Effective behavioral intervention strategies using mobile health applications for chronic disease management: a systematic review

Searches were performed to retrieve studies that were published in peer-reviewed journals from January 2005 to March 2016, and written in English; the following databases were used: PubMed, CINAHL, EMBASE, the Cochrane Library, PsycINFO, and Web of Science. We used combinations of the key words and indexing terms such as MeSH or Emtree linked to the search domains. An example of a PubMed search strategy is as follows: for mobile interventions, “Mobile Applications”[Mesh] OR “Cell Phones”[Mesh] OR “Computers, Handheld”[Mesh] OR “mobile health” OR “m-health” OR mhealth OR “mobile-health” OR smartphone* OR “smart-phone*” OR “mobile phone*” OR “mobile-phone*” OR “cellular phone*” OR “cellular-phone*” OR “smart device*” OR “smart-device*” OR “tablet* PC*” OR “tablet-based” OR “tablet* device*”; for chronic disease outcomes, “Disease Management”[Mesh] OR “Chronic Disease/prevention and control”[Mesh] OR “Chronic Disease/therapy”[Mesh] OR “disease* manag*” OR “disease* monitor*” OR monitor* OR “health promot*” OR Promot*, and for method, “Randomized Controlled Trial” [Publication Type] OR “Randomized Controlled Trials as Topic”[Mesh] OR “Controlled Clinical Trial”[Publication Type] OR randomized[Title/Abstract] OR randomised[Title/Abstract] OR randomly[Title/Abstract] OR “random* assign*”[Title/Abstract] OR trial*[Title/Abstract]. Then, those three groups of search results were combined with “AND” (see Additional file 1 for search strategy for each database).

The inclusion criteria were as follows: adult patients with chronic diseases (except diabetes) as the target population, an intervention that involved using a mobile application for smartphones or tablets, and assessing the health outcomes and process measures. The exclusion criteria were as follows: studies that focused on a healthy population, pregnant women, non-adults (i.e., adolescents and children), or healthcare providers (e.g., apps for physicians’ or nurses’ use only); studies that used only qualitative methods (e.g., focus groups or group/individual interviews) or simple usability tests; and studies that measured psychological outcomes only (Table 1). Two reviewers (MC and SAL) independently screened titles, abstracts, and full-text articles to decide whether an article was relevant to the review. In case of disagreement, a third person was consulted (JL). We excluded studies of diabetes management because several systematic reviews and integrated reviews have already been published to report the effectiveness of mHealth-based interventions [14‐17]. Only studies published in peer-reviewed journals were included.

Data were extracted from the selected articles and entered into an electronic data sheet. The contents of the data sheet included year of publication, research question or purpose, study design, types of disease, types of outcome and measurement, and the main results. In instances of disagreement, each case was discussed by the authors.

Selection bias (random sequence generation and allocation concealment), performance bias (blinding of participants and personnel), detection bias (blinding of outcome assessment), attrition bias (incomplete outcome data), reporting bias (selective reporting), and other biases (determined according to sample size calculation method, inclusion/exclusion criteria for patients’ recruitment, comparability of baseline data, funding sources, and any other potential methodological flaw that might have influenced the overall assessment) were assessed with the tool for risk of bias given in the Cochrane Handbook for Systematic Reviews of Intervention [19]. For each risk of bias item, the studies were classified as “unclear,” “low,” or “high” risk of bias respectively. Two reviewers (MC and SAL) assessed the trials independently and disagreements between two authors were resolved via discussion.

Studies on feasibility assess whether or not an intervention is appropriate for further testing, whereas studies on acceptability (which is a component of feasibility) determine how recipients react to that intervention [20]. Effectiveness is defined as an intervention study that shows statistical differences of one or more outcomes of interest measured between intervention and control groups. Health outcomes included physiological outcomes (e.g., gait and balance in patients with Parkinson’s disease or fatigue in patients with cancer) and psychological outcomes (e.g., quality of life, depressive symptoms, anxiety). Process measures [21] included participants’ adherence to, satisfaction with, and/or the level of engagement with mHealth systems. These process measures could be assessed via quantitative tools such as surveys or qualitative methods such as open-ended questions or a focus group interviews with intervention participants.

The majority of mHealth RCT-studies in this systematic review (10 out of 12 studies, 83.3%) showed statistically significant effects on health outcomes by incorporating mobile applications in managing chronic diseases. Those studies demonstrated improved physical functioning, adherence to prescribed medications, and/or ease of symptom evaluation and reports to care providers, as well as process measures including patient satisfaction with mHealth management and feasibility of smart-phone-based self-management interventions.

Kearney et al. [26] in the United Kingdom reported significant improvement in fatigue (odds ratio, OR = 2.29; 95% CI, 1.04–5.05; P = 0.040) and hand-foot syndrome (OR control/intervention =0.39; 95% CI, 0.17–0.92; P = 0.031) in patients with lung, breast, and colorectal cancer using a mobile phone-based remote monitoring of chemotherapy-related symptoms in comparison to the usual care group. In Norway, Kristjánsdóttir et al. [22, 23] showed a favorable effect on pain management in a 4-week follow-up (catastrophizing score lower for Intervention, M = 9.20, SD = 5.85, compared to Control, M = 15.71, SD = 9.22, P < 0.01, with a large effect size, Cohen’s d = 0.87) but not in the 5-month and 11-month follow-ups (outcome variables including catastrophizing, acceptance, functioning, and symptom level, all P > 0.1). In Spain, Garcia-Palacios et al. [27] developed an ecological momentary assessment (EMA) for chronic pain in fibromyalgia patients and found that patients with less familiarity with technology using a mobile EMA system via their smartphones showed higher levels of compliance than patients with a paper-based diary (complete record t = − 4.446, d = 1.02, reference Cohen’s d > =0.8, large effect). In Turkey, Cingi et al. [28] reported that patients with allergic rhinitis or asthma displayed better quality of life or well-controlled asthma scores by using the mHealth intervention compared to the control group (all P < 0.05). Dicianno et al. [24] demonstrated the feasibility of a mHealth intervention for patients with spina bifida to improve self-management skills and high usage of the mobile system was associated with positive changes in the self-management skills. In Sweden, Hägglund et al. [29] tested a tablet-based intervention in patients with heart failure (HF) and found improved self-care and health-related quality of life (HRQoL) and a reduction in HF-related hospital days (risk ratio, RR = 0.38; 95% CI, 0.31–0.46; P < 0.05). In Israel, Ginis et al. [25] conducted home-based smartphone-delivery automated feedback training for gait in people with PD, and found significant improvement in balance (F_(2,108) = 3.73, P = 0.04) from baseline to post-test.

Martin et al. [30] in the US found that an automated text message system increased physical activity to prevent cardiovascular diseases in phase 1 (weeks 2 to 3) and phase 2 (weeks 4 to 5), all P < 0.001. Piette et al. [31] in the US reported that the intervention group involving “CarePartners” connecting to a relative or friend living outside their home showed improvement in medication adherence and caregiver communication (all P < 0.05). DeVito Dabbs et al. [32] in the US conducted an RCT for patients with lung transplantation and reported improvement in self-monitoring (OR = 5.11; 95% CI, 2.95–8.87; P < 0.001), adherence to medical regimen (OR =1.64; 95% CI, 1.01–2.66, P = 0.046), and reported abnormal health indicators more frequently (OR = 8.9; 95% CI, 3.60–21.99; P < 0.001).

Two of the twelve studies reviewed (16.7%) showed similar or no effects of mHealth-based interventions on main outcomes of interest compared to control groups. In Finland, Vuorinen et al. [33] found no difference in the number of HF-related hospital days (incidence rate ratio, IRR = 0.812, P = 0.351). However, patients in the telemonitoring intervention group used more healthcare resources; increased number of visits to the nurse (IRR = 1.73; 95% CI, 1.38–2.15; P < 0.001), more time spent with nurses (mean difference = 48.7 min, P < 0.001), and increased number of telephone contacts initiated by nurses (IRR = 5.6; 95% CI, 3.41–7.63; P < 0.001). In Spain, Cubo et al. [34] reported a trend of lower PD functional status (the Unified Parkinson’s Disease Rating Scale, UPDRS I) in patients on home-based monitoring compared to patients in standard in-office visits (P = 0.06), while other outcomes (measured by UPDRS II, III, IV subscales) and HRQoL in PD did not show statistically significant differences between the intervention and control groups. Cubo et al. [34], however, explained that the approach of using home-based motor monitoring via mHealth applications compared to standard office visits in the 1-year follow up was cost-effective (incremental cost-effectiveness ratio, ICER, per unit of UPDRS subscales ranging from €126.72 to € 701.31).

Table 3 presents the details of mHealth interventions including duration, mobile app type, app content, and training methods. A quarter of the studies (3 out of 12) reported the feasibility of the mHealth intervention as a pilot study to assess the potential for successful implementation of the mHealth intervention to patients/participants [24‐26]. The majority of studies used smartphones as a mobile device, two studies [29, 34] used tablets for mHealth interventions, and two studies used telemonitoring wireless devices including weight scales for patients with HF [29] or gait detectors for patients with PD [25]. The length of interventions ranged from two weeks to twelve months; half of the studies (6/12, 50%) had more than six-month intervention periods and follow-ups. Most common components of mHealth interventions included remote symptom monitoring and self-assessment as well as tailored automated messages or self-care education to coach patients with chronic disease conditions that needed active disease management. One particular study by Kearney et al. was more inclusive in that it provided real-time feedback and tailored such feedback for symptom management depending on the severity, offering pharmacological, nutritional, or behavioral advice when needed [26].

The sample size of the studies reviewed was between 28 (patients with spina bifida) and 372 (patients with heart failure). In terms of subjects, studies included patients who were 18 years and above, with some studies limited to a particular age range such as 18–40 [24] or up to 69 years [30]. The mean age of participants in the mHealth intervention studies ranged from 30-year-old patients with spina bifida [24] to 75-year-old patients with HF [29]; the approximate average age group in this 12-study review was in the 50s. None of the studies reported effectiveness of mHealth intervention by age categories. These 12 studies also did not report participants’ prior experience with mobile devices such as smartphones or educational background, which may have affected the ability to use mHealth apps via mobile devices. Outcomes of interventions were measured either by mHealth systems directly or by paper-based questionnaires.

In terms of mHealth intervention training, either face-to-face information sessions at baseline or paper-based instructions were used in most studies. Kearney et al. [26] found that symptoms were reported differently on a paper-based questionnaire and mobile phone. Participants in the mobile group reported lower levels of fatigue compared to those in the paper-based group (OR _{non-mobile/mobile} = 2.29, 95% CI 1.04–5.05, P = 0.04) [26]. Reporting cancer toxicity symptoms (e.g., hand-foot syndrome and mucositis) in real time might allow for more accurate measurement [26].

None of the studies reported process measures, including adherence, level of engagement, and/or satisfaction with the mHealth systems. Potentially, there was limited information on such measures in the published articles with the authors possibly publishing the process measures elsewhere.

The risk of bias in the reviewed studies is summarized in Fig. 2 and shown for individual studies in Fig. 3. The intervention studies generally performed well in their risk of bias for random sequence generation (62% low risk), allocation concealment (23% low risk), blinding of participants and personnel (8% low risk), blinding of outcome assessors (15% low risk), incomplete outcome data (69% low risk), and selective reporting (92% low risk) (Fig. 2).

Considering the method of randomization, 5 trials did not present details of randomization [25, 27, 29, 30, 33]. Only 2 studies reported use of an adequate allocation concealment method [22, 32]. Although due to the nature of the mHealth intervention, it is almost impossible to blind participants and healthcare providers, one study described the blinding of participants by providing the same mobile phone application to both experimental and control groups with different functionalities. The experimental group was given communication, health status, or medication usage tracking, and a survey questionnaire, whereas the control group received only the survey questionnaire [28]. Only two trials were identified as blinding the outcome assessors. Four trials failed to provide a full description of participants and losses to follow-up during their trials [24, 28, 29, 31]. All studies had a low risk for reporting bias. Devito et al. had low risk for all items except blinding of participants and personnel [32]. Hägglund et al. had low risk only for reporting bias [29] (Fig. 3).