The Economist May 7th 2022 Technology Quarterly The quantified self 11
Thepulseof the people
A
s soon asthe covid19 pandemic began, several research insti
tutes around the world set up studies asking people to share
data from their wearable fitness trackers. On most devices, signing
up involved just a few clicks, and people did so enthusiastically.
The biggest study, the Corona Data Donation project set up by the
Robert Koch Institute in Germany, enrolled more than 500,000
people. Over 30,000 signed up for detect, a study by the Scripps
Research Institute in California.
When it comes to disease surveillance, the most useful bio
marker is fever, a direct sign of infection. But most wearables do
not measure temperature, because accurate readings are hard to
do. So a proxy had to be created using the standard things they do
measure, such as heart rate, sleep and activity level. Resting heart
rate, measured when people are sitting still, varies a lot from per
son to person—anything between 50 and 100 beats per minute
counts as normal—but each person’s rate is generally stable.
When the body fights an infection, however, the rate goes up, of
ten dramatically. With covid19, data from wearable devices
showed that this uptick happened four days before people felt any
symptoms. By one estimate 63% of covid cases could be detected
from changes in resting heart rate before the onset of symptoms.
Before covid came along, a team from Scripps led by Jennifer
Radin had shown that, in America, weekly changes in the propor
tion of people with abnormal results in heart rate, sleep and activ
ity—all measured from wearables—align neatly with the preva
lence of flulike symptoms as measured by established surveil
lance systems. These track flu outbreaks by canvassing doctors’
offices to find out if more people with such symptoms are starting
to show up. Because people usually seek care 38 days after symp
toms appear, by the time these data are collated, an epidemic is
usually at a different stage, possibly requiring different public
health measures. More timely insights are sorely needed.
That said, data from wearables have quirks of their own. One
day the Koch Institute team saw a sudden peak in the measure
ment derived from step count and heart rate they were developing
as a proxy for fever. It turned out that Apple had changed the algo
rithm that calculates resting heart rate on its devices. Such soft
ware updates have been a headache for the team because their data
come from about a dozen different devices. They also have to sort
out various gaps. Apple Watches are usually charged at night,
which means that they give no sleep data. Once through its teeth
ing problems, though, the project proved a success. “It is not 100%
accurate but it does a pretty good job,” says Dirk Brockmann, who
leads the team.
Other research teams have taken a different approach to popu
lationbased surveillance with wearable devices. They have devel
oped algorithms that examine deviations in each individual’s
metrics, based on whatever data their particular device collects.
They establish the person’s baseline levels of various biomarkers
and then look for changes that suggest he may be experiencing
some sort of anomaly in physiology. When lots of such changes
occur all of a sudden, different as they may be from person to per
son, it is reasonable to suspect that lots of people are falling ill,
and probably from the same thing.
One thing researchers now need to work out is whether the dis
easesurveillance algorithms based on wearable devices might
systematically miss what is happening with some types of people,
says Leo Wolansky from the Rockefeller Foundation’s Pandemic
Prevention Institute. For example, algorithms might unwittingly
be optimised for spotting outbreaks in wealthy areas where people
are more likely to have been using highend wearables for longer.
In poorer areas, where people may have different underlying
health conditions (which often affect digitalbiomarker measure
ments), the algorithm for wearables might be a lot more likely to
miss an outbreak. “As they often say in this field, ‘Garbage in, gar
bage out’, and we still have to better understand whether the data
we’ve captured has some garbage in it,” says Mr Wolansky.
Medical scans that look for a particular problem routinely turn
up other things, known as incidental findings. Something similar
has occurred with the mass scan of human bodies that has taken
place thanks to all these data from wearables. The German team
found that resting heart rate was higher in areas that had been in
East Germany than those in former West Germany. “We still don’t
know why this is,” says Mr Brockmann. “Is it because women work
more in East Germany? Or is it because people eat differently?”
Another mysterious finding is that Germans in all parts of the
country are sleeping less in 2022 than in 2020 and the resting
heart rate of the nation has gone up. One guess is that this may
have to do with the extra weight that people put on during lock
downs, but nobody really knows for sure. The data from wearables
has been “a question generator”, says Mr Brockmann, raising que
ries about health that would not have been asked otherwise.
The ability to examine lots of human bodies as they go about
their daily lives is also changing how clinical studies of new drugs
are done. According to iqvia, a research firm, 10% of latestage
clinical trials in 2020 used connected devices to monitor people,
up from 3% in 2016. A catalogue by the Digital Medicine Society, an
American organisation, lists more than 300 examples of digital
biomarkers that are used in trials.
Activity measures, such as step count, for example, are a for
Data from wearable devices are transforming disease
surveillance and medical research
Measuring the masses