We conducted searches in Cochrane Library, Web of Science, PubMed, Embase, and EBSCO for randomized controlled trials (RCTs) published from the inception of the databases up to December 28, 2022, without any language restriction. MeSH (Medical Subject Headings) and free search terms were both used in the literature search. The search terms included “cell Phones,” “telemedicine,” “smartphone,” “mobile applications,” “mobile phone*,” “telephone*,” “telehealth,” “tele-healthcare,” “electronic health*,” “application*,” “m-health,” “messaging,” “depression,” “depressive disorder,” “depress*,” “Major depression,” “sadness,” “late-life depression,” “LLD,” “aged,” “elder*,” “geriatric,” “senior people,” “RCTs,” etc. All titles, keywords, and abstracts have been reviewed in accordance with our search criteria. In this study, these research articles were exclusively published in English, focusing on telecare interventions for LLD. The specific search strategy is shown in Multimedia Appendix 2.

The inclusion criteria were the following: (1) studies were RCTs reported in full text with their title and abstract; (2) the average age of the study population was at least 60 years; (3) participants were diagnosed with depression in accordance with any established diagnostic criteria or with a score above a cutoff of any established depression rating scale at baseline; (4) the studies compare telecare (mobile phone, telephone, app, video, etc) participants with the control group receiving usual care (UC; routine, offline, or standard care); and (5) any health care professional providing care (ie, psychiatrists, family physicians, nurses, psychologists, etc).

Exclusion criteria were the following: (1) patients with manic or psychotic episodes or symptoms; (2) studies not related to the objective of this review and insufficient data, such as failure to report depression scale scores; and (3) books and studies without full text and studies in the format of abstracts of conference papers.

Two authors independently reviewed all the databases, with specific search strategies for the relevant articles (MW and CYL). The software EndNote X9 (Clarivate) was used to import all the references and remove duplicates. After removing duplicates, the relevance of the title and abstract of the articles was evaluated. Any disagreements were discussed until a consensus was reached. After screening the title and abstract, the articles were selected for the next step of a full-text review. The 2 authors screened the full-text articles independently (MW and CYL). Finally, eligible articles included in the study were processed based on inclusion and exclusion criteria. Any discrepancies that arose during the assessment were resolved by a third reviewer (FY). Two authors independently extracted data from the included studies and entered them into a predesigned data extraction form. Data extracted for this study included the following: first author, year of publication, country, sample size, mean age, intervention approach, duration, presence or absence of comorbid chronic conditions, depression degree, and outcome measurement tools (Table 1).

Two authors (MW and CYL) independently assessed the quality of the studies using the Cochrane Risk of Bias tool [40]. The assessment tool included 7 items (random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other bias), and authors judged each item individually as “low risk,” “high risk,” and “unclear risk.” The study was considered to be of high quality with a low risk score for at least 4 domains, of which 3 key areas had to be included (random sequence generation, allocation concealment, and incomplete outcome data). Consensus was reached by 2 authors (MW and CYL) through discussion with a third evaluator (FY).

Data were analyzed using Stata (version 16.0; StataCorp) and Review Manager (version 5.3; The Cochrane Collaboration). Intervention effects were estimated by calculating Cohen d standardized mean differences (SMDs) and 95% CIs [41]. All studies reported outcomes as continuous data. The Cochran Q statistic and I² statistic were used to assess the statistical heterogeneity between selected studies. Random-effects models were used when study heterogeneity was high (P<.10; I²>50%); otherwise, a fixed-effects model would be used. When heterogeneity identified across studies was high, we further performed subgroup analyses to explore possible explanations for heterogeneity. Publication bias was measured using a funnel plot and Egger linear regression analysis, and P<.05 on the Egger test indicated statistically significant publication bias [42].

The database search yielded 15,265 articles, of which 14,249 publications were excluded. A total of 1016 full-text articles were assessed for eligibility. Finally, only 12 studies were eligible for inclusion in this meta-analysis [21-23,31-39], all of which were RCTs published between 2009 and 2021. The PRISMA flow diagram is shown in Figure 1.

Overall, the quality of the included studies was moderate, of which 5 (41.7%) were of high quality. These studies show that the main bias in the blinding of participants and personnel may be caused by the nature of the intervention measures. All 12 articles reported adequate random sequence generation and, therefore, had a low risk of bias in this regard. In addition, 5 studies reported allocation concealment, which is a low risk of bias. As for detection bias, the assessors were blinded in 7 studies, the presence of blinding was unclear in 3 studies, and 2 studies were not blinded. The risks of study attrition bias and reporting bias were both low. Other risks of bias were also low but were unclear in 1 study. The specific risk of bias and quality assessment results are shown in Figures 2 and 3 [21-23,31-39].

The characteristics of the studies included are summarized in Table 1. A total of 1663 patients with LLD were involved, with an average age of over 60 years in each group. The sample size ranged from 25 [23] to 302 [31] participants. Studies were carried out across 6 countries, including the United States (n=6) [21,22,31,32,34,36], Korea (n=1) [23], Italy (n=1) [33], Australia (n=2) [35,38], China (n=1) [37], and Germany (n=1) [39]. Nine of these used telephone-based interventions, while the remaining studies used remote monitoring systems. Durations ranged from 10 weeks to 52 weeks. Depression, anxiety symptoms, and QoL were substantial influencing factors of treatment for older adults. Therefore, our primary outcome of interest was depression, and secondary outcomes were anxiety symptoms and QoL. Depression was evaluated using the Hamilton Depression Rating Scale, Patient Health Questionnaire-9 (PHQ-9), Beck Depression Inventory, Center for Epidemiological Survey, Depression Scale, Cardiac Depression Scale, Hospital Anxiety and Depression Scale, and Zung Self-Rating Depression Scale. Anxiety symptoms were assessed using Spielberger’s State Trait Anxiety Inventory, Hamilton Anxiety Rating Scale, and the 7-item Generalized Anxiety Disorder Scale. QoL was assessed using the 12-Item Short Form Survey and the 36-Item Short Form Survey. A higher score on the scales indicated better QoL and greater severity of depression and anxiety symptoms. The specific interventions are available in Multimedia Appendix 3.

A total of 12 RCTs involving 1663 participants were identified in this meta-analysis to calculate the effectiveness of telecare on depression, anxiety symptoms, and QoL in patients with LLD.

To compare the effects of telecare and UC in improving LLD, we included data from 12 of these studies. Our results show that telecare significantly reduced depressive symptoms in patients with LLD compared to those in UC (SMD=–0.46, 95% CI –0.53 to –0.38; P<.001). Fixed-effects model analysis revealed significant heterogeneity among the 12 included studies (I²=83.16%; P<.001; Figure 4) [21-23,31-39].

To address high heterogeneity, we performed subgroup analyses grouped by the type of scale (PHQ-9 or others), duration time (≤3 months or >3 months), device type (telephone-based or remote monitoring system), comorbid chronic diseases (presence or absence), and region (Europe and the Americas or others).

Random-effects models indicated that telecare significantly reduced depressive symptoms in patients with LLD compared to the UC participants (SMD=–0.59, 95% CI –0.80 to –0.38; P<.001). Results of subgroup analysis by duration showed that short-term (≤3 months) interventions (SMD=–0.72, 95% CI –1.16 to –0.28; P<.001) were more effective than long-term (>3 months) interventions (SMD=–0.52, 95% CI –0.75 to –0.29; P<.001); other scales (SMD=–0.65, 95% CI –0.96 to –0.35; P<.001) were more effective than the PHQ-9 (SMD=–0.53, 95% CI –0.83 to –0.22; P<.001); the remote monitoring system (SMD=–1.13, 95% CI –1.51 to –0.76; P<.001) was more effective than telephone-based interventions (SMD=–0.38, 95% CI –0.56 to –0.20; P<.001); the effect on patients with LLD with chronic diseases (SMD=–0.67, 95% CI –0.89 to –0.44; P<.001) was better than that on patients with LLD without comorbid chronic diseases (SMD=–0.10, 95% CI –0.41 to 0.20; P=.07); and telecare was more effective in Europe and the Americas (SMD=–0.73, 95% CI –0.99 to –0.47; P<.001) than in other regions (SMD=–0.22, 95% CI –0.35 to –0.09; P=.42; Table 2).

Meta-regression analysis showed that heterogeneity may not be related to the year of publication (P=.42), total sample size (P=.21), study area (P=.35), comorbid chronic disease (P=.47), duration (P=.75), and outcome measurement tools (P=.29). However, only the intervention device (P=.004) may have contributed to the heterogeneity.

The stability and reliability of the results of this meta-analysis and potential factors contributing to heterogeneity were explored by sensitivity analysis to assess the effect of the data of each study on the combined effect value (ie, SMD). The results of the sensitivity analysis showed that excluding each study individually had no significant effect on the combined effect value, and the study results were stable and reliable (Figure 5) [21-23,31-39]. Publication bias was assessed using funnel plots and Egger test indicators. The funnel plot was symmetrically distributed on both sides (Figure 6), and the Egger test showed no significant publication bias (P=.05).

To examine the efficacy of telecare in reducing anxiety compared with that of UC, we included 3 articles on patients with LLD. The results showed that telecare significantly reduced anxiety symptoms in patients with LLD (SMD=–0.53, 95% CI –0.73 to –0.33; P=.02; Figure 7) [22,33,38].

Six studies assessed the mental components of QoL by using the Medical Outcomes Study Short Form survey. Our meta-analysis shows that the QoL of patients with LLD improved, but, overall, it was not significant (SMD=0.30, 95% CI 0.18-0.43; P=.80; Figure 8) [22,31,35,36,38,39].

This meta-analysis shows that compared with UC, telecare significantly reduces symptoms of depression and anxiety but has no significant effect on improving QoL in patients with LLD.

This study still had some limitations. First, most of the studies included in the review lacked sufficient measure detail, leading to irreversible bias. Our study mainly included 2 interventions based on telephone and remote monitoring to reduce this bias. Second, the measurement tools used in this study lacked standardization and heavily relied on self-reports from participants, introducing a degree of subjectivity and concealment that is not as rigorous as structured interviews. However, we attempted to validate the effectiveness of the results by using authoritative scales. Third, differences in the specific intervention methods, frequency, and content among the included studies may lead to clinical heterogeneity across different studies.

Our meta-analysis shows that telecare has a positive impact on depressive and anxiety symptoms, despite high heterogeneity in depression symptoms. Therefore, studies with larger sample sizes and homogeneity were required to determine the effects of telecare in patients with LLD. Future research can continue to refine telecare systems and assess the specific needs of older vulnerable populations for more accurate evidence.