Evaluation of physical health status beyond daily step count using a wearable activity sensor

Wearable activity sensors, such as Fitbits, offer continuous remote monitoring of physical activity, but their use has largely been limited to daily step count. While increasing daily step count is associated with improved health outcomes, minute-to-minute data from these devices contain significantly more information. This study aimed to demonstrate that clinically relevant metrics beyond daily step count can be derived from wearable activity monitors. The researchers hypothesized that analyzing detailed minute-to-minute heart rate and step rate data would reveal additional insights into physical health beyond simple step counts. This is particularly important for patients with chronic conditions like PAH, where continuous monitoring can be crucial for effective management and treatment optimization. Existing methods for assessing physical health, such as questionnaires and timed walk tests, often have limitations in terms of information content and frequency of measurement. Wearable sensors offer a potential solution for overcoming these limitations by providing continuous, remote assessment of relevant parameters. The study focuses on PAH patients as they are a population where continuous monitoring of activity and cardiovascular function is particularly important for disease management.

The literature extensively supports the correlation between physical activity and health outcomes. Studies have shown that increasing daily step count is associated with decreased all-cause mortality. In hospitalized patients, low daily step counts are linked to poor outcomes such as readmissions. Other ambulation parameters, such as gait speed and timed walk tests, are also predictive of clinically relevant outcomes. While daily step count is a common metric from wearable sensors, the availability of minute-to-minute heart rate and step rate data opens up new possibilities for more comprehensive assessments. Existing research suggests that wearable devices like Fitbits can provide accurate and clinically relevant data in various patient populations, although concerns regarding accuracy in free-living settings and with atypical gait patterns remain. This study builds upon this existing literature by exploring the potential of a broader range of metrics derived from wearable sensor data to improve the assessment and management of PAH.

This prospective, observational study involved 22 PAH patients who wore Fitbit Charge HR devices between two routine clinic visits (average of 18.4 ± 12.2 weeks). At each visit, a maximum of 26 measurements were recorded (19 categorical and 7 continuous), including assessments of HRQOL, WHO Functional Class, symptoms, organ function, heart rate, 6MWT, RVSP, and blood biomarkers. Minute-to-minute step rate (SPM) and heart rate (BPM) data were downloaded from the Fitbit API. The researchers derived a variety of metrics from the data, including: 1. **Heart rate and step rate distributions:** Mean, standard deviation, and skewness were calculated for weekly step rate (excluding 0 SPM) and heart rate (separated into SR=0 and SR>0). 2. **Ambulation metrics:** Ambulations (≥2 min with SR ≥ 60 SPM) were identified to determine frequency, endurance (1/e of exponential fit to duration histogram), and intensity (standard deviation of step rate distribution). An ambulation product (frequency x endurance x intensity) was calculated. 3. **Fitness:** A metric analogous to the PWC170 test was derived from the slope of a heart rate vs. step rate plot. 4. **Device usage:** Usage was defined as the fraction of minutes with physiological heart rate (20 BPM ≤ HR ≤ age-predicted HRmax). Heat maps and maximum off-time were also calculated. 5. **Free-living 6MWD (FL6MWD):** The 6-minute window with the highest cumulative step count was identified and converted to distance using subject-specific step length estimations. 6. **Physical Health State (PHS):** PHS was calculated as FL6MWD divided by the predicted 6MWD for a healthy individual of the same age, gender, and BMI. Statistical analyses included thresholding, correlation analysis, Principal Component Analysis (PCA), and Latent Profile Analysis (LPA). Subgroups were defined based on thresholds and LPA results, and clinical parameters were compared between subgroups using the Mann-Whitney test.

The study revealed several key findings: 1. **Correlation with Clinical Parameters:** Several Fitbit-derived metrics demonstrated strong correlations with continuous clinical parameters. For example, albumin levels correlated with HR(SR=0) and HR(SR>0), while NT-proBNP correlated with HR(SR=0) and was inversely correlated with the fitness slope. Resting heart rate (RHR) at visits 1 and 2 correlated with HR(SR=0), HR(SR=0) skewness, and HR(SR>0). 6MWD correlated with FL6MWD, and RVSP at visit 1 inversely correlated with fitness slope. 2. **Thresholding and Subgroup Analysis:** Thresholding based on various Fitbit metrics resulted in statistically significant differences in clinical parameters between subgroups. For instance, subjects with an ambulation product P<1000 had lower 6MWD and experienced more pedal edema. Subjects with a fitness slope >0.15 had lower NT-proBNP levels. Subjects with average weekly usage <0.94 had more severe PAH, worse pulmonary health, and more difficulty breathing. Subjects with average FL6MWD <320m had lower 6MWD at both visits, more pedal edema, worse pulmonary health, and lower hemoglobin. 3. **PCA and LPA:** PCA identified subgroups based on physical activity (step rate parameters) and cardiovascular function (resting heart rate and heart rate variability). LPA identified three distinct groups of subjects with characteristic patterns of Fitbit metrics, consistent with PCA findings. 4. **FL6MWD:** The FL6MWD showed good agreement with clinic 6MWD in some subjects but was lower in others ('underperformers'), suggesting that daily activity in these individuals was below their capacity. Comparison of FL6MWD resulted in 6 statistically significant clinical parameters. Subjects with average FL6MWD < 320 m had lower 6MWD at visits 1 and 2, experienced more pedal edema at visit 2, had worse pulmonary health, and had lower hemoglobin at visit 2. 5. **PHS:** The physical health state (PHS) showed a wide range of values, indicating substantial variability in physical capacity among the subjects. The change in PHS during the trial varied, with some subjects showing improvement and others showing decline. 6. **Device Usage:** The study also highlighted diverse device usage patterns, with some subjects consistently wearing the device and others removing it overnight or for charging. Device usage was also correlated with clinical parameters.

This study demonstrates the potential of using minute-to-minute data from wearable activity sensors to assess physical health status beyond the traditional daily step count. The findings support the hypothesis that a richer array of data can be extracted from the raw data to provide more comprehensive insights into physical and cardiovascular function. The strong correlations between Fitbit-derived metrics and clinical parameters, along with the statistically significant differences between subgroups identified by thresholding and cluster analysis, highlight the clinical relevance of this approach. The diverse metrics used in this study capture different aspects of an individual's health, including physical activity, cardiovascular function, and overall health state. The FL6MWD provides a novel approach for assessing physical capacity in a free-living setting. The identification of subgroups and the associations between Fitbit metrics and clinical parameters provide valuable information for identifying individuals at higher risk of adverse outcomes or those who might benefit from more frequent monitoring or specific interventions.

This study successfully demonstrates that wearable activity monitors can provide valuable insights into physical health status beyond the commonly used daily step count. The various metrics derived from minute-to-minute data offer a more comprehensive assessment of physical activity and cardiovascular function. The strong correlations with clinical parameters and the statistically significant differences between subgroups identified in this study suggest that this approach holds significant potential for clinical applications. Future research should focus on validating these findings in larger, more diverse cohorts and exploring the use of these metrics in guiding treatment decisions and personalized care. Further investigation into the use of AI for activity recognition and the incorporation of additional clinical factors would also enhance the value of this approach.

This study has several limitations. First, the relatively small sample size (22 patients) limits the generalizability of the findings. Second, the lack of independent validation for heart rate and step rate measurements could introduce some uncertainty. Third, the study did not adjust for covariates such as medication use, which could have influenced the correlations between Fitbit metrics and clinical parameters. Fourth, the threshold values used for subgroup comparisons were either arbitrary or based on previous studies, and might require refinement in future research. Finally, the study was conducted at a single center, so the results might not be directly generalizable to other populations or settings.