Wearable sensor-based devices are increasingly applied in free-living and clinical settings to collect fine-grained, objective data about activity and sleep behavior. The manufacturers of these devices provide proprietary software that labels the sensor data at specified time intervals with activity and sleep information. If the device wearer has a health condition affecting their movement, such as a stroke, these labels and their values can vary greatly from manufacturer to manufacturer. Consequently, generating outcome predictions based on data collected from patients attending inpatient rehabilitation wearing different sensor devices can be challenging, which hampers usefulness of these data for patient care decisions. In this paper, we present a data-driven approach to combining datasets collected from different device manufacturers. With the ability to combine datasets, we merge data from three different device manufacturers to form a larger dataset of time series data collected from 44 patients receiving inpatient therapy services. To gain insights into the recovery process, we use this dataset to build models that predict a patient’s next day physical activity duration and next night sleep duration. Using our data-driven approach and the combined dataset, we obtained a normalized root mean square error prediction of 9.11% for daytime physical activity and 11.18% for nighttime sleep duration. Our sleep result is comparable to the accuracy we achieved using the manufacturer’s sleep labels (12.26%). Our device-independent predictions are suitable for both point-of-care and remote monitoring applications to provide information to clinicians for customizing therapy services and potentially decreasing recovery time.
Wrist-worn sensors have increased in popularity in health care settings. As the use of wrist-worn sensors increases, a better understanding is needed of how to detect changes in behavior as well as an ability to quantify such changes. We introduce a statistical method to address this need. In this study, we used Fitbit Charge Heart Rate devices with two separate populations to continuously record data. There were eight participants in the healthy control group and nine in the hospitalized inpatient rehabilitation group. We performed comparisons both within the groups and between groups on the gathered step count and heart rate data. The inpatient rehabilitation group showed improved step count changes between the first half of the study participation and the second half. Heart rate did not show significant changes for either the healthy control group or inpatient rehabilitation group across time. We conclude that our statistical change analysis applied to wrist-worn sensors can effectively detect changes in physical activity that provides valuable information to patients as well as their healthcare care providers.
1 November 2009 in Archives of Physical Medicine and Rehabilitation
Ethyl glucuronide in urine (uEtG) is a biomarker that objectively detects alcohol consumption. However, what uEtG values represent regarding the last drinking event remains unclear. This study investigated whether the recency or amount consumed at the last event are associated with multiple uEtG cutoffs indicative of light to heavy drinking. Adults ( n = 349) recruited for an alcohol use disorder clinical trial submitted urine samples and self‐reported alcohol measures at baseline and twice weekly during a 4‐week observation period prior to treatment. The ThermoFisher DRI immunoassay was used to quantify uEtG levels (0–2000 ng/mL). Self‐report measures included the number of hours since last drink (Hours), number of standard drinks at that event (Amount), and ratios of these measures (Hours:Amount and Amount:Hours). Three uEtG cutoffs were assessed (100, 500, and 2000 ng/mL) and generalized linear mixed modeling applied to compare self‐report measures below versus above each uEtG cutoff. A total of 1676 uEtG samples with self‐reported data were included. Comparisons across cutoffs were significant for Hours ( p < 0.001), while Amount was only significant at the 2000 ng/mL cutoff ( p = 0.001). Hours also had the strongest association with uEtG results across cutoffs (pseudo‐ R2 range = 0.474–0.530). Higher uEtG results indicated more recent drinking, rather than higher levels of alcohol consumption, and the period of detection of uEtG was shorter than previously reported. Even at the low threshold of 100 ng/mL, uEtG was best conceptualized as a measure of alcohol consumption in the prior 24 h, with higher uEtG levels being associated with fewer hours since last drink.