The protocol for this systematic review was registered with the International Prospective Register of Systematic Reviews (PROSPERO; registration number: CRD420251149977).

Eligibility criteria were established based on the PICOS (Population, Intervention, Comparison, Outcome, and Study Design) framework. The study population included patients diagnosed with SCI, with no restrictions on sex, age, injury level, or severity. Eligible interventions were defined as telemedicine-related modalities, including video consultations, mobile health apps, remote monitoring, and telephone follow-up. Comparators were not restricted and could include usual care or no tele-intervention. Outcomes were prespecified in five domains, including psychological health (eg, depression and/or anxiety), health-related quality of life, sleep-related functioning, functional independence, and rehabilitation outcomes, and pain intensity. Trials contributed to meta-analysis only when outcome data for at least one prespecified domain were available and extractable; otherwise, they were included in the descriptive synthesis. Only RCTs were eligible. To ensure that comparisons reflected allocation to mode of care delivery rather than self-selection, we included only trials in which participants were randomized to receive a telemedicine-delivered intervention versus a comparator condition (eg, usual care, standard follow-up, or in-person care). Studies in which telemedicine exposure was determined primarily by participant preference, acceptance of an offer, or other nonrandom allocation mechanisms were excluded. In line with the search strategy, studies were restricted to English-language full-text reports.

The exclusion criteria were as follows: (1) nonprimary research articles, including systematic reviews, narrative commentaries, conference abstracts, and case reports, and other non-RCT designs; (2) interventions not related to telemedicine; and (3) duplicate publications or studies with overlapping datasets. These exclusions were applied to ensure methodological rigor and extractable data.

Meta-analysis was performed when quantitative data were extractable; otherwise, findings were synthesized narratively. Trials not reporting any prespecified outcome domain were included in the descriptive summary (Table 1) but were not included in the quantitative synthesis. When outcomes were measured but insufficiently reported for extraction, authors were contacted where feasible, and missing results were documented.

To ensure comprehensive identification of relevant studies, we searched PubMed (via the NLM interface), Web of Science Core Collection (via Clarivate), Embase (via Ovid), MEDLINE (via Ovid), and the Cochrane Central Register of Controlled Trials (CENTRAL; via the Cochrane Library). The final search update was performed on February 17, 2026. In addition to database searching, we also searched trial registries and conducted supplementary searches (citation searching, websites and search engines, and hand searching). We additionally performed backward and forward citation searching using the reference lists of included studies and relevant reviews, and the “cited reference” function in Web of Science. Searches were restricted to English-language full-text reports. The search methods and reporting followed the PRISMA-S (Preferred Reporting Items for Systematic Reviews and Meta-Analyses–Search) extension [44].

A comprehensive strategy was developed by combining controlled vocabulary (eg, MeSH and Emtree terms) with free-text terms and adapting syntax to each database. Search concepts included (1) spinal cord injury and related terms, (2) telemedicine-related modalities (including telerehabilitation, telerehab, and closely related telehealth and telemedicine terms), and (3) an RCT component. To avoid overly broad retrieval, broad digital-delivery terms, such as online, telephone, app-based, web-based, and internet-based terms, were combined with rehabilitation- or therapy-related terms. An RCT filter was applied where appropriate, and animal-only records were excluded. The full, reproducible search strategies for all databases, copied exactly as run (including filters and limits), are provided in Table S1 in Multimedia Appendix 1. No natural language processing or text-mining tools were used to generate or refine search terms. No automated search translation tools were used; search strings were adapted manually to each database. The PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) flow diagram has been updated accordingly (Checklist 1). All records were imported into EndNote for record management and deduplication before screening. The electronic search strategy underwent internal checking for syntax and completeness. When outcome data were measured but incompletely reported, study authors were contacted when feasible and necessary to obtain missing information.

Two reviewers (WZ and LH) conducted the study selection process in a 2-phase approach, initial screening of titles and abstracts to assess relevance, followed by full-text evaluation to determine final eligibility. Reference management software (EndNote) was used to eliminate duplicate records, and all screening decisions and reasons for exclusion were systematically documented. Any discrepancies between reviewers were resolved through consensus discussions involving a third reviewer. No additional information was sought from investigators, and no automation tools were used in the selection process. No translation was required because searches were restricted to English-language full-text reports.

The data extraction process was independently conducted by 2 reviewers (WZ and LH) using a predesigned Microsoft Excel spreadsheet. Extracted information included basic study details (author, year of publication, and country), study design, sample size, participant characteristics (eg, sex and age), type of intervention and control group settings, outcomes, and follow-up duration. Any discrepancies in extracted data were resolved through discussion, with a third reviewer adjudicating when necessary. In cases of missing or unclear data, attempts were made to contact the original authors for clarification. All extracted data were cross-verified to ensure accuracy and consistency. No automation tools were used for data extraction, and no translation was required because inclusion was restricted to English-language full-text reports. No software was used to extract data from figures. When multiple reports corresponded to the same study, data were collated and extracted as a single study; where overlapping or inconsistent information was identified, the most complete report and/or the longest follow-up was prioritized.

We focused on five core outcome domains:

For each outcome domain, we extracted all relevant measurement instruments, assessment time points, and available analytical results. For meta-analysis, we prioritized postintervention values. For meta-analysis, we prioritized postintervention values at the end of the intervention; when multiple postintervention time points were available, we prioritized the end-of-intervention time point and summarized other time points narratively and explored them in sensitivity analyses when appropriate. Effect estimates were calculated respecting each instrument’s original scale without recoding. No changes were made to outcome domain definitions or selection rules after protocol registration. When required information was missing or unclear and could not be confirmed with the study authors, it was recorded as not reported, and no unverifiable assumptions were made.

The risk of bias was evaluated using the Cochrane Risk of Bias 2 (version 2; RoB 2; 22 August 2019 guidance), implemented using the RoB 2 Excel tool (ROB2_IRPG beta v9), a structured framework designed to assess the internal validity of trial outcome results. RoB 2 addresses five distinct domains: (1) bias related to the randomization procedure; (2) bias resulting from deviations from the intended intervention; (3) bias associated with incomplete outcome data; (4) bias in the assessment of outcomes; and (5) bias stemming from selective reporting of results. Each domain is appraised through a standardized decision algorithm, culminating in an overall judgment for each outcome, classified as “low risk,” “some concerns,” or “high risk.” Two reviewers independently assessed risk of bias for all included trials, with disagreements resolved by discussion and third-reviewer adjudication when needed. No adaptations were made to the RoB 2 tool, no additional information was sought from study investigators for risk-of-bias assessment, and no automation tools were used. Risk-of-bias judgments informed sensitivity analyses and the certainty of evidence appraisal.

Ethical approval is not applicable. This study is a systematic review and meta-analysis of previously published studies and did not involve the collection of individual-level data.

A total of 940 records were identified through comprehensive database searches, including PubMed (n=101), Web of Science (n=249), EMBASE and Ovid MEDLINE (n=418), and the Cochrane Central Register of Controlled Trials (n=168), with an additional 4 records identified from prior systematic reviews (n=4). After deduplication using reference management software, 531 unique records remained for title and abstract screening, of which 414 were excluded (reviews or meta-analyses, n=75; irrelevant interventions, n=19; other records not meeting the predefined inclusion criteria, n=320). A total of 117 reports were sought for full-text retrieval, and all were retrieved (n=0). After full-text assessment, 82 reports were excluded due to study protocols without results (n=54) or lack of an RCT design (n=28). Ultimately, we included 33 studies (35 reports) [9-43]. Two studies had multiple reports (Arora et al 2017 [9]/2017 [10]; Irgens et al 2022 [26]/2024 [27]). The study selection process is shown in Figure 1.

We included 33 studies (35 reports) [9-43]; 2 studies had multiple reports (Arora et al 2017/2017 [9,10] clinical and economic evaluations; Irgens et al 2022/2024 [26,27] clinical and cost-utility and environmental analyses), while all other studies were single reports. Sample sizes ranged from 13 participants to >400, with participants predominantly male, and reported mean or median ages generally spanning adulthood to older age. Intervention duration varied from single-session training to 8-12 weeks, to 3-6 months, and up to 2 years in community follow-up programs. Follow-up commonly occurred at the end of treatment and at short-term time points such as 1-3 months. Some studies extended follow-up to 6-12 months or longer. Full study characteristics are presented in Table 1.

The included studies used a diverse array of remote or hybrid intervention modalities. Telephone-based interventions were the most frequently used modality, accounting for 27.3% of studies (9/33) [9,11-13,20,35,36,39]. These interventions typically involved scheduled calls by health care professionals to deliver health consultations, psychological counseling, or disease-specific education. In some cases, telephone contact was supplemented with home visits to enhance personalization and continuity of care (9.1%, 3/33) [15,23,24]. Web-based or mobile app and social-platform interventions accounted for 18.2% of the included studies (6/33) [29-34]. These platforms, such as WeChat (Tencent), mobile apps, and dedicated websites, facilitated ongoing care, rehabilitation guidance, interactive education, and multidisciplinary support. These interventions were typically led by nurses or rehabilitation teams. Videoconferencing-based interventions were used in 15.2% of studies (5/33) [17,19,26,27,38,42], offering real-time interaction and visual feedback for psychological support, rehabilitation, and educational guidance. Tele-supervised rehabilitation or teleexercise programs also accounted for 15.2% of studies (5/33) [16,18,28,40,43] and were delivered through remote supervision and periodic reassessment, sometimes combined with adjunct technologies. A minority of studies investigated novel intervention modalities, such as transcranial direct current stimulation (tDCS) and functional electrical stimulation exercise training (FES-ET), reflecting the evolving landscape of remote therapeutic strategies (eg, Chantanachai et al [16] and Kowalczewski et al [28]). Psychological and behavioral interventions were widely represented, including cognitive behavioral therapy (CBT), mindfulness-based training, vocational planning, and electronic cognitive therapy (eg, Burke et al [14]; Hearn et al [22]; and Migliorini et al [37]). Full details of intervention modalities for each study are provided in Table 1.

The risk of bias across included studies was assessed using the Cochrane RoB 2 tool, and the results are summarized in Figure S1 in Multimedia Appendix 1. Multimedia Appendix 1 presents study-level judgments across the 5 RoB 2 domains (randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result) together with an overall judgment. Overall, a substantial proportion of included studies were rated as having some concerns, and a smaller subset was judged to be at high risk of bias, while approximately half were rated as low risk overall. Figure S1B Multimedia Appendix 1 shows the distribution of risk of bias judgments by domain. Missing outcome data had the highest proportion of low-risk judgments (approximately 94%), indicating that incomplete outcome data were generally well addressed in most trials. The randomization process domain also showed a high proportion of low-risk judgments (approximately 71%), suggesting that sequence generation and allocation procedures were relatively robust in many studies. In contrast, deviations from intended interventions and selection of the reported result were more frequently rated as some concerns, reflecting limitations related to intervention adherence, analytic approach, or reporting transparency. Measurement of the outcome showed a more mixed pattern, with a notable proportion of studies judged to be at high risk (approximately 23%), consistent with challenges related to blinding, outcome ascertainment, or reliance on self-reported measures. Taken together, these results indicate that concerns were driven primarily by issues related to deviations, outcome measurement, and selective reporting rather than missing data.

We assessed certainty of evidence for each critical outcome using GRADE and summarized results in a Summary of Findings table (Table S2 in Multimedia Appendix 1). Overall certainty ranged from high to low, with downgrading most commonly due to risk of bias and imprecision (small samples and wide confidence intervals), with detailed rationale provided in the Summary of Findings table footnotes. Evidence for depressive symptoms was of low certainty overall, although certainty varied by follow-up, including a high-certainty estimate for the >3-≤6 months subgroup and moderate or low certainty at other time windows. Evidence for anxiety was of low certainty, mainly limited by imprecision. For WHOQOL-BREF, certainty was moderate for the social relationships domain but low for the physical, psychological, and environment domains. For sleep quality assessed by PSQI, evidence at 1 month and 3 months was moderate certainty, downgraded primarily for imprecision because few trials reported sleep outcomes. For functional independence and participation, evidence was moderate for SCIM, while CHART domains ranged from moderate to low. For pain intensity (NRS), evidence was of low certainty.

Consistent with our objective to systematically review and meta-analyze the effects of telemedicine interventions on psychological health, health-related quality of life, sleep-related functioning, functional independence and rehabilitation outcomes, and pain intensity in patients with SCI, this review suggests that telemedicine may offer benefits for selected outcomes, but that these effects are domain-specific and time-dependent. The clearest signals of benefit were observed for the WHOQOL-BREF social domain and sleep quality at 3 months, findings that were supported by low between-study heterogeneity and moderate-certainty evidence. A more favorable effect on depressive symptoms was also observed in the >3-≤6 months follow-up subgroup, and although the certainty of evidence for this subgroup was high, this finding appears to reflect a time-specific benefit rather than a consistent overall effect across follow-up periods. By contrast, the overall pooled effect for depressive symptoms was not statistically significant, the prediction interval crossed the null, and certainty was low, suggesting that any overall benefit may be uncertain and may vary across settings. Evidence for anxiety, functional recovery, and pain intensity was also limited by imprecision, small sample sizes, lower-certainty evidence, and risk-of-bias concerns in a proportion of included trials, particularly related to deviations from intended interventions, outcome measurement, and selective reporting. Taken together, these findings suggest that telemedicine may serve as a useful complement to standard rehabilitation pathways in SCI, but that its effectiveness is likely to depend on the outcome targeted, intervention characteristics, follow-up window, and the strength and consistency of the underlying evidence across domains.

Telemedicine interventions may influence outcomes through two related but conceptually distinct pathways. First, remote delivery may be particularly suitable for individuals living with SCI by reducing barriers related to mobility, transportation, and geographic distance, thereby supporting continuity and timeliness of follow-up care [50-52]. Second, many telemedicine models embed enhanced care exposure, such as more frequent contacts, proactive monitoring, structured coaching, and facilitated self-management [53,54], which may improve outcomes irrespective of whether care is delivered remotely or in person. Because intervention dose and comparator intensity were variably reported and often differed across trials, the current evidence does not allow us to disentangle modality effects from care-intensity effects, and this should be considered when interpreting domain-specific benefits.

Psychological complications such as depression and anxiety are common complications following SCI [55]. A pooled analysis estimated the average prevalence of depression to be approximately 22% (95% CI 19%‐26%) [56]. Recent evidence syntheses highlight a substantial and persistent mental health burden after traumatic SCI, with depression being the most frequently studied condition [57]. Beyond prevalence, depressive symptoms have important clinical consequences. In a prospective follow-up study, higher depressive symptom burden was identified as a determinant of lower participation in inpatient rehabilitation [58]. Consistent with this, a large SCI Model Systems cohort study found that depression during inpatient rehabilitation was associated with poorer postdischarge outcomes at 1 year, including greater rehospitalization burden, worse pain severity, lower participation, and reduced life satisfaction, supporting the clinical value of early screening and timely intervention [59]. Positive affect is a crucial driver of self-management, whereas negative affect constitutes a significant barrier to self-care [60]. Our analysis further revealed that the most significant reduction in depressive symptoms occurred during the 3‐6 month follow-up period, whereas no significant effects were observed at 1‐3 months or beyond 6 months. We hypothesize that this may be attributed to the difficulty patients experience in accepting the reality of disability during the early stages of SCI, coupled with unstable treatment adherence. Over time, as patients gradually adapt and engage more actively with therapeutic measures, the benefits of telemedicine interventions become more evident during mid-term follow-up. Moreover, with the trend toward shorter hospital stays, patients with SCI now have fewer opportunities for inpatient rehabilitation and education, and higher adherence in the early postdischarge period may partially explain the absence of significant long-term differences [61,62]. Interestingly, one previous meta-analysis suggested no significant effects of telemedicine on depression at 3‐6 months (heterogeneity 59%) [8], whereas our findings demonstrated a significant benefit during the same period (heterogeneity 0%). This discrepancy highlights the need for additional high-quality studies to further clarify the actual impact of telemedicine on post-SCI depression. In addition, the contrast between the significant >3 and ≤6 months subgroup finding and the nonsignificant overall pooled effect indicates that any benefit for depressive symptoms should be interpreted as time-specific rather than as a consistent overall effect across follow-up periods, especially given the low certainty of the overall evidence. It is also important to note that different depression scales were used across the included studies, such as PHQ-9, HSCL-20B, BDI-II, CESD, DASS21-Depression, and HADS. In subgroup analyses restricted to HADS-based studies, the pooled estimate trended in the direction of improvement but remained statistically uncertain, highlighting limited power and the need for standardized assessment and reporting. Given that HADS has been widely used in medically ill populations and the depression subscale has supporting psychometric evidence [63], future trials using harmonized measures and clinically interpretable thresholds may help clarify whether the observed trends translate into consistent and clinically meaningful benefits.

With advances in medical care, the survival rate of patients with traumatic SCI has steadily increased [64]; however, this has been accompanied by long-term challenges such as mobility impairments, polypharmacy, bladder and bowel dysfunction, and restrictions in activities of daily living, all of which markedly reduce quality of life [65]. Observational studies using the WHOQOL-BREF consistently show substantially lower scores across domains in people with SCI compared with reference populations. In some cohorts, the psychological domain is among the lowest-scoring domains, underscoring the multidimensional and persistent burden of reduced quality of life in this population [66-68]. In our synthesis, telemedicine showed the most consistent benefit in the WHOQOL-BREF social domain at short-term follow-up, whereas evidence for the physical, psychological, and environmental domains remained uncertain and was supported by lower-certainty evidence. The social dimension encompasses interpersonal relationships, sexual life, and social support from friends, where telemedicine also showed favorable effects. In line with our findings, a previous meta-analysis reported that telemedicine interventions improved quality of life at 3 months, whereas the between-group difference disappeared at 12 months. However, given the limited sample size (only two studies), the representativeness of these results remains debatable [8]. Future trials should therefore report domain-level trajectories with adequate follow-up, document intervention intensity and engagement, and include clinically interpretable benchmarks to clarify whether early domain-specific gains translate into sustained improvements in broader quality-of-life outcomes.

Sleep disturbances are common after SCI and can exacerbate comorbidities that directly shape health-related quality of life, particularly mood, pain, and daytime functioning [69,70]. Recent syntheses indicate a substantial burden of insomnia symptoms in SCI and emphasize the need for routine sleep assessment across the continuum of care [71]. Among the studies included in our review, only two systematically assessed the effects of telemedicine using the PSQI. Our findings suggest that telemedicine had a limited impact at 1 month but more favorable effects at 3 months, although this interpretation should remain cautious because sleep outcomes were informed by only a small number of trials despite moderate-certainty evidence. The observed pattern of minimal short-term change with more favorable effects after sustained intervention is clinically plausible, as sleep improvements often require iterative adjustment and adherence over time. Consistent with this, a 12-week mHealth program incorporating multiple behavior-change strategies showed a favorable trend toward improved PSQI scores [72], and an earlier systematic review reported that telecounseling may improve sleep difficulties as part of broader comorbidity management, although the evidence base was small [54]. Until a larger and more methodologically consistent evidence base is available, conclusions regarding sleep benefits should remain cautious, and emphasis should be placed on standardizing definitions, follow-up windows, and reporting across trials.

Concerning functional independence, we did not observe significant improvements in pooled SCIM or CHART outcomes, and the overall evidence remained uncertain despite negligible heterogeneity for SCIM and moderate certainty for that outcome. Nonetheless, several individual trials provide signals that may help explain which intervention components are more likely to support functional gains. Nonetheless, individual trials provide signals that warrant interpretation. Swarnakar et al [40] reported substantial improvements in SCIM self-care and mobility domains after 8 weeks of telemedicine, with greater overall gains than controls. Similarly, Dallolio et al [19] found that 6 months of intervention significantly improved Functional Independence Measure (FIM) scores, particularly in grooming, dressing, and transfers, consistent with trends observed in SCIM. In contrast, Hossain et al [24] did not observe improvements in SCIM, which may reflect the intervention’s focus on complications such as pressure ulcers, urinary tract infections, incontinence, depression, and respiratory conditions rather than targeted functional training. Moreover, the absence of patient selection based on risk profiles or proactive help-seeking behaviors may have diluted the intervention’s impact [35]. Given that rehabilitation services in specialized centers are time-limited, telemedicine offers a valuable opportunity to extend long-term rehabilitation and professional guidance, particularly during the COVID-19 pandemic, when maintaining continuity of care remained crucial [40]. Taken together, these findings suggest that functional gains from telemedicine are more likely when interventions include clearly specified functional training components with sufficient intensity and when they are delivered to patients most likely to benefit.

Chronic pain is another common and debilitating consequence of SCI, with profound effects on daily functioning, mental health, and quality of life [73,74]. Oral pharmacological treatment for neuropathic pain after SCI is often inadequate, commonly resulting in only about a 20%‐30% reduction in pain intensity [75], while neuromodulation approaches (eg, transcranial direct current stimulation, transcranial magnetic stimulation) face limitations related to uncertain efficacy and accessibility [76]. Cognitive behavioral therapy–based pain management programs (CBT-PMP) have shown efficacy in improving mood, pain, and daily activities among individuals with SCI [77]. However, their broader implementation is constrained by mobility challenges and the need for specialized expertise [14]. In this context, telemedicine offers a practical route to extend access to structured psychological pain management. A Cochrane review of remotely delivered psychological therapies for chronic pain concluded that internet-delivered CBT can produce small improvements in pain intensity and functional disability, although benefits were not consistently maintained at follow-up, and evidence certainty varied across outcomes [78]. Telemedicine, particularly when integrated with CBT-PMP, is a promising approach to overcoming these barriers. Previous meta-analyses have demonstrated that internet-based psychological interventions can significantly alleviate symptoms in patients with chronic pain [79] and chronic diseases [80]. Although our review did not find significant improvements in SCI-related pain, the available trials suggested a possible benefit, but the evidence remained limited by small sample sizes, low certainty, and imprecision. Larger, multicenter RCTs with sufficient power are needed to clarify the true efficacy of telemedicine for pain management in SCI.

Several limitations should be considered when interpreting these findings. First, the number of included studies was relatively small, and some outcomes, such as sleep and pain, were reported in only a few trials, limiting precision and generalizability. Second, clinical and methodological heterogeneity was substantial, including variation in intervention modality, content, intensity, follow-up duration, and outcome measures, which likely contributed to imprecision and limited inference about which telemedicine approaches work best for specific outcomes. Third, many trials were underpowered and had relatively short follow-up, which constrains the assessment of durability and may underestimate longer-term effects that depend on sustained engagement. In addition, even within RCTs, differential uptake and engagement may occur after randomization. Participants with greater digital literacy, technology access, or comfort with telehealth may be more likely to adhere to and benefit from telemedicine interventions, whereas those with lower access or skills may disengage, which could introduce bias through deviations from intended interventions and differential attrition. Few trials reported baseline digital literacy or technology access, limiting our ability to assess effect modification by these factors and potentially constraining generalizability to settings with limited digital resources. Finally, subgroup analyses by delivery modality were limited, as most interventions relied primarily on telephone contacts, whereas emerging approaches such as virtual coaching, synchronous video, and app-supported monitoring may differ in effectiveness across domains.

This systematic review and meta-analysis offer an up-to-date, domain-based synthesis of randomized evidence on telemedicine interventions for SCI. Compared with earlier reviews, it incorporates additional recent trials and evaluates five prespecified outcome domains, enabling a more granular assessment of where evidence is most consistent and where uncertainty remains. Overall, telemedicine may be associated with improvements in selected patient-important outcomes, with the most consistent signals observed for social aspects of health-related quality of life and sleep-related functioning after sustained intervention exposure, whereas effects on depression, functional independence, and pain are less consistent and remain uncertain in several analyses. In real-world SCI care, these findings support telemedicine as a pragmatic adjunct to standard rehabilitation and follow-up, particularly for extending continuity and access for individuals facing mobility, geographic, or resource barriers. Future adequately powered randomized trials should standardize outcomes and follow-up, report intervention dose and engagement, and evaluate which telemedicine modalities and components work best for specific outcomes and patient subgroups to inform domain-specific, personalized tele-rehabilitation strategies.