Effectiveness of Robot Interventions for Cognitive and Psychological Outcomes among Older Adults with Cognitive Impairment: A Meta-Analysis

This review was performed to evaluate the effects of robot interventions on cognitive and psychological outcomes among older adults with cognitive impairment. Three databases (PubMed, Embase, and Cochrane Central Register of Controlled Trials) were searched for studies published in English between January 2015 and August 2021. We included studies that involved older adults with cognitive impairment, interventions using robots, outcome measures related to cognitive and psychological status, and randomized controlled trials. Ten studies included in the systematic review, and nine studies derived from these ten articles were included in the meta-analyses. The meta-analyses revealed that robot interventions significantly decreased anxiety and agitation but exerted no significant effects on cognitive function, neuropsychiatric symptoms, and quality of life. The subgroup analyses according to robot types revealed that pet-type robot interventions reduced anxiety and agitation. In addition, subgroup analysis according to the intervention format of robot interventions found that individual intervention was effective for improving agitation, but a group-based intervention was effective for improving depression. We suggest using robot interventions to improve psychological outcomes such as anxiety and agitation; however, further research is needed to determine whether robot interventions affect symptoms such as cognitive function, neuropsychiatric symptoms, and quality of life.


Introduction
Owing to the increasing aging population, dementia has been recognized as a public health priority worldwide.Currently, the global prevalence of dementia has reached approximately 55 million people and is expected to nearly triple to more than 152 million people by 2050 [1].Along with this rapid increase in prevalence, the health-economic cost of dementia is also expected to increase considerably [2].
Dementia is a degenerative neurocognitive disorder characterized by a decline in cognitive function, and its clinical features include cognitive impairment and clinically significant behavioral and psychological disturbances [3].The behavioral and psychological symptoms of dementia, including aberrant motor behavior, depression, appetite and eating changes, agitation, anxiety, sleep disturbances, night-time behavior, delusion, irritability, and hallucination, affect patients' quality of life and increase caregivers' burdens [4].Previous systematic reviews suggested that agitation and anxiety increased as the severity of dementia increased, and the prevalence of symptoms such as depression, agitation, and anxiety was as high as 20% or more for community-dwelling older adults with dementia [5].Meanwhile, various symptoms such as depression, delusion, hallucinations, apathy, and sleep disorders appearing in patients with dementia are grouped into a cluster called Healthcare 2023, 11, 2341 2 of 20 neuropsychiatric symptoms, and these symptoms negatively affect the quality of life of patients with dementia [6].
Dementia can be preceded by mild cognitive impairment (MCI).MCI is a transition stage between the cognitive decline expected in usual aging and the cognitive decline in dementia [7].According to a study by Mitchell and Shiri-Feshki [8], the rate of annual transition from MCI to dementia was 9.6% and 4.9% in specialist settings and general population settings, respectively.According to previous studies, it is estimated that 35-85% of older adults with MCI experience neuropsychiatric symptoms [9].Depression, anxiety, and agitation were reported as the most common symptoms [9].Although the effectiveness of non-pharmacological interventions for older adults with MCI on dementia prevention has not been fully proven, the rationale has been published that they reduce the risk of progression to cognitive decline or dementia [10].Therefore, recent studies are attempting to identify the effects of interventions during the MCI period on the prevention of dementia and the maintenance of cognitive function.
The increased care burden for older adults with dementia has led to the development of robotic assistive technology.As a result of prior review of older adults' experiences and perceptions of socially assistive robots, some older adults express negative attitudes or fears towards robots, but the older adults have relatively positive experiences about the function or usefulness of socially assistive robots [11].Socially assistive robot technology was designed to meet the social and psychological needs of older adults through humanrobot interactions [12].Socially assistive pet-type robots can act as companions for older adults by imitating the appearances and behaviors of animals, whereas socially assistive humanoid robots have human-like shapes and features; both types of robots can interact with older adults [13].Tanaka et al. [14] presented the effects of an intervention using a communication robot for older women and found that the intervention was effective in improving cognitive function, fatigue, and motivation.Additionally, Lee et al. reported that robot intervention for older adults with MCI effectively improved cognitive function and anxiety [15].Thus, these robots have been used to deliver cognitive or psychosocial support for older adults.
Considering that research on socially assistive robots has increased, evaluating their effectiveness is of great importance.A recent literature review focused on humanoid robots and comprehensively described their effects on patients with dementia; however, a meta-analysis for quantitative synthesis was not conducted [16].Other systematic reviews evaluated the effects of robot interventions through meta-analysis; however, the reviewed participants were restricted to patients with dementia [17,18].Furthermore, a systematic review evaluated the effects of socially assistive robots on older adults with dementia or cognitive impairment through meta-analysis; however, the reviewed outcomes were restricted to three psychological outcomes: agitation, quality of life, and depression [19].
We aimed to evaluate the effects of interventions that use robots on cognitive and psychological outcomes among older adults with cognitive impairment.We sought to resolve the unreviewed issues in previous literature reviews.Given that robot interventions may be more beneficial for older adults with MCI because they can be more responsive to robots than those with severe cognitive impairment [20], we reviewed studies involving older adults with cognitive impairment, including dementia and MCI, regardless of the severity of cognitive impairment.Moreover, we also reviewed interventions that used robots without restrictions on the type of robot.Additionally, we evaluated the outcomes related to cognitive and psychological status without restrictions to identify which outcomes were examined in previous studies and to examine the effects of robot interventions through a meta-analysis.We then conducted subgroup analyses to evaluate effects according to robot types and intervention format because previous studies reported variations in user experiences linked to factors such as the robot's appearance, design, and the specific tasks and methods employed when implementing robot interventions for older adults [11].

Design
This systematic review and meta-analyses were performed according to the Cochrane Handbook for Systematic Reviews of Interventions version 6.2 [21] and reported according to the standards of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 Statement [22].

Eligibility Criteria
The inclusion criteria were as follows: (a) studies that involved older adults with cognitive impairment, including dementia and MCI, regardless of severity; (b) studies that provided an intervention using robots; (c) studies that examined outcome measures related to cognitive and psychological status; and (d) randomized controlled trials (RCTs) or cluster randomized trials (CRTs) with a comparison group.
The exclusion criteria were as follows: (a) studies with samples including people without cognitive impairment; (b) studies that did not include outcome variables related to cognitive and psychological status; and (c) reviews, study protocols, conference abstracts, editorials, and observational studies.

Information Sources and Search Strategy
We searched three core databases (PubMed, Embase, and Cochrane Central Register of Controlled Trials) according to the Cochrane Handbook for Systematic Reviews of Interventions on 31 August 2021 [23].We developed a search strategy by combining controlled vocabulary search terms [e.g., Medical Subject Headings (MeSH) and Emtree] and text words [e.g., cognitive impairment, mild cognitive, MCI, cognitive decline, memory impairment, memory decline, dementia, Alzheimer, dementia (MeSH), robot*, robotics (MeSH), randomized, controlled clinical trial, randomized controlled trial, placebo, clinical trials as a topic (MeSH), randomly, and trial].Since 2015, experimental studies using robots to care for older people have increased [17,18].Therefore, the search was limited to articles published in English from January 2015 to August 2021.Supplementary Table S1 presents search strategies for databases.

Selection Process
The search results across databases were merged using EndNote ® (EndNote X9.3.3,Thomson Reuters, NY, USA).The duplicates were electronically and manually removed.Subsequently, the eligibility of the remaining articles was reviewed based on titles and abstracts.After removing irrelevant articles, the full texts of the remaining articles were assessed to determine study eligibility.Two independent researchers screened the articles and resolved any discrepancies through discussions.

Data Extraction
Data extraction was performed using a standardized data extraction sheet.Two independent researchers performed data extraction, and discrepancies were resolved through discussions.The extracted data included authors; year of publication; country; characteristics of participants, such as age, sex, sample size, recruitment settings, and cognitive status; intervention characteristics, such as intervention method, intervention duration, intervention frequency, and time per each session; comparison condition; outcomes related to cognitive and psychological status; measurement tools; measurement assessment time points; results; and risk-of-bias data.

Risk-of-Bias Assessment
The risk of bias was assessed using version 2 of the Cochrane risk-of-bias tool for RCTs and CRTs [24,25].Two independent researchers performed the risk-of-bias assessment.The tool for RCTs includes the following five domains: (a) randomization process, (b) deviations from the intended interventions, (c) missing outcome data, (d) measurement of the outcome, and (e) selection of the reported result.The tool for CRTs includes "timing of identification or recruitment of participants" as the sixth domain, along with the five aforementioned domains.Within each domain, there were signaling questions and response options.By using the answers to the signaling questions, judgments for each domain (i.e., "low risk of bias," "having some concerns," or "high risk of bias") led to an overall risk of bias for each study.Based on discussions and consensus, disagreements between two independent researchers were resolved.

Data Synthesis
To assess the effectiveness of robot interventions in improving cognitive and psychological outcomes among older adults with cognitive impairment, we conducted separate meta-analyses according to the type of outcomes.For meta-analyses of continuous data, we extracted the means, standard deviations (SDs), and sample sizes at the post-test in the experimental and control groups or the mean changes, SD differences, and sample sizes in each group.These numerical data derived from the included studies were entered into Comprehensive Meta-Analysis (version 3; Biostat, Englewood, NJ, USA).We calculated Hedges' g effect sizes and 95% confidence intervals (CIs).Given the existence of heterogeneity in the characteristics of the populations and interventions across the included studies, we applied DerSimonian and Laird's random-effects approach.If the lower and upper limits of the 95% CI included zero, the effect size of the meta-analysis result is not statistically significant.We provided forest plots displaying effect estimates and 95% CIs for both individual studies and the meta-analyses.Heterogeneity was assessed using Q, I 2 , and Tau 2 statistics.According to the Cochrane Handbook for Systematic Reviews of Interventions, a low p-value provides evidence of heterogeneity in the Q statistic with a p-value [26].The I 2 statistic result was interpreted as follows: 0-40% for possible unimportance, 30-60% for moderate heterogeneity, 50-90% for substantial heterogeneity, and 75-100% for considerable heterogeneity [26].Additionally, subgroup meta-analyses were performed based on the type of robot and intervention format.

Study Selection
Figure 1 shows a flow diagram of the study selection process.The initial search of the three databases identified 175 articles (i.e., PubMed = 47, Embase = 57, and Cochrane Central Register of Controlled Trials = 71), and 93 articles remained after 82 duplicates were removed.In the process of screening the articles' titles and abstracts, 73 articles were excluded, and 20 articles were retained.Subsequently, during the assessment of the articles' full texts, nine articles were excluded.The reasons for exclusion were as follows: not the population of interest [27], not the outcome of interest [28][29][30][31][32][33][34], and not the study design of interest [35].Of the remaining 11 articles, 2 [36,37] originated from one study.Conclusively, 10 studies reported in 11 articles were included in this review, and 9 studies derived from the 10 articles reporting sufficient numerical data to calculate effect sizes were included in the meta-analyses.

Study Characteristics
Table 1 shows the detailed descriptions of the included studies.Most studies (n = 8) adopted an RCT design, whereas two adopted a CRT design.Most studies (n = 6) compared one experimental group with one control group, such as usual care, standard care, and no intervention.By contrast, four studies used three groups for comparison.Specifically, a study compared a robotic seal with a nonrobotic plush toy and usual care [38]; one study compared robot-assisted cognitive training with traditional cognitive training and no intervention [39]; and another study compared a robot seal with a real dog and a soft toy cat [40].Furthermore, a study compared a humanoid robot, a pet robot, and standard care in the first phase and compared a pet robot, a real dog, and standard care in the second phase [41].

Study Characteristics
Table 1 shows the detailed descriptions of the included studies.Most studies (n = 8) adopted an RCT design, whereas two adopted a CRT design.Most studies (n = 6) compared one experimental group with one control group, such as usual care, standard care and no intervention.By contrast, four studies used three groups for comparison.Specifically, a study compared a robotic seal with a nonrobotic plush toy and usual care [38]; one study compared robot-assisted cognitive training with traditional cognitive training and no intervention [39]; and another study compared a robot seal with a real dog and a soft toy cat [40].Furthermore, a study compared a humanoid robot, a pet robot, and standard care in the first phase and compared a pet robot, a real dog, and standard care in the second phase [41].Ten studies, with a total of 1191 participants, were analyzed.Each study's sample size ranged from 24 to 415 [15,[20][21][22][24][25][26][27][28][29][30][31][32][33][34][35][36][37][38].Among the 10 studies, 7 recruited participants from long-term care facilities [38,[42][43][44] or nursing homes [36,37,40,41].The remaining three studies recruited participants from a hospital [15], daycare centers [20], and a dementia center [39].Regarding the cognitive status of the participants, most of the studies (n = 8) targeted individuals with dementia, one focused on those with MCI [15], and one study targeted those with MCI or subjective memory complaints [39].The mean age of the participants ranged from 74.0 [15] to 87.2 years [42].
Regarding intervention formats, five studies provided interventions using robots on an individual basis [15,38,40,42,44], and four studies conducted group programs [20,36,37,39,43].In one study, group sessions were provided, and PARO was used in a personalized manner at home; therefore, the study was classified as using a group program [20].One study employed group or individual sessions according to the cognitive function of the participants' dementia: group sessions for participants with mild to moderate dementia and individual sessions for those with moderate to severe dementia [41].

Risk of Bias in Included Studies
Supplementary Figure S1 presents the risk-of-bias results.Among the ten included studies, three RCTs and one CRT had an overall risk of bias of "some concerns", and five RCTs and one CRT had a high risk of bias.Four RCTs and one CRT had a high risk of bias because of blinding issues, and the deviations from the intended intervention were not balanced between groups.Three RCTs had a high risk of bias because of missing outcome variables and differences in the rates of attrition and dropout reasons between the groups.Additionally, one CRT had a high risk of bias owing to the recruitment of participants or the timing of identification because there was no information on whether all individual participants were identified and recruited prior to cluster randomization.Among the eight studies assessing depression, seven reported sufficient numerical data and were pooled using a meta-analysis [15,20,36,39,[42][43][44], and the results indicated no significant decrease in the intervention group compared with the control group (g −0.27; 95% CI, −0.54 to 0.00) and the heterogeneity scores were Tau 2 = 0.06, Q = 11.19,p = .08,and I 2 = 46.37%.

Postintervention Effects
Among the six studies assessing cognitive function, five reported sufficient numerical data and were synthesized quantitatively [20,39,[41][42][43].One study used two intervention groups of a pet-type robot and a humanoid robot with a control group and reported the cognitive function values of three groups [41].Therefore, the summary statistics of the two robot groups were combined into those of a single experimental group according to the Cochrane Handbook for Systematic Reviews of Interventions [23].The results showed no significant difference between the groups (g, 0.04; 95% CI, −0.16 to 0.24) and had unimportant statistical heterogeneity (Tau 2 = 0.00, Q = 0.94, p = .92,and I 2 = 0%).Among the eight studies assessing depression, seven reported sufficient numerical data and were pooled using a meta-analysis [15,20,36,39,[42][43][44], and the results indicated no significant decrease in the intervention group compared with the control group (g −0.27; 95% CI, −0.54 to 0.00) and the heterogeneity scores were Tau 2 = 0.06, Q = 11.19,p = 0.08, and I 2 = 46.37%.
Among the six studies assessing cognitive function, five reported sufficient numerical data and were synthesized quantitatively [20,39,[41][42][43].One study used two intervention groups of a pet-type robot and a humanoid robot with a control group and reported the cognitive function values of three groups [41].Therefore, the summary statistics of the two robot groups were combined into those of a single experimental group according to the Cochrane Handbook for Systematic Reviews of Interventions [23].The results showed no significant difference between the groups (g, 0.04; 95% CI, −0.16 to 0.24) and had unimportant statistical heterogeneity (Tau 2 = 0.00, Q = 0.94, p = 0.92, and I 2 = 0%).
A meta-analysis pooling three studies showed no significant difference in neuropsychiatric symptoms between the groups (g, −0.05; 95% CI, −0.31 to 0.22) and unimportant statistical heterogeneity (Tau 2 = 0.00, Q = 1.32, p = 0.52, and I 2 = 0%) [20,41,42].One study reported the means and SDs of neuropsychiatric inventory outcomes for two experimental groups (a humanoid robot group and a pet robot group) and one control group [41]; the means, SDs, and sample sizes across the two experimental groups were combined into those of a single experimental group according to the guideline of the Cochrane Handbook for Systematic Reviews of Interventions [21].
A meta-analysis pooling three studies showed no significant difference in neuropsychiatric symptoms between the groups (g, −0.05; 95% CI, −0.31 to 0.22) and unimportant statistical heterogeneity (Tau 2 = 0.00, Q = 1.32, p = .52,and I 2 = 0%) [20,41,42].One study reported the means and SDs of neuropsychiatric inventory outcomes for two experimental groups (a humanoid robot group and a pet robot group) and one control group [41]; the means, SDs, and sample sizes across the two experimental groups were combined into those of a single experimental group according to the guideline of the Cochrane Handbook for Systematic Reviews of Interventions [21].
Among the three studies assessing quality of life, one used a pet-type robot [37], another used a humanoid robot [42], and the third study [41] presented only the results on the quality of life using a pet-type robot.The results of a meta-analysis pooling two studies revealed no significant difference between a pet-type robot and control groups (g, −0.05; 95% CI, −0.51 to 0.42; Tau 2 = 0.05; Q = 1.80; p = .18;I 2 , 44.34%) [37,41].
Among the three studies assessing quality of life, one used a pet-type robot [37], another used a humanoid robot [42], and the third study [41] presented only the results on the quality of life using a pet-type robot.The results of a meta-analysis pooling two studies revealed no significant difference between a pet-type robot and control groups (g, −0.05; 95% CI, −0.51 to 0.42; Tau 2 = 0.05; Q = 1.80; p = 0.18; I 2 , 44.34%) [37,41].

Discussion
This study was designed to examine the effectiveness of robot interventions in improving cognitive and psychological outcomes in older adults with cognitive impairment.We identified 10 studies from 11 papers, and most targeted older adults with dementia, except for two studies that focused on individuals with MCI.The socially assistive robots used in the included studies were largely divided into pet-type and humanoid robots.All studies employing pet-type robots used PARO, which looks similar to a baby harp seal, whereas the studies using humanoid robots used three kinds of robots.Some interventions employed an individual approach or group formats.The most frequently measured outcome in the included studies was depression (measured in eight studies), followed by cognitive function (measured in six studies) and agitation (measured in four studies).Moreover, anxiety, neuropsychiatric symptoms, and quality of life were measured in three studies each.
Regarding the effects of robot interventions on anxiety, the meta-analysis pooling three studies showed that robot interventions had a small, significant effect on decreasing anxiety (g, −0.43), and the subgroup meta-analysis pooling two studies showed that pettype robot interventions had a significant, medium effect on decreasing anxiety (g, −0.53).Among the three studies, one reported that robot cognitive interventions using a tabletoptype robot reduced anxiety but stated a limitation of not being sure whether the effect was due to the use of the robots or the cognitive intervention itself [15].Among the remaining two studies using PARO, one reported a significant decrease in anxiety in the experimental group consisting of individuals who interacted with PARO for three months compared with that of the control group [43].By contrast, one study reported no significant  [15,20,36,38,39,[42][43][44].

Discussion
This study was designed to examine the effectiveness of robot interventions in improving cognitive and psychological outcomes in older adults with cognitive impairment.We identified 10 studies from 11 papers, and most targeted older adults with dementia, except for two studies that focused on individuals with MCI.The socially assistive robots used in the included studies were largely divided into pet-type and humanoid robots.All studies employing pet-type robots used PARO, which looks similar to a baby harp seal, whereas the studies using humanoid robots used three kinds of robots.Some interventions employed an individual approach or group formats.The most frequently measured outcome in the included studies was depression (measured in eight studies), followed by cognitive function (measured in six studies) and agitation (measured in four studies).Moreover, anxiety, neuropsychiatric symptoms, and quality of life were measured in three studies each.
Regarding the effects of robot interventions on anxiety, the meta-analysis pooling three studies showed that robot interventions had a small, significant effect on decreasing anxiety (g, −0.43), and the subgroup meta-analysis pooling two studies showed that pet-type robot interventions had a significant, medium effect on decreasing anxiety (g, −0.53).Among the three studies, one reported that robot cognitive interventions using a tabletop-type robot reduced anxiety but stated a limitation of not being sure whether the effect was due to the use of the robots or the cognitive intervention itself [15].Among the remaining two studies using PARO, one reported a significant decrease in anxiety in the experimental group consisting of individuals who interacted with PARO for three months compared with that of the control group [43].By contrast, one study reported no significant difference in anxiety between the intervention group consisting of individuals who interacted with PARO for six weeks and the control group.This indicates that the six-week period was inadequate to change an individual's mood [44].Overall, this review provides support for a significant reduction in anxiety among older adults with cognitive impairment through robot interventions.However, further studies are required to examine the underlying roles of robot interventions in decreasing anxiety and to determine the optimal duration of interaction with pet-type robots for the anxiety-reducing effect to occur.
All four studies assessing agitation used PARO, and PARO interventions had a small, significant effect in reducing agitation among older adults with cognitive impairment (g, −0.31).The subgroup analysis revealed that individual interventions using PARO had an effect with a small effect size on agitation (g, −0.45), whereas the effect of group interventions on agitation using PARO was insignificant.Leng et al. [45] suggested that personalized intervention considering an individual's abilities and needs is important to improve agitation in dementia patients through a review of non-pharmacological interventions.Lu et al. [46] reported that a high degree of personalization in interventions includes a comprehensive assessment and consideration of individual capacity, preferences, interests, and environments in the design and delivery of interventions.Therefore, to reduce the agitation of older adults with cognitive impairment, individual intervention using a robotic pet that includes a high degree of personalization can be considered.
The robot interventions using the individual approach had no significant effect on depression, whereas group interventions had a small effect in reducing depression (g, −0.39).These results suggest that robot interventions, especially group-based format, can help decrease depression.Studies that used the individual intervention format discussed reasons for the insignificant effect of robot interventions on decreasing depression.For example, one study using home-based robot cognitive interventions reported that the cognitive intervention was not significantly effective because the participants were not clinically significantly depressed at baseline [15].Another study that employed PARO using the individual intervention format reported that interactions with the robotic pet for 6 weeks were not significantly effective because the intervention period was inadequate to elicit mood changes [44].
Regarding the effects of robot interventions on cognitive function, neuropsychiatric symptoms, and quality of life, the meta-analyses showed no significant differences between the experimental and control groups.Regarding the effects of robot interventions on cognitive function, one study reported that robot interventions did not improve cognitive function because most participants had severe dementia [42].In contrast, a study targeting MCI participants or participants with subjective memory complaints reported that robot interventions significantly affected cognitive function [39].Consistently, a study found that the cognitive level of participants who expressed a positive feeling toward PARO was significantly higher than that of participants who expressed negative feelings toward the robot, suggesting that robot interventions are effective for participants with MCI who respond better to robots compared with those with severe cognitive impairment [20].However, only two studies targeted older adults with MCI; hence, further studies are needed to identify the effectiveness of robot interventions in improving cognitive function among older adults with MCI.
All three studies assessing neuropsychiatric symptoms found no significant difference between the experimental and control groups.However, one study suggested that robot interventions may reduce neuropsychiatric symptoms based on findings that the level of neuropsychiatric symptoms of the experimental group decreased during the robot intervention period; nonetheless, the level of neuropsychiatric symptoms returned to the baseline level during the withdrawal phase after the removal of the robot [42].Furthermore, one study reported that robot interventions significantly affected some neuropsychiatric symptoms, such as irritability, delusions, and apathy, although the total score was insignificant [41].Given this evidence, robot interventions could reduce neuropsychiatric symptoms among older adults with cognitive impairment.However, identifying the significant effects of robot interventions on overall neuropsychiatric symptoms is difficult because the tool for assessing neuropsychiatric symptoms consists of 12 sub-symptoms.
Regarding the effects of robot interventions on quality of life, two studies revealed no significant difference between the experimental and control groups [37,42], and one study reported that the quality of life of the robot intervention group was worse than that of the control group receiving conventional therapy [41].Therefore, the effectiveness of robot interventions in improving quality of life cannot be compared because of insufficient data.Quality of life is important for people with cognitive impairment for the rest of their lives [47].However, only three studies in this review assessed quality of life as an outcome, which is similar to that presented in a previous scoping review [47].In future studies that will evaluate the effectiveness of robot interventions, it is necessary to include quality of life as an outcome and reanalyze the integrated results of intervention effects on quality of life.

Limitations
This review had some limitations.First, this review restricted the search to articles reported in English.Second, despite the extensive search, only 10 studies met the inclusion criteria.Although two or more studies are sufficient for a meta-analysis [26], future metaanalyses combining further RCTs are needed to provide more robust evidence.Third, only two studies targeting those with MCI were included in this review.Thus, data from older adults with MCI are needed to identify the effectiveness of robot interventions in this patient population.

Conclusions
This review showed that robot interventions significantly decrease anxiety and agitation in older adults with cognitive impairment.However, given that this review showed that robot interventions had no significant effect on improving cognitive function, further research is needed to determine whether robot interventions affect cognition.Additionally, a pet-type robot effectively reduced anxiety and agitation in older adults with cognitive impairment.Humanoid robots were employed in only three studies, indicating a lack of research on the effectiveness of humanoid robots; thus, further research is necessary.Although subgroup analyses according to intervention formats showed that individual interventions reduce agitation and group-based interventions reduce depression, further research is needed to compare the effects of the individual approach with those of the group approach.

Figure 1 .
Figure 1.Flowchart presenting the study selection process.

Figure 1 .
Figure 1.Flowchart presenting the study selection process.

Figure 2
Figure 2 displays the effects of robot interventions on depression, cognitive function, agitation, anxiety, neuropsychiatric symptoms, and quality of life.

Table 1 .
Detailed descriptions of reviewed articles.