This systematic review protocol was registered with PROSPERO (ID: CRD42022339845) and follows the methodology of the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [33]. The PRISMA checklist can be found in Multimedia Appendix 1.

The search was completed on May 19, 2022, with inclusion criteria displayed in Textbox 1. The search strategy can be found in Multimedia Appendix 2. The search strategy included search terms for the older adult population undergoing in-home positioning or monitoring systems as their intervention. Keywords related to older adults included “Aged,” “Senior,” “Over 65,” or “Aging,” while terms on the setting included their home or house. For the technologies, terms included the purpose of monitoring (“Positioning” or “Localization”) and specific types of technologies, including wireless trackers (Bluetooth, Wi-Fi, UWB, or Zigbee), wearables (accelerometers or gyroscopes), camera, and audio systems. The search did not include comparator groups or outcomes to improve the sensitivity of the search. We included studies that had at least 4 patients to improve the sensitivity of the search. Articles focused on measuring life spaces outside the home (travel to appointments, shopping centers, or recreation centers) were excluded. Studies using wearables were only included if assessed within a home setting.

MEDLINE, Embase, CINAHL, and Scopus were searched, and articles were deduplicated. Abstract screening, full-text screening, data extraction, and quality appraisal were completed by 2 reviewers: the first author (AC) and 1 of 4 secondary reviewers (SP, BC, LQ, and JC). Reviewers were trained with 10 test abstracts and full-text articles, and then concordance was reviewed. At each stage, interrater agreement was calculated using the κ coefficient calculated by the following formula:

where Pr(a) represents the actual observed agreement and Pr(e) represents the chance agreement [34]. Disagreements were resolved by having both reviewers reassess articles for 2 additional rounds, and then the article was discussed to reach a consensus.

Data extraction included article demographics (country and year published), study design (clinical, usability, or technical study), population characteristics (age, gender distribution, clinical diagnoses, and comparators), types of localization interventions (wearable or ambient, data transmission, technology readiness level, and data analytics methods), and outcomes (types of activities monitored, clinical assessments and outcomes, acceptability, and reliability). Data were compiled into summary tables, presenting the population, technological intervention, and clinical outcomes of each study.

To assess risk of bias, the JBI checklist for case series critical appraisal tool was used, as we did not expect any high-quality randomized controlled trials related to in-home monitoring [35]. Criteria for appraisal were predetermined: studies with 7 or more “Yes” ratings were considered high quality, studies with 4-6 “Yes” ratings were considered moderate, and studies with fewer than 5 “Yes” ratings were considered low quality. No meta-analysis was planned, as we did not expect to find high-quality quantitative studies that would allow for heterogeneity to be assessed. Instead, the outcomes from each study were presented individually.

Clinical outcomes were summarized in summary statements, with only moderate- or high-quality studies considered. Evidence was summarized as positive if the majority of studies showed positive results, negative if the majority of studies showed negative results, and mixed if neither had a majority.

During the initial search, 1935 articles were found, with 1008 unique articles after deduplication. After abstract screening, 127 articles remained. After full-text screening, 46 articles were included in the final extractions: 36 technology-focused articles and 10 articles that included relevant patient populations. Agreement between reviewers at the abstract screening stage was 94.9% with a κ of 0.77, and agreement for the full-text screening was 95.8% with a κ of 0.71. Quality assessment agreement was 76% with a κ of 0.51. The PRISMA flowchart in Figure 1 maps out the excluded articles.

Table 1 displays the baseline characteristic for the 10 papers that included relevant patient populations. In total, 7 studies were conducted since 2019. In total, 5 studies were from North America, 3 from Europe, and 2 from Asia. In total, 2 studies were descriptive studies of the technology, 4 studies had a mixed design and qualitative study design, 2 had qualitative study designs, 1 was a mixed study with qualitative and quantitative outcomes, and 1 focused on quantitative outcomes. Only 1 study had more than 25 participants. In total, 8 studies had more female than male participants.

All studies had patient populations that included older adults, although only 7 specifically reported population characteristics. In total, 4 studies included older adults living at home with nonspecific functional challenges, 2 focused on adults with mild cognitive impairment or dementia, and 1 focused on older adults with frailty. Half of the studies (5/10, 50%) were considered low quality, 3 (30%) were considered moderate quality, and 2 (20%) were considered high quality.

Table 2 shows the technologies and localization methods used in the included studies, their setting, and the duration of monitoring. From a technology perspective, 2 used solely an ambient sensor design, 5 combined ambient sensors with wearables, and 3 used wearable-only designs. Ambient sensors included temperature sensors, magnetic door sensors, infrared motion sensors, light switch sensors, pressure detectors, and lidar sensors. Wearables included inertial measurement units (IMUs), electrocardiograms, heart rate meters, and wearable wireless tags (Wi-Fi or Bluetooth low energy). Of the 10 studies, 7 (70%) included technologies of unknown brand or model (3 only used non-branded devices), while 3 (30%) listed the brands of devices used.

Most studies (7/10, 70%) were done in the home setting, with 2 in a home-like laboratory setting and 1 in a laboratory setting. Studies in the laboratory-home involved monitoring sessions lasting between 1 hour and 7 days [28,29,41], while home-based monitoring ranged from 3 weeks to 18 months.

Table 3 displays the outcomes from studies that included patient populations. In total, 7 included technical outcomes, 6 included usability and acceptability outcomes based on patient or clinician surveys or interviews, and 3 included clinical outcomes. Room detection accuracy ranged from 50% to 88% across 3 studies [27,28,30], while 1 study reported failure rates of >15% for motion detectors and servers installed in the home [36]. One study reported linear accuracies of 1.5-2 m using wireless sensor networks within the home [29].

For studies that included usability and acceptability outcomes, surveys from 3 studies [29,36,38] showed positive results. One study focused on ease of setup of a smart home in a box design and found high ease of use, few concerns with devices, and highly efficient instructions [36]. Another found the highest satisfaction among older adults, followed by caregivers, and the lowest satisfaction with nursing staff [38]. One study found an average system usability scale score of 62.8, indicating below average usability [29]. Interview results from 4 studies [27,37,39,40] found improved safety and security with devices, but there was some perceived physical intrusiveness to ambient devices [39], and some patients changed their behavior due to monitoring [27]. One study found a tag-based system was highly acceptable [40].

Lastly, regarding clinical outcomes, 1 study provided qualitative observations on why patients behaved in certain ways in their home, finding certain activities are performed slower and some areas in the home are avoided, including staircases to avoid falls, depending on their functional level [37]. One study found no difference in behaviors between residents with mild cognitive impairment and those who were cognitively healthy, based on continuous monitoring of sleep and identifying frequency of forgetting to do activities [39], while another study was able to classify patients as frail, prefrail, or nonfrail with 80% to 87% accuracy using machine learning algorithms from Bluetooth-based wearable localization, being monitored for 1-7 days continuously in their own home while doing their own typical activities [41].

The primary objective of this systematic review was to review the clinical evidence for in-home localization technologies to support in-home monitoring of older adults. We found 36 articles that reported that their technology would be used for localization of clinical populations. Table 4 is a summary of the characteristics of studies focused on developing and evaluating in-home localization technologies for older adults.

Studies on ambient sensors were from North America (3/6 studies, 50%), wireless tags were most studied in Europe (6/6, 100% for Bluetooth or Wi-Fi and 5/7, 71% for other tags), and wireless tags alongside IMUs were solely studied in Asia (8/8, 100%). The majority (25/36, 69%) of studies were from after 2016. The stated purpose of monitoring was for older adults in a general sense in 27 (75%) out of 36 studies, while older adults with chronic diseases or disabilities were specified in 9 (25%) studies. The purpose of monitoring was mostly for health and safety monitoring (21/36, 58%).

The most common localization mode was to measure signal strength (23/36, 64%), followed by time-based localization (8/36, 22%), which calculates the time that it takes for a signal to travel from a tag to a reference point, and the least common was proximity sensing (5/36, 14%). Received signal strength involves estimating the distance between wearables and reference points based on the strength of the wireless signal. Localization accuracy was most reported as a linear distance (23/36, 64%), followed by classification of activities (13/36, 36%), room or area detection accuracy (6/36, 17%), and lastly accuracy in detecting multiple people in a space (5/36, 14%).

Table 5 summarizes the accuracies of different technologies, organized according to the method of localization and the type of accuracy reporting. Ambient sensors included infrared sensors, radiofrequency transceivers, and video feedback. Devices were primarily used for detecting people passing through spaces, with accuracies of 79% to 98% in differentiating people, and 92% accuracy in detecting presence in a room.

Bluetooth and Wi-Fi technologies can be used with either smartphones or individual tags, reducing the need for extra equipment for localization when compared to stand-alone tags. Accuracies ranged from 70-250 cm, with room detection accuracies of 70% to 87%.

Localizer tags include radio frequency identification (RFID), UWB, Zigbee, and GPS tags. Linear accuracies were superior to Bluetooth or Wi-Fi, ranging from 5 to 100 cm, with area accuracies of 90%. Tags were also used for fall detection and object detection.

Combining localizers with IMUs allowed for a combination of activity classification and localization. Accuracies ranged from 7.6 to 189 cm across 4 modalities (UWB at 7.6 cm, RFID at 10-40 cm, Zigbee at 83-189 cm, and Bluetooth low energy at 47 cm), while activity classification ranged from 89% to 100%, although reporting was not always clear on what activities were being classified. Area classification accuracies were between 86% and 90%.

Lastly, with unique technologies, including sound-based technology, GPS, vibration sensors, pressure pads, and triboelectric sensors, accuracies ranged from 20 to 30 cm with activity recognition at 92% to 99%.

From the 5 moderate- to high-quality clinical studies, 4 studies reported on acceptability of in-home localization systems. Results were mixed, with 2 high-quality studies indicating positive acceptability [38,39], 1 finding below average usability [29], and 1 finding a range of concerns over device obtrusiveness [27].

Two studies reported on clinical outcomes from in-home localization systems. One high-quality study showed no difference in behaviors in older adults with healthy cognition compared with those with mild cognitive impairment [39], and 1 moderate-quality study detected adaptive behaviors at home because of limitations to patient function [37].

This systematic review focused on the usage of localization methods to monitor older adults in their homes for any clinical application. While the primary objective was to evaluate the clinical evidence for localization technologies, a survey of technologies for in-home localization was also undertaken to understand upcoming technologies for localization.

Clinical utility of localization was mixed in this study. In the study by Rawtaer et al [39], cognitively healthy older adults (21 participants) and older adults with mild cognitive impairment (28 participants) were monitored and compared over 2 months using a custom set of motion sensors, proximity tags, a bed sensor, and wearables to capture sleep; activity levels; and forgetfulness regarding medications, keys, or wallets. Among typical activities, there was no difference in behaviors [39]. A second study, examining frailty, used in-home localization to detect frailty by measuring number of transitions, speed of transitions, and statistical features through machine learning algorithms, finding a classification accuracy of 82% to 85% when using random forest plots [41]. The model can be used in the future to detect frailty in the general population. The clinical evidence for using localization technology to support care of older adults is currently limited.

From an acceptability perspective, results were moderately positive [27,29,36,38,39]. Pais et al [38] discovered that ambient sensors garner greater acceptance compared to wearables. Moreover, they noted that older adults and caregivers exhibit higher acceptance levels toward both technologies in contrast to nurses. This trend could be attributed to the necessity for monitoring daily performance issues among older adults and their families. The acceptability of home monitoring has been thoroughly studied previously, finding that the trade-offs are critical to consider when developing these technologies [42-45]. These findings align with the present systematic review, with obtrusiveness being a major detractor for these localization technologies balanced by improved safety and security.

Common technologies for localization include ambient sensors; Bluetooth- or Wi-Fi–received signal strength; localizer tags using RFID, UWB, Zigbee, GPS; or IMUs with localizers. This review also found unique localization devices, including triboelectric trackers, ultrasound, floor vibration sensors, and pressure pads. Highest linear accuracies were found with UWB technologies at 5-20 cm compared with greater than 50 cm for most other technologies. UWB uses time-based localization, which involves measuring the time it takes for a signal to travel from a tag to a reference point and then trilateralizing the signal. Room detection accuracies were comparable across technologies, ranging from 75% to 92% using Bluetooth, Wi-Fi, RFID, radio frequency transceivers, or Zigbee with IMUs.

The current literature is limited as it focuses primarily on technical measures of accuracy. The shift needs to be made toward localization for activity identification that can then be used as evidence to provide an intervention. UWB positioning has the potential to make the shift from where a patient is in the home at a room level to a furniture level that can then allow identification of activities. Further exploration and development of algorithms to automatically detect activities are required before broader clinical usage.

This systematic review fills an important gap by including clinical results, user acceptability, and technological aspects of evaluating localization devices to support older adults to age in place. There remains little evidence for their usage for older adults, a finding that is supported by other systematic reviews. Lenouvel et al [31] reviewed sensors to measure and support activities of daily living for older adults in 2019. While they did not focus on localization, they found that passive and video sensor networks were used to assess activities of daily living across 13 studies out of their search of 10,782 studies, finding that sensors could detect changes in activity patterns but reported no clinical outcomes and that only 1 study assessed the acceptability of devices.

Another systematic review published in 2018 focused solely on technological aspects of human activity recognition supported with indoor localization. Cerón and López [32] described common localization technologies and data fusion methods, reporting accuracy of activity detection and localization accuracies without consideration for age of participants. Human activity recognition accuracy ranged from 72% to 99% across 27 studies, although the exact types of activities were not reported. Localization accuracies ranged from 0.8 to 7 m, depending on the type of technology, although the type of technology was not reported in the review. These values are comparable to this systematic review.

This systematic review had a strong search strategy, covering the major databases and having 2 reviewers screen, extract, and assess the quality of studies. Agreement between reviewers was high across screeners. The JBI quality assessment tool was used with a lower agreement with a κ of 0.51. Agreement was low due to inadequate training for the 10 clinical articles, a lack of specific definitions of how much clinical information was adequate for the study, and how follow-up was defined in the article. Each study was discussed between reviewers according to a standardized definition for the final results of this study.

While the methodology of this review was strong, the findings were not. There is limited clinical evidence for using localization to support monitoring older adults. It was surprising that there were also few studies that evaluated the acceptability of monitoring technologies. The quality of evidence also needs to be improved, with most studies having fewer than 25 participants with a case study design and the quality of studies being mixed.

Still, existing studies on acceptability of localization technologies form a strong basis for further development. Future studies should be located within the homes of participants, with sample sizes greater than 25 to demonstrate scalability and particular use cases in a broad range of home settings. From a study-design perspective, home monitoring as an intervention is a complex intervention that is challenging to capture in a randomized controlled trial. The recent guidelines by the Medical Research Council and National Institute for Health Research in the United Kingdom provide a new framework for assessment that includes considering the context, stakeholders, economics, and uncertainties in an intervention, grounding it in appropriate theory, and iterating to refine the intervention [46]. Nonrandomized designs, hybrid effectiveness-implementation designs, or n-of-1 trials may be more feasible.

Details around patient populations were scarce. Greater detail in medical histories, functional ability, and practical aspects (social supports, living spaces) need to be provided to generate profiles for how monitoring interventions have helped specific residents. A standard battery of activities of daily living needs to be established to allow accuracy in identifying and assessing activities of daily living to be comparable across studies in the range of indoor spaces being localized and the diversity of impairments common to older adults. There needs to be clearer reporting of the spaces being monitored, accuracy of devices, and types of activities of daily living being monitored to allow comparability. Lastly, study outcomes need to be shaped to demonstrate how monitoring technologies lead to clinically and personally relevant interventions that support aging in place. The 2 studies that looked at clinical outcomes in this systematic review focused on detecting dementia or classifying frailty. Perhaps the more important question is how minimally invasive interventions can be used to either prevent decline or intervene to support residents who are having greater challenges doing self-care activities.

The utility of localization techniques for health care is still untapped. While some initial work on detecting cognitive decline and frailty in the home setting has been documented in this review, further development and clinical evaluation of these technologies to determine potential use cases still needs to be undertaken. Development of these technologies requires a multipronged approach that combines understanding the limits of the technology, including the cost, the clinical applicability of localization for health management, and the acceptability of monitoring to enhance wellness. Technologies such as Bluetooth, Wi-Fi, and IMUs are already well established in the market for various quality of life use cases but not for health care.

Localization could be a powerful supporting tool for managing challenges with cognition, with interventions that take into account a user’s living patterns and reminders that are tailored to the home environment. Of upcoming technologies, UWB may be the most exciting, offering much higher accuracies than ambient sensors and wireless technologies such as Wi-Fi and Bluetooth. Cognition, mental health, and frailty could be more accurately measured longitudinally, rather than relying on snapshot clinical assessment tools when combined with collecting information on self-care and in-home activity levels. There is great potential for localization technologies to support wellness in the home.

There is no evidence for the usage of in-home localization technologies for any clinical outcomes and mixed evidence for the acceptability of localization technologies among older adults. However, there is a wide range of technologies available that have promising technical accuracy. The technology is ripe for monitoring devices to be tested clinically, providing data that can detect changes in cognition or frailty and drive interventions. Further study on the acceptability of these devices is also warranted to determine the least obtrusive and easier to use modalities that can bring the most benefit for older adults.