Dylan Richards’s scientific contributions

Introduction Motion-based wearable sensors, typically on wrist, have long been used for free-living sleep detection and quantification. However, it is hard to differentiate sleep from sedentary awake time by immobility alone. Vital signs, like heart rate and respiration rate, can greatly enhance determination of wake-sleep state, and are easily monitored with newer wearable sensors. Deep learning techniques are particularly adept at learning labeled physiological states. By combining movement plus vital signs in a deep neural network algorithm, improved sleep detection, fragmentation and sleep staging should be possible compared to activity alone. We report on performance of a deep learning sleep detection and REM/NREM algorithm providing 24-hour evaluation with high specificity using data from a torso-wearable patch sensor as compared to polysomnography (PSG). Methods Twenty-six healthy adults (mean age 53.7 years, 81% female) contributed 150 nights of PSG during laboratory visits, during which participants simultaneously wore a multi-day skin-adherent patch with continuous single-lead ECG and 3-axis accelerometer streams, as well as a wrist activity monitor. A pre-trained deep neural network algorithm generated epoch-level Wake/REM/NREM classification (Sleep equals REM plus NREM) using vital signs and movement derived from the patch sensor ECG and accelerometer waveforms and was then compared to expert human staging of PSGs. The wrist actigraphy sleep-wake determinations (Actiware) were also compared to PSG. Results Data includes 900 hours sleeping and 139 hours awake, of which 195 hours of sleep were in REM state. Using patch data, the deep neural net algorithm achieved 92% sensitivity and 85% specificity to detect sleep as compared to PSG; REM was detected with 85% sensitivity and 97% specificity. By comparison, the wrist motion-based algorithm only exhibited 33% specificity and 95% sensitivity, essentially overcalling immobile wake as sleep. Conclusion Sleep evaluation in free-living environments with wearable sensors can be greatly improved over conventional motion-based wrist sensors by leveraging continuous vital signs. Deep learning-trained neural network algorithms are particularly effective for use with such data, as demonstrated with this algorithm. Support (if any) R01 HL140580 and P01 AG011412

Introduction: The objective determination of a person’s cardiorespiratory fitness over time, especially in those with heart failure, can play a critical role in optimizing individual therapies, identifying early signs of decompensation, and accelerate the development of novel therapeutic interventions. However, existing methods, such as formal cardiopulmonary exercise testing (CPET), or surrogates like a 6-minute walk test (6MWT) are difficult for both patients and health systems and provide only sparse samples. Hypothesis: Using a wearable ECG-patch sensor and machine learning methods, cardiorespiratory fitness can be estimated daily using routine vital sign dynamics during free-living activities. Methods: CPETs and multiple days of ECG-patch sensor data were collected within 2 weeks of each other for 228 participants (146 normal, and 82 with Heart Failure). A model for VO 2 max estimation was developed using a two-stage procedure. (Figure) In the first stage, the model was pretrained on an unsupervised regression task on over 2000 unique patients’ data, representing over 200 patient-years of activity of daily living. The second stage fine-tuned the pretrained model to predict VO 2 max, using a K-Fold cross validation procedure with 50% train, 25% validation, and 25% testing for four folds. Results: CPET-determined VO 2 max values ranged from 7.1 to 69.1 ml/kg/min (median=28.9, IQR=24.5). The Pearson Correlation Coefficient (PCC) and Mean Absolute Error (MAE) of the estimated VO 2 max was 0.85 and 6.2 ml/kg/min, respectively. For the heart failure subgroup, the PCC and MAE are 0.73 and 3.7 ml/kg/min, respectively. (Figure) Conclusion: Our results show that estimated VO 2 max can be accurately and conveniently determined at high frequency in individuals with a wide range of cardiorespiratory fitness using data from a patch sensor during routine daily activities. Ongoing work in a large HF trial will evaluate its performance over time relative to 6MWT and surveys.

The COVID-19 pandemic has accelerated the adoption of innovative healthcare methods, including remote patient monitoring. In the setting of limited healthcare resources, outpatient management of individuals newly diagnosed with COVID-19 was commonly implemented, some taking advantage of various personal health technologies, but only rarely using a multi-parameter chest-patch for continuous monitoring. Here we describe the development and validation of a COVID-19 decompensation index (CDI) model based on chest patch-derived continuous sensor data to predict COVID-19 hospitalizations in outpatient-managed COVID-19 positive individuals, achieving an overall AUC of the ROC Curve of 0.84 on 308 event negative participants, and 22 event positive participants, out of an overall study cohort of 400 participants. We retrospectively compare the performance of CDI to standard of care modalities, finding that the machine learning model outperforms the standard of care modalities in terms of both numbers of events identified and with a lower false alarm rate. While only a pilot phase study, the CDI represents a promising application of machine learning within a continuous remote patient monitoring system.

Background: Implantable cardiac sensors have shown promise in reducing rehospitalization for heart failure (HF), but the efficacy of noninvasive approaches has not been determined. The objective of this study was to determine the accuracy of noninvasive remote monitoring in predicting HF rehospitalization. Methods: The LINK-HF study (Multisensor Non-invasive Remote Monitoring for Prediction of Heart Failure Exacerbation) examined the performance of a personalized analytical platform using continuous data streams to predict rehospitalization after HF admission. Study subjects were monitored for up to 3 months using a disposable multisensor patch placed on the chest that recorded physiological data. Data were uploaded continuously via smartphone to a cloud analytics platform. Machine learning was used to design a prognostic algorithm to detect HF exacerbation. Clinical events were formally adjudicated. Results: One hundred subjects aged 68.4±10.2 years (98% male) were enrolled. After discharge, the analytical platform derived a personalized baseline model of expected physiological values. Differences between baseline model estimated vital signs and actual monitored values were used to trigger a clinical alert. There were 35 unplanned nontrauma hospitalization events, including 24 worsening HF events. The platform was able to detect precursors of hospitalization for HF exacerbation with 76% to 88% sensitivity and 85% specificity. Median time between initial alert and readmission was 6.5 (4.2-13.7) days. Conclusions: Multivariate physiological telemetry from a wearable sensor can provide accurate early detection of impending rehospitalization with a predictive accuracy comparable to implanted devices. The clinical efficacy and generalizability of this low-cost noninvasive approach to rehospitalization mitigation should be further tested. Registration: URL: https://www.clinicaltrials.gov. Unique Identifier: NCT03037710.