Passive Data Challenge: Passive Data in Healthcare

Join our Meetup on August 2nd 2018 for the chance build your own innovation using passive data.

The winners of our passive data challenge will receive:

  • 3M NIS investment and incubation from the eHV
  • A free trip to Copenhagen to compete in LEO Innovation Lab’s Global Challenge
  • 10k EUR prize, 3 months of mentoring & incubation in Copenhagen


How Can We Use Passive Data in Healthcare?

In our previous post, we discussed the definition of passive data, and why we’re now at an inflection point to use such data to help patients.

We live in a time where smartphones permit, with the users’ consent, passive sensing: the capture of data about a person without extra effort on their part.  



… But the related technology, that enables passive sensing, has evolved. Physical activity, sleep and cardiovascular disease researchers have, for example, used passive sensing for decades through tech like pedometers, polysomnography and implantable electronic devices.

However, we are only now beginning to understand the power of passive data collected from smartphones. In fact, smartphones are of particular interest since they combine multiple sensors (e.g., GPS, accelerometer, microphone), can produce interesting device analytics (e.g., call logs, SMS patterns, apps being used), are ubiquitous in daily life, allow us to “plug into” a global network, and have ever-stronger computing power.

There are many companies out there that use the science behind passive data to help patients.

Some work; others… not so much.

So for this post, we’d like to explore some interesting scientific evidence and research focused on smartphone-based passive sensing for healthcare and wellbeing.

So strap yourselves in, and let’s get started!



One of the most popular uses of passive data is in the field of mental health.

Depression is a burdensome mental health disorder with high prevalence, but patients have to wait for several months to receive assessment and treatment. The good news is that in a 2016 study, Swedish researchers were able use smartphones to predict and detect depression using passive data, and were even able to provide intervention support.

So how did they accomplish such a seemingly magical thing?

Firstly, knowing that there’s a strong relationship between physical activity and depression, they used the smartphone’s acceleration sensor data to analyse general activity levels and walking time.

GPS information was then used every 15 minutes to get the subjects’ locations to see the distance they’ve traveled, and WiFi-based location logging was used to measure the time a subject stayed at home.

As previous studies showed that the higher a person’s depression levels are, the less likely they are to have social interactions, these researchers analysed the total time subjects used their phone, the number of incoming and outgoing calls and text messages, and the total number of contacts these communications were with as a benchmark of social activity.

Even the number of calendar events were analysed as a proxy for stress, which could have an influence on depression levels. Interestingly, calendar events in the evening could also represent another dimension for social activity, so they had to be careful to accurately categorise these.

Taking these passive data points together led to predicting depression with 61% accuracy (using the help of a support vector machine classifier for all you machine learning enthusiasts) – here’s the link to the study.

A similar passive data approach was recently also used by US researchers to predict bipolar disorder with 85% precision.

These researchers activated the smartphone’s microphone every two minutes to capture ambient sound. If human speech was detected, the microphone remained active, but to protect privacy, they did not record audio. Instead, they processed data in real-time to extract and store features like spectral content, regularity, and loudness.

Using these privacy-sensitive audio features, it was possible to estimate the number and duration of conversations an individual engaged in, and how much time a given individual spoke within a conversation along with speaking rate and variations in pitch, which have been used to detect social isolation – here’s the link to the study.

Other studies have also shown that smartphones could be used to detect Parkinson’s, where the accelerometer detects tremors and falls, the compass detects poor balance by looking for zig-zag movements, and even the microphone can detect slower monotone speech, which are symptoms of the disease.



Hundreds of biological clocks and oscillators in our body all work together so that we can function in the light and sleep in the dark (the ‘master clock’ in our brain is the Suprachiasmatic Nucleus for all you aspiring neuroscientists).

When these biological oscillators go awry, there could be some serious consequences, including cardiovascular disease, cancer, obesity and a variety of mental health problems.

That’s why researchers at Carnegie Mellon University decided to see whether smartphones can detect sleep and wake states, as well as sleep quality.

They used smartphone sensor data that might be relevant to sleep and sleep quality, including sound amplitude via the microphone, light via the ambient light sensor, and movement via the accelerometer.

As smartphones might be in a user’s pocket most of the time, making data collection from the light sensor tricky, the researchers were also able to collect ‘screen proximity sensor values’, like whether the screen was on or off, number of apps actively running on the phone, and the battery-charging state. The interesting thing about this study is the use of multiple factors in combination to try and detect sleep.

For example, if the screen is on, it’s safe to assume that a person is probably not asleep, but the screen is also sometimes automatically turned on for incoming calls or text messages, notifications and alarms. Thus, other data, such as motion, was combined with the screen state to detect people’s actual use of their devices.

Using such passive data resulted in the detection of sleep with 94% accuracy, and the detection of sleep quality with 84% accuracy using a Bayesian network model – here’s the link to the study.



One of the most common disorders among college students is social anxiety. 40% of students report feeling “overwhelming anxiety”, which is associated with impaired academic functioning, stressed relationships and avoidance of the social opportunities that college life can offer.

Using sequences of GPS coordinates as well as a knowledge of the neighbourhood, a team of University of Virginia researchers were able to recreate ‘GPS trajectories’ for the students, such as Home -> School; Food & Leisure -> Work; Work -> Church).

For each location visited, the researchers also measured the cumulative staying time and transition frequency from one type of location to another.

So what did they find?

While the researchers found that social anxiety levels are significantly correlated with places students visited (e.g., students that spent more time at religious locations reported much less social anxiety), the key idea of this study was the ability to recreate ‘GPS trajectories’ for broader and more complex uses – here’s the link to the study.

Location also comes into play when using passive data for nutritional wellbeing.

Researchers at Cornell University used similar ‘GPS trajectories’, together with some ‘active’ logging by users about what foods they ate, to learn a user’s physical activity and dietary behaviour.

Then, a deeply personalised recommendation was generated for the users to try and change behaviour towards a healthier lifestyle that maximises calorie loss.

These suggestions either continued building on the user’s positive activities, or made small changes in some situations, while also recommending the avoidance of negative activities. It could be context- and location-based walking tips like “walk near Garden Avenue” or “take a three minute walk near Juliette Lane”, as well as dietary tips like suggestions for a good meal in the neighbourhood, which helped users move towards a healthier lifestyle.

This approach, detailed here, showed a significant increase in physical activity and decrease in calorie consumption when participants received personalised recommendations.



The field of passive data from smartphones is a rapidly emerging one and, as you can see from the examples, the list for passive data applications grows every day.

Now it’s your turn.

In preparation for the upcoming challenge, we’ll be compiling a data pack with a list of the major types of passive data that can be collected by smartphones for inspiration.

So you don’t have to be a passive data guru to come up with a creative idea.

As you prepare for the challenge, read some of our content on passive data and, of course, come to the Meetup. You should be armed with the tools you need to make an impact and help better the lives of millions of patients!

See you at the August 2nd 2018 Meetup!