDLINE (Ovid), CINAHL (EBSCO), and Embase (Ovid) were searched on November 26, 2020, to identify literature on digital health interventions supporting the transition from hospital to home for older adults (Multimedia Appendix 1). These databases were selected because they (1) could identify health-related literature and (2) were determined by our research team (including a medical librarian HVC) to be appropriate for the scope of our search [43]. The review adhered to the PRISMA-S (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Literature Search) checklist [53]. The reference lists of some included articles (n=20) were hand searched, and content experts (n=6) were consulted to identify additional studies.

The search results were uploaded to the Covidence website. On the basis of the inclusion and exclusion criteria (Multimedia Appendix 2 [54,55]), each article’s title and abstract were screened by a single reviewer from the screening team (HS, TT, KK, RT, DP, MH, CMJ, AA, or JXN), followed by a full-text review conducted independently by 2 reviewers from the screening team. Any conflicts were resolved through team discussions. Studies were included if they tested a digital health intervention that supported a hospital-to-home transition for older adults and were published in or after 2010. For this review, an intervention that supported a hospital-to-home transition had to have recruited participants before their hospital discharge and continued in the home or community setting. The studies had to include ≥1 older adult but did not need to focus on older adults exclusively. No limits were imposed on study design to ensure that we included relevant studies, but articles had to report findings from empirical studies. Given that we intended to inform recommendations for the Digital Bridge intervention [45], a high-technology intervention for use in a “high-income country,” strictly telephone-based interventions, and interventions tested in a “low-income country” were excluded [54].

The following data were extracted from the articles using a customized form informed by the Template for Intervention Description and Replication framework [56]: author details; country and year of publication; sampling strategy; inclusion and exclusion criteria; the medium of technology, function of technology, and who provided the intervention; study findings; and limitations and future directions. Data regarding intervention effectiveness were not extracted, reported, or synthesized in this review because this was outside its purpose, scope, and intent [57].

We descriptively reported study characteristics and qualitatively analyzed data using a thematic analysis [51]. We first analyzed each study’s discussion using data-driven codes to identify lessons learned. We then coded data deductively by grouping the technology functions according to the 3 categories of continuity of care described by Haggerty et al [52]. The technology functions and providers involved in intervention delivery were compared with the provider roles and functions identified in phase 1.

The phase 2 database search identified 29,359 articles. Additional articles (n=10) were identified from hand-searching reference lists of the included articles. After removing duplicates, 81.88% (24,048/29,369) remained for the title and abstract screening and 4.02% (967/24,048) met the criteria for full-text review. Of these 967 articles, 142 (14.7%) met the study inclusion criteria (see Figure 1 [111] for the PRISMA [Preferred Reporting Items for Systematic Reviews and Meta-Analyses] flow diagram). Table 1 provides details of the study characteristics.

The studies were conducted in multiple countries, most of them in the United States (Table 1). They were published between 2010 and 2020, with a growing rate of publications over the years (Figure 2).

Of the 142 interventions, 47 (33%) were classified into multiple categories of intervention types (N=193 intervention classifications). Of the 6 intervention-type characterizations, smartphone, tablet, or web-based interventions (91/193, 47.2%) were the most common than telemonitoring and wearables, clinical documentation system (45/193, 23.2%), clinical documentation systems (29/193, 15%), automated telephone calls or automated SMS text messaging (14/193, 7.3%), email interventions (10/193, 5.2%) or other interventions (eg, television video; 4/193, 2.1%).

As shown in Table 2, a total of 35.9% (51/142) of the interventions used multiple provider roles (n=202 provider roles identified) in the implementation of the digital health intervention, with nurses (64/202, 31.7%) and physicians (61/202, 30.2%) being the most common providers of digital health interventions. Discharge-specific personnel such as a transition coach, nurse care transition coordinator, discharge facilitator, advanced practice nurse, and systems navigator were less common (18/202, 8.9%).

Some interventions had designated a study-specific health care provider to carry out the digital health intervention activities, whereas others added the responsibility onto a provider’s existing workload. The responsibilities of providers also differed based on the type and purpose of technology and whether communication between patients and providers was initiated by patient or provider. Among some interventions with patient-initiated communication, providers had to always be available for consultation during the intervention period.

The digital health interventions were most commonly used up to 7 days after discharge (29/142, 20.4%) or between 31 and 90 days after discharge (39/142, 27.5%). It was less common for the interventions to continue for 91 days to <6 months after discharge (18/142, 12.7%) or beyond 6 months after discharge (7/142, 4.9%).

In terms of the technology functions that supported hospital-to-home transitions, most (116/142, 81.7%) of the technologies fell into multiple categories (ie, 57/142, 40.1%, fell into 2 categories and 59/142, 41.5%, fell into 3 categories).

Of the 317 total technology functions within the included interventions, 142 (44.8%) were related to management continuity, including following up, assessing, and monitoring patients’ status after hospital discharge, as well as facilitating referrals. Some technologies could identify values outside a predefined range during follow-up, assessment, and monitoring of patients’ status. However, others required human resources to review all data to identify abnormal values. In both cases, if values fell outside the range, a human resource (eg, provider or study personnel) had to follow-up and provide appropriate guidance and immediate treatment or the technology instructed a patient to initiate contact with a provider. Informational continuity was supported among 32.2% (102/317) of the identified technology functions, including facilitating communication (eg, between inpatient and outpatient providers or between patient and provider) and educating patients and caregivers. Relational continuity (eg, counseling and rapport building) was least supported by the technologies (73/317, 23%).

Of the total outcomes of interest (n=315) examined in the articles, more than half of the outcomes evaluated the effect of the intervention on patient-level factors (eg, disease knowledge, quality of life, and changes in physical or psychological functioning) and technology-user interactions (eg, use of technology, patient satisfaction with technology, and the perceived value of technology) at 28.6% (90/315) and 28.3% (89/315), respectively. Of all outcomes, 17.5% (55/315) related to health care use, examined through health care–related costs and hospital readmission rates or emergency department visits at various time points (eg, 30, 60, 90, and 180 days after discharge). The intervention effect on provider-related outcomes (eg, changes in provider workflows, provider burden, and clinical documentation accuracy), implementation-related outcomes (eg, compliance; 9/315, 2.9%), and caregiver- and family-related outcomes (eg, caregiver stress; 3/315, 1%) were less commonly examined (23/315, 7.3%). Other outcomes (eg, documentation time, economic evaluations; 46/315, 14.6%) were measured.

Among digital health interventions, researchers reported usability issues with the technology because of patients’ physical condition [177], patients’ or providers’ lack of technical skills and experience [159,176,177,202], and the technology not fitting into patients’ or providers’ routine and workflow [119,173]. Regarding fit with workflows, researchers emphasized that health care providers operate within regulated environments; ethical, legal, and professional considerations related to providers’ roles and care settings had to be accounted for in the design of digital health technology interventions [148,159,227]. Technical issues such as internet connectivity issues, software updates, or immaturity of the prototype [148,152,177] also decreased usability and interfered with the technology’s function (eg, restricted data transfer and alert failure) [152,202]. In addition, some researchers encountered compatibility issues with older devices and other organizations’ electronic documentation systems, which created usability issues [44,145,148,150,192,210,212].

In terms of the technical content, researchers found that some participants perceived the patient-facing content as problematic because of the technical language, tone, or complexity [152,162], as well as hypertext links that caused confusion [39,152]. Patient-initiated technology functions also presented a challenge because not all participants could use the functions or follow the instructions as intended [196]. Researchers also found it challenging to set alerts that would be appropriate for all patients because excessive alerts caused “alert fatigue” and resulted in less attention being paid to the alert or ignoring it altogether [40,44,152,157].

A few researchers emphasized that designing digital health interventions to address and improve multiple components of the transition process may enhance functionality [119,145]. In addition, they indicated that technology accessibility, adaptations, or customization could accommodate individual preferences and increase applicability to different populations [39,140,205,227]. Researchers indicated that increasing accessibility could start with providing the technology to participants without a personal device to reduce disparities of access based on technology ownership [150]. Researchers suggested using participants’ devices to enhance usability when possible, which may require compatibility across multiple data and operating systems [122,167,212]. In addition, technical setup and training on using the technology and engaging caregivers in the intervention could support the usability and intervention quality, safety, and adherence [112,135,157]. Moreover, building self-directed functions might help overcome logistical barriers associated with scheduled interventions [136]. This finding extends to timely feedback because researchers found that participants wanted to be notified when providers had reviewed their responses [150].

To address the technology’s fit with workflows, workloads, and buy-in, some researchers believed that participants, family, caregivers, and providers should be engaged in helping to design the intervention [136,148,228]. Researchers found that interventions that placed high accountability and responsibility on health care providers and added additional work to their workload resulted in provider-related usability issues because providers “struggled to find time in their day” to carry out intervention activities [40,150,195]. Researchers noted that identifying functions rather than provider roles may enable the technology to accommodate differences among jurisdictions (eg, country and institution) and changing scopes of practice over time [227]. Alternatively, if human resources are limited, interventions using automated telephone calls or central monitoring centers for multiple institutions could be considered to reduce the number of personnel and time required for monitoring [188]. Thus, understanding how technology could be integrated into the day-to-day practices of all stakeholders was an essential task for technology developers, along with helping providers envision ways to implement the technology in practice [119,181].

Recruitment challenges and high attrition were commonly reported within the studies [135,246]. As several interventions had a small sample size, researchers acknowledged limitations, including being unable to explore the relationship between participants’ profiles and adherence and compliance information or conduct subgroup analyses [112,135,157,205]. Researchers reported that sampling bias could have had an impact on the generalizability of their results because the samples were small [115,141,165] and homogenous (ie, primarily White) [150] and could have been exacerbated because inclusion was limited to participants with internet-enabled devices [201]. Missing data was another concern reported by researchers that may have affected the reliability of the intervention results [158].

Beyond data collection, researchers reported that interventions conducted at a single site may have reduced generalizability to other settings [160,181,218,226,241]. The study by DeVito Dabbs et al [140] indicated that outcome measures such as rehospitalization and survival may not be sufficiently sensitive to identify the impact of a technology intervention.

Researchers found that effectively integrating the technology in clinical environments would likely require early engagement with patients and providers, support from senior leadership, integration within existing electronic systems [119,144,148,166], and testing of technologies in real-world settings to identify implementation barriers [140]. Finally, researchers of a digital health technology intervention that operated across settings or institutions reported challenges with coordination among providers in hospital, primary care, and community settings [148].

Several researchers recommended more extensive and diverse participant samples in future digital health interventions [150,171,173,247] and consideration of low compliance rates within sample size calculations [173]. They believed that providing participants with an explanation of the potential benefits and utility of the technology may also enhance study participation [173]. In addition, comparing the characteristics of participants with those of individuals who declined participation gave researchers insight into selection bias and the intervention’s generalizability and acceptability [151,169,231,237].

Regarding outcomes of interest, researchers advised carefully considering which outcome measures (eg, objective or subjective) [173,224] and end points to use [186,193]; multicenter studies with longer follow-up time (ie, >30 days after discharge) might be required to observe the intervention’s effect on patient-clinician relationships [160,181]. Opportunities identified by researchers to improve data include analyzing technology log data for objective data on patients’ and providers’ use of technology [186,193], beginning the intervention within the hospital setting, and incorporating the follow-ups into routine patient visits to potentially minimize data lost during follow-up [141,148].

This review demonstrates that many existing technologies that support hospital-to-home transitions encounter implementation-related barriers. The health care system is complex, and the discharge process is often “busy, rushed and emotional” [249]. During hospital-to-home transitions, patients move from one setting to another and provider functions and responsibilities become unclear because communication often fails to cross boundaries [250]. Thus, a critical lesson from this review is that digital health interventions should emphasize the provider functions that the technology supports rather than focusing on how professional groups can use solutions because roles and functions can differ by organization and care setting.

We have highlighted that many providers involved in transitions tend to have overlapping functions. We have outlined specific provider functions that could be built into the design of digital health interventions to support transitional care workflows and potentially reduce provider burden. These functions may address the factors presently limiting uptake of digital health interventions, including poor fit with providers’ functions and provider perceptions of low degree of usefulness [41,42,248]. To meaningfully support hospital-to-home transitions, digital health interventions may need to address multiple functions involved in patient care beyond primarily supporting functions related to management continuity (eg, monitoring) and informational continuity. On the basis of the findings from this review, technology functions related to relational continuity warrant further exploration. These are the components that are appropriate for technology to address and support and the ones that rely on the interface between people and technology.

Moreover, technologies should be designed to minimize the burden on providers and be designed in such a way that they can support provider functions. Although technologies demonstrate their ability to support specific provider functions such as remote monitoring and patient education, they add functions and place high levels of accountability on single providers. For instance, remote monitoring technologies could yield large quantities of data that providers then become responsible for sorting through and acting on, adding another function to their workload [120,153]. Integration of such technologies in clinical practice could be unfeasible because the added provider functions are among the prominent barriers to the uptake of technologies [41].

Perceived usefulness may be improved by highlighting how the purpose and function of the technology fit with the functions of providers during hospital-to-home transitions and whether it could result in time savings and the workload reduction of providers and by outlining the responsibilities of providers in the delivery of digital health interventions [140]. Furthermore, as technologies integrate more advanced and automated functions, the burden on providers may be reduced. For instance, automated reminders may reduce demands on providers [34]. However, advanced technologies may not be suitable for all patients and these individuals may require training to recognize red flags and when to re-engage with providers [152]. Sharing responsibility with, and facilitating more active involvement of, patients and caregivers (when appropriate) or adding trained volunteers may be another way to reduce the added responsibility faced by health care providers [251].

Older adults have unique transitional care needs that the providers strive to meet through their functions. Provider functions to achieve relational and informational continuity of care have been deemed necessary to achieve high-quality hospital-to-home transitions for older adults [22,252]. However, we identified these functions to be a gap in the existing digital health interventions supporting hospital-to-home transitions because these functions were least supported by technology. We believe that these should be integrated within technology functions of future digital health interventions.

Of note, this review revealed that digital health interventions were rarely designed to meet the unique needs of older adults or exclusively tested with older adults. Thus, we contend that future technology functions should be designed to meet these specific transitional care needs while also accounting for design considerations related to older adults’ complex needs, including physical, cognitive, and sensory needs [253-255]. Moreover, new strategies may be needed to recruit and retain older adults with poor health status. Using human-centered design principles, including co-designing and testing with clinicians and older adults with complex care needs, may enhance the use and effectiveness of interventions [41,248] and could reveal how better to integrate relational management into the technology functions. Furthermore, critical investigations of how the functions of existing digital health interventions may have contributed to the exacerbation of inequities are necessary to highlight new insights and guidance for functions of future interventions to eliminate such disparities [255,256].

We recommend that those leading digital health interventions examine outcomes of interest at multiple levels, including the patient, provider, organization, and system levels. Most transitional interventions examined the impact of digital health interventions on patient-level outcomes. However, not all studies had examined why participants declined or dropped out of digital health interventions, which would have provided valuable insights for future work. Provider-, organization-, and system-level outcomes were less common but are essential to consider. Although patient-level outcomes are helpful, costs and benefits need to be assessed for health care organizations and health systems, including economic feasibility and quality measures [257]. In particular, exploring patient-level outcomes can provide insight into whether the technology functions effectively support the provider functions.

Moreover, evaluating specific technology functions may provide insights into which ones may need to be refined. Researchers may also further explore the feasibility and benefits of transition-specific roles to support digitally enabled transitions because these studies were limited. In addition, reporting research-level outcomes, including insights and reflections from the research teams, may contribute valuable knowledge that could guide future interventions.