Digital therapeutics
“I
t looks likeyour back is a bit tight today. Let’s modify your
workout.” The voice is gentle yet commanding, the instruc
tions rolled out in the signature cadence of a physiotherapist. It is
also unmistakably robotic. The aiphysio issues her commands
straight from a smartphone’s speakers. A phone with a camera is
all she needs to do her job: select the exercises to suit the patient’s
injury, guide him through each session and order corrections
when he is not doing something right (bending a knee at the
wrong angle, for example). An aialgorithm marks up his body as
he moves, which is how it knows when a joint is hurting or the
back is stiffer than the day before.
This digital therapist, developed by Kaia Health, a German
startup, is, by many measures, as good as a human therapist. One
trial, which involved 552 exercises by osteoarthritis patients,
found that human therapists agreed with the corrections to exer
cise suggested by Kaia’s app as often as they agreed with the cor
rections suggested by other human therapists. In clinical trials pa
tients with back pain using the app improved more than those
who got inperson physiotherapy. Making people with injuries
bend and twist carries some risks. On that, too, Kaia’s app is no
worse than human experts. Less than 0.1% of nearly 140,000 users
of the app in studies reported adverse events.
The app is registered as a medical device by America’s Food and
Drug Administration (fda) and also in the European Union. Since
2017 the fda has approved more than 40 other health apps for pro
Treatment—not just diagnosis—is being delivered by apps and
devices connected to them
Software as treatment
8 Technology QuarterlyThe quantified self The Economist May 7th 2022
As too often with the design of new tech products, behavioural sci
ence is an afterthought. When she began researching her book,
Natasha Schüll, the cultural anthropologist from nyu, found try
ing to decode the logic behind many healthtech products very
frustrating. Then she realised that there was no logic. It was “a
pinch or two of positive psychology thrown in with the infrastruc
ture of a punitive Skinner box, thrown in with some other notion
of the brain” ending up as “a hotchpotch of adhoc things that was
not that studied or scientific”. The typical design strategy, she con
cluded, “was just throwing things at the wall”.
This may be why many of the apps fail to establish themselves
in people’s lives (see chart on previous page). There are more than
400,000 health and wellness apps on the Apple and Google app
stores with 250 new ones added daily. Appetite for them is healthy,
with around 5m app downloads per day. But 95% of those down
loads will be deleted within 24 hours.
The problem is that people do not just need a product that is
well designed. They need a product that is well designed for them,
says Liz AshallPayne from Orcha, a British organisation that eval
uates the quality of health apps for clients like the National Health
Service. As she points out, buying a pair of trousers online is made
easy by filters for size, colour and style, but no such system exists
on the app stores. A teenager seeking help for anxiety will need a
different type of app to his grandparent wanting the same thing.
A lot of dead wood
Orcha has rated 7,000 health apps on three criteria: privacy, user
experience and evidence that they work. Only about a quarter of
them meet its quality threshold on all three. Mentalhealth apps
are particularly weak. But quality is improving, says Ms Ashall
Payne. She attributes that to the guidelines for health apps that
were set recently by British health authorities, which gave devel
opers clarity on what “good” looks like. As the market matures the
bad products will die out, “but it’s going to take time”.
Making people stick to healthy behaviours is probably the big
gest challenge in public health. That makes innovations which
boost compliance particularly exciting. Technology developed by
Sweetch, an Israeli startup, is making the advice dispensed by
health apps more practical and personalised. Sweetch’s aibased
algorithm is a hybrid of a personal secretary and a motivational
coach. It keeps track of whatever it is the user must do, such as
walking a certain number of steps or checking their weight week
ly, and finds the best time to suggest they do it.
The vigilant bot may spot, for example, that you have a 20min
ute gap between meetings and suggest you pop out for a walk to
the coffee shop a few blocks away to do your steps. It changes your
activity goals up and down depending on how you are doing, to
prevent you from becoming demotivated and ditching the whole
thing. The prompts are available in 33 different tones of voice:
combinations of words that can be friendly, begging, command
ing and so on. Yoni Nevo, Sweetch’s chief executive, says that the
algorithm evaluates about 700m possible combinations of things
to say to an individual on any given day. It takes about four to five
weeks for the algorithm to learn what makes you tick by trying dif
ferent combinations of all these things.
Device manufacturers, for their part, are starting to realise that
when it comes to giving people health data, less is sometimes
more. Many people are happier when they are simply told whether
things are fine or not, rather than being flooded with all sorts of
data to make sense of before breakfast. Some smart scales now just
buzz approvingly when you step on them to confirm that your
weight and bodyfat composition are in your target range, rather
than giving you the metrics. Wearables like the Oura ring are now
able to tell you not just what your heart rate, sleep or temperature
trends are, but also what that means—and what changes might
improve things. And when the data show that things have gone off
kilter, novel digital treatments may come to the rescue. n