Real-World Postural Transitions as Biomarkers of Functional Impairment in Duchenne Muscular Dystrophy Open Access

Duchenne muscular dystrophy (DMD) is a severe, progressive muscle disorder caused by mutations in the dystrophin-encoding DMD gene. DMD is a rare condition with an average global incidence of approximately 1:5,000 live males [1]. Common presentations of the disease include delayed motor development, muscle weakness, and abnormal gait patterns. Typical symptom onset occurs in early childhood around 3–5 years of age, with progressive muscle weakening over time. Evolution of the disease leads to muscle atrophy, loss of ambulation, and death due to cardiac/respiratory complications often occurring between 30 and 40 years of age [2]. Early diagnosis and intervention are likely to improve quality of life [3].

Recent experimental therapeutic approaches have aimed to restore the missing dystrophin protein [4]. However, clinical trials for these therapies are complicated by the need for sample sizes large enough to overcome heterogeneity in disease progression which are difficult to obtain with the rare nature of the disease. Additionally, current clinical trial designs that necessitate travel for onsite assessments can result in additional patient burden, impact recruitment and retention, and increase costs. Together, these factors disincentivize investment in research and therapeutic development of the field. Further, most standard clinical assessments are limited by their effort-based and subjective nature as well as their inability to capture the totality of a patients’ symptomatic experience across an extended timeframe [5].

The ability to remotely assess movement quality alleviates the assessment burden associated with current clinical trial practice and can offer potential objective endpoints. Wearable devices have recently gained traction for their potential to identify objective digital biomarkers of DMD and present a novel assessment tool to capture DMD-related real-world functional impairment. With a primary disease presentation of progressive muscle weakness and symptoms involving abnormal gait and cardiac complications, wearable sensors can assess features inherently related to the disorder. Gait, cardiac function, activity, and postural transition-related measures can all be captured remotely via wearable sensors [6‒8] and could serve as potential markers of DMD-related functional impairment.

With cardiac complications including sinus tachycardia being common in the DMD population, heart rate has been previously examined as both a signal of cardiomyopathy in DMD and a potential biomarker of the disease [9]. Participants with DMD exhibited higher resting heart rates than control subjects, and higher heart rates were associated with progression to cardiomyopathy. These findings collectively underscore the significant likelihood of elevated heart rate among DMD participants, which may be attributed to the disease’s propensity to lead to cardiomyopathy.

There has been limited work exploring ambulation in DMD using sensors, with 15 papers published from 2000 to 2022 [6]. Much of the previous work has focused on examining physical activity levels and gait-related metrics. Free-living physical activity in DMD individuals (categorized to include sitting, walking, standing, and lying) has been quantified and described using a single, chest-worn sensor [5, 10]. Research of gait-related metrics in DMD patients has primarily involved temporal gait features, specifically stride velocity obtained through accelerometer and gyroscope sensors [11]. The 95th centile of stride velocity (SV95C) has been shown in both free-living and laboratory settings to be a salient indicator of DMD disease progression, and was found to be the most sensitive indicator of functional impairments in DMD [12]. In 2019, SV95C became the first wearable-collected digital outcome approved by the European Medicines Agency for use as a secondary endpoint in DMD clinical trials, and in 2023, it was approved as a primary endpoint [12, 13]. Despite the early successes of gait-related measures of functional impairment, there is a relative dearth of exploration of additional approaches for characterizing real-world impairment. For example, there are a variety of gait-related measures (e.g., acceleration, variability) and alternative measures of functional impairment, such as postural transition quality, that have yet to be explored in DMD. Additionally, little is known about the relationships between different types of functional impairment markers which may be particularly informative for tailoring therapies.

In this study, we aim to build upon the aforementioned work using remotely collected data from wearable devices to characterize cardiac function, activity, gait, and postural transition-related biomarkers of DMD in free-living data. We propose new metrics to capture free-living measures of gait acceleration and variability, and to evaluate movement smoothness, balance, and postural control during sit-to-stand and stand-to-sit transition periods. We quantify the relationships among these potential biomarkers of disease state function, their associations with clinical outcome measures, and their ability to identify DMD-related functional impairment to gain a deeper understanding of disease presentation and to build a body of evidence for alternative measures of clinical interventions.

Eleven DMD and six age-matched healthy male participants were enrolled into the study designed to evaluate the feasibility, wearability, and participant satisfaction of novel assessment tools (ClinicalTrials.gov NCT05657938; SB-00022-001). Three participants chose to withdraw from the study early, with one citing skin sensitivity to the adhesives, leaving nine DMD and five healthy participants upon study completion (Table 1). Participants were selected based on specific inclusion and exclusion criteria. The inclusion criteria required that participants are male at birth and between 4 and 12 years of age. DMD participants needed to provide additional proof of a confirmed diagnosis via genetic testing or clinical records. Additionally, all DMD participants had to be on a stable glucocorticoid dose for at least 3 months prior to the study and be ambulatory, defined as the ability to walk down a hallway at home without assistance or support. Exclusion criteria for DMD participants included enrollment in any interventional studies such as gene therapy treatments and medical conditions that could affect their performance in the study, recent major surgeries, allergies to adhesives, or active implants. For the control group, exclusion criteria included musculoskeletal injuries, illnesses that preclude functional testing, recent major surgeries, medical conditions, allergic reactions to adhesives, or implants.

Data collection occurred over 3 weeks. In the first and third week, three BioStamp nPoint sensors (Medidata Systems) were secured directly to the skin of participants daily using an adhesive. The sensors were worn from waking to bedtime and positioned on the distal lateral shank, the anterior thigh over the rectus femoris, and the chest in a lead II configuration by the patient or a caregiver. In the second week, participants wore the chest and thigh sensors during sleep for three nights of their choice. The sensors were set to record triaxial accelerometer (31.25 Hz ± 16G), gyroscope (31.25 Hz ± 500°/s), and surface biopotential (250 Hz) measurements. Participants were instructed to complete four user-reported tasks at the beginning of each wear session: walking, standing, sitting, and lying down. Timing of those activity completions was logged by a caregiver using the BioStamp Link app. In addition to raw accelerometer, gyroscope, and surface biopotential data, the BioStamp system provides FDA-cleared measures of activity levels, in terms of activity counts in five-second epochs, as well as heart rate in beats per minute in five-second epochs [14]. Reported heart rates were normalized by age-adjusted maximum heart rate prior to analysis [15].

Functional ability and disease severity were evaluated using both a clinical assessment gold standard and an innovative emerging measure, the North Star Ambulatory Assessment (NSAA) [16, 17] and the Duchenne Video Assessment (DVA) [18], respectively. Two DVAs were recorded and scored by three raters during the first and third week, when wearing the sensors. The average data across the three raters were used for comparisons. The DVA is a standardized assessment of compensatory movement strategies that capture quality of movement. Tasks are assigned based on participants’ functional abilities, designed to reveal the quality of movement as it relates to ease of movement. The DVA was designed specifically as a remote measure of movement quality to capture functional ability in the context of everyday life.

A single modified NSAA video assessment with a trained physical therapist was conducted as well during either the first or the third week, flexible to their choice. The NSAA is a clinical assessment scale consisting of 17 scored activities and two timed tests designed to measure functional ability in ambulant patients with DMD. The NSAA-like measurement performed in the context of this study was virtually conducted once by physical therapists via video conference. This assessment is referred to as a modified NSAA as the participants were not sent standardized materials to their home to perform the standard NSAA. The utility of the NSAA in this study was to provide a directional evaluation of functional capacity to compare against the results of the DVA, which was also performed in a home environment, and results are not intended to be directly compared to historical NSAA data.

One DMD participant’s data were removed from analysis due to substantively poor protocol adherence. A total of 12 protocol deviations occurred with participants whose data were included in analysis. The most frequent deviation came from unmarked NSAA (9), followed by the DVA being completed outside the measurement window (2), skipped prescribed activities (2), and failure to wear the BioStamp devices (1).

Accelerometer data from the BioStamp devices were analyzed using a processing pipeline developed in prior work for quantifying free-living measures of balance and mobility impairment [7, 8, 19‒23]. At a high level, the pipeline leverages data from chest and thigh accelerometers and a deep learning model to detect periods during daily life when participants are walking, sitting, standing, and lying down. Consecutive windows of these activities are used to identify bouts of those activities as well as postural transitions (e.g., sit-to-stand, stand-to-sit). Validated parameters that describe how walking, standing, and postural transitions are completed are then extracted for further analysis [24, 25]. The deep learning activity recognition model used in this pipeline was trained on more than 100,000 unique four-second observations from a variety of patient populations both with and without balance and mobility impairment and achieves an accuracy of 96.7% on held-out data [23, 26]. Herein, activity classifications were confirmed for each participant based on caregiver-labeled and investigator-confirmed bouts of walking, standing, sitting, and lying down. As described in prior work [7, 8, 19‒23], the pipeline reports a variety of mobility parameters that capture how participants engage in their daily physical activities. Based on prior work in DMD, and the expected impact of disease-relevant muscle degeneration, we focus primarily on measures of gait and postural transition performance as they are expected to require muscular contributions that may help expose disease-relevant deficits. We also examine heart rate and overall activity levels as candidate biomarkers but anticipate that they may not prove as sensitive as their activity performance-related counterparts.

For gait, we considered measures of gait acceleration (root-mean-square of chest acceleration in the anterior-posterior direction, lower values associated with impairment [27]), variability (frequency dispersion of chest acceleration in the medial-lateral direction, higher values associated with impairment [27]), and speed (stride velocity, lower values associated with impairment [11]). The proposed measures of gait acceleration and variability can be computed directly from raw chest accelerometer data during detected walking bouts. However, estimates of stride velocity require additional computation to resolve. We estimated stride velocity following established approaches [4, 28], which leverage accelerometer and gyroscope data from the lateral shank. During each detected gait bout, stance phases were determined from foot contact and foot-off events identified from the medial-lateral angular rate signal [29]. Shank orientation was first computed at a mid-stance instant during the most static quarter of the initial stance phase (at the beginning of the stride) using only accelerometer data [30]. A forward estimate of instantaneous shank orientation after this instant and up to the same mid-stance instant of the subsequent stance phase (at the end of the stride) was determined via time integration of the quaternion kinematic equations [30]. We made an additional estimate of shank orientation during the subsequent stance phase using the same methods as for the initial stance phase. This estimate was fused with the forward estimate using a Rauch-Tung-Striebel Kalman smoother [31] and used to express the accelerometer signal in the global frame after which gravitational acceleration was subtracted. The shank translational velocity was obtained by time integration of the global shank acceleration. We used the fact that the shank velocity is approximately zero during the initial and subsequent stance phases (at the mid-stance instants described earlier) to correct velocity estimates (i.e., a zero-velocity update) using a Rauch-Tung-Striebel Kalman smoother. Stride length was computed as the net translational displacement magnitude of the shank following time integration of the corrected velocity estimate. Stride velocity was computed as the stride length divided by the stride time. Outliers were removed as they likely were characterizing non-gait events. Our activity classification approach considers temporal windows of data which, even if classified as walking, may also contain behaviors other than walking that occur during the window (e.g., short period of standing after the conclusion of walking bout). The presence of these behaviors can lead to the detection of spurious gait events and strides. Our gait event detection and stride segmentation algorithms include gait parameter thresholds (e.g., minimum stride times) to remove these strides from analysis. An average of 29% of detected strides were removed per participant (∼2,965 strides per participant).

For postural transitions, we considered measures of transition smoothness (transverse-plane jerk of chest, calculated as the time derivative of the accelerometer signal, during the stand-to-sit transitions: sitting jerk and sit-to-stand transitions: standing jerk), reactive balance (transverse-plane jerk of chest during the standing period following sit-to-stand transition: reactive jerk), and anticipatory postural control (transverse-plane jerk of chest during the standing period preceding stand-to-sit transition: anticipatory jerk). Like gait, postural transitions represent a somewhat constrained and repeated daily activity that is key for independent function and impacted directly by muscle degeneration characteristic of DMD [17]. While not previously explored in DMD, we hypothesized that these specific metrics relating to the smoothness of the transitions as well as an individual’s ability to prepare for, and recover from, postural transitions will be sensitive to the presence of DMD as well as the degree of functional impairment.

Measures of overall activity level and heart rate are reported directly by the FDA-cleared BioStamp system [14]. For overall activity level, we considered the combined activity count, which is the sum of the activity counts reported by the chest and thigh devices and aggregated as the mean across the recording period for each participant [32]. Heart rate recordings that are reported throughout the wear period were considered within specific activity bouts of walking, standing, sitting, and lying down.

Consistent with prior work [6], we aggregated observations across the entire study period using 95th percentile rather than at daily level; therefore, no minimum daily aggregation thresholds were set a priori. The minimum amount of post-processed daily sensor data for any participant was 3.2 h. For the aggregation of biomarker values, the minimum number of total days of sensor wear was 15 days. Since not all days included measurements for each biomarker, the minimum number of days included in the aggregation for the least frequent biomarkers (gait measures) was 3 days. Distributions of each measure are visualized (online suppl. Fig. S1; for all online suppl. material, see https://doi.org/10.1159/000545617) and tested for normality (online suppl. Table S1) in the online supplementary material. This approach to aggregation over the entire study period is intended to focus our analysis on a measure of functional capacity or worst-case examples within these tasks that help cut through some of the inherent noise in free-living measurement of these parameters. The final, sensor-derived measures considered in this analysis and their relevance to functional capacity or worst case are reported in Table 2.

Associations of candidate biomarkers with one another and with the established outcome measures were examined via Spearman’s correlation coefficients. These associations were used to down-select parameters for further comparison and to identify relationships between these candidate measures of functional impairment and accepted clinical measures (NSAA, DVA). ANOVA tests were used to evaluate differences between DMD and control groups in these potential biomarkers, controlling for age where appropriate. Level of significance was set a priori at ɑ = 0.05. Effect sizes of significant differences were estimated with Cohen’s d, which can be considered as a measure of the sensitivity of a given biomarker to DMD-related functional impairment.

Spearman correlations of candidate biomarkers with one another and the outcome measures are displayed in the heat map of Figure 1. Significant correlations are indicated with (*) sign. Many of the activity, cardiac, gait, and transition measures were moderately to very strongly associated (|r| = 0.58–0.83, p = 0.049 to p < 0.001) with at least one of the clinical markers of functional impairment.

Reactive jerk demonstrates the strongest association with DVA (r = −0.82) and NSAA (r = 0.83), followed by stride velocity (r = −0.75) and (r = 0.82) with DVA and NSAA, respectively. There is also some overlap in measures within and between domains.

Examining candidate measures from these domains more deeply, the DMD and control groups exhibited similar proportions of time spent lying, sitting, standing, and walking throughout the monitoring period (Fig. 2a). However, activity intensity was significantly lower in participants with DMD than controls when controlling for age (p = 0.005, Fig. 2b) in this sample. These results agree with prior work demonstrating low activity intensity in participants with DMD [33].

As shown previously with clinical measurements in ambulatory DMD patients with mild to moderate cardiac involvement [34], we observed higher heart rates in participants with DMD compared to controls. This difference, detected from ECG captured in the free-living environment for what is potentially the first time, was particularly pronounced during detected standing bouts (p = 0.03), as shown in Figure 3).

Potential gait biomarkers exhibited several significant differences between participants with DMD and controls. Specifically, participants with DMD were shown to have reduced gait speed (stride velocity) (p = 0.02, Fig. 4a) and acceleration capacity (root-mean-square of anterior-posterior acceleration [RMS AP]) (p = 0.02, Fig. 4b) and increased gait variability (mediolateral frequency dispersion) (p = 0.04, Fig. 4c) relative to controls. These results align with prior work demonstrating reduced gait speed and particularly SV95C, in those with DMD [11]. We extend upon results in prior work herein to consider additional measures of walking capacity as related to gait acceleration and gait variability as potential sensitive and objective biomarkers of disease in DMD.

We have identified free-living measures of postural transition performance as potential biomarkers of functional impairment in those with DMD. We observed that worst-case examples of sitting and standing transition smoothness, reactive jerk following standing, and anticipatory postural control preceding sitting are all significantly lower in those with DMD than controls (Fig. 5, p = 0.03 to p < 0.005). Interestingly, smoother transitions and more limited postural sway immediately following or preceding the transition were associated with slower transitions suggesting that those with DMD may be choosing to transition in a more controlled and slower fashion. This could reflect the relative effort required to complete these challenging daily activities for those with DMD-associated functional impairment and the related care they elect to take in their daily movements.

Looking across candidate biomarker domains, measures from each domain were able to differentiate participants with DMD from controls with strong to very strong effects (Fig. 6, Cohen’s d = 0.96–2.80). The radar plot of Figure 7 highlights the phenotypic profiles of the DMD and control groups, further illustrating these differences. Notably, while each measure independently demonstrates an aspect of functional impairment in DMD, it could be that considering these data together in a wearable-derived composite marker could be more sensitive to individual impairments and better inform treatment planning by identifying and triaging domains to be targeted.

The purpose of this study was twofold: to identify candidate biomarkers using real-world data and to characterize the relationships between these parameters and standard clinical assessments. We have found nine total candidate biomarkers that differentiate DMD-related functional impairment, all exhibiting large to very large effect sizes. Each candidate biomarker was moderately or strongly associated with standard clinical measures of function in DMD. Of the nine candidate biomarkers, we present six as novel and understudied in this population: chest RMS AP, chest ML FD, sitting jerk, standing jerk, anticipatory jerk, and reactive jerk.

Large effect sizes between control and DMD groups were found in the postural transitions, notably in movement smoothness. Both movement smoothness metrics exhibited similar effect sizes to those seen in the standard, effort-based, and subjective clinical measures of function in DMD (NSAA, DVA). These findings highlight the potential that digital assessment of free-living postural transition features offers as novel objective biomarkers of the disease, and in capturing similar information as current clinical endpoints using remote, continuous assessments.

While gait-related metrics have been relatively well examined in DMD populations, postural transition metrics have been largely unstudied. Postural transitions may be more taxing on full-body musculature than gait and thus could serve as a more sensitive biomarker by exposing DMD-related deficits earlier or more consistently in the heterogenous course of disease. Interestingly, DMD participants exhibited significantly smoother transitions and less postural sway directly preceding and following postural transitions. These smoother transitions and increased postural control are associated with slower transition times (jerk metrics: standing jerk, reactive jerk, sitting jerk, and anticipatory jerk are negatively correlated with transition time with r = −0.32, −0.46, −0.49, and −0.31 respectively). This pattern suggests a preference among individuals with DMD for controlled and deliberate movement during postural transitions. This behavior likely reflects the increased effort required for these activities due to DMD-associated functional impairments, and the compensatory care taken.

Many of the digitally captured biomarkers discussed herein showed strong and significant correlations with current clinical endpoints. Walking speed capacity (SV95C) showed some of the strongest associations with the two clinical measures of function in DMD. This aligns with previous research showing temporal gait parameters to be associated with function and supports the recent approval of SV95C as a primary endpoint for DMD clinical trials by European regulatory bodies. This approval is encouraging for the potential of other digital biomarkers to be considered as trial endpoints, particularly given that postural transition metrics measured in this study exhibited similar associations with the clinical assessments when compared to SV95C. Sit-to-stand postural transition metrics (standing jerk, reactive jerk) had the highest correlation with the NSAA. Interestingly, while walking speed capacity had high correlation with the DVA, reactive jerk within the sit-to-stand postural transition domain had slightly higher correlation with both NSAA and DVA followed by other metrics from this domain.

Building on the strength of the SV95C, we introduce 95th centile RMS AP as a novel gait biomarker for DMD that may provide additional sensitivity in measuring peak ambulatory function. Walking speed capacity has been studied previously in DMD [11, 12] and is a complex estimation requiring a sensor to be worn on the ankle. Importantly, computation of the 95th centile RMS AP only requires a chest-patch sensor. Intuitively, stride velocity and acceleration are linked, and this intuition is supported in the parameter associations. The 95th centile RMS AP and SV95C are significantly, albeit moderately, correlated (r = 0.63) and exhibit similar effect sizes. Thus, while stride velocity and RMS AP are related, RMS AP likely captures some independent information.

Clinical measures of functional impairment (NSAA, DVA) consider a comprehensive picture of functional domains that could be impacted by DMD. The candidate biomarkers identified herein suggest that a similar composite approach may be beneficial when considering wearable-derived markers of function. Figure 1 highlights that markers from some domains (e.g., transition and activity) have less overlap with each other than others (e.g., gait and transition) despite being related to composite clinical measures and sensitive to the presence of DMD (Fig. 7). These markers may each provide a unique picture of the functional impairment experienced by those living with DMD and could form phenotypes that help capture the heterogenous presentation of DMD symptoms. Taken together, there is likely utility in considering a wearable-derived composite marker of functional impairment in DMD that mirrors clinical assessments and is composed of these objective markers within each subdomain. Future work should consider these markers independently and together and explore how they evolve over time with the progression of DMD symptoms and could be used to inform tailored therapies.

These findings suggest new potential digital biomarkers and offer valuable insight into the presentation of real-world DMD-related functional impairment. The biomarkers proposed in this work are effective in differentiating healthy controls from DMD and are strongly correlated with clinical measures of motor function (NSAA and DVA). These results provide persuasive preliminary evidence for the candidate biomarkers and suggest a clear path for future research. Further studies in larger samples with more severe mobility impairments (e.g., [5]) are needed to fully assess the potential of these candidate biomarkers in DMD and their sensitivity to different DMD severities, within-individual changes over time, and between-group differences in response to treatment. Additional examination of the minimum amount of data needed to capture reliable estimates of these candidate biomarkers and exploration of other potential barriers and facilitators to their implementation in research and clinical practice should be addressed in future work.

There are several limitations that are important to note when considering this work. First, these data are from a small sample of patients living with DMD and controls that are captured over a relatively short measurement period. Findings need to be replicated and further explored in a larger cohort tracked longitudinally to confirm generalizability and the ability of the candidate biomarkers to capture clinically important changes. This is particularly relevant given the heterogeneity in DMD symptom progression, treatment regimens, and associated responses. Similarly, it is possible that three nights of sleep was not sufficient to establish habitual sleep patterns for participants. Longer periods of sleep monitoring should be considered in future studies. There may also be unexplored opportunities for considering additional data sources (e.g., surface electromyography) that could be considered in future work.

This study protocol (SB-000-22-001) was reviewed and approved by Advarra IRB (initial approval September 30, 2022). Written informed consent was obtained from the parent/legal guardian/next of kin to participate in the study, and child participant assent was obtained where appropriate.

R.S.M. is an inventor on patents and patent applications related to digital health technologies; has advisory relationships with Epicore Biosystems, Pfizer, and Solid Biosciences; is co-founder of Panic Mechanic and Biobe; reports research funding from NSF, NIH, Solid Biosciences, Medidata Systems, Kern Family Foundation, and MassMutual; has affiliation with Stanford University and the University of Hawaii at Manoa; and is associate editor at PLOS Digital Health and npj Digital Medicine. E.W.M. is an inventor on patents and patent applications related to digital health technologies; is co-founder of Panic Mechanic and Biobe; and reports research funding from NSF, NIH, and MassMutual. D.M.R. is an inventor on patent applications for downloadable software using artificial intelligence (AI) for healthcare communication quality measurement, research, and training; serves as a member of Vermont Senate Committee’s AI Task Force charged with making recommendations on the responsible growth of Vermont’s AI technology markets, and possible regulation of AI in state government; and reports research funding from NSF, NIH, USGS, NASA, DoD, and USDA primarily in the areas of environmental sensor monitoring and analysis. R.D.G. is an inventor on patents and patent applications related to digital health technologies and reports research funding from the Wu Tsai Human Performance Alliance at Stanford University and the Joe and Clara Tsai Foundation. B.M.M., P.D., and J.F.B. are employees of Medidata Solutions and A Dassault Systèmes Co., and have stock ownership/options as a result of this employment. J.L.M. is an employee of Solid Biosciences and has stock ownership/options as a result of this employment. C.M. is an employee of/consultant for Solid Biosciences. J.M. is a prior employee of Solid Biosciences and a current employee of Chroma Medicine. D.D.K. and C.J.G. declare no conflicts of interest.

This research was funded by Solid Biosciences. The funders contributed to the design, data collection, and reporting of this study as well as manuscript planning, writing, and decision to publish. The funders did not contribute to the data analysis.

C.J.G., D.D.K., B.M.M., P.D., J.F.B., and D.M.R. contributed to analysis and interpretation of data for the work; drafting the manuscript and reviewing it critically for important intellectual content; and final approval of the manuscript. J.L.M., C.M., and J.M. contributed to the conception and design of the work; acquisition and interpretation of data for the work; drafting the manuscript and reviewing it critically for important intellectual content; and final approval of the manuscript. E.W.M., R.D.G., and R.S.M. contributed to the conception and design of the work; acquisition, analysis, and interpretation of data for the work; drafting the manuscript and reviewing it critically for important intellectual content; and final approval of the manuscript.

Cailin J. Gramling and Dheeraj Dhanvee Kairamkonda contributed equally to this work.

The data that support the findings of this study are not publicly available due to their containing information that could compromise the privacy of research participants but are available from the corresponding author R.S.M. upon reasonable request.