Digital Health Tools Applications in Frail Older Adults - A Review Article

Introduction

Aging society is an inevitable ongoing trend in in the world. World Health Organization reported that within 2030 1 in 6 people in the world will be aged over 60 years and by 2050 number of persons aged over 60 years old is expected to reach 426 million (1,2). This trend is common in many countries and is attributed to a combination of factors such as improved healthcare and advancements in medical technology, which have allowed people to live longer (3). The world public health is now facing the challenges and opportunities that come with an aging society since older adults are linked to increased multiple chronic diseases, comorbidities and mortality (4–6).

Frailty is a clinical condition highly prevalent in the aged population in which a frail individual is more vulnerable to health-related risk exposure (7,8). Studies showed that this condition has been linked to increased hospitalizations, Emergency Department (ED) visits (9–12), poorer quality of life (13), impaired cognitive function (14), increased morbidity and mortality (15). Frailty is commonly defined by Fried et. al. 2001 using unintentional weight loss, gait speed, exhaustion, grip strength and physical activity as a clinical diagnostic criteria (16).

There has been an increase in studies on frailty in recent years since frailty could be decreased or reversed with a long-term-based exercise intervention (17,18). Fairhall et. al. 2014 conducted an randomized controlled trial of 241 community-dwelling older adults in Australia where the findings showed that exercise and nutrition intervention could significantly improve frailty status in the treatment group (19). The result agreed with Nakamura et. al. 2022 where 111 community-dwelling older people in Japan were randomly assigned to perform a home-based training during Covid-19 pandemic (20).

Several studies have employed digital tools to help diagnose and treat frailty as technology has improved and become more accessible (21,22). We believe that information technology can help us recognize frailty earlier, and that the earlier we identify this condition, the better healthcare providers can treat the patient with a better prognosis and health outcomes. This review aims to identify and summarize prospective characteristics, diagnostic models, and therapeutic studies in utilizing digital health technologies in community-dwelling frail older persons.

Search Strategy

We used PubMed as our main source of published literature for our search strategy. The combinations of search terms were (“frailty*” OR “frail” OR “frail elderly”) AND (“digital” OR “machine learning” OR “smartphone*” OR “AI” OR “artificial intelligence” OR “deep learning” OR “device*”) AND (“older adults” OR “elderly” OR “elder” OR “old”). The selected publications in our review were limited to English publications and publications within 2012-2023. Additional literature found in systemic reviews and meta-analysis were manually selected to include in this review. Inclusion Criteria: 1.) The study recruited older adults aged at least over 40 years old; 2.) The study applied digital health tools to find association, causal relationship or make prediction between frail and non-frail population. Exclusion Criteria: 1.) The study was not written in English; The study did not demonstrate a comparison of result between frail and non-frail population; The study did not utilize digital tools.

Characteristics studies

We found 9 relevant frailty characteristics studies and summarized them into Table 1. Most studies used various types of digital sensors to measure surrogate outcome of frailty and are categorized into 1.) Trunk parameter 2.) Gait parameter and 3.) Non-gait parameters

Trunk parameter

Most studies’ methods involved researchers instructing volunteers to perform physical function tests while wearing a digital sensor that measures characteristics that likely represents frailty. Galan-Mercant et. al. (2013) studied 30 community-dwelling volunteers over the age of 60 who performed a sit-to-stand, stand-to-sit test while wearing an iPhone4 attached to the chest to assess 3D acceleration, angular velocity, and trunk displacement during the turn transition (23). The findings revealed all factors differed significantly between frail and non-frail subjects.

Parvaneh et. al. 2017 conducted a study using a wearable necklace-like sensors located at the chest of 120 community-dwelling participants aged over 70 years old to monitor and assess postural transition differences among frailty levels for 24 hours, and the results showed that the number of Stand-to-walk and total postural transitions were significantly different between groups (26). Millor et. al. (2013) asked 47 community-dwelling volunteers over the age of 50 to perform stand-up and sit-down from a chair as many times as they could in 30 seconds while wearing an inertial orientation tracking sensors on their lumbar spine (24). The study showed that healthy participants outperformed frail people with less sway on the sit-to-stand cycle.

Therefore, the parameters derived from sensors attached to the trunk such as 3D acceleration, velocity and postural sway while doing physical function tests could discriminate frail and robust in the community-dwelling older adults.

Gait Parameter

Gait assessment was another method used by researchers to analyze diagnostic variables in frail older adults. Zhou et. al. 2019 investigated whether parameters from an instrumented trail making task (iTMT) and gait sensors worn on both shins to measure gait speed and iTMT derived parameters could distinguish between frail and robust participants (28). The findings revealed that gait speed and iTMT velocity were significant parameters that could help classify frailty status among the outpatient care population. Moreover, Jasen et. al. 2019 carried out an intervention research which 112 community-dwelling older persons were requested to wear a wearable sensor in a shirt while undertaking a walking test under two conditions: 1.) Walk a distance of 4.57 meters at your own speed; and 2.) Walking a 10-meter distance as rapidly as possible (29). The findings correlated with the previous studies, which suggested that the proportion of time spent walking and standing, the maximum steps in one test bout, and walking speed might all be potential predictors of frailty classification (30).

Non-gait Parameters

To determine frailty status, other variables could be used in addition to those mentioned above. Toosizadeh et. al. (2015) studied the association between frailty status and non-gait parameters using a wearable gyroscope sensor attached to the upper arm and wrist of 117 community-dwelling adults over 65 years old to measure elbow function while performing a 50-second trail of elbow flexion in a seated position (25). The results revealed that the speed of elbow flexion, flexibility, and power of elbow flexion differed significantly between robust and frail participants. In the study by Castaneda-Gameros et. al. (2018), moderate-to-vigorous physical activity measured by a sensor that records acceleration and gyroscopic data worn on the hip for 7 days was associated with frailty status in community-dwelling old adults (27). Additionally, Kikuchi et. al. 2020 found the association of intensity-specific physical activity. The results showed that sedentary behavior and physical activity (moderate-to-vigorous) were significantly different between frail and robust in 511 Japanese community-dwelling participants aged over 65 (31).

As a result of the mentioned studies, there are multiple potential variables that could represent characteristics of frailty. Non-gait parameters appeared to have the highest clinical feasibility if researchers could integrate a model into a smartwatch since a wrist-worn device is simple to use and most older adults are already accustomed to wearing a smartwatch.

Diagnostic Studies

Frailty identification is a clinically relevant topic since it is a condition that may be reversed from frail to robust. Several studies are being conducted to develop tools and diagnostic models for classifying frail and non-frail older adults. According to the authors’ evaluation of the published evidence in this field, there are two types of frailty diagnostic tools that use technology: 1.) Clinical Data; and 2.) Data derived from wearable devices and biological sensors which are summarized in Table 2.

Clinical Data

Aznar-Tortonda et. al. 2020 collected data from 621 community-based participants in a cross-sectional observational study utilizing an Android mobile device application. Sex, age, polypharmacy, hospitalization, and diabetes history were chosen characteristics and employed in a logistic regression model (32). This model obtained an AUC of 0.78, suggesting that a brief clinical history might be utilized to classify frailty in older persons. Sajeev et. al. 2022 used 20 anthropometric, environmental, social, lifestyle, and physiologic variables from 656 community-dwelling adults aged 40–65 years old to develop and internally validate four machine learning models, including logistic regression, linear discriminant analysis, support vector machine, and random forest (33). With an AUC of 70.8, the random forest model achieved the highest discrimination performance. This study found that machine learning models could be used to diagnose frailty. However, the large number of variables in the purposed models could make it difficult to implement them in clinical practices and community settings, and the selected features appeared to be more difficult to measure and more complicated than the standard diagnostic criteria for frailty. A future study is required to demonstrate the real-world application of a frailty diagnostic machine learning model based on clinical characteristic data.

Data derived from wearable devices and biological sensors

Most studies for biological sensors and wearable devices employ criteria similar to characteristics research. We divided the parameters into three major categories: 1.) Physical Function test; 2.) Gait and balance test; and 3.) Non-gait-related test.

Therapeutic studies

Based on the current evidence summarized in Table 3, pre-frail and frail older adults are recommended for multi-component physical activity program and progressive resistance training program. Multiple studies have shown improved cognitive function, physical function, and frailty status in older adults after physical exercise intervention. Therefore, our review selected frailty therapeutic studies that integrated the use of technology to improve frailty state in the elderly.

Daniel et. al. 2012 conducted a study where 23 community-dwelling pre-frail volunteers aged over 70 years old were randomized into one of three groups: control, seated exercise, or Wii®-fit. The findings showed better outcomes for all intervention groups (40). Wii-fit exercises and seated exercises were both superior to the control group in maintaining or improving physical functions. Liao et. al. 2019 recruited a randomized controlled trial of 52 prefrail and frail elderly where the participants were divided into two exercise intervention 1.) Exergaming group and 2.) Combined resistance, aerobic and balance exercise group for 36 sessions over 12 weeks (41). The results revealed both gaming exercise and combined exercise groups improved frailty status among the elderly. The study correlated with Moreira et. al. 2021 where an RCT of 66 pre-frail older adults were assigned to either exergaming intervention and traditional multicomponent exercise (42). The findings showed that both programs were clinically effective for delaying frailty status and improving physical and cognitive function.

Exergaming have shown positive health outcomes in terms of enhancing physical function, cognitive function, and frailty status. The programs could be done in a home setting, making exercise intervention easily accessible. However, the majority of frail people are older adults, who may face challenges using technologies because of their lack of digital literacy and technology acceptance. One of the studies cited above had a dropout rate of over 30%, which suggests that a portion of older persons might not find the use of a digital intervention tool appropriate.

Conclusion

From our review, we found that there are many potential etiognostic factors that could help diagnose frailty status using digital tools from trunk, gait, upper-extremity, and physical activity parameters. Researchers had used these parameters to create multiple well-performing models to classify frailty status in the older adults. We found non-gait parameters the most appealing variables for future research as a frailty diagnostic model integration into a wearable device. However, the model classification results should be interpreted with caution because these models may be overfitting due to a lack of external validation studies.

Regardless of the tools used, studies have shown that exercise can improve frailty status. Rather than developing a single standalone exercise platform, digital health technology developers should focus on how to implement these therapeutic platforms with health care providers or coaching platforms that could encourage and motivate prefrail and frail old adults to engage in more physical activity.

However, there are gaps between frailty screening and care delivery for patient at risk. We did not find a study that combines screening and therapeutic digital health tools into one. To prevent frailty early on, we propose that future studies should focus on how to integrate a screening/diagnostic tool with personalized therapeutic/prevention recommendations to users.

Supplementary Materials

None

Author Contributions

The review’s first draft was written by N.I. and W.S. N.I. and W.S. reviewed and edited the manuscript, assisted in the formulation of the research question, and advised on study objectives and methodologies. N.I. and W.S. contributed to the review’s conceptualization, design, and development. The final version of the manuscript was reviewed and approved by all authors.

Funding

This research received no external funding. The APC was funded by Chiang Mai University, Thailand.

Informed Consent Statement

Not applicable.

Data Availability Statement

The data presented in this study are available on request from the corresponding author.

Conflicts of Interest

The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of manuscript, or in the decision to publish results.