A systematic review and meta-analysis on digital mental health interventions in inpatient settings

Mental health disorders are highly prevalent in Europe (15–40% annually) and impose substantial personal and economic burdens, yet fewer than one-third receive adequate treatment. Structural barriers (e.g., provider shortages, long wait times, stigma) limit access and effectiveness. E-mental health (EMH) interventions—delivered via web, apps, SMS, video, or digital monitoring—can bridge supply gaps, reduce waiting periods, and offer anonymity, and are effective and accepted in outpatient care across disorders such as depression, anxiety, eating disorders, PTSD, and work-related stress. Inpatient care addresses more severe cases and may benefit from EMH to blend with inpatient therapy, bridge waiting times, and improve stabilization and relapse prevention through aftercare. Prior evidence includes RCTs on EMH as inpatient add-ons and aftercare, and a systematic review suggesting efficacy of EMH aftercare, but no meta-analyses have synthesized EMH efficacy in inpatient settings or aftercare. This study conducts a meta-analysis of RCTs to estimate overall EMH effects in inpatient contexts, compare blended versus aftercare implementations, examine effects by disorder and therapy type, assess follow-up stability, evaluate risk of bias, and perform post-hoc analyses on EMH medium and control group type.

Prior meta-analyses demonstrate EMH efficacy in outpatient settings for depression, anxiety, eating disorders, PTSD, and workplace stress, with generally positive acceptance by patients and clinicians. Limited and dated systematic reviews exist for EMH in inpatient contexts, with one review supporting EMH aftercare but constrained by few studies and older literature. Existing RCTs have examined EMH as add-ons during inpatient care and as post-discharge aftercare. Gaps remain regarding synthesized inpatient-specific evidence, disorder-specific inpatient effects (e.g., psychosis, eating disorders), and the comparative impact of therapy models (CBT, psychodynamic), EMH media (web, app, SMS), and control conditions.

Design and registration: Systematic review and meta-analysis conducted per PRISMA 2020 and Cochrane Handbook guidelines; not preregistered. Search strategy: Databases searched included ScienceGov, PsycInfo, PubMed, and CENTRAL; ProQuest for dissertations; ICTRP and ClinicalTrials for trial registries. Primary search (February 2024) used: ("digital" OR "online" OR "e-mental health" OR "technology-based" OR "web-based" OR "internet-based" OR "mobile-based") AND ("psychotherapy" OR "psychiatric" OR "psychosomatic") AND ("inpatient" OR "ward patient" OR "hospitalized") AND ("RCT" OR "randomized controlled trial"). Searches were in English and German by two researchers. A secondary search (July 2024) expanded to “e-health”, “mhealth”, “telemedicine” and MeSH terms (“digital health”, “telemedicine”, “psychotherapy”, “psychosomatic medicine”, “inpatients”) for PubMed and CENTRAL; no additional eligible studies found. Eligibility criteria: Inclusion—RCTs (including cluster and pilot RCTs) evaluating EMH during inpatient treatment (blended) or aftercare post-inpatient; or psychiatric symptoms in somatic inpatient populations; reporting mental health outcomes (e.g., symptom severity, relapse/readmission, quality of life). Exclusion—non-EMH interventions; non-inpatient populations; outcomes not mental health related (purely somatic or purely acceptability; neuropsychological/cognitive-only outcomes excluded); non-RCT designs; insufficient data for effect extraction; high risk of bias per RoB 2. Study selection: Three independent raters screened and selected studies with consensus resolution. Risk of bias: Assessed using Cochrane RoB 2 across domains (randomization, allocation concealment, blinding of participants/personnel, blinding of outcome assessment, incomplete outcome data, selective reporting). Ratings: low/medium/high risk; high-risk studies excluded. Common concerns included attrition (high/uneven/unclear), non-random allocation (e.g., alternating), and inadequate blinding information. Outcomes and data extraction: Extracted variables—author, year, country, treatment type (blended vs aftercare), disorder category, somatic comorbidity, EMH medium, therapy type, control group type (active, passive, additive), outcome measures, follow-ups, sample sizes, and summary statistics (means, SDs, ORs, effect sizes). One rater extracted data; two raters verified. Measures included clinical symptoms and psychosocial performance (e.g., relapse/readmission, symptom severity, general psychopathology, well-being/quality of life, employment when relevant, and related psychosocial metrics). Studies served as random effects when multiple outcomes were included. Variable categorization: Treatment type (blended vs aftercare). Disorder types per ICD-10/DSM-5: anxiety (F40–F41), eating (F50), mood (F3), psychotic (F2), substance use (F1x.2), transdiagnostic, somatic comorbidity, and return to work. Therapy type classified per study (CBT, psychodynamic, NA if not specified). Effect size computation: Hedges’ g used. Transformations: from Cohen’s d using g = d(1 − 3/(4df − 1)) and v_g = v_d(1 − 3/(4df − 1))^2; from odds ratios via d = log(OR)√(3/π) then to g. Random-effects models (study as random factor) were used. Heterogeneity was anticipated and assessed. Total effect analyzed twice: all included outcomes and clinically relevant outcomes only. Publication bias assessed via funnel plots (regression of effect on SE) and p-curve. Confidence intervals from standard errors. Data and code availability: Data (search lists, included studies, extracted data, RoB assessments) and R code available at OSF: https://osf.io/bc59e. Analyses in R (RStudio 2021.9.1.0; R 4.1.2) using dmetar and metafor.

Study selection and characteristics: 26 RCTs with 123 effects (n=6112; intervention n=3041; control n=3071) were included after excluding 4 high risk-of-bias studies. Five studies implemented blended EMH during inpatient stay; 21 implemented EMH as aftercare. Most frequent patient groups: eating disorders (k=7), mood disorders (k=6), transdiagnostic (k=4), psychotic disorders (k=3), return-to-work (k=2), somatic comorbidity (k=2), anxiety disorders (k=1), substance abuse (k=1). Control groups: 13 passive, 8 active alternative, 5 additive (EMH added to active control), and 1 both. EMH media: 18 web-based (e.g., SUMMIT, IN@, HEINS, Deprexis, GSA Online, EDINA), 5 app-based (e.g., MCT & More, Mindshift), 3 SMS-based. Most studies conducted in Germany (17); others in Sweden (2), USA (2), and one each in Hungary, Iran, Finland, Canada, Australia. Risk of bias: Of the included studies, 19 had medium risk and 7 low risk. High-risk studies were excluded due to unclear/high/uneven attrition and inadequate randomization/blinding. Publication bias: Funnel plot regression indicated asymmetry across all effects (z=3.6, p<0.001; i=123). After excluding 4 effects with largest SEs (i=119), asymmetry was not significant (z=1.89, p=0.058) and the total effect size was unchanged (g=0.33 [0.20, 0.46], p<0.001). P-curve of 109 significant effects showed strong right-skew (P<0.001), indicating a true effect; flatness tests were non-significant, suggesting no p-hacking. Overall effects: Across all outcomes, EMH showed a small, significant effect (g=0.30 [0.20, 0.39], p<0.001; k=118). Limiting to clinically relevant outcomes increased the effect (g=0.36 [0.22, 0.50], p<0.001; k=83). Heterogeneity was significant (all outcomes: Q(117)=408.25, p<0.001; clinically relevant: Q(82)=647.91, p<0.001). By treatment type: Blended EMH during inpatient care: g=0.42 [0.27, 0.58], p<0.001 (k=19). Aftercare EMH following inpatient care: g=0.29 [0.24, 0.34], p<0.001 (k=99). By disorder: Eating disorders g=0.19 [0.07, 0.32], p=0.003 (k=17); Mood disorders g=0.38 [0.28, 0.49], p<0.001 (k=22); Psychotic disorders g=0.43 [0.27, 0.58], p<0.001 (k=10); Return-to-work g=0.21 [0.12, 0.30], p<0.001 (k=24); Transdiagnostic g=0.40 [0.31, 0.49], p<0.001 (k=34). Non-significant: Anxiety disorders g=0.35 [−0.22, 0.93], p=0.23 (k=3); Somatic comorbidity g=0.19 [−0.02, 0.39], p=0.072 (k=6); Substance abuse g≈0.006 [−0.27, 0.28], p=0.964 (k=2). Post-hoc focusing on anxiety symptoms across studies showed a significant effect (g=0.39 [0.18, 0.59], p<0.001). By therapy type: CBT-based EMH g=0.26 [0.18, 0.34], p<0.001 (k=43); Psychodynamic EMH g=0.35 [0.27, 0.43], p<0.001 (k=39). Follow-up stability: No significant moderation by time to measurement (t(61)=−0.97, p=0.337). Effects observed up to 24 months post-treatment showed no decline. Post-hoc by EMH medium: Significant effects for web-based tools (g=0.32 [0.25, 0.37], p<0.001) and multimedia (g=0.79 [0.29, 1.29], p=0.002; single study). App-based and SMS-based effects were not significant. Deprexis (used in 3 studies) showed g=0.61 [0.46, 0.77], p<0.001. Post-hoc by control condition: EMH outperformed passive controls (g=0.29 [0.19, 0.39], p<0.001), active alternative treatments (g=0.32 [0.24, 0.40], p<0.001), and improved outcomes when added to active treatment (g=0.30 [0.22, 0.39], p<0.001).

This meta-analysis provides the first quantitative synthesis of EMH efficacy specifically in inpatient contexts and post-discharge aftercare. Findings demonstrate small but significant benefits overall, with larger effects for clinically relevant outcomes and when EMH is blended with inpatient care. Effects are consistent across major therapy modalities (CBT and psychodynamic) and multiple diagnostic categories, notably psychotic, mood, and transdiagnostic populations. The absence of effect moderation by follow-up time (up to 24 months) suggests stability of benefits. The results support integrating EMH into inpatient treatment pathways and aftercare to address access barriers, enhance stabilization, and prevent relapse, complementing evidence from outpatient settings. Web-based interventions have the strongest evidence base among EMH media; specific platforms like Deprexis show robust effects. Although psychosis showed relatively strong effects, the literature is mixed and limited; interactive, multi-module platforms (e.g., HEINS, Horyzons) may be necessary, and EMH should augment rather than replace in-person care due to potential concerns (e.g., paranoia, engagement). For eating disorders, especially anorexia nervosa, evidence remains sparse; promising signals warrant caution given severity and somatic risks. Non-significant subgroup findings for anxiety, somatic comorbidity, and substance use likely reflect underpowered subgroups; a post-hoc analysis showed meaningful improvements in anxiety symptoms. Overall, EMH can enhance usual care and function as a scalable aftercare option where treatment as usual is minimal, helping to close supply gaps. Nevertheless, heterogeneity and medium RoB in many studies emphasize cautious interpretation and the need for more rigorous trials.

EMH interventions are effective for inpatient mental health care when blended with in-person treatment and as aftercare, with small-to-moderate effects across diagnoses and therapy models. Benefits appear stable up to two years post-treatment and hold against both passive and active control conditions. Web-based tools have the most consistent support; specific platforms (e.g., Deprexis) show strong effects. Future research should prioritize: larger, high-quality RCTs; disorder-specific inpatient evaluations (e.g., psychosis, anorexia nervosa, substance use); direct comparisons of EMH media and specific components; standardized, clinically meaningful outcomes (e.g., remission/relapse); longer-term follow-up beyond 24 months; and broader geographic representation to improve generalizability.

- Small number of included studies for several subgroups (e.g., anxiety disorders, substance abuse, whole health), limiting subgroup inference and power. - Majority of studies rated as medium risk of bias; four high-risk studies excluded due to attrition and methodological issues. - Significant heterogeneity across studies in designs, populations, and outcomes. - Outcomes predominantly symptom questionnaires; fewer studies reported clinically decisive endpoints (e.g., remission, relapse/readmission); only six studies reported such outcomes across varied disorders. - Meta-analysis not preregistered; despite two search waves (Feb and Jul 2024), relevant studies may have been missed due to search terms/databases. - Regional concentration in Western/Northern Europe (especially Germany); limited data from other regions reduces generalizability. - Follow-up effects only assessed up to 24 months; longer-term durability unknown. - Engagement/adherence issues and their influence on outcomes remain unclear; mixed evidence on usage-intensity effects. - EMH implementation risks include variable quality standards, data privacy concerns, and digital literacy barriers for clinicians and patients. - Exclusion of neuropsychological/cognitive-only outcomes narrows scope to mental health symptoms and psychosocial measures.