The global population is aging rapidly, with an estimated 1.6 billion adults aged 65 and over by 2050 [1]. This demographic shift is accompanied by a shrinking caregiver-to-older-adult ratio: in the United States, the ratio for adults aged 80 and older is projected to decline from 7:1 in 2010 to 3:1 by 2050, with fewer caregivers available for the growing aging population [2,3]. This imbalance presents the urgent challenge of how to support older adults in aging safely and independently at home.

Health monitoring smart homes (ie, sensor-based health monitoring at home) offer a promising solution [4,5]. These systems typically consist of ambient sensors (eg, motion, door, humidity, and more) connected to a central hub that transmits data to a secure platform. Algorithms analyze movement patterns to detect anomalies such as prolonged inactivity, disrupted sleep, or changes in kitchen or bathroom use, all potential indicators of emerging health concerns. Alerts can be communicated via email, SMS text message, or phone calls, and escalated to caregivers or clinical teams for follow-up.

However, most smart home health monitoring has been developed and tested in higher-income settings, leaving critical gaps in understanding their feasibility and acceptability among older adults experiencing poverty [6,7]. Older adults in subsidized housing or with limited income face distinctive barriers to technology adoption, including limited internet access, low digital literacy, mistrust of systems, and cultural expectations around caregiving [8-12]. Without deliberate attention to these barriers, smart home innovations may risk widening health disparities rather than closing them [13,14].

This study aims to identify and understand barriers to adopting smart home health monitoring occurring within a co-designed community support system among older adults experiencing poverty. To address these aims, a low-cost, community-in-the-loop “smart health system” prototype was developed, featuring continuous ambient health monitoring capabilities. Using mixed methods, we evaluated how older adults interacted with the system, the challenges they encountered, and the features that supported trust, engagement, and equitable access. Here, the term smart home refers broadly to ambient sensing in the home for the purpose of monitoring residents’ health. The prototype deployed for this study is referred to throughout as a smart health system because the smart home was embedded in a community-in-the-loop support system where older adults received a smart home along with systematic support from self-identified family and friends (ie, a support person) and from appointed community health workers (CHWs) and registered nurses (RNs).

This paper shares the findings from trialing the community-in-the-loop approach to health monitoring in resource-constrained settings that use smart homes and discusses the potential for smart home adoption by older adults living in poverty. The purpose of this paper is to provide insights into how, and if, continuous ambient health monitoring is possible in older adult populations experiencing poverty. The low-cost system is discussed in detail in Multimedia Appendix 1.

Smart home systems leverage ambient sensors, Internet of Things (IoT) devices, and data analytics to monitor daily activities and detect health-related changes [14-16]. Prior studies demonstrate their potential to support aging in place by identifying deviations in mobility, sleep, and self-care routines. Rantz et al [17,18] showed that sensor-based systems can detect early signs of health decline. Pioneering work by Cook and the team at the Center for Advanced Studies in Adaptive Systems (CASAS) developed lightweight, easy-to-install systems for activity recognition and health monitoring, including nurse-in-the-loop models and pain-related behavior detection [5,19-22]. More recent work has advanced IoT-based frameworks for secure remote monitoring [23] and patient-centric agents using blockchain on 5G networks [24]. Furthermore, work has been done that specifically identifies gaps in chronic disease applications [4].

Smart home systems show promise in detecting early signs of clinically relevant conditions such as cardiovascular disease, congestive heart failure, cancer, urinary tract infections, sleep disturbances, and cognitive decline [4,21,25]. These conditions disproportionately affect older adults experiencing poverty who often face delayed diagnoses and limited access to care [26-29]. Sensor-based alerts, such as changes in mobility, sleep patterns, or hygiene routines, can serve as proxies for emerging health issues and prompt timely interventions. These studies confirm the feasibility of unobtrusive monitoring but often remain focused on higher-income populations.

While smart home technologies are well established, their adoption among diverse older adults remains understudied. Prior work has primarily focused on for-profit senior living settings, including Tiger Place [17,18], Touchmark [21,30,31], and early collaborative aging research using technology (CART)–related studies [32] where adoption has not faced many barriers. In contrast, adoption barriers among low-income older adults persist, including privacy concerns, perceived intrusiveness, system complexity, digital literacy challenges, limited internet access, and inadequate technical support [4,7,33-35]. Similar inequities appear across digital health interventions, such as mobile health and telehealth [36], for chronic disease management [37-39].

Key facilitators for older adults and the health care team include ease of use, perceived usefulness, and a system’s ability to enhance safety and independence [4,9,40-42]. Trust in the technology and the organizations deploying it strongly shapes adoption, particularly amid concerns about data security and reliability [4,9,43]. This study extends prior work by examining barriers and facilitators to adoption among older adults with multiple chronic conditions living in poverty, with particular attention to affordability, usability, cultural relevance, and community co-design as key determinants of trust and engagement.

Co-design and community-based participatory research methodologies have proven effective in tailoring digital health interventions to the needs of diverse populations [44,45]. Involving end users in design and testing enhances trust, usability, and perceived ownership of technology [46,47]. However, few studies have applied these principles to smart home health monitoring systems serving low-income older adults with multiple chronic conditions.

This study embedded community engagement across all phases of system development and evaluation [48]. We brought together low-income older adult residents in subsidized housing and living in the community, subsidized housing managers, CHWs from diverse cultural and ethnic communities, and nursing and engineering researchers to co-design features including notification preferences, sensor placement, and privacy safeguards. Adapted from the “nurse-in-the-loop” model, this community-in-the-loop approach ensured alignment with participants’ cultural, social, and socioeconomic contexts. Iterative feedback revealed preferences for low-burden devices, clear feedback loops, and direct human contact during alerts, showing how co-design can make smart home health systems more inclusive, acceptable, and scalable in resource-constrained settings.

This study was approved by the Washington State University Institutional Review Board (IRB) for human participants’ research under IRB#19135. All participants provided informed consent in their preferred language. Seven languages were represented in the participant pool. Each participant received a US $50 gift card and a results summary after study completion.

A community-engaged, prospective, mixed methods design was used to prototype and evaluate a low-cost smart health monitoring system for older adults managing multiple chronic conditions while experiencing poverty. The study was conducted in collaboration with local housing authorities, culturally based health organizations, and a free community health clinic in the US Pacific Northwest.

The prototype system was designed as a closed-loop model to help families, communities, and health care teams support older adults aging in the place of their choosing (Figure 1). The system operated through three integrated components: (1) ambient sensing, (2) a rules-based algorithm developed for this study that continuously processed sensor data and automatically generated alerts for clinically relevant changes in daily behavior, and (3) community members who worked alongside the technology to respond promptly to alerts, including facilitating engagement with the health care team.

The prototype system included 6 alert types, with alert messaging designed not to imply judgment or provoke stress or fear. The system alerted for social isolation (Homebody), decreased activities (Daytime Resting), kitchen use (Eating Routines), decreased hygiene (Bathing), sleep disturbance (Restless Night), and sleep patterns (Changing Sleep Routine). The participant-facing alert terms associated with clinically relevant behavior changes are italicized. Multimedia Appendix 1 contains detailed descriptions of alerts, including defined thresholds.

A convenience sample was recruited with the assistance of subsidized housing building managers. More than 80 residents across 7 subsidized housing properties attended information presentations about the smart health system, after which interested individuals met privately with a trained research assistant or the lead primary investigator (PI) to review consent forms in their preferred language. Residents were informed that the system was a research prototype, that it was not intended for emergency detection or intervention, and that the absence of alerts would not imply health or clinical stability. They were instructed to use usual health care and emergency services. Adults aged 50 years and above with 2 or more chronic conditions and income at or below the federal poverty level (US $14,580 annually for a single-person household) [49] were included. The lower age threshold was informed by community partner input, which emphasized that chronic conditions often develop earlier among racially and ethnically diverse populations. Older adults living alone were prioritized for recruitment, although homes with fewer than 2 residents met inclusion criteria. A target was set for recruiting >80% single-resident homes. Exclusion criteria included dementia, severe mental illness, or pets that could interfere with sensor readings.

Data collection combined continuous, ambient, home-based sensing with participant-facing and team-documented narratives, and in-the-moment surveys associated with health alerts. Table 1 contains the data elements of the study with descriptions. Alerts were triggered when anomalies exceeded preset thresholds (eg, no door use for 48 hours, reduced kitchen activity, sleep disruption). To minimize alert fatigue, some overlapping alerts were combined into a single alert, yielding 4719 novel alerts. For example, a Restless Night alert and a Changing Sleep Routine alert on the same night would be combined into one alert requiring one survey response. Because it was not possible to have continuous daily CHW coverage for the study duration, the full-time research RN or lead investigator (an RN) assisted in responding to alerts, and alert notifications were delivered only during business hours. Business hours were 9:00 AM to 6:00 PM Pacific Time, Monday to Friday. Nighttime sensing resulting in alerts saw the alerts delivered at 9:00 AM on weekdays, and weekend alerts were delivered at 9:00 AM on Mondays. No alerts were discarded.

Each alert included a secure survey link for reporting current health status. Unacknowledged alerts prompted CHW follow-up and escalation to an RN as needed. Narrative data were drawn from three sources: free-text survey responses, CHW/RN field notes after alert-related contacts, and semistructured exit interviews. Although 6 additional languages (Cantonese, Mandarin, Korean, Vietnamese, Spanish, and Russian) were spoken beyond English, multilingual CHWs communicating in preferred languages documented all responses in English.

At a conceptual level, the smart health system generated alerts using 3 components: sensing, rule-based interpretation, and filtering. Ambient motion and door sensors continuously captured movement throughout the home, and these raw signals were aggregated into higher-level behavioral events representing where and when a participant was active (eg, time in the kitchen, nighttime location, frequency of door use). A set of rules, co-designed with community partners and the PIs who were RNs, compared each participant’s recent behavior against their own established baseline routine to identify clinically meaningful deviations, such as prolonged inactivity, disrupted sleep, reduced kitchen use, or missed bathing patterns. When a deviation crossed a predefined threshold, an alert was generated. To prevent alert fatigue, repeat alerts of the same type within a 1-week window were suppressed.

Multimedia Appendix 1 describes the full technical specifications of the sensors, processing pipeline, rule definitions, and figures exhibiting the health-monitoring dashboard (deidentified) and the clinical research dashboard, which shows an exemplary behavioral anomaly where a participant did not exit their home for nearly 3 days.

Quantitative data, including participant characteristics, alert metrics, and response pathways, were summarized descriptively. Response trajectories were prospectively examined to identify alerts that elicited participant, support person, and CHW responses and those that resulted in subsequent RN action.

Qualitative data included free-text survey responses, follow-up narratives documenting health changes, and semistructured exit interviews. The data were analyzed using qualitative descriptive methods [50] and reflexive thematic analysis [51]. Coding was inductive [52] and iterative, with repeated review of written text-based data to refine codes, categories, and themes.

To strengthen rigor, the analytic team included RNs, a family nurse practitioner, bilingual CHWs, and engineering or computing expertise selected to reflect the linguistic, cultural, and clinical heterogeneity of the study population. The team met across 4 synchronous analytic circles and 8 asynchronous review sessions to discuss emerging patterns, compare coding perspectives, and refine thematic structure. After each session, written analytic summaries were synthesized into a shared memo for iterative team review. Related subthemes were consolidated into higher-order concepts through consensus discussion. An audit trail of analytic decisions, supporting written text-based excerpts, and reflexive memos was maintained throughout to enhance transparency and cultural sensitivity.

A total of 63 older adults consented to participate in the study, of whom 46 completed 6 months of monitoring and 2 completed 4 and 5 months of monitoring, respectively. These 2 individuals were included in the total sample because they completed two-thirds or more of the study. Fifteen participants withdrew, with reasons including privacy and system invasiveness, discomfort with the perceived invasiveness of questions asked by the Internet Service Provider if the internet was not already connected in the home (free internet was available in subsidized housing), and time constraints. The total sample included 48 older adults (ie, 48 smart home sites). Figure 2 summarizes participant flow from initial information sessions through to the final analytic sample.

Sample demographics are presented in Table 2. Participants ranged in age from 54 to 94 years, with 77.1% (n=37) identifying as female. Educational attainment varied, with 44% (n=18) holding a high school diploma and 17% a college degree. Seven languages were represented across 4 racial groups and 8 ethnicities. Most participants lived alone (n=44, 92%) and had access to the internet before consenting to participate (n=38, 79%). Ten (21%) participants received internet access via study-provided hotspots. The most prevalent conditions among participants were heart (n=23, 47.9%), musculoskeletal (n=12, 25%), and gastrointestinal (n=6, 12.5%) conditions.

Participants received 4719 emails and SMS text message alert reminders when baseline movement patterns changed, breaching thresholds of the custom rules-based monitoring algorithm. If a participant did not respond to their email alert within an hour, they received an SMS text message reminder. If they still did not respond, their support person received an email followed by an SMS text message alert. If the support person did not respond, the alert was automatically escalated to a CHW who then contacted the participant as soon as possible by phone. After making contact, the CHW responded to the alert on behalf of the participant and either documented a nonissue or supported the health situation as needed, including escalating a health issue to the research team RN as needed per protocol. CHW and RN responses to alerts included phone outreach and asking targeted questions related to 1 of 6 alert types (Homebody, Daytime Resting, Eating Routines, Bathing, Restless Night, Changing Sleep Routines). For example, for a Homebody alert, the participant would be asked how they were feeling and how long it had been since they last left their house or had visitors over. For a Restless Night alert, the participant would be asked about changes in bedtime, morning wake time, and the number of sleep interruptions that included getting out of bed, as well as the participant’s perspective on the cause of sleep interruptions (eg, needing to go to the bathroom to void, experiencing pain, or worrying). If appropriate, the participant was asked to take their vital signs on equipment provided by the study team. Across the 6-month monitoring period, CHWs and nurses completed 1060 participant contacts, resulting in 72 CHW or RN interventions combined (nonurgent and urgent). Commonly, CHW interventions included therapeutic listening and recommendations to avoid dehydration and get enough rest. Common RN interventions included advice to take medications as prescribed, check blood sugar, take vital signs (blood pressure, heart rate, oxygen saturation), or call their support person. For 93% (4389/4719) of the alerts, no action was required. In six instances, the RN suggested the participant call their provider the same day (n=6, <1%). In 5 instances (n=4, <1%), the RN recommended urgent care.

Participant digital response rates were 1.57% (n=74) for initial email alerts and 7.79% (n=368) for follow-up SMS text message alerts. Support persons responded to 2.37% (n=112) of the email alerts and 5.21% (n=246) of the SMS text message alerts. CHWs responded to 78.36% (n=3698) of the alerts, including 4.68% (n=221) within 1 hour, with the remaining responses occurring after 5 hours. In contrast to the very low response rates via email (n=186, 3.94%) and SMS text message (n=614, 13.01%) from participants and support persons, CHWs demonstrated substantially higher engagement, responding to 78.26% (n=3698) of the alerts.

Participants in the study completed a likelihood-to-recommend rating during their end-of-study interview using a 0‐10 scale (0=would never recommend the health monitoring system and 10=would highly recommend it to family or friends). Fifteen participants did not provide a rating. One response knowingly exceeded the scale maximum and was recoded to 10 prior to analysis in accordance with the instrument’s scale limits.

Among respondents (n=33), the mean likelihood-to-recommend rating was 8.68 (SD 1.68), the median was 9.0 (IQR 8.0‐10.0), and scores ranged from 3 to 10.

Following a standard Net Promoter Score methodology, respondents were then classified as promoters (9-10), passives (7-8), or detractors (0‐6). Most respondents were promoters (22/33, 67%), while 21% (n=7) were passives and 12% (n=4) were detractors, resulting in a Net Promoter Score of +55.

This study identified several barriers to smart home adoption among older adults managing multiple chronic conditions while experiencing poverty. Beyond technical feasibility, adoption was shaped by digital access barriers, identity theft concerns, cultural caregiving expectations, environmental constraints, and structural inequities tied to poverty. Across qualitative themes, Aloneness, Trust, and Human Connection emerged as central determinants of engagement.

A key finding was that participants consistently preferred human-mediated communication, particularly with CHWs and RNs, over automated alerts. Although direct digital response rates were low, CHW engagement was high, suggesting that integrating trusted community and clinical personnel into the health care-technology-patient loop may be critical for sustained use in low-income settings. Future research should test how to reduce the staffing intensity of this model while preserving the human contact participants valued, including through acuity-based tiering, machine learning personalization, and language-concordant outreach.

This study also introduces digital distress, a novel concept describing stress that arises when technology feels intrusive, confusing, insufficiently personalized, or disconnected from linguistic and cultural context. Digital distress was most evident when participants were uncertain how the system worked, whether they were using it correctly, or who was responsible for monitoring health changes.

Collectively, these findings indicate that smart health systems designed around human relationships, cultural responsiveness, and contextual adaptation, and implemented with trusted CHW and nursing support, may better promote equitable aging in place among older adults managing multimorbidity while experiencing poverty.

This study has several limitations. The sample was relatively small and drawn from a specific population of older adults managing multimorbidity while experiencing poverty. Despite the extant literature defining older adulthood as age 65 years or above, the study sample included immigrant adults aged 50 years or above who were managing multimorbidity, which commonly affects these populations at younger ages due to the impacts of social determinants of health. This may limit transferability across populations, settings, and cultural contexts. In addition, the smart health system was a prototype that underwent iterative refinement during the study, and its rules-based alerting approach may have contributed to irrelevant or missed alerts. The intensive involvement of CHWs and RNs, while central to equitable implementation, may also limit scalability in other care settings. Because digital distress emerged inductively within this context, additional research is needed to examine its relevance across populations and technologies. Future work should evaluate scalable machine learning approaches, multilingual and culturally responsive interfaces, improved alert accuracy and data visualization, integration with clinical workflows, and strategies for communicating automated messages and engaging older adults and families with sensor-derived health data while preserving simplicity and trusted human support.

Smart home health monitoring systems hold promise for supporting aging in place and helping older adults maintain independence, but their success among those experiencing poverty depends on more than technological innovation. If systems are not designed with attention to the unique barriers faced by this population, they risk reinforcing existing health inequities. This study demonstrates that adoption is shaped by a complex interplay of digital literacy, cultural values, trust in institutions, and the physical realities of low-income housing. A community-in-the-loop approach that integrates ambient sensors, automated unobtrusive monitoring, and human-centered care networks may reduce isolation, foster trust, and maintain human connection while offering real-time health insights. To be effective and equitable, future smart health systems must be co-designed with communities and prioritize personalization, cultural sensitivity, and low-burden engagement. In addition, technology-enabled health monitoring for the home must include intentional, explicit strategies to mitigate digital distress. These findings offer a roadmap for researchers, designers, and policymakers seeking to close the digital divide and promote health equity through technology-enabled care.