The study selection criteria were explicitly defined using the well-established PICO (Population, Intervention, Comparison, Outcome) framework:

Population (P): Critically ill patients aged ≥18 years who were formally admitted to an ICU for a period exceeding 24 h.

Intervention (I): Occupational therapy, delivered either as a standalone therapeutic regimen or as a core, clearly definable component within a broader multidisciplinary intervention protocol where the distinct OT contribution could be accurately described.

Comparison (C): Routine ICU care, which could encompass standard physical therapy and/or nursing care as directed by the primary medical team, but explicitly excluded the specific, protocolized OT intervention under investigation.

Outcomes (O): The pre-specified primary outcome was the level of independence in activities of daily living (ADL), measured using validated instruments such as the Barthel Index or the Functional Independence Measure (FIM). Secondary outcomes of interest included the incidence of delirium (assessed using the Confusion Assessment Method for the ICU [CAM-ICU]), grip strength (objectively measured in kilograms using a standardized hand dynamometer), and the total duration of mechanical ventilation (recorded in hours or days).

Study Design: Only randomized controlled trials (RCTs) were considered for inclusion to maximize internal validity. Exclusion criteria comprised: duplicate publications, non-Chinese/English language publications, unavailable full-text articles, conference abstracts, dissertations, case reports, and non-randomized observational studies.

A comprehensive and systematic search strategy was developed in collaboration with an experienced academic librarian. Two reviewers independently performed systematic electronic searches in the following databases from their inception to August 5, 2025: PubMed, Embase, Web of Science, CINAHL, Cochrane Central Register of Controlled Trials, CNKI, Wanfang Database, and SinoMed. The search strategy strategically combined relevant controlled vocabulary terms [e.g., Medical Subject Headings (MeSH) in PubMed, Emtree in Embase] with extensive free-text keywords related to three core concepts: (1) intensive/critical care, (2) occupational therapy, and (3) randomized controlled trials. No restrictions were applied concerning publication date or status. The complete, detailed search strategy utilized for PubMed is provided in Supplementary Table 1 and was logically adapted for syntax and terminology across all other databases.

Two reviewers, both formally trained in evidence-based practice and systematic review methodology, independently screened all retrieved records for potential eligibility. Following the removal of duplicate citations using EndNote 20 software, the titles and abstracts of all remaining records were screened for preliminary relevance. The full-text articles of all potentially eligible studies were then procured and thoroughly assessed against the pre-specified inclusion and exclusion criteria. Any disagreements arising at any stage of the selection process were resolved through iterative discussion or, when necessary, by consulting a third senior reviewer. A standardized, piloted data extraction form was employed to systematically collect the following information from each included study: (1) basic study characteristics (title, first author, country, year of publication, trial registration number); (2) detailed participant information (total sample size, specific ICU type, patient demographics, baseline illness severity scores); (3) comprehensive intervention characteristics (precise description of both the OT and control interventions, including timing of initiation, session frequency, total intervention duration, and provider credentials); and (4) complete outcome data (means, standard deviations, and event counts for all relevant outcomes at all reported assessment time points). Where data were incomplete or unclear, corresponding authors were contacted via email on two separate occasions to request the missing information. Five studies were excluded due to unavailable data despite our attempts to contact the authors.

The methodological quality and risk of bias of each included RCT were independently evaluated by two reviewers using the Cochrane Risk of Bias tool for randomized trials (RoB 2.0) (10). This rigorous tool assesses potential bias across five key domains: bias arising from the randomization process, bias due to deviations from the intended interventions, bias due to missing outcome data, bias in the measurement of the outcome, and bias in the selection of the reported result. For each study, explicit judgments for each domain were made, culminating in an overall risk of bias judgment, categorized as ‘low risk of bias’, ‘some concerns’, or ‘high risk of bias’. Any disagreements between reviewers were resolved through consensus to reach a final rating.

Meta-analysis was performed utilizing Review Manager (RevMan) software, version 5.3. For continuous outcomes, the mean difference (MD) was calculated when the same measurement instrument was used consistently across studies; conversely, the standardized mean difference (SMD) was employed when different, though conceptually similar, measurement tools were used, with conventional interpretations of SMD values of 0.2, 0.5, and 0.8 representing small, medium, and large effect sizes, respectively. For dichotomous outcomes, the relative risk (RR) was calculated as the measure of association. All pooled effect estimates are reported with their corresponding 95% confidence intervals (CIs). Statistical heterogeneity among the included studies was quantitatively assessed using the Chi² test (employing a conservative significance level of p < 0.10 to indicate potential heterogeneity) and the I² statistic, which quantifies the proportion of total variation attributable to heterogeneity. An I² value of less than 50% was considered to indicate low heterogeneity, in which case a fixed-effect model was applied for pooling. An I² value of 50% or higher was interpreted as indicating substantial heterogeneity, warranting the use of a more conservative random-effects model (utilizing the DerSimonian and Laird method) and prompting the performance of pre-specified subgroup analyses to explore potential clinical or methodological sources of heterogeneity [e.g., type of OT intervention (standalone vs. combined within multidisciplinary care), nature of the control group comparison, specific assessment tool used]. Sensitivity analyses were conducted by sequentially removing each individual study from the meta-analysis to assess the robustness and stability of the pooled results. A formal assessment of publication bias using funnel plots and statistical tests such as Egger’s test was originally planned but was subsequently deemed methodologically inappropriate due to the small number of studies (<10) included in each separate meta-analysis, as the statistical power of such tests is unacceptably low to reliably distinguish chance from real asymmetry under these conditions.

We also conducted a pre-specified subgroup analysis to evaluate the efficacy of OT in critically ill patients with stroke versus those without stroke, where data were available. This subgroup analysis aimed to explore whether the benefits of OT were consistent across these clinically distinct populations.

The initial execution of the systematic search strategy across all designated electronic databases yielded a total of 3,115 unique records. Following the removal of 652 duplicate citations, 2,463 records underwent preliminary screening based on their titles and abstracts. From this initial screening, 33 full-text articles were deemed potentially eligible and were retrieved for a detailed, in-depth eligibility assessment. Of these, 24 studies were excluded for the following reasons: interventions were inconsistent with the review’s definition of OT (n = 16), study design was not an RCT (n = 3), and essential data could not be extracted (n = 5). This process culminated in the final inclusion of nine RCTs that fully met all inclusion criteria, encompassing a total study population of 819 critically ill patients. The complete, sequential study selection process is delineated in the PRISMA flow diagram (Figure 1).

The key methodological and clinical characteristics of the nine included RCTs are comprehensively summarized in Table 1. Collectively, the studies randomized 409 patients to experimental (OT) groups and 410 patients to control groups. The mean age of participants, where reported, was predominantly over 50 years across the included study cohorts. All described interventions were delivered within a hospital-based ICU setting. The control groups uniformly received some form of routine care, which, depending on the specific ICU protocol, could include elements of physical therapy. The experimental interventions were categorized as consisting of OT alone in four trials (11–14) and OT integrated with physical therapy in five trials (15–19). The most frequently reported core component of the OT interventions was ADL training (11–15, 17–19) (implemented in 8 RCTs, n = 757 patients), typically involving practical, graded exercises such as dressing, grooming, beading, or manipulating buttons. Cognitive interventions (11–13) (employed in 3 RCTs, n = 249 patients), including structured activities like card games, multisensory stimulation, and memory retraining exercises, were also commonly incorporated. One trial (16) did not provide specific details regarding the constituent OT activities. Methodological quality indicators revealed that five of the included trials (11, 13, 15–17) had prospectively registered their study protocols, four trials (11, 13, 15, 17) explicitly reported conducting an a priori sample size calculation, and five trials (12, 13, 15–17) incorporated a designated follow-up period beyond the immediate ICU stay.

The results of the methodological quality assessment using the Cochrane RoB 2.0 tool are presented graphically in Figure 2. Concerning the domain of bias arising from the randomization process, five trials (11, 13, 15–17) were judged to be at low risk, typically due to the reported use of computer-generated random sequence generation and adequate allocation concealment. The remaining four studies (12, 14, 18, 19) provided insufficient methodological detail regarding the randomization procedure and were consequently rated as having ‘some concerns’. Due to the inherent nature of the complex OT intervention, blinding of participating patients and intervening therapists was not pragmatically feasible in any of the included studies, leading to an unequivocal rating of ‘high risk’ for the domain of bias due to deviations from the intended interventions across all studies. The blinding of outcome assessors was explicitly reported and judged to be at low risk of bias in five trials (11, 13, 15, 17, 18) for objectively measured outcomes such as grip strength and ventilation duration; however, for more subjective, patient-reported outcomes like ADL, the risk of measurement bias was generally considered high. Incomplete outcome data were adequately addressed with appropriate statistical methods in seven studies (11–13, 15–17, 19), whereas two studies (14, 18) demonstrated notable imbalances in missing data between the intervention and control groups, raising some concerns regarding potential attrition bias. Selective reporting of results was deemed to be at low risk for all included studies, as the pre-specified primary outcomes outlined in the methods sections were consistently reported in the results. The overall risk of bias judgment was consequently ‘high’ for all nine studies, a determination primarily driven by the unavoidable lack of blinding for participants and personnel.

Sensitivity analyses, performed by systematically omitting one study at a time from each meta-analysis, confirmed that the pooled estimates for all primary outcomes remained statistically significant and stable in direction and magnitude, with the exception of the ADL outcome. For ADL, while statistical significance (p < 0.05) was consistently maintained, the precise point estimate of the SMD exhibited expected minor fluctuations, a common occurrence when substantial heterogeneity is present. This suggests that no single study was entirely responsible for the observed heterogeneity or the overall significant finding, though the result should be interpreted with an understanding of its underlying variability. As pre-specified in the protocol, a formal quantitative assessment of potential publication bias using funnel plots and associated statistical tests (e.g., Egger’s test) was not undertaken. This decision was based on the well-established methodological principle that such tests possess unacceptably low statistical power and are prone to misleading conclusions when the number of contributing studies in a meta-analysis is fewer than 10 (Supplementary Tables 2a,b).

The overall certainty of evidence for each primary outcome was assessed using the GRADE framework and is summarized in the Summary of Findings table (Supplementary Table 3).

Activities of Daily Living (ADL): The evidence was rated as low certainty. It was downgraded by one level due to serious risk of bias (high risk of performance bias across all studies) and by one further level due to serious inconsistency (substantial unexplained statistical heterogeneity, I² = 56%).

Incidence of Delirium: The evidence was rated as moderate certainty. It was downgraded by one level due to serious risk of bias (high risk of performance bias) but was not downgraded for other domains, as inconsistency, indirectness, and imprecision were not deemed serious.

Grip Strength: The evidence was rated as moderate certainty. It was downgraded by one level due to serious risk of bias (high risk of performance bias) but was not downgraded for other domains.

Duration of Mechanical Ventilation: The evidence was rated as moderate certainty. It was downgraded by one level due to serious risk of bias (high risk of performance bias) but was not downgraded for other domains.

Publication bias remained undetectable for all outcomes due to the limited number of included studies (<10).

The conclusions of this review should be interpreted in light of its limitations, which can be categorized into those inherent to the primary included studies and those specific to this systematic review itself.

Notwithstanding these acknowledged limitations, the cumulative evidence synthesized in this review suggests that the timely integration of occupational therapy into the standard, multidisciplinary care of critically ill patients represents a promising and viable strategy for improving multidimensional recovery and potentially mitigating the burden of PICS. Based on the current evidence, we recommend that occupational therapy be considered as a core component of early rehabilitation bundles in the ICU, particularly for high-risk populations such as stroke survivors, older patients, and those requiring mechanical ventilation. Interventions should include protocolized, task-oriented ADL training and cognitive stimulation, initiated as early as clinically feasible (39). For current clinical practice, our findings provide a compelling evidence base to support the active consideration and implementation of early, protocolized OT interventions as an integral component of bundled care strategies aimed at preventing and managing the complex sequelae of critical illness. Occupational therapists, with their unique expertise in functional recovery, cognitive rehabilitation, and psychosocial support, should be considered integral members of the modern ICU multidisciplinary team. Future research in this field should unequivocally focus on conducting larger, rigorously designed, multicenter RCTs that employ standardized, yet potentially adaptable, OT intervention protocols to confirm and extend these promising findings. These future trials should prioritize the implementation of robust blinding procedures for outcome assessors and should place a strong emphasis on measuring patient-centered outcomes that matter most to survivors, such as long-term quality of life, successful return to work, and community reintegration (40). It would be particularly valuable to investigate the optimal timing of OT initiation (e.g., very early vs. early), the ideal daily frequency and total duration of sessions, and its differential impact on specific, well-defined patient subgroups (e.g., surgical versus medical ICU patients, patients with versus without pre-existing baseline cognitive impairment). Exploring the long-term sustainability of the benefits conferred by ICU-based OT interventions, particularly their effect on functional status and quality of life at 6 and 12 months post-discharge, represents another critical and clinically relevant avenue for future investigation. Additionally, well-conducted economic evaluations and formal cost-effectiveness analyses are urgently warranted to determine the economic value and potential resource implications of implementing dedicated ICU OT programs within different healthcare systems (41). Finally, while the current review demonstrates the clinical benefits of OT, the economic implications of implementing such services in the ICU remain largely unexplored. Future research should include formal cost-effectiveness analyses to evaluate the long-term economic value of OT interventions, considering potential reductions in ventilator days, ICU length of stay, and the burden of long-term disability (42), which would provide crucial information for healthcare policymakers and administrators.