Patients were recruited from VA Cleveland Medical Center outpatient clinics between June 2009 and January 2012 and were followed for 12 months. To be eligible, patients had to have an HF diagnosis, New York Heart Association classification of II or III, and a documented ejection fraction <40% (see Multimedia Appendix 1). Patients also had to have attended at least one VA outpatient visit within the previous 12 months, have a VA primary care provider, and be able to participate in automated telephone calls in English. Patients needed to nominate an eligible CarePartner, that is, a relative or friend living outside their home. Patients were excluded if they lived in a skilled nursing facility; were prescribed oxygen supplementation; were receiving palliative care; had a life-threating condition such as lung cancer; or had ICD-9 coded diagnoses indicating dementia, bipolar disorder, or schizophrenia.

Potentially eligible patients identified from electronic medical records were sent an invitation letter, followed by a screening and recruitment call. Eligible and interested patients were mailed informed consent forms and were assisted in identifying potential CarePartners using the Norbeck Social Support Questionnaire (NSSQ) [39]. To be eligible, CarePartners had to live outside the patient’s home, speak English, have access to a telephone and email, and report at least monthly contact with the patient. CarePartners provided verbal consent to participate.

After completing baseline surveys, patient-CarePartner dyads were randomized by a research associate to a patient-focused mHealth service (standard mHealth) or a service that included feedback to patients’ CarePartners (mHealth+CP). Pairs were randomized within strata defined by whether the patient had an in-home caregiver. Sealed randomization envelopes were created by the study coordinator in blocks using an online random number generator. It was impossible to blind patients to their random assignment because patients and CarePartners were aware whether the CarePartner received email feedback.

Patients, CarePartners, and in-home caregivers (when present) randomized to standard mHealth were mailed information about HF self-care [37]. Patients received weekly IVR monitoring and self-management support calls for 12 months. Up to nine call attempts per week were made at times the patient indicated were convenient. IVR calls included recorded information and questions that patients answered using their touchtone keypad. The IVR calls were developed by a panel including primary care physicians, cardiologists, nurses, and experts in health behavior change and mHealth. Calls lasted roughly 10 minutes and followed a tree-structured algorithm to ask about overall health, HF symptoms, and self-management behaviors. Patients received pre-recorded information tailored to their reported symptoms and self-care practices. See Figures 1 and 2 for screenshots of the website used for enrollment and call scheduling.

When patients reported an urgent issue via IVR (ie, worsening shortness of breath or a significant weight increase), the system automatically issued a fax notification to their clinician. A significant weight increase was defined as a 5-lb increase over 1 or 2 weeks, a 7-lb increase over 3 weeks, or an average gain of 2 lbs per week since the last automated call if more than 3 weeks had elapsed. Actions taken by clinicians based on the faxes were not tracked.

The mHealth+CP intervention was based on self-regulation theory, which emphasizes communication of expectations of behavior (“standards”), promotion of motivation to meet standards, and monitoring with feedback regarding the gap between behavior and standards [40,41]. Patients and CarePartners randomized to mHealth+CP received identical intervention elements described above.

mHealth+CP CarePartners were automatically emailed a structured report after each completed IVR call. CarePartner reports were sent to their personal, individual email addresses, which were stored in the system’s secure database at the University of Michigan. Reports described in lay language what patients’ responses meant in terms of risk for HF exacerbations and included suggestions for how CarePartners could support self-management. Email reports referred to the patient using gender-specific pronouns, for example, “Your partner did not weigh himself last week”, but were otherwise de-identified. Reports included feedback about the patient’s most recent issues as reported during their IVR call, including shortness of breath, medication adherence, salt, and fluid intake, and increases in weight. CarePartners were asked to call their patient-partner weekly to review the reports and address identified problems.

CarePartners received guidelines about how to communicate in a positive motivating way, avoid conflict by respecting boundaries, include in-home caregivers, and respect confidentiality. Patients received a notebook including reminders and tips for their weekly patient-CarePartner calls. CarePartners received logbooks for tracking IVR reports, upcoming patient contacts, clinical encounters, and medication refills.

The sample included all patients with 12-month surveys plus 22 patients for whom 6-month survey data were carried forward. Initial analyses compared the baseline characteristics of patients who did versus did not have 12-month data in the imputed sample. Subsequent analyses compared patients and CarePartners across arms in the sample with outcome data. IVR call completion rates were calculated using one record per week of attempted IVR calls, that is, 52 call-weeks per patient minus weeks in which the patient was on vacation or hospitalized. Logistic models were used to predict patients’ likelihood of completing each weekly call as a function of arm, baseline characteristics, and the number of weeks since enrollment. Statistical tests for the analyses of call completion rates were adjusted for clustering of call-weeks within patients.

The primary outcome was change in HF-specific quality of life between baseline and 12 months. The study was powered to detect a medium/small effect (d=.351) assuming a 20% loss to follow-up, similar to that observed in the prior HF trial by Sisk et al [51]. All outcomes were analyzed on an intent-to-treat basis. The xtmixed and logistic regression commands in Stata version 13.1 [52] were used to identify intervention effects on patients’ HF-related quality of life, self-care, and patient-CarePartner communication. Predictors for each analysis included an indicator for arm, time (baseline, 6-month, and 12-month), and an arm-by-time interaction. Effect estimates represent differences across arms adjusted for baseline values. To examine differences across arms in IVR reports of medication adherence and symptoms, graphical displays were created illustrating the proportion of patients reporting a given outcome each week, separately by arm. Logistic regression models were fitted to predict patients’ weekly IVR-reported outcomes, with weekly reports clustered within patient. Models included the following predictors: arm, time, and an arm-by-time interaction term. Variances for the estimated intervention effects were adjusted for the within-patient correlation of IVR reports across weeks [53-55]. To illustrate the magnitude of intervention-control differences in IVR-reported outcomes, the probability for each outcome at week 26 and 52 was predicted based on the logistic model separately for mHealth+CP and standard mHealth groups.

Among patients with chronic medical problems, depressive symptoms may influence their perceived health status even more than objective symptoms and impairments resulting from their medical condition [56]. Depressed patients often attribute their difficulties to insufficient social support [57,58], and CarePartners’ support may have counteracted their tendency to over-generalize health problems [59-61]. To test this hypothesis, we examined a potential interaction between patients’ baseline level of depressive symptoms (CES-D) and arm, with respect to IVR reports of excellent/very good health. Specifically, patients’ unadjusted frequency of reporting excellent/very good health was examined graphically as described above, within subgroups defined by baseline CES-D scores. Because graphical displays suggested an inflection point with two very different slopes, we fit logistic models separately for patients with CES-D scores that were low (0-4) versus high (5-10). Each model included terms for arm, baseline CES-D score, an arm-by-CES-D interaction, time, and an arm-by-time interaction.

The study protocol was approved by the Ann Arbor VA Human Subjects Committee, and all patients provided written informed consent. Patients and CarePartners received US $20 for completed surveys; patients did not have financial incentives for completing IVR calls. None of the authors had any financial conflict of interest.

See Multimedia Appendix 4 for the CONSORT-EHEALTH checklist [62].

A total of 4140 potentially eligible patients were identified from electronic medical records. Of these, 372 were randomized, and 331 (89%) had outcome data at 12 months (see Figure 3). Patients lost to follow-up were less likely to report at baseline that they spoke with their CarePartner at least twice per week (43.9% versus 65.9%, P=.006) and had better baseline HF self-care scores as measured by the HFSCB (P=.002) but were not significantly different from patients with follow-up data on any other characteristic shown in Table 1 (see also Multimedia Appendix 5).

Patients in both arms had similar baseline characteristics, except that mHealth+CP patients were more likely to have a high school education or less (Table 1). Education was included in outcome analyses as a control variable, although analyses not including education as a covariate produced essentially the same results. There were no significant baseline differences across arms in measures of patient-CarePartner communication or in CarePartner characteristics. As expected in a VA population, most participants were male. Patients were on average 67.8 years of age (SD 10.2), 77.0% (255/331) were white, 48.0% (159/331) had a high school education or less, 32.6% (108/331) lived alone, and 87.6% (290/331) were retired or unemployed. While most patients (65.9%, 218/331) reported at baseline that they talked with their CarePartner by phone at least twice per week over the prior 6 months, 44.8% (147/328) reported regularly experiencing one or more negative emotions during those conversations, and 21.5% (71/331) agreed that it was difficult to talk with their CarePartner about their illness. Compared to patients, CarePartners were younger, more likely to be female, more likely to be employed, and had more years of education. A total of 41.4% (137/331) of CarePartners were the patients’ daughters/daughter-in-laws, 20.2% (67/331) were sons/son-in-laws, 11.2% (37/331) were friends, 9.1% (30/331) were sisters/sisters-in-laws, and the remaining 18.1% (60/331) were other family and social network members.

Patients participated for a total of 15,709 call-weeks, during which they completed 14,175 calls, for a completion rate of 90.2%. IVR completion rates were essentially the same between mHealth+CP and standard mHealth arms (90.8% versus 89.7%), and there was no change in patients’ likelihood of completing IVR calls throughout follow-up (P=.19). The likelihood of call completion was unrelated to patients’ baseline HF-specific quality of life (MLHFQ) scores, HF self-management scores, CES-D scores, or measures of patient-CarePartner relationship quality (all P values ≥.15). IVR calls generated fax notifications to clinicians 1606 times (11.3% of completed calls), including 743 for weight gain, 774 for shortness of breath, and 89 for both problems. There were no differences in the number of fax alerts to clinicians between arms (P=.52).

In this randomized comparative effectiveness trial, no group differences were identified at 6 or 12 months in the primary outcome of HF-specific quality of life or the composite measure of HF self-care. However, a number of potentially important differences in the process and outcomes of care were identified that favored mHealth+CP compared to standard mHealth. For example, in both follow-up surveys, a greater proportion of mHealth+CP patients reported perfect medication adherence, and mHealth+CP patients were consistently more likely throughout the 1-year follow-up to report via IVR that they took their HF medications as prescribed during the prior week. mHealth+CP patients also had a significantly greater decrease in their likelihood of reporting shortness of breath via IVR and were less likely to report clinically significant weight gains.

The strongly negative association between patients’ baseline depressive symptoms and IVR reports of perceived general health that we observed in the standard mHealth group was not apparent among patients who were randomized to mHealth+CP. In particular, patients with more severe depressive symptoms at baseline were relatively likely to make positive self-assessments about their health via IVR if they were in the mHealth+CP arm. This finding (as well as the feedback from CarePartners presented here) is consistent with studies suggesting that social support can have powerful impacts on patients’ well-being over and above the concrete benefits in terms of specific self-management behaviors [58].

These intervention effects represent positive impacts in some of the most fundamental areas of HF-related self-care and morbidity. Medication adherence is vital for HF patients, and poor adherence is a major predictor of acute events [63]. Shortness of breath and rapid weight gain are correlates of patients’ functional decline and used as sentinel events to identify patients at high risk for acute episodes. If these risk factors can be effectively addressed via mHealth services such as this one that focus on increasing caregiver support instead of the use of costly medical services, it would represent a major advance.

It is important to emphasize that these intervention effects were observed in a comparative effectiveness trial, over and above potential changes in health and self-care among patients receiving an active control intervention. All participants identified a CarePartner prior to randomization, and control patients and CarePartners received considerable information about HF self-care and self-management support. Control patients also received weekly IVR monitoring and self-management support calls with feedback to their clinician.

A soon-to-be-published companion paper using survey data from CarePartners in this same trial provides additional positive information consistent with the patient information reported here [38]. Compared to CarePartners in the standard mHealth arm, those randomized to mHealth+CP reported greater involvement in the patient’s medication adherence at both endpoints (both P<.05). mHealth+CP CarePartners also were more likely to report attending the patient’s medical visits at the 6-month follow-up. Importantly, CarePartners reporting the most symptoms of depression and strain at baseline had those symptoms significantly reduced if the CarePartner was in the mHealth+CP versus standard mHealth arm. These CarePartner reports as well as the qualitative feedback from CarePartners reported here suggest that involvement in mHealth information exchange may significantly improve relationship quality and self-management assistance for patients with chronic health problems. The qualitative feedback presented in Table 3 and Multimedia Appendix 3 is particularly interesting—since many mHealth+CP CarePartners volunteered that the intervention served to strengthen their relationship with their patient-partner.

The 1-week reporting interval used for the IVR-based outcome measures may have been more sensitive to intervention effects than the 6- and 12-month surveys. Differences across arms in IVR-reported adherence were consistent over the 12-month follow-up, and improvements in shortness of breath and weight became evident only after several months of program participation. This suggests that the pattern of effects is not the result of biased reporting, which tends to be immediate and short-lived [64]. Also, improvements in health and self-care measured via IVR were consistent with patients’ improved medication adherence reported in both follow-up surveys and with reports of more frequent supportive communication with CarePartners. More generally, reports of health behaviors are more reliable when reporting intervals are brief, avoiding the biases associated with longer periods of retrospective recall [44,45,49,50,65].

This trial had several limitations. It is possible that patients were biased about their medication adherence reporting in order to avoid burden for their CarePartner or conflict in the relationship. However, prior studies have shown that patients’ medication self-reports are highly correlated with objective measures of medication use, especially when the recall interval is short and the measure is designed to identify even mild forms of non-adherence [44,45]. Also, other positive reports from both patients and CarePartners in this trial corroborate patients’ IVR reports of medication adherence when randomized to mHealth+CP. Nevertheless, it would be important to confirm these findings with medication refill data. Similarly, it would be useful to verify patients’ self-reported weights using data-storing electronic scales. Another limitation is that the trial was conducted among VA patients, nearly all of whom were men. Caregiving dynamics differ by patients’ demographic and clinical characteristics, and future studies should determine whether results can be replicated in other populations, including non-VA patients and women. Some important clinical information about participants was not collected during the trial. For example, we do not know whether patients underwent cardiac surgery, resynchronization therapy, or revascularization. While we have no indication that randomization was unsuccessful, and patients in both groups were well matched on a wide range of baseline characteristics, it remains possible that unobserved differences in patients’ clinical status at the time of enrollment may have contributed to the intervention effects observed. Finally, our study had several outcomes measured at two time points, and multiple comparisons may have contributed to the findings. However, results were consistent with the study’s theoretical framework, and significant results were consistently in the same direction, that is, favoring mHealth+CP over standard mHealth.

This comparative effectiveness trial suggests that, although not all outcomes were different across arms at follow-up (notably HF-specific quality of life and a composite measure of HF self-care), providing caregivers with automated updates and guidance on self-care support may enhance the beneficial effects of mHealth for HF patients’ health and self-management. Given increasing numbers of patients with chronic illness and the growing strains on clinical resources, health systems using mHealth approaches should consider creative ways to engage patients’ social supporters to play a more active role. Finally, trials such as this one that include frequent mHealth monitoring may uncover intervention effects that are missed through more intermittent surveys and lengthy retrospective recall intervals.