As the global aging situation becomes more serious, dementia has become a highly prevalent disease in the older population [1]. The characteristic of dementia is the degradation of memory, cognition, behavior, and daily activity abilities [2]. Symptoms such as agitation and depression increase the risk of secondary problems like fractures and falls, which can seriously impair patients’ quality of life. In addition, it puts more strain on those who provide care, raising the expense of nursing and medical care. According to the August 2020 update of the “Guidelines for Dementia Prevention, Intervention, and Care” by the Lancet Committee, psychotropic medications typically have little effect on neurological or mental symptoms and can have substantial side effects [3]. As a first line of treatment for behavioral and psychological symptoms of dementia, nonpharmacological therapies are advised due to the drawbacks and safety concerns associated with medication therapy [4].

Multiple studies have shown that music therapy and multisensory stimulation intervention have significant effects in improving behavioral and psychological symptoms of dementia [5,6]. The senses enable patients to interact effectively with the environment and communicate better with others. The effects of pet therapy applied in the geriatric population are mainly in the form of improved social interaction, improved emotional state, enhanced cognitive level, and improved problems. Studies on patients with dementia have demonstrated the beneficial effects of pet therapy [7]. Based on the above evidence, using intelligent robots for intervention in patients with dementia may have the following advantages:

In recent years, intelligence robot interventions have caught the interest of scientists studying dementia. A variety of intelligent robots, including pet robots, companion robots, social assistance robots, and humanoid robots, are being used for neuropsychiatric symptoms, cognitive function, and pain in patients with dementia [8,9]. The results of intelligence robot intervention vary, nevertheless. For instance, a mixed methods study revealed that following a 6-week intervention, the robotic companion dog or cat group’s feelings of melancholy and loneliness considerably dropped as compared to the control group [10]. The study’s qualitative findings revealed that the robotic companion pet experience was viewed positively by the participants, their families, and professional caregivers, who felt that the robot enhanced communication and offered companionship [10]. Conversely, another study indicated that a social robot intervention did not significantly improve positive affective states [11]. There is currently conflict over the efficacy of interventions on agitation and quality of life, as well as a lack of data for objective indicators like physical activity and sleep duration, despite the publication of a few meta-analyses of intelligence robot intervention [8,9,11-13]. Subgroup analyses of intervention settings, such as the kind and purpose of intelligent robots and the length, the location, and the format of the intervention, have not been done in previous studies. New trials have been published in the last two years; thus, an updated review of this material is still necessary.

Given this, this study aimed to examine how intelligence robots affect both subjective and objective markers in patients with dementia by a thorough assessment of the literature and meta-analysis. Furthermore, the type of intelligent robots, as well as subgroup analysis based on intervention times, are discussed. Overall, the findings will offer support for further relevant studies and therapeutic applications.

We conducted this systematic review and meta-analysis by following the PRISMA (Preferred Reporting Items in the Systematic Review and Meta-Analyses) guidelines (Multimedia Appendix 1).

WF and RZ conducted a comprehensive search in February 2024 across the following databases: Web of Science, CINAHL, PubMed, Embase, Cochrane Library, and CINAHL. We found additional relevant publications by searching the bibliographies of the included papers and previous relevant systematic reviews. Details about the customized search approach for each database can be found in Tables S1-S6 in Multimedia Appendix 2. The following search terms were used: “dementia,” “Alzheimer’s disease,” “AD,” “robotics,” “robotic,” “robot,” and “robot-assisted.”

The included studies satisfied the following criteria: (1) adult patients were diagnosed with dementia; (2) the study was a randomized controlled trial (RCT); (3) the intervention group received intelligence robot intervention, with no restrictions on the kinds of intelligent robots; (4) the control group was given standard treatment; and (5) the study was published in English. The exclusion criteria are as follows: (1) absence of data for the network meta-analysis; (2) duplicate data; and (3) qualitative research, books, review studies, conference abstracts, or study protocols.

EndNote 20.4 (Clarivate Analytics) was used to eliminate duplicates before deciding if the searched records were eligible. Initial screening of literature by two independent authors (WF and RZ). WF and RZ discussed and settled any differences between them until they came to an agreement. A third author (XL) was consulted if a consensus could not be achieved. The first author, the country, the year of publication, the type of study, the analyzed sample size, the average age of the participants, the specifics of the intervention (methods, frequency, and duration), and assessment time points were all extracted using a sheet by two independent authors, WF and XL.

Two authors (WF and RZ) independently applied the Cochrane Collaboration risk of bias tool [14] to assess the quality of the included studies.

The meta-analysis was conducted using Stata software (version 16.0; StataCorp). To address potential bias caused by differences in indicator levels between the intervention and control groups at baseline, we calculated the mean difference and SD between postintervention and baseline values to reflect within-group changes.

The effect size, which quantifies the difference in the magnitude of change between groups, was expressed as either the standardized mean difference (SMD) or the weighted mean difference (WMD). The WMD was used when studies measured outcomes on the same scale; otherwise, the SMD was applied.

Heterogeneity was considered significant when the P value of the Cochran Q test was less than 0.10 and the Higgins I² statistic was greater than 50%, and a random effects model was used to account for heterogeneity between studies; otherwise, a fixed effects model was used [15]. A forest plot and a funnel plot were generated for the analysis. The calculation of Egger regression tests was done to assess publication bias.

Forest plots were generated, and sensitivity analyses were performed to further assess the stability of the study results. The calculation of Egger regression tests, visualization of the funnel plot, and trim-and-fill method were used to evaluate publication bias [16].

Subgroup analyses were conducted to explore the influence of study design factors (intervention duration) and intervention conditions (types of intelligent robots) on the effect of the intervention and to assess possible sources of heterogeneity.

The PRISMA flowchart of study identification and screening is outlined in Figure 1. A total of 1319 records were extracted from the electronic database search, and 3 records were identified through hand searching of reference. Following the elimination of duplicates, we went over the abstracts and titles of 797 (60.29%) publications to weed out any that were not relevant. In the end, after verifying compliance with the inclusion criteria and exclusion criteria, a total of 15 (1.13%) studies were included in the systematic review, and 12 (0.91%) of these studies were included in the meta-analysis.

The characteristics of the included studies in the systematic review are shown in Table 1. A total of 15 studies were conducted in 8 countries, including Australia (5/15, 33%) [17-21], Norway (3/15, 20%) [22-24], China (2/15, 13%) [25,26], Spain (1/15, 7%) [27], New Zealand (1/15, 7%) [28], Japan (1/15, 7%) [29], the United Kingdom (1/15, 7%) [30], and the United States (1/15, 7%) [31]. All studies were published after 2015, and the number of participants in each study ranged from 22 to 175, with intervention periods ranging from 6 weeks to 3 months. Participants were mainly female, with the proportion ranging from 64% to 88%, and the average age of participants was reported to be from 83.4 to 89.0 years in 13 (87%) [17-26,29-31] studies. The venues where the interventions were implemented included long-term care facilities, nursing homes, and other venues. Only 2 (13%) [27,29] studies were single-center studies, while the remaining 13 (87%) [17-26,28,30,31] studies were multicenter studies.

Each included RCTs’ bias risk is displayed in Table S7 in Multimedia Appendix 2. A computer-generated random list was used to generate the randomization sequence in 6 of the RCTs [17,19,23,25,27,28,30], while 1 RCT [31] included coin tossing. The risk of selection bias was deemed questionable since the randomization procedure was not reported in detail in 2 other RCTs [28,29]. However, because it was challenging to blind staff members and participants in the intelligent robot interventions, only 1 RCT [19] was deemed to have a low risk of performance bias.

The research findings of the five outcome measures, namely cognitive function, neuropsychiatric symptoms, agitation, depression, and quality of life, indicate that the point estimates of the combined effect sizes were obtained after removing 1 study at a time and analyzing the remaining studies. The results showed that the exclusion of a particular study does not significantly alter the overall results, indicating that an evaluation of the 5 outcomes had considerable stability (Figures S9-S13 in Multimedia Appendix 2).

Funnel plots and Egger tests were performed to evaluate publication bias of 5 outcome measures, namely cognitive function, neuropsychiatric symptoms, agitation, depression, and quality of life. The results showed no publication bias in cognitive function, neuropsychiatric symptoms, depression, and quality of life (Figures S14-S17 in Multimedia Appendix 2). For agitation, the results of the Egger test showed possible publication bias. There was no significant difference between the effect size estimates obtained using the trim-and-fill method and those obtained by meta-analysis, indicating the results were reliable (Figures S18 and S19 in Multimedia Appendix 2).

A total of 3 studies [21-22,25] were left out of the meta-analysis due to the inability to combine the inclusion of several datasets. Jøranson et al [22] looked into how the intelligent robot Paro affected the sleep habits of patients with dementia living in a nursing home. This study used wrist actigraphy to assess objectively sleep-wake patterns and showed that an intelligent robotic intervention significantly increased the percentage of sleep efficiency and reduced the frequency of nocturnal awakenings in patients with Alzheimer disease. Ke et al [25] assessed the effects of the humanoid social robot (Kabochan) on technology acceptance among older individuals with dementia; they found that being exposed to intelligent robots may enhance their perceptions of the technology’s utility and attitudes toward it. Moyle et al [21] showed that participants in the PARO group were more verbally and visually engaged than participants in plush toy.

This study indicates that intelligence robot interventions effectively influenced agitation, anxiety, and length of time lying down during the daytime. The current meta-analysis did discover, however, that intelligence robot interventions had no discernible effects on depression, quality of life, cognitive function, neuropsychiatric symptoms, sleep duration, wakefulness during the day or night, step count, or frequency of physical activity.

In this study, intelligent robot intervention significantly reduces levels of anxiety in patients with dementia, which is consistent with meta-analytic reviews of Saragih et al [8]. Anxiety is highly prevalent across dementia stages, with an overall pooled prevalence of 39% [32]. In recent years, various clinical applications of intelligent robots have been used to provide high-quality emotional support and companionship [33]. Intelligent robots can provide emotional support and companionship to patients with dementia, easing anxiety by interacting with them and sharing stories, music, or simple games. In future studies, the effects of different design elements in intelligent robot interventions on the anxiety levels of patients with dementia can be further explored to understand the preferences of patients with dementia for the appearance characteristics and interaction contents of intelligent robots, so as to improve the references for the design of intelligent robots and clinical application programs, and to give full play to their effectiveness in alleviating the anxiety levels of patients with dementia.

Agitation improved in the intelligent robot intervention group. One study has shown that providing patients with dementia with sensory stimulation, mainly through sight, sound, touch, taste, and smell, can help to improve agitation [34]. Intelligent robots integrate the application of visual, tactile, and auditory multiple sensing technologies, such as the social robot (Paro) covered with artificial fur, capable of moving and emitting sounds to achieve multisensory stimulation, which can put patients with dementia in a multisensory enriched environment and alleviate their agitated state. It has been shown that agitation prolongs the hospital stay of people with dementia and imposes physical, psychological, and financial burdens on caregivers [35]. More research is needed in the future to evaluate whether these interventions also benefit shorter hospital stays, better caregiver psychological health, and lower costs.

The intelligent robot intervention was not significant for overall depression levels in patients with dementia, which is consistent with the findings of Abbott et al [36]. Subgroup analyses based on the length of the intervention were carried out, and the findings indicated that when the intervention lasted for 12 weeks, the intelligent robot intervention helped patients with dementia experience lower levels of depression; however, no such beneficial effect was observed when the intervention lasted for shorter than 12 weeks. Depression in people with dementia often involves deeper emotional issues, and the provision of emotional support and cognitive stimulation by intelligent robots may take longer to produce significant improvements [12]. Subsequent investigations ought to delve deeper into the impact of intervention duration on the efficacy of intelligent robot interventions. The humanoid and animal robots did not differ from one another in subgroup analyses depending on the type of intelligent robot, and neither had a significant impact on the depression levels of patients with dementia. The usefulness of humanoid and animal robot treatments has not been the subject of as many studies, and this study’s sample size was limited. In order to provide some guidance for the appearance design and function setting of intelligent robots, the sample size should be increased in subsequent research, and various types of intelligent robots should be thoroughly examined.

Cognitive function and neuropsychiatric symptoms in patients with dementia did not significantly improve with intelligent robot interventions. The cognitive deterioration involved in dementia is usually associated with structural changes in the brain [37], and although intelligent robots can provide social interaction and behavioral stimulation, these may not be sufficient to reverse or significantly improve the cognitive impairment caused by neurodegenerative diseases. Intelligent robots lack the human caregiver’s capacity for emotional depth, empathy, and complex decision processing, and their role in neuropsychiatric symptoms in patients with dementia may be limited.

The relevant data for this study came from wearable devices, and the included studies reported difficulties in wearing wearable devices in patients with dementia. The results did not suggest a significant positive effect on sleep and physical activity. Given that individuals with dementia were reticent to wear armbands, particularly at night [17], this conclusion should be interpreted cautiously. To increase patient happiness and participation, more appropriate techniques for tracking sleep and physical activity in patients with dementia should be investigated in the future.

In the group receiving intervention, the quality of life did not improve. Studies have shown that one of the main factors influencing the quality of life of patients with dementia is their capacity to carry out activities of daily living [38]. Patients with dementia also tend to be less capable of taking care of themselves, and some even experience difficulties walking. Although intelligent robots can provide a certain degree of life guidance and support, they cannot completely replace human care and assistance. The cognitive abilities, neuropsychiatric symptoms, the stress of being the primary caregiver, and other factors all affect the quality of life of patients with dementia [39].

Subgroup analysis according to the length of intervention showed that intelligent robot intervention for 12 weeks or more had a more positive effect on the depression level and agitation of patients with dementia, and there was no significant difference in neuropsychiatric symptoms and quality of life. In patients with dementia receiving 12 weeks or more of intervention, symptoms such as depression and agitation may change more significantly due to improvements in emotional regulation and behavioral interventions. However, neuropsychiatric symptoms and quality of life are affected by a combination of more complex physical, psychological, and social factors, and the effect of the intervention may not be significantly different in these dimensions. This suggests that although long-term interventions have a positive impact on mood and behavior, longer or more comprehensive interventions may be needed to improve neurodegenerative lesions and quality of life. A subgroup analysis was performed according to whether the intelligent robot was a humanoid robot or a pet robot. The results showed that there was no significant difference in the improvement of cognitive function, neuropsychiatric symptoms, depression, and quality of life in patients with dementia between the two types of robots. Although humanoid robots and pet robots differ in appearance and behavior, their core role in providing intervention is often similar, that is, to improve the mood and behavior of patients through interaction, emotional companionship, cognitive training, and other means. This may be the reason for the similar intervention effects of the two.

This study has several practical implications for practice. First, interventions based on intelligent robots can be considered for inclusion in the long-term care of people with dementia. Second, the shape characteristics of intelligent robots may have no effect on the intervention effect on people with dementia, and other design characteristics need to be further explored. Third, when using wearable devices to assess sleep, physical activity, etc, in people with dementia, the wearing comfort and compliance of people with dementia should be considered.

The following limitations of this study need to be taken into account when interpreting the results. First, the sample size included in this study was small, and further analysis of the impact of factors such as the age, gender, and geographical location of participants on the effectiveness of the intervention was lacking. Second, only original studies published in English were included, and there is a possibility that relevant studies published in other languages were missed. In addition, not all studies were assessed by professional blinders, which may introduce bias into the study.

Despite these limitations, this review aims to comprehensively evaluate the impact of intelligent robot interventions on people with dementia. Seven outcome indicators, including objective indicators of sleep and activity, were meta-analyzed. The impact of the intervention duration and robot type on smart robot interventions was further explored to provide a reference for the future optimization of smart robot development and intervention program design.

This meta-analysis comprehensively assessed the impact of intelligent robot interventions on people with dementia and showed that intelligent robot interventions can reduce agitation and anxiety, with a limited role in improving depression, anxiety, cognitive functioning, neuropsychiatric symptoms, and quality of life, and that sleep- and physical activity-related outcomes need to be analyzed with caution. Subgroup analyses showed better results for the longer the duration of the intelligent robot intervention. The types of intelligent robots included humanoid robots and animal robots, and subgroup analyses showed no difference in their effectiveness. The use of intelligent robots in nursing is still in its infancy, although a number of nations have implemented intelligent robotic interventions for patients with dementia. These interventions have primarily involved women and have small sample sizes and have been concentrated in high-income nations, which may be related to the epidemiological features of dementia. Subsequent research endeavors ought to enhance the conceptual framework of sentient robots and delve deeper into the ways in which the nature of the sentient robot and the length of the intervention affect its efficacy. When using wearable technology to gather physiological data, researchers should take patient compliance into account.