We conducted a subgroup analysis using data from the MORE-PC randomized clinical trial and manual chart review. Study reporting followed the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines (Multimedia Appendix 1) [16].

The MORE-PC trial took place between March 2022 and January 2023 within an academic health system in Philadelphia, Pennsylvania. Participants in the intervention arm received automated SMS text messages for 30 days alongside standard transitional care management. Participants in the control arm received standard transitional care management. Full details on the study design can be found in the original manuscript [15].

Patients included in the MORE-PC trial were adults who received care within a University of Pennsylvania Health System’s primary care practice and had been discharged home from any regional acute care facility. Patients in the intervention arm who had a hospital revisit (return emergency department [ED] visit or readmission) within 30 days of index discharge were included. We included all patients who met these criteria in our analyses.

This study was considered a part of the larger MORE-PC study, which was reviewed and approved by the University of Pennsylvania Institutional Review Board (protocol number 24-0431). The trial was conducted under a waiver of informed consent based on meeting criteria that both arms receive usual care, SMS text messages involved no more than minimal risk to patients, patients were free to opt out at any time after enrollment, and requiring informed consent would limit external validity of the study [15]. All data were stored on the Penn network to protect privacy. In addition, following completion of chart reviews, patient identifiers were removed for subsequent analyses. Trial participants received no compensation.

The primary independent variable of interest was mHealth engagement, defined as using the MORE-PC SMS text messaging platform or patient portal, identified through MORE-PC data and patient charts. We used data collected on patients’ age, sex, race, ethnicity, insurer, and comorbidities from MORE-PC data files. Data on intervisit health care interactions—mHealth use, phone calls, and clinic visits—were collected from MORE-PC data files and manual chart review. We also assessed 3 primary outcome variables: time to revisit, relatedness of index hospitalization, and revisit and predictability of revisit. Time to revisit was calculated as days from discharge to revisit and was a classified as a continuous variable. Relatedness and predictability were classified as ordinal variables. Relatedness was defined as any connection between index and revisit diagnoses, as determined through chart review. For example, 2 visits for decompensated heart failure would be considered directly related. Similarly, an index visit for deep vein thrombosis and revisit for gastrointestinal bleed after starting anticoagulation would be considered directly related. An admission for pancreatitis and an admission for alcohol withdrawal would be considered indirectly related as alcohol use disorder underlies both presentations. Predictability describes the extent to which a revisit could have been anticipated based on information available to the outpatient team through review of medical records considering issues remaining at discharge and concerns brought up between visits. For example, if a patient was discharged after experiencing a bleeding gastrointestinal ulcer, reported recurrence of blood in the stool at a follow-up primary care visit, then represented to the ED, this would be considered predictable. By contrast, if a patient with an index hospitalization for pyelonephritis who had no noted complications posthospitalization, then represented to the ED for an acute stroke, this would be considered unpredictable.

We developed a template for systematically reviewing and extracting data from electronic health records of the study population (Multimedia Appendix 2) in an effort to minimize bias in this process. We tested the template on 15 records to identify areas of ambiguity. Subsequently, 3 researchers (SL, AA, and KP) reviewed the same 5 records to ensure agreement across data elements. Remaining records were divided among these researchers for data extraction.

Classification of relatedness and predictability were determined through review of index visit, intervisit, and revisit documentation.

We examined patient-level and encounter-level characteristics, comparing those engaged in mHealth and those not engaged. We assessed differences in baseline characteristics and outcome variables across groups using Wilcoxon tests for continuous variables and χ²or Fisher exact tests for categorical variables. For comparing intervisit interactions across groups, we performed 2-sample proportion testing. Patients with missing data for a given variable were excluded for that analysis. Analyses were conducted using SAS Studio (version 3.81; SAS Institute) and R software (version 4.3.2; R Foundation for Statistical Computing).

A total of 496 patients in the intervention arm of the MORE-PC trial had a hospital revisit within 30 days of discharge. Of these, 296 (60%) were female, 217 (44%) were aged ≥71 years, 283 (57%) were Black, and 315 (64%) were insured by Medicare (Table 1). A total of 185 (37%) patients engaged with mHealth prior to their revisit.

Patients who engaged with mHealth were younger (n=47, 26% aged <50 years vs n=41, 14% of those not engaged; P=.004 across age groups) and more likely to have commercial insurance (n=43, 23% vs n=35, 11%; P=.01). Patients in both groups had similar rates of ED visits and hospitalizations in the year prior to index admission (ED visits, mHealth users vs nonusers: 3.08 vs 3.89, P=.45; hospital admissions: 1.66 vs 1.63, P=.88).

Across both groups, hospital readmissions were nearly twice as common as standalone ED returns (117/185, 63% of revisits among mHealth users vs 185/311, 59% among nonusers). Revisits were more predictable among patients who engaged with mHealth (105/185, 56% highly or somewhat predictable vs 152/311, 49% among non-mHealth users; P=.04). A similar proportion of revisits were directly or indirectly related to the index admission among mHealth users and nonusers (98/185, 53% vs 173/311, 55%; P=.09). mHealth users and nonusers had similar rates of ED referrals from their outpatient teams (66/185, 36% vs 94/311, 30%; P=.21). On average, those engaged with mHealth had more time between their index discharge and revisit (15.2 vs 11.3 days; P<.001).

Among mHealth users, similar numbers of patients used SMS text messaging versus the patient portal (111/185, 60% vs 107/185, 58%; Figure 1, Multimedia Appendix 3). mHealth users were more likely than nonusers to engage with practices via non-mHealth modalities (telephone: 54% vs 44%, P=.03; clinic visits: 112/185, 61% vs 131/311, 42%; P<.001). Both groups had similar rates of home care visits (72/185, 39% vs 114/311, 37%; P=.60).

We examined patient, hospitalization, and intervisit health care interaction features among patients enrolled in an mHealth program, MORE-PC, with 30-day hospital revisits. We found that despite active outreach offered by the MORE-PC program, a minority of patients in the intervention arm used mHealth to communicate with outpatient providers before returning to the hospital. Notably, the parent MORE-PC trial found no significant reduction in 30-day readmissions, suggesting that the presence of an mHealth outreach program alone may be insufficient to reduce revisits without sufficient patient engagement. mHealth users tended to be younger and more likely to have commercial insurance. Predictability of revisits was high among those who used mHealth, and time from index discharge to revisit was longer as compared to those who did not use mHealth. In addition, those who used mHealth were also more likely to use other means of communication with outpatient providers.

Similar to our study, previous research has shown disparities in mHealth use by age, socioeconomic status, and insurer [17,18]. One potential explanation for the differences in usage by age may be that older patients have more difficulty using technology or be less inclined to adopt new technology [19,20]. Insurance and socioeconomic differences likely go hand in hand, and may be associated with mHealth use through mechanisms such as education or affordability of technology [18]. Interestingly, we did not find significant differences in mHealth use by race or ethnicity, though prior work has shown disparities by race and ethnicity [17,18]. With future development and expansion of mHealth interventions, it is crucial to consider ways of ensuring equitable implementation of mHealth interventions given early evidence of disparities across age, racial or ethnic, and socioeconomic groups.

Unsurprisingly, we found higher levels of revisit predictability among those who used mHealth. Though expected with greater communication with practices, this is encouraging as it suggests increased awareness of issues prompting a revisit. Though predictability may not have any bearing on preventability of revisits, timely identification of areas for potential intervention may help to avert a subset of revisits. In line with this, we found time from discharge to revisit was, on average, longer for mHealth users, suggesting mHealth use may delay but not prevent revisits. In line with previous work, we found no major difference in relatedness of index hospitalizations and revisits [21].

mHealth users interacted with outpatient teams by telephone and clinic visits at higher rates than nonusers. This raises concern that mHealth offerings may be additive, rather than substitutive, to traditional communication modalities, benefitting engaged groups more than reaching disengaged groups. For example, although it appears promising that mHealth use is associated with increased predictability and prolonged time to revisit, it is possible that this is more reflective of patient engagement than true advantages of mHealth. Previous research has explored such challenges, highlighting engagement of diverse patients for development of health applications and usage of community health workers to encourage uptake, though more work is needed in this area [22,23].

First, this study included patients within a single health system, and thus results may not be generalizable. Second, chart review methods may be subject to bias. We created data extraction templates and reviewed an initial sample as a group to try to minimize discrepancies across reviewers but recognize this bias cannot be entirely eliminated. Third, we categorized mHealth engagement as responding to SMS text messages or sending portal messages, but definitions of mHealth vary, and it is possible that those who engage via SMS text message may be different from those who engage via portal. Moreover, defining engagement through a binary variable may not capture meaningful differences across a spectrum of mHealth engagement. Finally, this study focuses on a single sample of mHealth users and nonusers with 30-day revisits but does not capture items such as patient satisfaction or strength of provider handoffs that may also be used to define success mHealth in transitions of care.

In this secondary analysis of patients enrolled in an mHealth intervention who experienced 30-day revisits, mHealth engagement was limited. mHealth users were more likely to be younger, commercially insured, and to interact with their practices by traditional modalities (telephone and clinic visits), raising concerns about equity of mHealth interventions and their potential to increase disparities. We also found revisits among mHealth users occurred farther from their index discharge and were considered more predictable, suggesting potential benefits of mHealth engagement. Further research is needed to better define meaningful engagement with mHealth, identify outcomes most important and relevant to mHealth use, and understand how to leverage mHealth to equitably engage patients during transitions of care.