A PRISMA (Preferred Reporting Item for Systematic Reviews and Meta-Analyses) chart summarizes the process of study selection (Figure 1). The search (date: July 13, 2022) returned 180 articles, of which 122 were related to the age groups +65 and +80 years. We excluded reviews (n=15), surveys (n=19), viewpoints (n=10), and other irrelevant studies (ie, those that used technologies that did not have any ICT or those whose main objective was to test a functional domain in older adults using a nonassistive ICT used for data collection or stimulation presentation; n=31). We then read the remaining 43 articles and further excluded those that did not include qualitative or quantitative research methodologies. This left us with 31 articles that were analyzed qualitatively (for content, theme, and discourse).

Table 1 summarizes the reviewed literature with respect to the sampling, study design, and research objectives. See Multimedia Appendix 1 [67-97] for the descriptions of these studies.

Table 2 provides an overview of the scope of the methods, objectives, and outcomes in the reviewed literature.

The studies reviewed included qualitative focus groups (9/31, 29%), mixed methods (10/31, 32%), quantitative interventions (10/31, 32%), and 2 observational and 1 participatory studies. Furthermore, 4 studies were longitudinal and 2 were multinational.

Recruitment procedures were not described in all articles; however, among those that did (n=13), only 1 study involved a random sampling strategy from a general population pool. The remaining (12/31, 39%) used selective and targeted methods of recruitment by reaching out to a specific group of older adults whose needs were a priori known to clinical partners or organizations that care for older adults.

Participant characteristics were often biased by the inclusion criteria, which either included those older adults without cognitive or physical disabilities or those with such conditions. Only 14 studies specified the health conditions of the participants in the study sample. Eight studies recruited only healthy older adults; 3 studies recruited participants with cognitive impairment, 2 of which also had healthy older adults as controls; and 1 included older adults with disabilities.

Gender was also a biasing factor, with women being overrepresented in the study samples among the 10 studies that reported the gender ratio. In addition to limitations in the representativeness of the samples, further biases were noted in retention and attrition, which will be addressed in the Discussion section.

Most of the articles reviewed framed the research in the context of the burdens of aging (cognitive deficit, physical disability, dependence, frailty, and isolation). Among the articles included in this review, 39% (12/31) of articles began by describing concerns about the rising costs of the growing older population; 45% (14/31) began by discussing age-related disabilities such as dementia and frailty; 39% (12/31) focused on desire for independent living or aging in place; and 29% (10/31) framed the study in the context of preventive measures to promote successful aging.

In terms of the objectives of research, 4 general categories emerged: studies that aimed to assess A-ICT–based interventions to promote healthy aging by physical activity (n=10); studies that evaluated the acceptance and feasibility of introducing A-ICTs such as robots and remote health monitoring systems to promote independent living (n=9); studies that validated the ability of A-ICTs to detect age-related risks caused by physical or mental deficits (n=5), and several multistakeholder focus groups or ethnographic studies about the future of A-ICTs in health care for older adults (n=8). See Multimedia Appendix 1 [67-97] for the descriptions of these studies.

This was not the case in most studies reviewed here. Only 2 studies referred to age as an explanatory variable to validate automated measurement systems with the assumption that older adults have functional deficits compared with young adults [78,87].

We found that some of the quantitative studies framed within the discourse of successful aging included psychometric instruments specific to deficits of older populations, such as assessing cognitive deficits (eg, the Montreal Cognitive Assessment test and the Mini Mental State Examination) [76,84-89,92,95], geriatric depression [72,76,85,86,88], chronic illness and disability [68,75,76,85,87], risk of falls [67,68,72,75], and other instruments that measured presumed age-related variations such as loneliness [73], self-efficacy, and self-reliance [67,73,77] quality of life and life satisfaction [67,72,73,77].

Given that sampling in the reviewed reports was not entirely random, it is plausible to assume that examiners possessed some perspective on the life stages of older adults. Among the quantitative studies in this review, awareness of life stage was operationalized in terms of the quantitative assessment of aspects of individual life being affected by age (as described above). Among qualitative and design-related studies in this review, attention was focused on capturing the existing and emerging needs and attitudes of older adults. This manifested in terms of multistakeholder study designs [90,94], evaluating factors such as digital literacy [76,90,91,93], history of ICT use [90,91,96], technology acceptance [69-71,78,81-83,89-91,93,97], and tailoring the interventions to individual needs [75,77,78,93,94,96], which implied that older adults were less likely to be aware of new technological developments.

Some statements in the conclusion may be construed as value-laden. For example, referring to the aging population under study as “elderly” is no longer culturally accepted. The word “elderly” was in the abstract, keyword, and conclusion of some studies [68-70,73,85,86,89,94]. Referring to the population as “older people” objectifies and segregates them. The following examples, all from studies involving technology development, demonstrate this objective distancing, which makes the aging body part of the machinery that was invented and tested:

Another possibility for value-laden conclusions is the emphasis on deficits and generalizing it to a population. These deficits may be related to individual conditions, such as dementia or learning disabilities, which are the most feared stigmatizing notions. Examples of such statements include:

The majority of studies in this review expressed a general concern for the impending costs of aging and recruited based on age criteria. However, age was not used as a predictive or explanatory variable. Only 1 study conducted random sampling. The rest performed convenience or selective sampling within existing pools of individuals who participated in other geriatric care programs. Even within these pools, the rates of participation and attrition varied. Studies that specifically focused on medical conditions such as frailty and dementia were purposively recruited too.

This observation reinforces our concern that participation in age-related A-ICTs may be biased by the research questions and methodology. We identified the following factors to be more important than age.

The objectives of promoting healthy or successful aging, cost reduction, independent living, and risk reduction were the primary motivations for the studies reviewed here. However, the findings of most studies are not strongly conclusive, especially those of randomized controlled trials (RCTs). The variability in the results was mainly related to acceptance and adherence, but not age.

For example, in a large-scale clinical trial of ICT-coached cardiovascular risk management, the largest effects were observed in those who had the ability to access and engage with technology [85]. A similar RCT investigating the impact of tablet-based physical training reported that attrition was related to individuals’ appraisal of what they expected from the app as well as to the motivational factors related to social interactions [75]. In other words, the context in which an ICT-based intervention took place was more effective in promoting adherence than ICT functionality.

The circumstances and contexts of individual experiences seemed to shift the statistical results. For example, an RCT of an ICT-based exercise program tried for 8 weeks by home-dwelling older adults reported that group differences were not due to improvements in the intervention group but due to a decline in the control group [67]. A subgroup analysis revealed that the benefits of the intervention in the high-adherence group were more pronounced than in others, thus calling for future studies to explain individual differences in adherence to the intervention. Another RCT involving a web-based multidomain lifestyle intervention did not find any significant functional improvements in participants and observed that, despite accepting the conditions of the trial, adherence (especially with physical exercises) requirements were not met. They speculated that this was linked to motivational factors not measured in the trial [76].

Insights from the multistakeholder focus group study of Welfare Technologies that offer assistive robots to the Swedish older population [83] point to another important source of bias: individual perceptions of agency and control. This study found that not age but the perceived absence of ethical and governance frameworks, as well as lack of collaboration and health spending, were impediments to acceptance, access, and successful adoption of the proposed technologies.

In fact, tailoring interventions to individual needs was found to be important in a longitudinal RCT that introduced a wearable ActiGraph to monitor the physical activity of prefrail older adults [68]. In another RCT of an ICT-based frailty prevention study, physical fitness in the intervention group did not increase, but quality of life in the control group declined. The study did not have any qualitative data to offer any explanation of the change in the control group but called for future examination of the neurobiological mechanisms of this effect [72].

A multinational acceptance study (local communities in Greece, Finland, United Kingdom, and Denmark) of an “intelligent” telecare system for independent living and self-care of older adults with mild cognitive impairment explained the challenges of recruitment (achieving <25% of the targeted sample size) [89]. Interestingly, this study observed significant regional differences in service use, as well as regional and personal variations in service appreciation. It was noted that learning to operate a new automated system for those with cognitive impairment is impractical, and if such interventions are offered, they need to be offered in a personalized manner to individuals who can benefit from them.

This study is laudable for providing a very detailed picture of the challenge of recruiting representative samples within the predefined clinical criteria and explaining deviations from initial study design and recruitment caused by the reality of heterogeneity not only among the needs of individuals with dementia, but also heterogeneity of the technology literacy of caregivers (family or nurses) and the health care systems within which they receive care [89]. In conclusion of their report, the authors have noted the following:

A user-centered phenomenological study of older adults’ experience during assistive technology device (ATD) delivery [79] revealed that the framing of the study in the context of age-related deficits was in and of itself a biasing factor in recruitment:

In this study, the perception of self-deficits and fear of ageist attitudes contributed to reluctance to participate:

This study concluded that satisfaction with assistive technology is not easily measurable and that “there are several complicating reasons for older individuals not to acknowledge unsatisfactory experiences in the service delivery process...related to expectations, disappointments, fear, and abandonment but also hope, mastery, and resourceful and dynamic self-management of care.”

Another qualitative study across 2 sites (in the United Kingdom) with 30 older adults and 39 nonfamily caregivers undertook purposive sampling to capture the diversity in minor or major home adaptation and its funding source: age range (65-74, 75-85, and >85 years), ethnicity, gender, household composition, house type, and tenure. This study noted that stigmatizing notion of aging (equating it with vulnerability and disability) was an impediment to learning about and seeking technologies to adapt homes to the needs of older adults. Participants in their study showed awareness of the “ageist” attitudes, and some expressed a fear that to use assistive technologies (as neutral as a staircase railing) would make them “appear old” or signal frailty and disability, which would lead to stigmatization [96].

Refusal to participate in the study of a home-monitoring installation study by 54 of 127 eligible candidates, and the completion rate of 12 of 21 in those who enrolled in a year-long trial point to other sources of bias [95]. Refusal to participate may also be implicit, for example, by not receiving a response from more than 39,000 of the 45,466 invited individuals (with 696 explicit refusals) [85]. Therefore, it is not surprising that positive results are reported within a highly motivated sample who showed interest in adopting home monitoring strategies.

Notwithstanding the limitations in sampling discussed above, several studies within this review have strived to conduct participatory and inclusive research, especially regarding the future of A-ICTs. Strategies used to accomplish this included conducting focus groups involving different stakeholders [92-98].

In recognition of the fact that age does not capture the heterogeneity of needs within older populations, several studies in this review took a need-based approach to studying the affordances of A-ICTs [77,78,92,94].

Conducting research across different national or regional jurisdictions was an important strategy to demonstrate not only the diversity of older adults’ needs but also the differences in institutional and socioeconomic contexts in which they lived [89,91,93,96].

Longitudinal follow-up in several of the studies reviewed here ranged from 6 to 12 weeks to 12 to 18 months. The longer the duration of the study, the greater the possibility of examining the evolving relationship between the users, caregivers (family or professional, such as nurses or social workers), and the technologies presented [75-77,79,82,89,95,98].

Presenting different real [84] or hypothetical [94] options to users or evaluating A-ICTs within simulated environments such as living laboratories [81,82,88] added flexibility to the research frameworks. The more options and the greater the opportunities to “play,” the greater the chances of retaining participants and recording positive experiences.

Tailoring the intervention individual needs [75,77,78,93,94,96], “technology matching” [78], or calibrating the intervention to the cognitive or physical ability of the participants [68,71] were important in addressing the heterogeneity within the sample.

Adding interviews within a safe space encouraging participant’s candid and critical views also led to the discovery of factors that could bias the research, such as motivations [74], tensions such as fear of stigmatization [79,83], or fear of losing human touch [80], thus extending the findings beyond age-related explanatory variables, such as physical and cognitive ability, and technology use.

Clearly, the question of how to create an ICT to assist the aging population poses one of the most complex questions at the intersection of many fields: medicine, interaction design and communications, health psychology, sociology, and engineering. This inherent interdisciplinarity in and of itself presents challenges in developing an account of complex systems dynamics [101].

This review indicates that the simpler the targeted behavior (eg, increasing physical activity), the more likely it was that the study reported satisfactory outcome [67-70,74,75]. However, the introduction of other technologies such as assistive robots or the introduction of technologies for independent living resulted in ambiguity in the interpretation of findings. The inconsistencies were caused by individual differences in attitudes, abilities, and expectations tied to prior knowledge, experience, and care systems [73,80-86,90,91,94,95,96]. Physical activity is beneficial for health and is widely accepted irrespective of age. That robots and machines may replace humans is received with ambivalence and therefore more variations in acceptance are expected.

The inclusion of both quantitative RCTs and qualitative observational or focus group studies in this review uncovered an interesting disciplinary gap to bridge. For example, quantitative studies on the benefits of interventions for healthy aging often control for variations in emotional, physical, and psychological factors. However, these studies rarely accounted for participant’s views, cultural backgrounds, lifestyles, and socioeconomic resources available to them. Conversely, design-oriented and participatory studies that focused on contextual and socioeconomic factors that shape attitudes rarely controlled for cognitive, physical, and affective variations that shape one’s particular needs.

Interestingly, a discourse analysis of the language used to describe research objectives and conclusions revealed that overly reductionist studies are more likely to deviate from the recommendations of the Non-Ageist Guidelines for Research. Studies designed with qualitative and participatory components were generally more careful in discussing the needs of older adults, primarily focusing on the desire to live independently in later years of life. In contrast, more technical articles used a more “objective,” and potentially objectifying tone in describing the research rationale and results. For example, in a study that utilized machine-learning approaches to identify risky drivers, the authors included age as a clinical variable and concluded that age-related decline was predictive of unsafe driving [88]. In another study, to validate a virtual reality–based diagnostic tool for the early detection of dementia, researchers included 2 groups of young and healthy older adults on the presumption that the system would be sensitive to detecting age-related deficits in the latter group [87].

The concept of assistive ICTs for health care for older adults is new and still in the design phase, and tensions in personalizing or generalizing care are pervasive [94]. Therefore, interdisciplinary and cocreation approaches may achieve a higher research impact and accelerate the development of more effective solutions [81].

RCTs are preferred by scientists seeking evidence for valid, reliable, reproducible, and effective interventions. Given this wide acceptance, it is perhaps surprising that even in purely medical interventions, contemporary researchers have questioned the validity of making inferences from RCTs [102,103]. As Fink and Keyes [104] have shown, ignoring complexity, especially in public health science, can lead to erroneous inferences and shift the ground on which scientific inquiries claim truth: reproducibility. This review included 6 RCTs, 4 of which focused on physical activity [67,68,72,75] and 2 on general behavioral coaching to promote a healthy lifestyle [76,85]. Overall, recruitment, retention, and adherence of participants presented challenges. For decades, medical anthropologists and other social scientists have pointed out the limitations of RCTs in studying interventions that depend on culture and context, especially interventions that target the psychosocial well-being of populations [105]. The fact that a larger proportion of studies in this intersectional study were non-RCTs points to the complexity of the problem and the fact that reductionist methodologies are not sufficient.

In RCTs, it is difficult to account for all the personal factors that shape one’s behavior, let alone their relationship with new technology. Among older adults, Birkland [106] has synthesized data from extensive interviews to categorize users under labels Enthusiasts, Socializers, Practicalists, Traditionalists and Guardians—with the latter groups being the most likely to find new technologies stressful [107]. These factors influence attrition. For example, in a mixed methods study of an ICT-based behavioral intervention to increase physical activity [74], those who completed a 6-week intervention were highly interested in the intervention. The study reported that the mean daily step count increased from 9443 (SD 3952) steps before the intervention to 10,773 (SD 2659) steps after the intervention, with a mean increase of 14%. It is important to note that these effects were not statistically significant. In this study, one participant decreased their step count from 15,611 to 14,772 (already higher than the commonly recommended 10,000 daily steps). These numbers, as well as comments recorded from the participants, indicate that participants whose data acquisition was successful (8/16, due to technical failure) were highly motivated to engage with this activity, a reality that is not considered in statistical power calculations to recommend the intervention for RCT.

Similar to clinical researchers, who seek objective evidence to support the universality of the effectiveness of their intervention, designers strive to satisfy the universal design principles of equity (useful strategies for a wide range of users independent of age), flexibility (accommodating a range of preferences and methods), simplicity (independence from literacy, skills, or language), perceptibility (clearly communicating their purpose and use case), failure-safety (minimizing risks of error), and accessibility [108]. For this reason, holistic and participatory research practices help refine applications through recursive evaluations and improvements [92,109,110].

In the Introduction, we pointed out that double-binds may be caused by miscommunication, which can be corrected and negotiated over time. Considering the dynamics of change in behavior requires flexibility in methodologies that can capture patterns of change. For instance, longitudinal studies within this review illustrate that technology acceptance is a dynamic process that begins with recruitment (ie, who chooses and refuses to participate) and evolves through experimentation and trial, as study participants establish relationships with the researchers. Indeed, in a study where the acceptance of robots for medication delivery was investigated [80], despite growing acceptable over time, ultimately, the users did not find such technology to be a suitable replacement for human care. Monitoring attrition over time, as well as monitoring the relationship dynamics during research, provides insight into the context of how needs, perspectives, and levels of engagement change [72,75-77,79,81,82,89,91,95,96].

A phenomenological study of older adults’ experiences during assistive technology service delivery further underlines the importance of taking a recursive approach to research [79]. Before receiving the assistive technologies that the participants had applied for, they were optimistic and hopeful that the technology “would make life easier and would enable them to perform their desired activities. The participants were confident they would be able to manage using the ATD.” However, after receiving the devices “their encounter with their ATD and the person who delivered it either confirmed the participants’ positive expectations (if they could manage to use it correctly) or surprised them when their expectations that the ATD would ease their everyday life were not met.” Further examination of explanatory variables (such as the self-described personality of users who called themselves assertive) showed between-individual differences in the process of adoption (taking charge and seeking help or putting up and being dissatisfied).

Conducting research over time allows researchers to examine how individuals cope with the technostress of new experiences. Consider learning as an example. Several studies in this review found that, in addition to acceptance, the desire or ability to overcome the challenge of learning can themselves become sources of dissatisfaction and potential stress [78,82,89,90,93,97]. Independent of age, a large proportion of eHealth solutions suffer from high rates of dropout, discontinuation of use, or nonadoption [111]. Therefore, it is important to develop tools to match gerontechnologies to individual needs is important [78]. Incorporating Lazarus’ transactional model for stress and coping offers an empirical framework for evaluating the influence of personal factors in the appraisal, adoption, or rejection of new interventions [112,113].

Capturing individual preferences in real-life situations is important. A mixed methods study involving the introduction of 6 robotic services in a realistic living environment [84] provides an example of how playful and recursive approaches may be informative. In this study, participants were given the freedom to become familiar with the robots before starting the experimental session to feel more confident in testing them. Interestingly, this study found that older adults enjoyed the anthropomorphic design of the robot (with a moving blinking head), which is conducive to interaction. According to the authors, “Only two users did not get pleasure in testing the Robot-Era system because they claimed to see the robotics system as an appliance that is used for its usefulness and not for pleasure.” It is plausible to suggest that the playful nature of the task of evaluating various types of robots, without pressure to adapt them to their real lives, was conducive to better interaction with the systems [84].

Besides providing an opportunity to learn from simulation, adding playfulness to the research itself mitigates the discomfort that arises from perceived social evaluative threats in researcher-participant relationships [67,70,83]. When technology is introduced through a serious health care context, the risk of self-censoring owing to the fear of losing potential privileges increases [79]. Conversely, playfulness can help free expressions of actual needs and attitudes. For example, introducing assistive robots with the specific aim of assisting individuals with mild cognitive impairment resulted in rejection of the notion that such services could replace humans for caregiving, whereas those with mild cognitive impairment seemed to enjoy the playful features of the technology in proving distraction while also being concerned about the image portrayed by using a machine for companionship [81].

Adopting a playful framework mitigates performance stress, increases enjoyment, and provides an opportunity to make observations about choice, socialization, mastery, and self-efficacy to help envision more inclusive and user-centric interventions.