Wearable sensors can alert you when you are getting sick, Stanford study shows

January 18, 2017

Current versions of three of the devices used for heart-rate and peripheral capillary oxygen saturation measurements in the study (credits left to right: Scanadu, iHealth, and Masimo)

Fitness monitors and other wearable biosensors can tell when your heart rate, activity, skin temperature, and other measures are abnormal, suggesting possible illness, including the onset of infection, inflammation, and even insulin resistance, according to a study by researchers at the Stanford University School of Medicine.

The team collected nearly 2 billion measurements from 60 people, including continuous data from each participant’s wearable biosensor devices* and periodic data from laboratory tests of their blood chemistry, gene expression, and other measures, and related this data to a range of normal (baseline) values for each person in the study compared to when they were ill.**

Participants wore between one and seven commercially available activity monitors and other monitors that collected more than 250,000 measurements a day. The team collected data on weight; heart rate; oxygen in the blood; skin temperature; activity, including sleep, steps, walking, biking and running; calories expended; acceleration; and even exposure to gamma rays and X-rays.

There were three participant groups: Participant #1 wore seven portable devices for large segments of this study; 43 individuals wore Intel’s Basis to measure activity (steps), heart rate, sleep, and skin temperature, with data securely uploaded to the cloud; and 16 individuals wore either iHealth or Masimo finger devices for sensing heart rate and SpO2 (peripheral capillary oxygen saturation).

Identifying health problems in advance

Wearable devices used by participant 1 (credit: Xiao Li/PLOS Biology)

The study, led by Michael Snyder, PhD, professor and chair of genetics, and senior author of the study, was published online Jan. 12 in open-access PLOS Biology. It demonstrated that, given a baseline range of values for each person, it is possible to monitor deviations from normal and associate those deviations with environmental conditions, illness, or other factors that affect health. Distinctive patterns of deviation from normal seem to correlate with particular health problems. Algorithms designed to pick up on these patterns of change could potentially contribute to clinical diagnostics and research.

The results of the current study raise the possibility of identifying inflammatory disease in individuals who may not even know they are getting sick.

For example, in several participants, higher-than-normal readings for heart rate and skin temperature correlated with increased levels of C reactive protein in blood tests. C reactive protein is an immune system marker for inflammation and often indicative of infection, autoimmune diseases, developing cardiovascular disease or even cancer. Snyder’s own data revealed four separate bouts of illness and inflammation, including a Lyme disease infection and another that he was unaware of until he saw his sensor data and an increased level of C reactive protein.

The wearable devices could also help distinguish participants with insulin resistance, a precursor for Type 2 diabetes. Of 20 participants who received glucose tests, 12 were insulin-resistant. The team designed and tested an algorithm combining participants’ daily steps, daytime heart rate and the difference between daytime and nighttime heart rate. The algorithm was able to process the data from just these few simple measures to predict which individuals in the study were likely to be insulin-resistant.

The study also revealed that declines in blood-oxygen levels during airplane flights were correlated with fatigue. Fortunately, the study showed that people tend to adapt on long flights; oxygen levels in their blood go back up, and they generally feel less fatigued as the hours go by.

The future of wearable devices: monitoring human health continuously

During a visit to the doctor, patients normally have their blood pressure and body temperature measured, but such data is typically collected only every year or two and often ignored unless the results are outside of normal range for entire populations. But biomedical researchers envisage a future in which human health is monitored continuously.

“We have more sensors on our cars than we have on human beings,” said Snyder. In the future, he said, he expects the situation will be reversed and people will have more sensors than cars do. Already, consumers have purchased millions of wearable devices, including more than 50 million smart watches and 20 million other fitness monitors. Most monitors are used to track activity, but they could easily be adjusted to more directly track health measures, Snyder said.

The work is an example of Stanford Medicine’s focus on “precision health,” whose goal is to anticipate and prevent disease in the healthy and to precisely diagnose and treat disease in the ill. With a precision health approach, every person could know his or her normal baseline for dozens of measures. Automatic data analysis could spot patterns of outlier data points and flag the onset of ill health, providing an opportunity for intervention, prevention or cure.

Researcher Elizabeth Colbert, of the Veterans Affairs Palo Alto Health Care System, is also a co-author. This research was funded by the National Institutes of Health, a gift from Bert and Candace Forbes, and Stanford’s Department of Genetics.

* “After evaluating more than 400 available wearable devices at the beginning of the study, we selected [seven] for participants to use. The criteria for selection [were] (1) ability to access the raw data from the manufacturer, (2) cost, (3) overlap in measurement of at least one component with another device to assist in reproducibility, and (4) ease of use, reasonable accuracy, and had a direct interface for raw data. These devices collectively measure (a) three physiological parameters, including heart rate, peripheral capillary oxygen saturation, and skin temperature, (b) six activity-related parameters, including sleep, steps, walking, biking, running, calories, and acceleration forces caused by movement, (c) weight, and (d) total gamma and X-ray radiation exposure.” — PLOS Biology paper authors

** “In this work, we investigate the use of portable devices to (1) easily and accurately record physiological measurements in individuals in real time (or at high frequency), (2) quantify daily patterns and reveal interesting physiological responses to different circadian cycles and environmental conditions, (3) identify personalized baseline norms and differences among individuals, (4) detect differences in health states among individuals (e.g., people with diabetes versus people without diabetes), and (5) detect inflammatory responses and assist in medical diagnosis at the early phase of disease development, thereby potentially impacting medical care.” — PLOS Biology paper authors


Abstract of Digital Health: Tracking Physiomes and Activity Using Wearable Biosensors Reveals Useful Health-Related Information

A new wave of portable biosensors allows frequent measurement of health-related physiology. We investigated the use of these devices to monitor human physiological changes during various activities and their role in managing health and diagnosing and analyzing disease. By recording over 250,000 daily measurements for up to 43 individuals, we found personalized circadian differences in physiological parameters, replicating previous physiological findings. Interestingly, we found striking changes in particular environments, such as airline flights (decreased peripheral capillary oxygen saturation [SpO2] and increased radiation exposure). These events are associated with physiological macro-phenotypes such as fatigue, providing a strong association between reduced pressure/oxygen and fatigue on high-altitude flights. Importantly, we combined biosensor information with frequent medical measurements and made two important observations: First, wearable devices were useful in identification of early signs of Lyme disease and inflammatory responses; we used this information to develop a personalized, activity-based normalization framework to identify abnormal physiological signals from longitudinal data for facile disease detection. Second, wearables distinguish physiological differences between insulin-sensitive and -resistant individuals. Overall, these results indicate that portable biosensors provide useful information for monitoring personal activities and physiology and are likely to play an important role in managing health and enabling affordable health care access to groups traditionally limited by socioeconomic class or remote geography.