This systematic review and meta-analysis were conducted according to the Cochrane Handbook for Systematic Reviews of Interventions [23] and reported following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) [24] (Multimedia Appendix 1) and PRISMA-S (Preferred Reporting Items for Systematic Reviews and Meta-Analyses literature search extension) [25] guidelines. The protocol was registered in PROSPERO (International Prospective Register of Systematic Reviews; CRD42023479314). No deviations from the predefined protocol occurred.

We included studies based on the following criteria: (1) randomized controlled trials (RCTs); (2) adults aged ≥18 years with chronic diseases, defined here as long-lasting conditions requiring ongoing management over time, typically characterized by slow progression and the need for sustained self-care activities; (3) participants recruited in any health care or community setting; (4) “digital health interventions for person,” according to the WHO framework, defined as “the capabilities of digital technology that can be implemented to achieve objectives that are targeted toward persons,” including members of the public who are potential or current users of health services and caregivers [10]. Eligible interventions included mHealth and eHealth approaches (eg, telemonitoring, video consultations, mobile apps, and electronic health records), software-based solution, and emerging technologies (eg, artificial intelligence, big data, and genomics) compared with any type of control condition; (5) self-care abilities, including treatment adherence, measured using validated quantitative questionnaire or instruments; and (6) studies published in English. Studies focusing on patients with cancer, mental illness, pregnant women, or in prisons were excluded.

We searched 4 electronic databases, PubMed (via PubMed), CINAHL (EBSCOhost), Scopus (Elsevier), and PsycINFO (EBSCOhost). We included studies published between January 1, 2013, and December 31, 2025, following the release of 2 key international and European policy documents on digital health in 2012 [26,27]. The search strings are reported in Multimedia Appendix 2. These were developed iteratively based on preliminary scoping searches and refined in consultation with an expert librarian. Additionally, the reference lists of included trials, gray literature, and registries were screened for any additional eligible studies.

All retrieved records were imported into Rayyan (Rayyan Systems Inc), an AI-assisted tool for systematic reviews, which was used to remove duplicates and manage the screening process. Titles and abstracts were independently screened by 2 reviewers (JL and FF). Full texts of potentially eligible studies were assessed independently by 2 reviewers. In this second step, the level of agreement was Cohen κ=0.83 (95% CI 0.71-0.93), demonstrating good interrater reliability. Disagreements were resolved by discussion or consultation with a third reviewer (DP).

Two reviewers (JL and FF) independently extracted the following data using a standardized Excel form, which was piloted in 2 studies, after the screening was concluded. The data were authors, year, objectives, design, country, intervention, comparator, setting, participant characteristics, and outcomes (quantitative and narrative data), including follow-up duration and measurement instruments. No disagreements were detected. We contacted 6 corresponding authors of the included studies to request clarification or additional information when data were missing.

The methodological quality of included studies was assessed using the Cochrane Risk of Bias 2.0 tool for RCTs [28]. Discrepancies were resolved by consensus or by consulting a third reviewer. The quality of evidence for each outcome of interest was independently assessed by 2 reviewers using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) system with the GRADEpro software [29]. GRADE is an internationally recognized and widely used framework that provides a transparent, reproducible, and systematic method for rating the certainty of evidence at the outcome level. It considers 5 key domains, risk of bias, inconsistency, indirectness, imprecision, and publication bias, to determine the confidence that the estimated effect is close to the true effect. By evaluating and combining these domains, the overall quality of evidence for each outcome is rated as high, moderate, low, or very low. Language to report confidence in results according to the certainty of evidence was adherent to the Cochrane Handbook [23].

The primary outcome was self-care measured with a validated questionnaire. Where sufficient data were available, we performed meta-analyses on comparable outcomes, grouped by outcomes and questionnaires, and homogeneity in digital interventions and conditions. We used standardized mean differences (SMDs, Hedges adjusted g) with 95% CIs because different versions of the same questionnaires were used across studies to measure the same outcome. We interpreted an SMD of 0.2, 0.5, and 0.8 as a small, medium, and large effect size, respectively. A random-effects model was used to account for clinical and methodological heterogeneity. Heterogeneity was assessed visually (forest plots) and statistically (I² statistic), with thresholds interpreted as follows: 0%-14% negligible, 15%-29% low, 30%-60% moderate, 50%-90% substantial, and 75%-100% considerable heterogeneity [23]. We planned to conduct subgroup analyses if any significant heterogeneity emerged in the pooled results and sensitivity analyses if high-risk-of-bias studies were included in the meta-analysis. Small-study effects were investigated through the visual inspection of the funnel plots and the Egger test when at least 10 studies were available. Prediction intervals were calculated with at least 5 studies and no asymmetry in the funnel plot [23]. No double-counting of participants occurred in the meta-analyses. Each study contributed only independent data, and the same participants were not included more than once within any single analysis. Analysis was performed with RevMan (version 5; Cochrane) and R software (R Foundation for Statistical Computing) with the package metafor [30].

When meta-analysis was not feasible (eg, insufficient data and heterogeneous outcome reporting), results were synthesized narratively.

We included 55 RCTs [31-85] (Figure 1), of which almost half (n=26, 47.3%) were published in the past 5 years. The sample sizes ranged from 27 [79] to 1571 [39] (Multimedia Appendix 3 [31-85]). Most studies were conducted in the United States (n=11) [33,48-50,58,63,65,69,79,82,84] and in China (n=10) [43,45,47,61,64,70-72,76,77], followed by Iran [44,53,68], and Australia (n=3) [42,59,75] (Multimedia Appendix 3).

The most common chronic conditions were heart failure (n=13) [37,39,42,44,46,50,55,62,72-75,79], type 1 or type 2 diabetes (n=10) [31,33,35,41,45,47,49,53,59,61], and acute cardiovascular diseases, including myocardial infarction or stroke (n=10) [32,34,36,38,51,56,57,64,65,67], representing the most frequently targeted chronic conditions across the included studies.

Among the included studies, 5 involved participants with multiple chronic diseases [48,58,68,77,78], whereas atrial fibrillation was the primary focus in 4 others [40,52,76,82].

Other less frequent conditions included asthma (n=3) [43,60,84], COPD (n=2) [66,71], hypertension (n=2) [63,85], chronic hepatitis B [54], HIV infection [69], and epilepsy [70].

All studies obtained ethical approval and respected the privacy regulations.

Characteristics of interventions and control care are reported in Table 1 and in Multimedia Appendix 3. Across the 55 included studies, digital health interventions for persons fell within the WHO [10] functional categories of “Targeted communication to persons,” “Personal health tracking,” and “Person-based reporting.”

Mobile or web-based applications were the most frequently used technology (n=39, 70.9%), followed by telemonitoring systems (n=24, 43.6%) and connected medical devices such as blood pressure monitors, glucometers, weight scales, electrocardiogram devices, or activity trackers (n=24, 43.6%). Text-message programs, including SMS text messages, WeChat-based messages, and email/SMS reminders, were used in 18 (32.7%) studies, whereas tablet-based apps were reported in 6 (10.9%) studies.

Overall, interventions were predominantly multicomponent, with the most common configuration combining mobile or web-based applications, telemonitoring, and connected devices (n=12, 21.8%). Stand-alone mobile or web-based applications were used in 10 (18.2%) studies, whereas mobile/web-based application plus text-message programs were reported in 8 (14.5%) studies. Telemonitoring combined with connected devices but without apps was used in 6 (10.9%) studies, and text-message–only interventions were also used in 6 (10.9%) studies.

Interventions can also be characterized according to their main behavior change technique categories [86]. The most common category was prompts/cues, operationalized through reminder messages, push notifications, interactive voice response messages, or automated prompts, which were present in 54 (98.2%) studies. Self-monitoring of behavior or outcomes was also frequent and was usually implemented through apps or connected devices, allowing patients to enter or transmit symptoms, medication use, or physiological parameters; active patient interaction was reported in 44 (80%) studies. Feedback on behavior/biofeedback, often delivered automatically or after clinician review of transmitted data, was commonly paired with alerts for symptoms or high-risk values, which were reported in 33 (60%) studies. A further recurring behavior change technique category was instruction on how to perform the behavior together with information about health consequences, usually delivered through educational modules, videos, app-based content, or structured tele-education, and identifiable in 48 (87.3%) studies. Goal setting, action planning, and behavioral reinforcement were observed in 24 (43.6%) studies, primarily in multicomponent coaching or structured self-management programs. Social support features, such as peer interaction, online discussion boards, caregiver involvement, or shared monitoring functions (eg, “medfriend”), were less common and were identified in 10 (18.2%) studies.

These components were delivered at different levels. Phone calls were incorporated in 21 (38.2%) studies, mainly for follow-up, motivational counseling, or technical support, and face-to-face sessions were included in 12 (21.8%) studies, usually for initial training or reinforcement. All interventions included patient-level components (n=55, 100% studies), such as self-monitoring, reminders, education, and behavioral support. In addition, provider-level involvement, including monitoring of transmitted data, follow-up contacts, or clinical feedback delivered by nurses, pharmacists, physicians, or multidisciplinary teams, was identified in 43 studies (78.2%). In several cases, health care professionals also contributed to treatment adjustment or clinical decision-making based on remotely collected data.

Of the 55 RCTs, 34 (61.8%) were rated as having some concerns, 20 (36.4%) as high risk, and 1 (1.8%) as low risk (Multimedia Appendix 4 [31-85]).

Bias arising from the randomization process: most studies adequately described the random sequence generation and allocation procedures. A total of 40 (72.7%) trials were rated as low risk, while 15 (27.3%) were judged as having some concerns.

Bias due to deviations from intended interventions: in this domain, 35 (63.6%) studies were assessed as low risk, 18 (32.7%) as some concerns, and 2 (3.6%) as high risk.

Bias due to missing outcome data: most trials reported adequate follow-up and outcome data. A total of 37 (67.3%) studies were rated as low risk, 16 (29.1%) as some concerns, and 2 (3.6%) as high risk.

Bias in the measurement of the outcome: the majority of studies were judged as having some concerns in this domain (n=53, 96.4% studies), while 2 (3.6%) studies were rated as high risk.

Bias in selection of the reported result: most studies were judged as low risk (n=39, 70.9% studies), whereas 16 (29.1%) were rated as having some concerns.

This systematic review and meta-analysis evaluated the effectiveness of digital health interventions in improving self-care and medication adherence among adults with chronic diseases. A total of 47 RCTs were included, encompassing a wide range of chronic conditions, geographic settings, and digital modalities. Overall, the evidence suggests that digital interventions can support patient self-care and adherence behaviors; however, the effects were heterogeneous and not consistently statistically significant.

For diabetes, all SDSCA domains (foot care, diet, physical exercise, and glucose monitoring) showed small, nonsignificant effects with low to very low certainty, indicating that confidence in these estimates is limited. Notably, subgroup analyses in glucose monitoring revealed significant effects in studies conducted in China compared to those conducted in the United States and the United Kingdom, suggesting that contextual factors, such as program intensity, cultural adaptation, and professional involvement, may influence outcomes. Across studies not included in the meta-analysis, digital interventions, particularly those integrating education, monitoring, and pharmacist or clinician support, tended to show improvements in multiple self-care domains (eg, diet, glucose monitoring, and foot care). However, these studies are likely affected by methodological limitations highlighted in the risk of bias assessment. Our meta-analysis findings are inconsistent with the existing body of evidence, although the results from the narrative synthesis are consistent with those reported by Liu et al [87], who found significant improvements in hemoglobin A_1c, blood pressure, and diabetes self-management activities, especially in short-term interventions (<6 months) and younger populations, and with Shrivastava et al [88], who observed general improvement trends but limited statistical significance due to small samples and unclear risk of bias.

For heart failure, pooled analyses suggested small to moderate improvements in self-care monitoring when measured with the SCHFI instrument, particularly in studies conducted in Canada, the United States, and Australia, compared to those in China, while little to no effects were found in maintenance, nor on self-care when measured with the EHFScB scale. However, certainty of evidence ranged from very low to low, indicating limited confidence in these estimates and that future evidence is likely to change the direction or magnitude of the observed effects. Our narrative synthesis indicates heterogeneous and inconsistent findings; interventions incorporating clinician oversight and interactive components were associated with improvements in specific self-care behaviors, although these findings should be interpreted cautiously, given the generally high or unclear risk of bias across several domains. The absence of an effect on self-care maintenance likely reflects the current capacity of available technologies to support the detection and monitoring of clinical parameters, while remaining insufficient to effectively influence self-care maintenance, defined as behavioral change in key habits [6], such as physical activity, diet, and medication adherence. Overall, our results highlight the uncertainty of the current evidence reporting mixed findings across studies, suggesting that effectiveness remains uncertain, although interactive, feedback-driven models may offer potential benefits [89,90].

This is also supported by the pooled estimate from 17 RCTs on medication adherence across different conditions, which showed no significant overall effect, with very low certainty, indicating that the true effect is likely to be substantially different from the observed estimate and that confidence in this finding is very limited. This was supported by the wide prediction interval, which indicates substantial variability in effects across different settings, suggesting that future studies may plausibly show benefit, no effect, or even harm. No further evidence emerged from subgroup analyses by intervention type and country. This is supported by the narrative synthesis that showed contrasting results, even though findings are predominantly derived from studies at high or moderate risk of bias, requiring attention in interpreting the results. This contrasts with Lanke et al [21] and Kim et al [20], who found significant improvements, especially when apps included interactive functions, enhanced reminders, or data sharing. Shrivastava et al [88] similarly observed improvement trends but noted that few studies reached significance, largely due to methodological limitations. In stroke, Zeng et al [91] reported significant adherence gains from mHealth apps and messaging interventions.

Differences between our findings and prior meta-analyses may reflect broader inclusion criteria, greater heterogeneity in populations and interventions, and the inclusion of multiple digital modalities (apps, telemonitoring, SMS text messages, and web portals), which may have diluted specific adherence effects. Many included trials were small, short-term, and used heterogeneous adherence measures (MMAS, MARS, ARMS, and Medication Event Monitoring System), which reduced the precision and certainty of the evidence.

In other chronic conditions, narrative findings suggest that multicomponent digital interventions may improve self-care in some diseases, particularly cardiovascular conditions, chronic hepatitis B, epilepsy, and hypertension, where improvements were reported in domains such as symptom monitoring, treatment adherence, and overall self-management. In contrast, no significant effects were observed in COPD, while mixed results emerged for Parkinson disease, where improvements were limited to specific domains such as monitoring and maintenance but not management, and for patients with multimorbidity, among whom no significant changes were reported across self-care domains. However, given that the majority of studies were assessed as having a high or moderate risk of bias, these results should be interpreted with caution.

Overall, the inconsistency in our findings may be explained by broader inclusion criteria, greater heterogeneity in populations and outcomes, and diverse intervention modalities. Our review encompassed multiple chronic conditions and technologies (apps, telemonitoring, SMS text messages, and web portals), which may dilute specific effects on medication adherence. Additionally, many included studies were small, of short duration, or used heterogeneous measurement tools (MMAS, MARS, ARMS, and Medication Event Monitoring System), leading to lower precision and downgraded certainty. Furthermore, despite efforts to ensure methodological consistency, self-care remains a complex and multidimensional construct to measure [92], and its measurement varies substantially across studies. The use of different validated instruments, each capturing distinct dimensions of self-care, may limit comparability and contribute to heterogeneity in the findings.

In terms of clinical implications, this review both corroborates, contrasts, and extends prior literature. It confirms that digital health interventions may hold potential for improving self-care, especially in heart failure and monitoring, while also indicating that their effects are uneven and context-dependent [14,15,20,21]. Specifically, due to inconsistent self-care maintenance results across diabetes, heart failure, and other conditions, it remains necessary to support the implementation of digital health interventions with human components to ensure long-term engagement of the person, such as in-person visits and motivational interviewing [93,94]. Other challenges should also be considered in implementing digital health interventions. Equity and digital health literacy remain critical determinants of effectiveness that are not thoroughly addressed in the included studies. Evidence from Turnbull et al [95] showed that the impact of digital self-care interventions is not uniform across populations. Their review showed that web-based programs can benefit disadvantaged or minority groups (eg, ethnic minorities and individuals with lower socioeconomic status), although digital access, health literacy, and usability continue to pose substantial barriers. Ge et al [96] highlight that digital exclusion among older adults is a multidimensional phenomenon that includes resource, skill, and motivational constraints limiting engagement with digital health interventions. Sociodemographic, functional, and psychological factors, such as limited confidence and technology-related anxiety, also reduce participation [97,98] but were rarely analyzed as moderators in the studies included in this review. Therefore, this systematic review highlights the importance of integrating digital interventions into routine care through structured implementation strategies. These include ensuring education for both patients and professionals, tailoring interventions to individual levels of digital literacy, and embedding digital tools within existing clinical pathways to enhance continuity rather than create parallel workflows. Moreover, co-design approaches and ongoing user engagement should be prioritized to improve usability, acceptability, and adherence [10,93]. Health systems adopting digital self-care solutions should also consider equity-oriented implementation frameworks to prevent the widening of existing disparities and to ensure that innovations benefit all patient populations.

As regards research implications, several priorities emerge. First, future trials should adopt standardized outcome measures to enhance comparability and enable robust meta-analyses. The proliferation of heterogeneous instruments (eg, multiple versions of SDSCA, MARS, or MMAS) hampers synthesis and reduces external validity. Second, longer-term trials are needed to assess the sustainability of intervention effects, as most included studies had follow-up durations shorter than 12 months. Third, equity-focused research is essential. Few RCTs stratified outcomes by socioeconomic status, age, or eHealth literacy, despite evidence that these factors critically determine effectiveness. Fourth, cost-effectiveness analyses are urgently required, given the resource implications of scaling digital health interventions within health systems. Fifth, future studies should investigate advanced technologies, including artificial intelligence, wearable sensors, and adaptive platforms, to determine whether they can deliver more consistent benefits. Finally, there is a need to conduct studies on self-care in chronic diseases grounded in robust theoretical frameworks, such as the middle-range theory [5,6]. Recent versions of measurement instruments incorporate self-care management as a third component, alongside maintenance and monitoring, which is crucial for developing individuals’ ability to manage their health autonomously.

A limitation of this review is the inclusion of self-care outcomes only when measured with a validated instrument to ensure consistency across measures. However, many self-care behaviors can also be evaluated through alternative methods, such as step counts for physical activity or pill counts and prescription records for medication adherence. Future research could therefore expand on these findings by adopting a broader approach to measuring self-care behaviors. In addition, the search was conducted on 4 databases and was limited to articles published in English, which may have led to the exclusion of relevant studies and introduced language and publication bias.

In conclusion, current evidence suggests that digital health interventions may provide some benefit in self-care monitoring in heart failure, while showing no clear or consistent effects on self-care maintenance or when assessed with alternative instruments. In diabetes, effects across all self-care domains (diet, physical activity, foot care, and glucose monitoring) were small and nonsignificant. In other chronic conditions, results varied. Positive changes were observed in some cardiovascular conditions, chronic hepatitis B, epilepsy, and hypertension, whereas no significant effects were found in COPD and multimorbidity, while mixed results emerged in Parkinson disease. Similarly, no clear overall effect was observed for medication adherence across chronic conditions. Overall, the certainty of the evidence is predominantly low to very low, limiting the ability to draw firm conclusions. Results should therefore be interpreted as suggestive but not definitive, emphasizing the need for larger, methodologically robust trials with standardized outcomes and longer follow-up to clarify the true impact of digital health on self-care and adherence. This review synthesizes evidence across multiple chronic conditions, highlighting that the effects of digital health interventions on self-care remain variable and context-dependent. From a practical perspective, integrating digital tools with human support and adapting interventions to patient characteristics may help enhance their effectiveness in routine care.