Introduction
Digital therapeutics represent a novel approach to patient management and treatment. They are defined as therapeutic interventions that utilize software programs to manage patients’ medical conditions [1]. With the rapid digital transformation of medical systems worldwide, digital therapeutics are garnering increasing attention. They are a form of digital technology with a proven treatment effect, designed to directly treat and manage patients’ diseases and disabilities. Since the U.S. Food and Drug Administration approved ‘reSET’ for the management of substance use disorder in 2017, a variety of digital therapeutics have emerged and are now widely used in the clinical field [2].
Currently, digital therapeutics are being utilized in a variety of prospects, including hypertension, diabetes, and even oncology. Several studies have reported that digital therapeutics demonstrate economic benefits in the management of hypertension and diabetes [3,4]. In the field of oncology, digital therapeutics have shown effectiveness in patient management.
Gait involves numerous physical systems, including the cardiovascular, pulmonary, neurological, and musculoskeletal systems [5]. Some research has observed gait disturbances in chronic diseases such as chronic obstructive pulmonary disease and chronic kidney disease (CKD). Additionally, gait disturbances can serve as a predictor for cardiovascular disease [6–8]. From a neurological perspective, gait function demonstrates a correlation with neurological diseases such as cerebral palsy and Parkinson disease. This correlation extends to fundamental areas such as cognitive function, and gait function can be utilized as a significant factor in confirming these diseases [9–11]. Gait speed can also be useful in screening for sarcopenia [12]. Another report suggests that sarcopenia is associated with cognitive impairment, which may be attributed to a slower gait speed [13]. Therefore, gait can be considered a significant biomarker, indicative of an individual’s health status.
Globally, CKD has a prevalence of 13.4%, necessitating renal replacement therapy for millions of patients. CKD imposes a significant burden on global health and mobility, primarily due to its impact on cardiovascular risk and end-stage kidney disease. Additionally, the rise in diabetes, hypertension, obesity, and aging patients contributes to an increase in CKD cases. To manage the economic burden of the growing number of CKD patients, it is crucial to enhance disease surveillance and prevention [14,15].
Patients undergoing hemodialysis who are unable to walk or can only walk slowly have higher mortality and hospitalization rates than those who can walk normally [5]. Dialysis patients often exhibit gait disturbances due to metabolic factors that contribute to muscle loss. There are also observable changes in the brain area that controls motor function, leading to walking disorders [7]. Furthermore, these gait disturbances can contribute to falls, which are associated with increased mortality and morbidity, and pose an economic burden on society.
As CKD advances, there is a higher rate of muscle loss, which is associated with increased disability in daily life and a reduction in walking speed. This muscle loss has also been linked to a higher rate of vascular calcification and vascular access failure in patients undergoing hemodialysis [16].
We hypothesized that gait can serve as a proxy indicator of patients’ disease status (e.g., sarcopenia or CKD) and that health interventions for those patients will improve gait. Thus, our interventions have the therapeutic goal of improving gait parameters. We will also conduct bioelectrical impedance analysis (BIA) measurements to analyze changes in muscle mass.
Increasingly many researchers are interested in identifying ways to improve gait abnormalities in CKD patients, particularly in light of studies indicating that musculoskeletal health is linked to kidney function. However, few studies have specifically investigated abnormal gait in CKD patients. Nonetheless, interventions that have an impact on gait—including exercise, nutritional support, and the management of underlying conditions—may exert a beneficial effect on CKD patients’ physical function and quality of life.
CKD exerts substantial negative impacts on the structure, activity, and function of bone and skeletal muscle, reducing the quality of life of CKD patients and making a contribution to morbidity and mortality. Exercise has positive effects on the metabolism and function of muscle and bone, which has prompted suggestions that regular physical exercise could serve as a therapeutic modality for patients with muscle and bone-related disorders [17]. Muscle weakness can contribute to gait abnormalities, including slower walking speed, changes in stride length, and impaired balance. CKD patients might also develop peripheral neuropathy as a result of metabolic imbalances and uremic toxins. Neuropathy can affect motor and sensory nerves; this can impair gait through poor coordination, muscle weakness, and changes in sensations [18]. Abnormal gait in CKD patients could potentially be improved indirectly through the treatment of complications such as peripheral neuropathy and the management of factors contributing to wasting and muscle weakness. Chu and McAdams-DeMarco [19], in this context, pointed out that exercise training, in addition to cognitive training, can lead to better quality of life and long-term health outcomes.
In CKD patients, gait parameters can be improved through regular exercise—in particular, balance exercises and resistance training—since these activities have positive impacts on coordination, balance, and muscle strength. Sufficient protein intake and nutritional support may avert muscle wasting and weakness, which could also lead to improved gait in patients with CKD.
The aim of this study is to compare the gait of patients with CKD to that of healthy individuals in order to identify any distinctive features of pathologic gait in CKD patients. We conducted an intervention targeting pathologic gait through lifestyle modifications, such as dietary changes and exercise interventions, and we will conduct long-term follow-up to determine whether these can lead to improvements. Ultimately, we plan to use gait as a baseline measure and biomarker for diagnosis and treatment within the realm of digital therapeutics.
Methods
This study received approval from the Institutional Review Board at Hallym University Chuncheon Sacred Heart Hospital (No. 2022-06-009). All participants provided informed consent.
Subjects (chronic kidney disease patients and healthy individuals) and the study protocol
To investigate the characteristics of pathological gait in patients with CKD, we recruited 300 CKD patients and 220 individuals from a non-CKD healthy population. The healthy population was defined as those who do not currently have acute or chronic diseases or take regular medications, including for hypertension and diabetes, and have no noteworthy medical history in our hospital’s electronic medical records. We excluded those who were unable to walk, had pacemakers, or declined health intervention services. Consequently, 276 CKD patients and 217 individuals from the healthy population were included in this study. In both groups, skeletal and inertial measurement unit (IMU) measurements were conducted as the primary examination.
For CKD patients, we collected information on age, sex, primary renal disease and the length of ESRD. The CKD patient group underwent a primary workup that included a complete blood count (CBC), routine chemistry, bone mineral density (BMD), BIA, and a handgrip test. The healthy individuals, meanwhile, received a primary workup consisting of BIA and a handgrip test.
Both groups participated in an 8-week health intervention, after which they were retested using skeleton and IMU measurements, similar to the primary examination. Concurrently, as part of a secondary evaluation, CKD patients underwent testing for CBC, routine chemistry, BIA, and handgrip strength, while the control group was tested for BIA and handgrip strength (Fig. 1).
Structure of the gait analysis room
Tests were carried out in the gait analysis room established at Hallym University Chuncheon Sacred Heart Hospital. The laboratory measured 9 × 4 m and was equipped with Azure Kinect cameras (Microsoft Corp.) centrally positioned for recording inspections. A rectangular line was delineated with tape to guide subjects in their walking patterns as per the instructions. The laboratory was staffed with three inspectors and a nurse, who were responsible for conducting tests and managing any unforeseen circumstances.
Initially, the subjects completed a 10-step walk, followed by a 6-m walk. This 6-m walk was performed twice in total. Gait data was collected during the middle of these walks, specifically excluding the acceleration and deceleration phases. In addition, the Timed Up and Go (TUG) test and Tinetti test, which are widely used in studies of walking and falls, were conducted as performance-oriented mobility assessments [20,21]. Participants engaged in activities such as sitting on chairs, rising to their feet, following walking instructions, or standing with their eyes closed. Their stability or instability during these tasks was then scored. Based on these tests, we categorized participants into high-risk fall groups. Additionally, while the participants were performing these tasks, we collected data on their walking patterns (Fig. 2).
Skeleton analysis and artificial intelligence models
We used Azure Kinect and its software development kit to acquire RGB images and depth data [22]. This equipment enables us to capture three-dimensional (3D) skeleton information from subjects. We have also implemented an automatic quality inspection framework to filter out noisy data automatically. Furthermore, by using 3D skeleton refining tool, we were able to generate refined skeleton data, as depicted in Fig. 3A [23].
While participants walked, between 80 and 120 frames of skeleton data were collected at a speed of 20 to 25 frames per second. The gait parameters were then extracted from this joint three-axis data. The organization of the gait parameter, based on the joint three-axis data, is displayed in Table 1. Our gait analysis provides a total of 44 parameters, including movements of the head, spine, hip, knee, and foot joints. The skeleton data were analyzed using various artificial intelligence (AI) classification models, including GRU (gated recurrent unit) and RNN (recurrent neural network), to distinguish the healthy population and CKD patients. The derived data were then compared with diagnostic labels, which were based on medical diagnostic data. The analysis was conducted using accuracy, sensitivity, specificity, precision, recall, and F1 score, metrics that have proven accurate in previous studies [24]. Ultimately, our goal is to develop a model capable of distinguishing between patients with CKD and those without, by analyzing the gait parameters of each subject.
Inertial measurement unit analysis and artificial intelligence models
Before the walk test, subjects were equipped with devices to collect IMU data. This process involved gathering information from the three-axis gyroscope and three-axis accelerometer, and applying a pre-processing technique to eliminate noise using digital filter algorithms. Only data collected during the walking phase were used in the study, excluding measurements taken before the walk commenced and after it had ceased.
1) Inertial measurement unit data selection and preprocessing method
In this study, three-axis acceleration data and three-axis gyroscope data were generated as input to the model. Only the IMU data measured during 10-second walking and 6-meter walking were analyzed. Data that were missing due to errors at the time of measurement or were determined unsuitable for analysis were excluded. The data measured at 50 Hz by IMU were segmented into 3-second windows (150 samples) to generate the data, and the windows were slid by 12 samples. Windows with missing data (missing packets due to communication problems) were not used for segmentation. There were 7 labels: 0 for patients without CKD, 1 to 5 according to CKD stage, and 6 for hemodialysis or peritoneal dialysis.
All data were normalized using the standard scale, with missing data excluded. Subjects in each class were divided into a training set and a testing set, and the model’s generalization performance was evaluated on a subject-by-subject basis. Furthermore, after the window split, the ratio of training sets to testing sets in each class was approximately 8:2, as shown in Fig. 4A.
2) Methods and architecture of artificial intelligence model analysis
The CKD classification network consists of a one-dimensional (1D) convolutional neural network (CNN). Four distinct models were developed for each sensor placement site (wrist, waist) and each gait measurement method, with training subsequently conducted. The 1D CNN model is structured with feature extraction part, which is composed of convolution layers, and classification part, which is composed of fully connected layers. The information gleaned from the feature extraction is then relayed to the classifier, which consists of dense layers and a softmax layer. In the context of CNN, the input signal is processed as a time series signal, necessitating the use of a 1D convolution operation. This operation is employed to analyze the characteristics of the gait signal (Fig. 4B). We split the training and validation sets using a stratified approach and used k-fold cross-validation to evaluate performance comprehensively in a way that does not depend on specific targets. Specifically, when this method is practically applied, 1) a developed model with training data will be fine-tuned with a validation set to strengthen the model, or 2) a model will be learned with not only training data but also all datasets.
Protocol of the bioelectrical impedance analysis measurement
Subjects were instructed to consume meals 2 hours before BIA measurement to prevent interference from digestion. For dialysis patients, BIA measurements were performed within 10 to 30 minutes after dialysis. Once subjects arrived to the gait analysis room, they were instructed to stand for 5 minutes with light clothes and BIA was measured using the 790 device (InBody) in an upright position. They underwent BIA measurements with the same machine before and after the health intervention.
Protocol of the health intervention program
The protocol of the planned health intervention program is shown in Fig. 5.
Personal health information was gathered using smartphone applications, and Health Cube (Health Bridge) models were employed to assess the health status of members. Following an 8-week healthcare intervention service, the performance of the intervention was managed. Additionally, life logs, including diet photos, exercise details, health coaching records, and self-reported health information, were collected. The detailed methods are described below.
Health cube classification
We divided the participants into 27 categories based on their classification of physical activity (high, medium, and low), disease (hyperlipidemia, diabetes, and hypertension), and weight (normal, obese, and underweight). The health goal was set to be one level higher than the subject’s current health status, and exercise and nutritional management were implemented.
Eight-week health intervention
Using the developed app, strength training and aerobic exercise were performed once or twice a week and nutrition was appropriately managed, depending on the subject’s Health Cube classification and health goal. We analyzed and provided customized health coaching to subjects based on their activity, step count, InBody diagnosis results, and meal photos.
Study variables
The variables collected in this study included the following: the appendicular skeletal muscle index (ASMI), which was obtained using BIA to determine sarcopenia, and a handgrip test (Table 2). Sarcopenia was categorized into three stages based on the criteria outlined in Table 3. Probable sarcopenia was identified by criterion 1. The diagnosis was confirmed by additional documentation of criterion 2. If criteria 1, 2 and 3 are all met, sarcopenia was considered severe (Table 3) [25]. To align the gait analysis with existing test methods, we conducted 10-second walk, 6-meter walk, TUG, and Tinetti tests. Blood and urine tests were performed for patients, and electronic medical records were reviewed to investigate medication use and disease codes for all subjects. Additionally, BMD was measured using a dual X-ray absorptiometry scan for patients only (Table 2).
Gait analysis schedule
The detailed schedule for gait analysis was as follows. Initially, the patient was escorted to the examination room, where they took a 10-minute rest, provided their informed consent, and completed a health-related survey and BIA study. Subsequently, after donning IMU sensor, their gait was measured (Table 4).
Outcome measurements
All CKD subjects in this study are regular outpatients at Hallym University Chuncheon Sacred Heart Hospital. Therefore, any incidents of falls, fractures, hospitalizations, and deaths will be confirmed at the 1- and 3-year marks. Laboratory tests will be conducted during each outpatient visit (every 1 to 6 months). Additionally, BMD tests will be performed annually.
Results
Baseline characteristics of the participants
Out of 493 participants, 217 did not have any disease including CKD, while 276 did. The age and sex of the two groups were not matched, as the study’s aim was to gather data on both normal and pathological gait. The CKD group was older (CKD, 65 years vs. normal, 38 years) and had a higher proportion of men than the normal population (CKD, 157 [64.3%] vs. normal, 87 [35.7%]). The mean BMI was higher in the CKD group than in the normal control group (CKD, 24.98 kg/m2 vs. normal, 23.63 kg/m2). The ASMI was not significantly different between the two groups, but the prevalence of sarcopenia was higher in the CKD group (CKD, 73 [26.5%] vs. normal, 31 [14.3%]). As anticipated, the CKD group had a higher number of slower TUG subjects (those who took 10 seconds or longer to perform the test; CKD, 204 [73.9%] vs. normal, 129 [59.4%]) and exhibited weaker hand grip strength compared to the normal population (CKD, 204 [73.9%] vs. normal, 129 [59.4%]). In the Tinetti test, a biomarker for balance and falls, the CKD group had a higher number of subjects at risk (Table 5).
We enrolled numerous young people to collect normal gait, and we plan to perform 1:1 matching by age and sex in the future for additional analyses to confirm differences with CKD patients.
Discussion
As society continues to age, the incidence of fall accidents is on the rise, leading to a significant economic burden globally. Consequently, there is a growing demand for technology that can prevent such accidents [26]. Falls can be attributed to a variety of factors, including muscle loss, balance impairment, nutritional deficiencies, comorbidities, and medication. Notably, gait plays a substantial role in the occurrence of falls [27,28].
Gait disorders can negatively impact numerous areas, such as patient quality of life, long-term survival, and the occurrence of additional complications, including falls [29,30]. In the realm of modern medicine, where there is an increasing focus on quality of life and daily health care, gait is emerging as a crucial factor [31]. Various factors, including aging, cognitive impairment, dementia, sarcopenia, and musculoskeletal disorders, are involved in gait disorders in complex, mutually interdependent ways [32].
Gait analysis and interventions to improve gait have been researched in various medical fields, including sports medicine, pediatrics, orthopedics, and neurology, as well as in the context of sarcopenia. However, there has not been extensive research or standardization related to gait analysis and interventions specifically tailored for CKD patients, including those with ESRD. Nonetheless, it is reasonable to hypothesize that gait performance could be indirectly improved by addressing underlying factors, such as CKD complications, nutritional deficiencies, and muscle weakness, through physical therapy, exercise, and other strategies for disease management. However, further studies investigating gait analysis and interventions specifically developed for CKD patients are necessary in order to develop more targeted strategies.
Previous studies have underscored the significance of walking and suggested the necessity for quantitative standards for walking data, to facilitate the analysis of walking disorders [33,34]. Therefore, this study conducted a gait analysis, utilizing patient skeleton and IMU data, to lay the groundwork for future research into gait disorders. The hope is that this analysis will offer a scientific foundation for a clinical comprehension of patients’ pathological gait mechanisms, and supply objective, quantifiable data for evaluation.
We also encourage lifestyle modifications in all subjects through the use of a smartphone application. The data collected will be subsequently utilized for gait analysis via ontology-based analysis. This will aid us in gaining a precise understanding of gait, provided that the subjects incorporate the use of IMU into their daily routines.
However, when analyzing gait, the presence of numerous confounding factors must be kept in mind. Gait can be affected by orthopedic problems, cognitive and neurologic disorders, medication, and nutritional status, as well as by demographic characteristics such as sex and age. Nonetheless, gait has the potential to serve as a biomarker for digital therapeutics because it can be measured longitudinally at regular intervals and is linked to an individual’s health status. A limitation of our study, though, is that it was not feasible to perform BMD measurements in individuals with normal clinical profiles, although these people could have underlying health problems.
Our study not only aims to evaluate gait disorders through gait analysis but also to utilize baseline data in the development of digital therapeutics. Despite the rapid growth of the digital therapeutic market, gait remains a highly attractive area of focus [26,27]. However, as previously mentioned, numerous factors influence walking, and there are insufficient standards for gait analysis data, particularly for CKD subjects, that can differentiate between normal and pathological gaits. Consequently, there is a growing demand and investment for large-scale clinical data and AI learning data by group. We also plan to conduct a subgroup analysis matched for age and sex.
Thus, in this study, we conducted health interventions on subjects using patient medical data. The aim was to establish AI learning data capable of distinguishing between normal and abnormal walking patterns. Throughout this process, we also collected data on the subjects’ daily activities.
We also aim to assess the efficacy of health interventions utilizing AI models, drawing on historical data. Over the course of follow-up, our goal is to analyze both normal and pathological gait in Koreans, as well as falls, hospitalization rates, and mortality following the display of pathological gait. Ultimately, we intend to use the gait itself, along with related elements, as digital biomarkers. We will then apply this data to the development of digital therapeutics, which can predict high-risk fall groups, prevent falls, and aid in the rehabilitation of CKD patients (Fig. 6).
In conclusion, gait serves as a significant biomarker in the progression of CKD and can be utilized as an assessment variable after health interventions. Through this study, we aim to differentiate between pathologic and normal gait, and ascertain whether the stage of CKD can be determined by gait. Additionally, we will explore the potential of gait as a diagnostic and therapeutic biomarker for digital therapeutics. This will be achieved by examining the prevalence of hard outcomes based on patients’ gait and observing any improvements in their gait following our health intervention.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
Print
