Care Robots as Emerging Health Technologies: Systematic Review and Meta-Analysis of Randomized Controlled Trials

Clinical Snapshot

Key Findings

• P Value: Neuropsychiatric symptoms: P=.04 (fragile; P=.10 after exclusion of Petersen 2016). All other domains: non-significant

• Effect Size: Neuropsychiatric symptoms: Hedges g=0.44 (small-to-moderate effect). Other domains: QoL g=0.15, depression g=0.20, agitation g=0.32, stress/pain g=0.53, social-communicative skills g=0.45, cognitive function g=0.18 — all non-significant

• Primary Outcome: Neuropsychiatric symptoms: the only domain reaching statistical significance under HKSJ-corrected random-effects pooling (k=6 RCTs)

• Nnt Or Sensitivity: No NNT calculable given non-significant or fragile pooled effects and very low GRADE certainty. Sensitivity analysis (leave-one-out) for neuropsychiatric symptoms: significance lost in 5 of 6 iterations, confirming result fragility. GRADE certainty: low for neuropsychiatric symptoms; very low for all remaining six domains

• Confidence Interval: Neuropsychiatric symptoms: 95% CI 0.03–0.84 (marginally excludes null). All other domains: 95% CIs cross zero. Prediction intervals encompass the null for all seven domains (e.g., neuropsychiatric PI: −0.42 to 1.30; stress/pain PI: −1.57 to 2.62)

Clinical Application

Care robot deployment faces significant implementation barriers: high acquisition and maintenance costs, need for trained facilitators to support interactions, technological heterogeneity across platforms, and infrastructure requirements. The review's conclusion that robots should function as 'facilitator-supported adjuncts' rather than autonomous care substitutes has direct workforce and cost implications. Feasibility is further constrained by the absence of robust evidence to justify capital expenditure at scale. In Australia, care robots are not currently listed on the Prostheses List (PLDB) or subsidised through the Pharmaceutical Benefits Scheme (PBS), meaning costs fall entirely to facilities or consumers. The Aged Care Quality and Safety Commission's focus on person-centred care and the Royal Commission into Aged Care Quality and Safety recommendations emphasise human connection and staffing adequacy — contexts in which care robots may serve as adjuncts during workforce shortages but must not substitute for mandated human care ratios. The TGA classifies interactive therapeutic devices under the medical device regulatory framework; companion robots such as PARO may require assessment depending on therapeutic claims made. RACGP guidelines do not currently address care robot prescribing. Australian aged care providers considering care robot procurement should note that this meta-analysis provides insufficient evidence to support routine adoption, and any implementation should occur within an evaluation framework aligned with the Australian Commission on Safety and Quality in Health Care's digital health safety standards. Older adults (particularly those with dementia or neuropsychiatric symptoms in residential aged care) and paediatric populations requiring social or therapeutic support. The evidence base is most relevant to institutional settings (aged care facilities, paediatric wards) rather than community or home-based deployment. Cross-population pooling limits precision for any single clinical group.

Abstract

BACKGROUND: Global aging and health care workforce shortages are increasing demand for therapeutic support among older adult and pediatric populations. Care robots, an umbrella term for socially assistive robots, companion and pet platforms, and therapeutic humanoids sharing a social interaction component, have been proposed as embodied digital health adjuncts, but prior syntheses have remained population- or platform-specific, leaving cross-population and cross-platform effects unquantified. OBJECTIVE: This review quantifies the pooled effects of care-robot interventions across 7 prespecified outcome domains, examines robot platform, target population, and intervention-duration moderators, and grades evidence certainty. METHODS: Following PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020, we searched 5 databases (PubMed, including MEDLINE, Embase, Cochrane CENTRAL, CINAHL, APA PsycINFO) and 2 trial registries (ClinicalTrials.gov; World Health Organization International Clinical Trials Registry Platform) through April 29, 2026, with no language restriction. Eligible studies were randomized controlled trials of embodied care robots versus standard care, active controls, or waitlist. Random-effects meta-analysis used Hedges g with Hartung-Knapp-Sidik-Jonkman correction; for domains with k≥4, we report 95% CIs and prediction intervals (PIs). Risk of bias used the Cochrane Risk of Bias 2 tool; certainty used the Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework. RESULTS: A total of 34 randomized controlled studies (n=2476; 17 countries; 2015-2024) met inclusion criteria; 20 contributed to at least 1 meta-analysis, and 14 entered narrative synthesis. Under Hartung-Knapp-Sidik-Jonkman pooling, only neuropsychiatric symptoms reached significance: Hedges g=0.44 (95% CI 0.03-0.84; 95% PI -0.42 to 1.30; k=6; P=.04). The remaining 6 domains were favorable but nonsignificant: quality of life Hedges g=0.15 (95% CI -0.41 to 0.71; 95% PI -0.86 to 1.16; k=5); depression Hedges g=0.20 (95% CI -0.08 to 0.49; 95% PI -0.36 to 0.76; k=7); agitation Hedges g=0.32 (95% CI -0.07 to 0.71; 95% PI -0.26 to 0.89; k=4); stress and pain Hedges g=0.53 (95% CI -0.48 to 1.53; 95% PI -1.57 to 2.62; k=6); social-communicative skills Hedges g=0.45 (95% CI -0.52 to 1.42; 95% PI -1.66 to 2.56; k=5); and cognitive function Hedges g=0.18 (95% CI -0.62 to 0.98; 95% PI -1.11 to 1.47; k=4). All 7 PIs encompassed the null, indicating no assured effect in new settings. The neuropsychiatric symptom result was fragile: significance was lost in 5 of 6 leave-one-out iterations and on excluding Petersen 2016 (data extraction ambiguity), yielding a Hedges g of 0.40 (P=.10). GRADE certainty was low for neuropsychiatric symptoms and very low for the remaining 6 domains. CONCLUSIONS: Across cross-population, cross-platform randomized controlled trial pooling, the evidence does not yet support routine clinical adoption of care robots. Potential benefits are narrow, of low certainty, and contingent on adequately powered multicenter confirmatory trials with PI-informed sample sizes. Care robots are best framed as facilitator-supported adjuncts that augment, rather than substitute for, human-delivered care.