Standardized search strategies, eligibility evaluations, and data extraction procedures were used (detailed below and in Multimedia Appendix 1). This review was registered with the International Prospective Register of Systematic Reviews (PROSPERO CRD42021224407), in accordance with PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analysis) guidelines (Checklist 1).

As this was a systematic review, institutional review board approval was not required.

A literature search in the PubMed, Scopus, Embase, CENTRAL, Web of Science, and PsycINFO databases identified potentially relevant articles published from 2015 through 2020, to enhance relevance to current technology. Given our group’s focus on CF, 2 categories of search terms were used: “CF-specific” and “other chronic conditions,” which included conditions identified by the study authors as having similar adherence and self-management characteristics to CF (eg, conditions with complex daily medical regimens and diseases often diagnosed in childhood, thus involving caregivers in self-management tasks).

Peer-reviewed, English language articles published between 2015 and 2020 reporting original empirical findings of mHealth interventions for selected medical conditions and targeting adherence and self-management were included. The mHealth interventions must be accessed on a mobile device (smartphones, cell phones, or tablets, including internet browser programs) and used by a person managing a medical condition or their caregiver.

Study records were compiled in a database; duplicates were removed based on DOI number or title. Reviewers (CKE, E McWilliams, DY, TS, and Brandi Blackshear) evaluated each study record (title and citation; blinded double review) for eligibility criteria. The reviewers screened studies for final inclusion and data abstraction using a REDCap (Research Electronic Data Capture; Vanderbilt University) [21,22] form developed for this study.

Statistical analyses were conducted in Stata 15 software (StataCorp LLC). Abstracted data were summarized using frequencies and percentages. Subgroup analyses examined differences in study characteristics, including BCTs used, age group (adult only [≥18 years and older] vs adolescent and young adult [11-25 years or sample characterized by authors as “adolescents and young adults”]), study design (RCT vs non-RCT), and whether the intervention was theoretically derived. We also conducted an exploratory subgroup analysis to examine which BCTs appeared the most often in interventions explicitly focusing on medicine taking compared to interventions focusing on self-management and other adherence targets. The results highlight BCTs (1) appearing in ≥5% of studies and (2) with a difference of >10% between positive effects versus no impact on adherence and self-management outcomes. This does not mean that rarely used BCTs are ineffective or that 10% is a verified benchmark of clinically meaningful difference. This pragmatic decision supported the interpretation of a large number of BCTs and comparisons. Statistically significant (P<.05) differences in the number of BCTs based on the direction of results were tested using 1-way ANOVA. No effect measures, missing summary statistics or data conversions, or meta-regression were used for this systematic review.

Figure 1 presents this review’s PRISMA diagram. The initial search returned 14,889 articles. After removing duplicates and clinical trial registrations, 7400 titles were screened for initial eligibility, 303 articles were potentially eligible, and 122 manuscripts met the criteria for data extraction (see Table S1 in Multimedia Appendix 1 for all included studies and characteristics).

The most represented medical conditions were type 1 or 2 diabetes and asthma (Table 1). Only 6 (4.9%) out of 122 studies involved people with CF. Most studies were published in 2020 (32/122, 26.2%), were conducted outside of the United States (64/122, 52.5%), and used an RCT design (75/122, 61.5%). Nonrandomized study designs were primarily observational pre-post (23/122, 18.9%), observational without pre-post measurement (6/122, 4.9%), or mixed methods (6/122, 4.9%) studies. Study sample sizes ranged from 10 to 14,085 (median 92, IQR 44-179) participants. Most studies involved adult-only (81/122, 66.4%) or adolescent and young adult–only samples (22/122, 18%), followed by child, adolescent, young adult (11/122, 9%); child, adolescent, young adult, and adult (6/122, 4.9%); and child-only (2/122, 1.6%) samples.

Adherence and self-management outcomes were typically evaluated with objective health outcomes or subjective behavior measures (61/122, 50% for each; Table S2 in Multimedia Appendix 1). mHealth interventions were delivered via app (75/122, 61.5%), SMS text messaging (34/122, 27.9%), or website (30/122, 24.6%). Nearly all studies presented mHealth tools used by patients (118/122, 96.7%), but many included health care providers (48/122, 39.3%) or caregivers (21/122, 17.2%). Only 22 (18%) interventions were clearly informed by scientific theory. mHealth interventions most often targeted taking medication (68/122, 55.7%), diabetes self-management activities (46/122, 37.7%), dietary recommendations (32/122, 26.2%), exercise and physical activity (27/122, 22.1%), asthma self-management activities (13/122, 10.7%), managing disease activity and symptoms (11/122, 9%), and general “self-care” behaviors (5/122, 4.1%). One (0.8%) study targeted airway clearance therapy.

Study results were characterized as having a positive effect on the outcome or outcomes (59/122, 48.4%), followed by mixed results (40/122, 32.8%) or no impact (23/122, 18.9%). No studies were characterized as having negative effects (Table 1). Comparing studies reporting positive effects to no impact, 34% (20/59) of the positive studies used objective behavior adherence measures compared to 13% (3/23) of no-impact studies (Table S2 in Multimedia Appendix 1).

Across all reviewed studies, 32 different BCTs were used (mean 4.30, SD 2.32). Table S3 in Multimedia Appendix 1 provides the frequencies, definitions, and examples of BCTs appearing in ≥5% of reviewed studies.

Table S1 in Multimedia Appendix 1 reports each study’s risk of bias rating. No study was excluded due to bias rating. For RCTs, 57% (43/75) received an overall risk of bias rating of “Some concerns,” 39% (29/75) had “High” concerns, and only 4% (3/75) had “Low” concerns. “High” concern ratings were generally due to deviations from the intended interventions (18/29, 62%), the randomization process (11/29, 38%), or missing outcome data (11/29, 38%). For nonrandomized studies, 82% (36/44) received an overall risk of bias rating of “Serious” concerns, 9% (4/44) had “Critical” concerns, and 2% (1/44) had “Moderate” concerns. No nonrandomized study had “Low” risk of bias. “Serious” or “Critical” ratings were generally due to confounding (38/40, 95%) or deviations from intended interventions (13/40, 33%).

Our literature review of mHealth adherence and self-management interventions returned 122 studies, from which we identified discrete behavioral strategies using the BCT Taxonomy [6] with promise to promote adherence and self-management for people living with medical conditions requiring complex, daily self-management activities. The BCT Taxonomy provides, to date, the most rigorously tested, standardized method to identify cross-cutting BCTs with potential applicability across chronic medical conditions with overlapping adherence and self-management demands. The BCT Taxonomy also helps compare mHealth interventions and provides a shared language about BCTs for clinicians, researchers, mHealth innovators, and other key stakeholders such as patients and caregivers. As technological advances can quickly outdate mHealth, focusing on BCT principles, rather than the technology to deliver them, enhances the research’s relevance and potential generalizability to a range of complex medical conditions, including rare diseases (an area of focus for our group), which often have significant need for such tools in contrast to the finite resources available to conduct large-scale, multistep mHealth design and evaluation studies.

Consistent with prior research [25,26], using more BCTs was associated with improved adherence and self-management. However, 6 BCTs appear particularly promising: self-monitoring of behavior, self-monitoring of outcomes of behavior, feedback on behavior, feedback on outcomes of behavior, credible source, and goal setting. Self-monitoring of behavior and outcomes of behavior involve tracking health behavior engagement (eg, logging in an app when medicine is taken) or outcomes of behavior (eg, using a Bluetooth-enabled glucometer to monitor blood glucose levels), whereas feedback on behavior and outcomes of behavior involve providing users with a summarized interpretation of the tracked data (eg, providing in-app graphical representations of one’s daily step count over the past month). Consistent with our results, a prior meta-analysis showed that monitoring medication adherence and providing feedback improve medication adherence [27]. These strategies may build awareness for when the mHealth user engages in a health behavior, provide opportunity to reflect on successes and challenges, and ultimately help the user make informed behavior changes. Credible source involves providing expert-generated information about managing the user’s medical condition (eg, the app contains information about etiology, symptoms, and treatment), which presents users with knowledge to understand the condition and its management. Goal setting (behavior) involves setting measurable and attainable goals for a target health behavior (eg, set a goal for number of days to exercise in a month), which can help the mHealth user focus on key health behavior and build self-efficacy as goals are met.

Developmental differences emerged between adult samples and adolescents and young adult samples. In adult-only studies, prompts and cues (reminders) were associated with positive outcomes, consistent with reviews showing that reminders are associated with a 2- to 3-fold increase in adherence [28,29], but they were less effective in adolescent and young adult studies. Indeed, a pre-post study of children and adolescents with CF found that adherence did not change after delivering reminders only (therefore excluded from this review) for 6 months [30]. Adolescents and young adults may benefit from improving knowledge (information about health consequences), improving skills (problem-solving), and building motivation (material reward [behavior]). Given the small number of adolescent and young adult studies, these results and interpretations should be seen as hypothesis generating.

Differences emerged between interventions explicitly targeting medicine taking versus those focused on disease self-management and other adherence targets. In interventions targeting medicine taking, prompts and cues and credible source were associated with positive outcomes. Reminders and expert information may be the most effective when focused on discrete, clearly defined behaviors rather than complex, multicomponent self-management activities. In interventions focused on self-management and other adherence targets, instruction on how to perform the behavior, goal setting (behavior), and action planning were associated with positive outcomes. Over three-quarters (43/54, 80%) of studies focused on self-management and other adherence targets involved people with diabetes or asthma, which are relatively common yet complex medical conditions involving self-management behaviors that extend beyond simply taking medicine. Skills training, behavioral goals, and assistance with creating a detailed plan for managing a complex medical condition may be the most effective for multicomponent self-management activities that may involve monitoring and intervening upon changes in disease activity (eg, managing fluctuations in blood glucose levels for people with diabetes or managing asthma exacerbations) and self-managing lifestyle and environmental considerations (eg, diet in diabetes and environmental triggers in asthma). Careful consideration of the intervention target will likely help to further guide appropriate BCT selection from the BCTs found to be associated with improved adherence and self-management in our review.

This review has limitations. A meta-analysis was not conducted due to heterogeneous outcomes [17-19], thus we could not conclude which BCTs were statistically the most effective. Our risk-of-bias assessment highlighted methodological concerns across the studies reviewed. No-impact studies were more likely to be RCTs, and positive studies were more likely to be nonrandomized, raising concerns about publication bias toward positive results irrespective of study quality. We excluded reminder-only interventions; thus, most studies incorporated more than 1 BCT. Our reported average number of BCTs is likely higher than that of all adherence-promoting mHealth interventions. Although we identified some BCTs that may be effective, others may be as or more effective in supporting disease self-management but were rarely used in the reviewed studies. Moreover, no BCT was found to do harm. Thus, mHealth innovators should continue to integrate and evaluate how a wide variety of technology-delivered BCTs may support people living with chronic diseases, including rare diseases such as CF. An inherent limitation of conducting literature reviews is that a cutoff date must be selected, yet scientific literature is constantly being published; there may be utility in conducting an updated systematic review of this topic in the future. We only included studies that were published in peer-reviewed journals to focus on interventions with clear evidence of scientific evaluation; however, expanding our review to “gray literature” may have provided more insight into the most current interventions and reduced publication bias. Our review characterized mHealth BCTs generally. Other metrics including digital literacy and socioeconomic barriers to mHealth were not evaluated. Future researchers should evaluate these factors to support sustained mHealth use among diverse audiences. Additionally, BCTs were analyzed across the included chronic medical conditions given the disproportionate number of studies in diabetes and asthma compared to other medical conditions. Although the BCTs are disease agnostic, intervention developers and researchers should carefully consider the applicability of the BCT to the target patient population.

Our review identified discrete BCTs that may have broad cross-cutting applicability across chronic diseases with complex medical regimens, including people with CF, the community with which our group primarily works with. We consider our systematic review approach to be a model for gathering key findings from the extant scientific literature to inform the development of multicomponent behavioral mHealth interventions tailored for a patient population that may be smaller and with less existing research, yet has significant self-management needs warranting further research, such as CF [9,10,31,32]. Research involving people with chronic medical conditions following complex treatment regimens should prioritize the design and evaluation of mHealth interventions incorporating cross-cutting, evidence-based, and age-appropriate BCTs to promote adherence and self-management. Such an approach could help accelerate mHealth intervention design and evaluation to create effective products that may be efficiently disseminated to communities with significant need for such tools.

Accelerating mHealth design and evaluation by taking a cross-cutting approach to BCT selection would also help answer remaining “unknowns” about mHealth BCTs and strengthen mHealth intervention quality. For example, although interventions including more BCTs appear to have greater benefit, the optimal number, type, and combination of BCTs to include in mHealth interventions have not been determined. For BCTs demonstrating potential to promote adherence or self-management, the ideal delivery method must be determined (eg, should the BCT self-monitoring of behavior be delivered via manual data entry of treatment completion or using an electronic monitoring device to automatically track data?). An overrepresentation of certain BCTs (eg, prompts and cues and self-monitoring of outcomes of behavior) and underuse of other, potentially more effective techniques (eg, feedback on behavior and goal setting [behavior]) highlight mHealth’s focus on simpler technologies at the expense of innovation and efficacy. Collaborations between behavioral scientists, care teams, patients, caregivers, and industry could answer these questions and produce mHealth solutions that are transformative and effective.

When incorporating BCTs that are expected to effectively and appropriately generalize to a range of complex medical conditions and associated regimens, mHealth intervention developers must still consider the unique needs of the target population. In CF, for example, highly effect CF transmembrane conductance regulator modulator therapies have the potential to simplify the regimen and reduce treatment burden [33,34]. The implementation of key BCTs may need to be adapted as new therapies roll out, although the core theory behind the BCT itself is not expected to change. It is critical to build the scientific evidence base for effective adherence and self-management mHealth interventions that maintain pace with rapidly advancing medical management across complex medical conditions.