Socially Assistive Robots for People Living with Dementia in Long-Term Facilities: A Systematic Review and Meta-Analysis of Randomized Controlled Trials Open Access

The global population is rapidly aging, and the incidence and prevalence of living with dementia continue to increase. The World Health Organization (WHO) estimates that more than 55 million people worldwide are currently living with dementia. It is estimated that the number will reach 139 million in 2050 [1]. This shows that living with dementia has become a common disease in an aging society. Dementia is mainly caused by the continuous degeneration of the cerebral cortex [2]. This degeneration results in single or combined declines in cognitive functions such as attention, executive function, memory, or language, and prevalent behavioral and psychological symptoms of dementia (BPSD). BPSD includes emotional, perceptual, and behavioral disorders [3]. BPSD affects up to 90% of people living with dementia and causes multiple health issues [4]. It also indirectly increases long-term care needs and related costs in manpower and resources [5, 6]. The development of dementia cannot be stopped or reversed. Therefore, appropriate and developed BPSD interventions should be explored. One such intervention is socially assistive robots (SARs), which can assist older adults living with dementia and provide support to mitigate the adverse effects.

The development of robots, guided by advancements in science and technology, has advanced to the point where they can meet the needs of different groups in public healthcare. One of the major groups is older adults living with dementia [7]. Interacting with robots having healthcare functions may yield positive effects on the psychosocial problems [8] of these patients and reduce costs [9]. Under the restriction of social distancing during the COVID-19 pandemic, the flexibility and convenience of robots have the mission to replace or support human workers [10], and may have advantages such as reducing the risk of infection, reducing BPSD, and relieving pressure on caregivers [11, 12]. Although older adults do not easily accept new technological products, this fact does not exclude the possibility of interacting with robots. It is still believed that the use of robots as dementia care tool can meet the needs of and changes in clinical practice [13].

As confirmed by research results, dementia care related to robots has been actively implemented [14]. SARs are the most common type of healthcare robots, which are generally defined as all robots that assist humans through social interaction. The types of robots include pet robots, companions, and service robots [15]. These robots can assist humans with movement, self-care, human interaction and relationships, physiological support, and non-physiological support [16]. In the treatment of older adults living with dementia, socially assistive robotics can be applied to neuropsychiatric symptoms (NPSs), including delusions, hallucinations, agitation/aggression, depression, anxiety, euphoria, and apathy [17‒19]. Multiple studies have reported improvements in BPSD. One meta-analysis only noted significant improvements in BPSD and negative mood related to SARs [20]. Studies have also pointed out that SARs can also improve social interaction [18] or quality of life (QoL) [21]. Thus, SARs intervention has been previously used for therapy [22, 23]. There have been systematic reviews of SAR interventions in older adults living with dementia, but there has been no analysis of their effects [7, 24]. In addition, two meta-analyses of SARs intervention randomized controlled trials (RCTs) have recently been published, but the results are inconsistent [23, 25]. Current research continues to look into the effects of SARs therapy on dementia patients [8]. As mentioned above, RCTs on alleviating symptoms of dementia or slowing the progression of dementia still continue as the number of people living with dementia increases. Although a few studies have explored the effectiveness of dose for robotic-assisted therapy, there is still a lack of evidence for the sustained effects of this therapy.

However, no studies have analyzed the effects of robots on positive emotions or followed up on the intervention outcomes. In view of this, the purpose of this research was to explore the effects of SARs on BPSD and NPS in older adults living with dementia through a systematic literature review and meta-analysis. This article presents a complete meta-analysis of the outcomes and effects of SARs on depression, agitation, anxiety, BPSD, QoL, positive emotional experiences, and social interaction, along with subgroup analysis based on intervention times to determine the immediate and follow-up effects of each outcome variable. The overall results will provide evidence for future related research and clinical practice.

This research was constructed following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and registered on PROSPERO (CRD42020169340). We used the PRISMA checklist from the PRISMA statement websites (http://www.prisma-statement.org/) to improve the reporting of systematic reviews and meta-analyses. The PRISMA checklist is shown in online supplementary Table 1 (for all online suppl. material, see www.karger.com/doi/10.1159/000529849).

This research was carried out from February 15, 2022. The search by two researchers was based on data from nine English electronic databases: CINAHL, Cochrane Library, EMBASE, IEEE Digital Library, MEDLINE, PsycINFO, PubMed, Web of Science, Scopus, and the Chinese Electronic Periodical Service. The included literature was searched by PICO with English keywords of MeSH synonyms (medical subject headings) and entry terms (using the PubMed query as an example). Dementia* OR robots* OR emotional robots OR companion robots OR therapeutic robots OR robotic pets OR social commitment robots OR neuropsychiatry inventories OR anxiety OR cognitive function OR sleep hours OR pleasure OR happiness OR social interaction were all search terms. This strategy was used to search all databases. The inclusion criteria were as follows: (1) population: people living with dementia or older adults with mild cognitive impairment; (2) intervention: emotional robots, companion robots, therapeutic robots, robotic pets, and social commitment robots; (3) comparison: usual care or no intervention; (4) outcomes: QoL, depression, agitation, neuropsychiatry inventory (NPI), anxiety, cognitive function, sleep hours, pleasure/happiness, and social interaction. Exclusion criteria were as follows: (1) participants without dementia; (2) failure to investigate the efficacy of SAR research; (3) non-RCT; (4) trials that did not compare the effects of intervention between groups. Only full-text articles published between 2011 and 2022 were included. Duplicate articles will be discarded. Two researchers screened titles and abstracts independently based on the inclusion and exclusion criteria. Three researchers then reviewed the full text, and the discussion resolved disagreements. One researcher extracted information for inclusion articles from the full texts and entered it into a Word spreadsheet (e.g., author, publication year, country, participants’ mean age, intervention detail, follow-up times, and outcome measurement tools).

This review was performed according to the Cochrane Handbook for Systematic Reviews of Interventions. The assessment items included the following: (1) random sequence generation (selection bias), (2) allocation concealment (selection bias), (3) participant and personnel blinding (performance bias), (4) outcome assessment blinding (detection bias), (5) incomplete outcome data (attrition bias), (6) selective reporting (reporting bias), and (7) other biases. When an assessment item is clearly and completely described, it is classified as low risk. When it is not described or incompletely described, it is classified as high risk, and when the description is unclear or incomplete, it is classified as unclear bias risk. The risk of bias was independently assessed by two authors. The corresponding (third) author was responsible for resolving any disagreements in the risk of bias assessment.

A meta-analysis was performed in RevMan 5.4.1 software (Cochrane Collaboration, Copenhagen, Denmark). The mean and standard deviation with 95% confidence interval (CI) were chosen to calculate the effect size of continuous outcomes. We will provide summaries of intervention effects for each study by standardized mean differences for outcome data. Statistical significance was considered at a p value of less than 0.05. The accuracy of data extraction was verified by two researchers. The results of this research are limited to the included studies, and the number of studies limited by each outcome variable is small. I² was used to summarize the impact of heterogeneity. In addition, the 95% CI was used to determine the accuracy to avoid underestimating the variability between interventions. Because it was assumed that all studies shared the same common effect, a fixed-effects model was used for analysis [26]. We will use subgroup analysis to explore heterogeneity in outcome measurement tools. Because there were fewer than 10 inclusion articles for each outcome, publication bias was not assessed using a funnel plot and sensitivity analysis.

This systematic review and meta-analysis screened 405 articles from 9 English databases according to the inclusion criteria. A total of 13 articles with duplicates were removed. 338 articles were excluded after careful screening of the research design, title, and abstract. After full-text and criteria assessment again, 17 articles were excluded due to participant types and defects in study design. Finally, a total of 14 RCTs were retained for systematic review. Of these, nine articles were included in the meta-analysis (shown in Fig. 1).

A total of 14 articles were included [27‒40] (Table 1). Most (84.62%) provided detailed information on random sequence generation and were considered low risk [27, 29‒32, 34‒37, 39, 40]. Half (50.00%) of the articles did not describe the blinding and allocation concealment under randomization to reduce selection bias [28, 31, 33, 36‒39]. The majority (71.43%) did not explain the blinding of subjects and researchers to avoid executive bias and were considered unclear risk of bias [27‒30, 33, 36‒40]. Just over half (57.14%) did not describe blinding of the outcome assessors to avoid detection bias and were considered unclear risk [28‒30, 33, 36‒39]. The majority (71.43%) adopted intention-to-treat analysis or had attrition rates lower than 20% to avoid attrition bias [27, 29, 30, 33‒39]. A similar number (78.57%) published research plans and completed registration to reduce reporting bias and thus was considered low risk [27, 29‒37, 40], and an equal number (78.57%) provided explanations of financial assistance and avoidance of regulatory bias and thus was considered low risk [27, 29‒36, 38, 40]. The results of the risk of bias assessment outcome are shown in Figure 2.

A total of 14 articles included the intervention methods and outcome measures of SAR interventions for older adults living with dementia [27‒40] (Table 1). The research subjects came from six countries: Australia [28, 32‒35], China [27], Denmark [38, 39], Norway [29, 30], New Zealand [31, 37], Spain [40], and the USA [36]. The sample sizes ranged from 17 to 455. A total of 1,712 older people living with dementia received SAR interventions in long-term care facilities [27, 28, 33‒35], nursing homes [29, 30, 38‒40], day care centers [31, 40], residential aged care facilities [32], urban secure dementia units [36], and rest homes [37].

In the research design section of the experimental group, the dose of the SAR intervention was most commonly set at 15 min [28, 33‒35] or 30 min per session [29‒31], followed by 10 min [38, 39], 20 min [36], around 30 and 40 min [40], 45 min [32], and 60 min per session [37]. The frequency of the SAR intervention was most commonly set at 2 times a week [29, 30, 37, 40] and 3 times a week [28, 32‒36], followed by being held 2 or 3 times a week irregularly [31], and once every other week [38, 39]. The duration of the SAR intervention was most commonly set at 12 weeks [29, 30, 36‒40], followed by 10 weeks [28, 33‒35], while other interventions were conducted for 5 weeks [32], 6 weeks [31], or 8 weeks [27]. The majority of the SARs’ intervention design was about 15 min per session, 3 times per week, for 12 weeks. During the intervention period for the control group, most control participants continued to receive usual care [27‒31, 33‒37, 40], but a few engaged in reading activities [32], played with soft toys [38, 39], or interacted with animals [40].

The purpose of interventional research on SAR intervention is to focus on the improvement of dementia and explore the significance of changes in dementia symptoms in the long term. This systematic review and meta-analysis identified follow-up on depression in six articles [27, 29, 31, 32, 36, 37], agitation in three articles [29, 31, 35], anxiety in two articles [32, 36], BPSD in three articles [27, 31, 40], QoL in five articles [27, 30, 32, 37, 40], positive emotion – pleasure/happiness in three articles [31, 32, 35], and social interaction – talk/verbal in three articles [31, 35, 37].

The Geriatric Depression Scale (GDS) and Cornell Scale for Depression in Dementia (CSDD) are commonly used to measure depression in older adults living with dementia. The GDS evaluates the depression experienced by older adults in the past week, with higher scores indicating more severe depression. The GDS has good reliability and validity and is often applied in gerontology-related research [41, 42]. The CSDD is regarded as the gold standard for assessing depression in older adults living with dementia [43]. Similarly, it evaluates the depression experienced by older adults in the preceding week. A score of more than 10 on the CSDD indicates possible depression, and a score of more than 18 indicates severe depression. Three articles reported the GDS score [27, 32, 36, 37], while the other three reported the CSDD score [29, 31, 36].

Six articles reported significant impacts on depression during SAR intervention for older adults living with dementia (MD = 4.73, 95% CI [4.54, 4.93], Z = 47.60, p < 0.001), with extreme heterogeneity (I² = 97%) [27, 29, 31, 32, 36, 37]. In addition, subgroup analysis of the GDS indicated that three articles found non-significant impacts on depression during SAR intervention for older adults living with dementia (MD = 0.08, 95% CI [-0.66, 0.81], Z = 0.21, p= 0.83) [27, 32, 36, 37], Figure 3 and subgroup analysis of the CSDD indicated that three articles found significant impacts on depression (MD = 5.08, 95% CI [4.88, 5.29], Z = 49.30, p < 0.001) [29, 31, 36]. CSDD has significant heterogeneity (I² = 91%). However, subgroup analysis of the CSDD revealed that two articles showed non-significant impacts on depression in the follow-up after SAR intervention for older adults living with dementia (MD = −0.11, 95% CI [−2.76, 2.55], Z = 0.08, p = 0.94) [29, 31] (shown in Fig. 3).

The Cohen-Mansfield Agitation Inventory-Short Form (CMAI-SF) is a common assessment tool for measuring agitation in older adults living with dementia [44]. It evaluates the agitation experienced in the two past weeks, with higher scores indicating more severe agitation. The CMAI-SF has good reliability and validity and is often applied in dementia-related research. Three articles reported the CMAI-SF score [29, 31, 35].

Three articles reported non-significant impacts on agitation during SAR intervention for older adults living with dementia (MD = −2.07, 95% CI [−4.93, 0.78], Z = 1.42, p = 0.15) [29, 31, 35]. In addition, subgroup analysis of the CSDD revealed that two articles showed non-significant impacts on agitation during SAR intervention for older adults living with dementia (MD = −1.51, 95% CI [−4.68, 1.65], Z = 0.94, p = 0.35) [31, 35] and two articles reported non-significant impacts on agitation in the follow-up after SAR intervention for older adults living with dementia (MD = −2.83, 95% CI [−7.27, 1.62], Z = 1.25, p = 0.21) [29, 31] (shown in Fig. 4).

The anxiety levels of older adults living with dementia were measured with the Rating for Anxiety in Dementia (RAID) [45]. The RAID has good reliability and validity for measuring anxiety, with higher scores indicating more severe anxiety. Two articles reported the RAID score [32, 36]. Two articles reported significant impacts on anxiety during SAR intervention for older adults living with dementia (MD = 2.80, 95% CI [2.70, 2.90], Z = 54.06, p < 0.001) [32, 36], with substantial heterogeneity (I² = 61%) (shown in Fig. 5).

The NPI is a common assessment tool for measuring the BPSD of older adults living with dementia [46]. The NPI has good reliability and validity and is often applied in dementia-related research. Three articles reported the NPI score [27, 31, 40].

Three articles reported non-significant impacts on BPSD during SAR intervention for older adults living with dementia (MD = −0.63, 95% CI [−2.24, 0.97], Z = 0.77, p = 0.44) [27, 31, 40]. In addition, subgroup analysis of the NPI revealed that two articles found non-significant impact on BPSD during SAR intervention for older adults living with dementia (MD = -0.61, 95% CI [-2.28. 1.06], Z = 0.71, p = 0.48 ) [27, 31] (shown in Fig. 6).

QoL can be measured as an indicator of the magnitude of the impact of a SAR intervention [47]. The QoL-Alzheimer’s Disease (QoL-AD) Scale is a common assessment tool for measuring QoL in older adults living with dementia [48], with higher scores indicating better QoL. QoL-AD has good reliability and validity in older adults [48]. In addition, some articles applied the Quality of Life in Late-Stage Dementia (QUALID) Scale to older adults living with dementia [49]. The QUALID evaluates the QoL experienced in recent weeks, with higher scores indicating good QoL. Three articles reported the QoL-AD score [27, 32, 37], while the other two reported the QUALID score [30, 40].

Five articles reported non-significant impacts on QoL during SAR intervention for older adults living with dementia (MD = −0.38, 95% CI [−1.92, 1.16], Z = 0.48, p = 0.63) [27, 30, 32, 37, 40]. In addition, subgroup analysis of the QoL-AD revealed that three articles found non-significant impacts on QoL during SAR intervention for older adults living with dementia (MD = −0.22, 95% CI [−2.07, 1.62], Z = 0.24, p = 0.81) [27, 32, 37], and subgroup analysis of the QUALID indicated that two articles found non-significant impacts on QoL MD = −0.74, 95% CI [−3.52, 2.04], Z = 0.52, p = 0.60) [30, 40] (shown in Fig. 7).

Positive emotion is derived from behavioral records, the Observed Emotion Rating Scale, and the Video Coding Protocol-Incorporating Observed Emotion Scheme [31, 32, 35]. The behavioral record was based on behavioral tracking methods for facial expressions, e.g., smiling, sadness, and fear, which is a rating tool used in research for people with dementia [31]. The Observed Emotion Rating Scale measures mood states during a session using observation [32]. The Video Coding Protocol-Incorporating Observed Emotion Scheme is a quantitative measure during recorded observations [35]. Three articles reported significant impacts on positive emotion (pleasure/happiness) during SAR intervention for older adults living with dementia (MD = −2.44, 95% CI [−3.91, −0.97], Z = 3.25, p < 0.01) [31, 32, 35], with substantial heterogeneity (I² = 70%) (shown in Fig. 8).

Measuring social interaction for people with dementia is done using behavioral records, the Video Coding Protocol-Incorporating Observed Emotion Scheme, and observations of social behavior [31, 35, 37]. The behavioral record was based on behavioral tracking methods for social interactions, e.g., talking to others, cooperation, and reciprocity [31]. Incorporating an observed emotion scheme is a measure of social engagement during recorded observations [35]. Social behavior is calculated by observing how often SAR is discussed during the sessions [37]. Three articles reported significant impacts on social interaction (talk and verbal) during SAR intervention for older adults living with dementia (MD = −5.07, 95% CI [−7.80, −2.34], Z = 3.64, p < 0.001) [31, 35, 37], with substantial heterogeneity (I² = 74%) (shown in Fig. 9).

Symptoms of living with dementia can affect cognitive behavior and the ability to organize thoughts, leading to complex BPSD. Therefore, conditions such as cognitive function, psychological state, environmental equipment, and personal needs must be considered in the design of appropriate dementia interventions. The primary goal is to reduce NPSs [50]. SAR intervention has become an emerging dementia care strategy to complement drug therapy [51].

This systematic review and meta-analysis present the latest information on the effectiveness of SAR intervention in mitigating the psychological symptoms of living with dementia. A total of 14 articles examining the application of SAR as an intervention and in routine care were included in the systematic review. Nine of these articles were included in the meta-analysis. Results of this analysis, based on SAR as an intervention and compared to routine care, show that it can improve NPS of depression and anxiety, positive emotions (pleasure/happiness), and social interaction (talk/verbal) of older adults living with dementia.

This finding is similar to that of Sung et al. [52], who found that older adults in long-term facilities improved their communication skills and social interaction after SAR intervention. Lane et al. [53] observed that SAR intervention in older veteran populations improved positive emotions and behaviors, and also that it decreased negative affective behaviors such as social isolation, anxiety, sadness, and agitation, which are similar to the findings of our study. Rouaix et al. [54] applied robots in psychomotor therapy programs for older patients in wards, and they found that the negative emotional responses of anxiety or sadness (tears, depression) were not exhibited during the program implementation. They also noted obvious happiness-related responses (well-being, obvious happiness), which are also similar to the results of our study.

However, a meta-analysis by Pu et al. [23] found no significant difference in anxiety and depression among older adults who received SAR intervention. In our meta-analysis, we discovered that the older adults had different characteristics, the number of articles included in the outcome variable analysis was relatively small, and the sample sizes were also relatively small. Therefore, it cannot be confirmed whether all studies found the same common effect. The analysis with a random-effects model may have caused differences in the results of our study. Nevertheless, through the results of our study and previous articles, the advantages and effectiveness of SAR have been demonstrated.

In the previous review, the format, dose, frequency, and duration varied greatly [23, 24]. In our review, the vast majority of SARs are low dose (15 min at a time), high frequency (3 times per week), and long-lasting (12 weeks). It is also unclear over what time frame SAR works and what dosages of the intervention work best in people with dementia [23], although one study on pet robots for people with dementia showed that both individual and group-format SAR significantly ameliorated BPSD [25]. However, individuals with SAR interventions were more acceptable and applicable [23]. Take Jones’ study as an example. The robotic pet was designed for people suffering from dementia. Participants received individual, 15-min sessions with SARs three per week for 10 weeks, and there were fewer observed instances of agitation at week 10 [28]. The Petersen study found that those exposed were exposed 3 days a week, with each session lasting 20 min and sessions continuing for 3 months of intervention with the robotic seal, PARO, in long-term care facilities. The result showed that older people with dementia had decreased stress and anxiety [36]. Due to insufficient data, we are still unable to provide guidance for SAR intervention.

The results of our analysis showed that SAR did not reduce agitation, overall BPSD, or QoL in older adults living with dementia. In addition, the follow-up after the interventions also found that SARs had no lasting effect on improving depression and agitation in older adults living with dementia. This is in alignment with the meta-analysis by Pu et al. [23], who also showed that SARs did not have significant impacts on agitation, overall BPSD, or QoL in such adults. Although no relevant studies have explored or followed up on the research effects of SARs in older adults living with dementia, emotional behavioral disorders such as depression, agitation, and anxiety in BPSD [55] may indirectly affect the research outcome. Nonetheless, the fact that agitation was not improved by SAR intervention may indirectly explain the reasons for the failure to improve BPSD.

From the results of our study, it can also be found that the areas that can be improved are related to the emotional states of psychology. QoL is not a single dimension. Rather, it is a multi-dimensional concept that can include physical health, mental state, degree of independence, social relationships, personal beliefs, and the environment [56]. This may explain why our study did not show a significant improvement in QoL. Furthermore, the disadvantages of applications of SAR in healthcare interventions include difficulties in operation and control, lack of flexibility in the system, high technical failure, and being unable to adjust the behavior of the robot for individual needs, etc., from the mixed methods study [57]. The predicament of robustness and reliability in these systems could also explain the reasons for influencing the results of the study.

Several limitations of our systematic review and meta-analysis must be noted. First, five articles did not present outcome variable data and could not be analyzed. The lack of such data indirectly limits the strength of the evidence. Second, most of the articles in the meta-analysis had small sample sizes, which may lead to a reduction in effect size. Third, all the included articles on SAR design were divided into groups and individuals in two ways. The dose, frequency, duration, and control group setting of the SAR interventions were varied, as was also the absence of a long time to follow-up. Fourth, the SAR types were divided into toy animals and humanoid robots, but the effectiveness of different types has not been explored. Fifth, countries currently conducting SAR research have not included certain races or ethnicities, such as oceanic or African countries. This lack of inclusivity may marginalize minority countries or populations.

In this review, the limited number of trials limits the strength of the evidence and results in a poor interpretation of some studies that do not present outcome variable data. However, we believe that as SAR advances, researchers will continue to investigate its effectiveness for people living with dementia in the future. Given the challenges of the research process, regardless of its status, it should be adequately powered by enrolling a large enough number of participants to provide overall confidence in the evidence. To evaluate high-quality outcomes, future trialists should follow an organized framework and theories based on groups and individuals, SAR interventions, and long-term follow-up. Furthermore, more stringent RCT standards are required to assess the impact of the seemingly disparate SAR types on the reported outcomes. There were little data on other underrepresented races or ethnicities, demonstrating the need for further research in this area.

In any event, our systematic review and meta-analysis still have established advantages. They were based on evidence-based research, for only RCTs were included. Objective research results and a clear causal relationship could be obtained. All the articles were assessed for risk of bias according to the Cochrane Handbook for Systematic Reviews of Interventions. Simultaneously, the team members shared a common focus on research methods and the risk of bias impacting performance to reach conclusions after a thorough and rigorous evaluation. This all contributes to the development of knowledge on SAR intervention for older adults living with dementia.

Science and technology are rapidly evolving, and there is a pressing need to cut costs. The new dementia care model is impacting and revolutionizing treatment methods through technology-assisted measures. The results of our systematic literature review and meta-analysis show that SAR intervention can mitigate depression and anxiety in older adults living with dementia, as well as promote positive emotions and social interactions. SAR intervention is still in the developmental research stage. At present, several countries have carried out research on SARs for living with dementia population. Especially under the limit of social distancing during the COVID-19 pandemic, the flexibility and convenience of SARs to replace workers may have advantages such as reducing the risk of infection and BPSD. Therefore, rigorous designs and standardized protocols must be used to accumulate results and to develop precise conclusions. It is important for our statement that relevant studies on SAR in people living with dementia be processed in a rigorous manner with regular updates for future research. We still need to collect future findings in SAR for guidance, practices, and policies.

We thank the National Taipei University of Nursing and Health Sciences for the support.

An ethics statement is not applicable because this study is based exclusively on published literature.

The authors have no conflicts of interest to declare.

The authors have no funding to declare.

C.J.H. initiated the idea for the meta-analysis and supervised the data collection, statistical analysis, and writing of the paper and critically revised the text for its content. P.S.L., C.H.W., S.L.L., T.C.H., and C.M.T. collected the data. C.J.H. conducted the statistical analyses and wrote the paper.  C.J.H., P.-S.L.,helped write the text. All authors have read and approved the final manuscript.

All data generated and analyzed during this study are included in this article and its online supplementary material files. Details can be consulted from the corresponding author.