We searched MEDLINE, Scopus, ClinicalTrials.gov, and the Tufts Medical Center Cost-Effectiveness Analysis Registry for randomized or quasi-randomized clinical trials that used their data to evaluate the cost-effectiveness of telemedicine vs standard care for the management of type 2 diabetes mellitus, hypertension, and heart failure. The meta-analysis was limited to these 3 diseases as they all represent chronic cardiovascular conditions that are measurable through standardized tests and characterized by similar long-term care requirements and for which both cost-effectiveness evaluations and telemedicine interventions are well established in the literature, with RCTs dating back to before 2005 [20,21]. Indeed, several telemedicine trials conducted throughout the years have enrolled patients with multiple chronic diseases or risk factors, most often diabetes, hypertension, and heart failure [22-24], and suggest that telehealth could be efficacious in controlling these conditions [13]. To be included in this study, trials had to estimate the ICER per quality-adjusted life year (QALY) gained or the ICER per another primary outcome (1% reduction in hemoglobin A_1c [HbA_1c] for diabetes and 1–mm Hg reduction in systolic blood pressure [SBP] for hypertension) using a follow-up equal to or longer than 3 months [16].

Further details on the identification and selection processes are reported in Multimedia Appendix 1 [24-30].

The primary outcome was the ICER per QALY gained, which is considered a foundational metric to evaluate the value of a health intervention and allows for cross-disease comparisons of health benefits by quantifying gains in both duration and quality of life [31,32]. The ICER was defined as the difference in costs divided by the difference in QALYs between telemedicine and standard care [33], indicating telemedicine cost per additional QALY gained. Secondary outcomes were ICER per 1% reduction in HbA_1c and ICER per 1–mm Hg reduction in SBP.

When available, we extracted the 95% CIs or uncertainty intervals, computed the SE of the ICER, and computed a summary estimate of the mean difference in ICER across individual studies using a random-effects [34] generic inverse variance approach to account for between-study heterogeneity [35]. This approach was adopted as it is cited in the literature among those appropriate for meta-analyzing ICERs [36]. All costs were adjusted to 2025 USD, and between-study heterogeneity was quantified using the I² metric. Specifically, as inflation adjustment was reportedly performed in most (>80%) of the included studies, the current conversion simply aimed to adjust for the inflation that occurred between the time of publication and the year 2025 using the US consumer price index. Inflation adjustment was performed after currency conversion as practically all studies were conducted in high-income countries and the US inflation rate was considered sufficiently representative of all the study settings [37].

Meta-analysis was performed by pooling only studies with results in the same quadrant of the cost-effectiveness plane (ie, when ICERs indicated that the telemedicine intervention had a higher efficacy for an increased cost compared to control [36]). When most trials did not report an estimate of ICER SE, P value, or CIs, missing data imputation was inappropriate, and head-to-head meta-analysis could not be performed [38]. The health economic findings were synthesized narratively and presented using tables and graphics [2] in which the weighted mean of the ICER was shown. ICER means were weighted for the sample size, separately reported for each outcome, disease, and date of publication (excluding studies published before 2015). Given the variability in follow-up length of the included studies, in an attempt to reduce individual study variability, the estimates of the RCTs that used different follow-ups were recomputed assuming (1) constant effectiveness over time and (2) linear increase in costs with time (eg, the ICER of a trial with a 6-month follow-up was doubled). A 12-month follow-up was commonly assumed to be a sufficient time frame to detect meaningful changes in disease control from a clinical and organizational point of view. While this approach may overestimate the interventions’ setup costs, it was selected over more complex modeling approaches to simplify the interpretation of the results given that most of the included studies (14/27, 51.9%) used a 12-month follow-up. As no meta-analysis included more than 7 trials, potential publication bias could not be assessed through formal testing of funnel plot asymmetry [39].

All meta-analyses were carried out using RevMan (version 5.4; The Cochrane Collaboration).

The PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) flow diagram of the selection process is summarized in Figure 1 [25]. Of the 4183 papers initially retrieved, a total of 27 (0.6%) RCTs used trial data to estimate and report one of the selected outcomes and were included in the meta-analyses.

The main study characteristics are reported in Multimedia Appendix 2 [40-65] stratified by disease: diabetes (10/27, 37% of the RCTs) [40-49], hypertension (11/27, 40.7% of the RCTs) [50-60], and heart failure (6/27, 22.2% of the RCTs) [58,61-65]. The first trial was published in 2005; 55.6% (15/27) of the studies were published after 2015 and were defined as “more recent.” In total, 29.6% (8/27) of the RCTs were carried out in the United States; 18.5% (5/27) were carried out in the United Kingdom; 25.9% (7/27) were carried out in other European countries; and the remaining studies (7/27, 25.9%) were carried out in Australia, Bangladesh, Brazil, Russia, and Singapore. The sample sizes ranged from 117 to 1665 (564 on average, SD 444), and the duration of follow-up ranged from 3 to 60 months, although most of the trials (14/27, 51.9%) followed patients for 12 months. Almost all RCTs computed ICERs from the health care provider (13/27, 48.1%) or health care system (13/27, 48.1%) perspective; the interventions included telemonitoring in all trials on hypertension management (11/11, 100%), whereas most studies on diabetes and heart failure used some form of telephone coaching or support. Interestingly, 40.7% (11/27) of the studies did not report a willingness-to-pay (WTP) threshold, and among the studies providing a WTP threshold for QALYs, this was highly variable: from US $5000 to US $100,000. Only 7.4% (2/27) of the studies cited a threshold for 1–mm Hg reductions in SBP (US $27 to US $4071) [52,57], and 3.7% (1/27) of the studies cited a threshold for 1% reductions in HbA_1c (US $1000) [40].

The methodological quality of the included RCTs is reported in Multimedia Appendix 3 [40-65]. Most RCTs (25/27, 92.6%) had an intermediate risk of bias, and only 7.4% (2/27) were judged to have a high risk of bias. The most common sources of concern were inadequate or insufficiently described methods to handle missing data (21/27, 77.8% of the RCTs), handle deviations from the intended interventions (20/27, 74.1%), and measure outcomes (17/27, 63%). The randomization process and the selection of the reported results were less concerning, especially in view of the wide standardization of outcomes for diabetes (HbA_1c) and hypertension (SBP).

In total, 48.1% (13/27) of the RCTs estimated the ICER per QALY gained (Multimedia Appendix 2), but only 3 of the trials on heart failure management reported sufficient data (95% CI, P value, or SE) to be included in the meta-analysis. Overall, based on a total sample of 1825 patients, the summary estimate of ICER per QALY gained of telemedicine vs standard care on a 12-month follow-up was US $45,741, with a large, nonsignificant 95% CI (–US $31,050 to US $122,533; P=.24). In the more recent RCT by González-Guerrero et al [65], the telehealth intervention was more effective and cost-saving compared to standard care.

The weighted means of the 48.1% (13/27) of the studies reporting the ICER per QALY gained are reported in Multimedia Appendix 4 [42,43,45,48,49,57-59,61-64]. The average ICER largely varied by disease, being highest for the management of diabetes (5/13, 38.5% of the studies, 2949 patients; US $44,725, SD 18,925), intermediate for heart failure (5/13, 38.5% of the trials, 3171 patients; US $40,271, SD 22,933), and lowest for hypertension (3/13, 23.1% of the studies, 1314 patients; US $18,966, SD 13,289).

One out of the 11 hypertension trials did not calculate an ICER as higher effectiveness was found in the control arm; therefore, it could not be used for either SBP reductions or QALY gained [60]. In total, 90% (9/10) of the studies on hypertension computed the ICER per 1–mm Hg SBP reduction, and 70% (7/10) of the RCTs on diabetes reported an ICER per 1% HbA_1c decrease. However, only 50% (5/10) of the trials on hypertension, including a total of 1986 patients, reported sufficient data and were included in a meta-analysis (Figure 2). Overall, the summary estimate of ICER per 1–mm Hg reduction in SBP of telemedicine vs standard care for a 12-month follow-up was US $27 (95% CI US $6.4-$47.7; P=.01). The weighted mean for the same outcome from the 90% (9/10) of the trials on hypertension that reported an ICER was US $132 (SD 227; [50-57,59] Multimedia Appendix 4).

With regard to the ICER per 1% HbA_1c reduction, no meta-analyses were possible as no trial reported a measure of dispersion from the mean, and the weighted mean of the 7 trials that reported an ICER, including a total of 4157 patients, was US $3349 (SD 2249; Multimedia Appendix 4 [40,41,43-46,48]). When only the 2 more recent trials were considered, the average ICER decreased to US $1629 (SD 743).