#Dataforbetterhealth: A Hackathon by @DigitYzer


Author: Sonja Peters


Four of the five members of the VisionWorks Data Science team took up the challenge to analyze and visualize data at the @DigitYzer hackathon.

Perfectly timed after a busy work week, and a warm-up prior to Ordina's New Year party, the VisionWorks Data Science Team pitched their tents at @DigitYzer in Brussels for the #dataforbetterhealth Hackathon.

You might be able to take this analogy literally; camping is allowed on the spot for this 26-hour challenge, presented by Minister of Social Affairs and Public Health Maggie De Block. One of her creeds is 'transparent governance', and she adds the word to the word and makes a first data source of RIZIV available to the participants.

After a short briefing, several things become clear:

The participating teams will not lack anything: super fast WiFi, helpful employees, work, rest and sleeping places, food and drinks. Everything has been thought of. Thus, the participating teams began to dig into the data. The teams ranged from unrelated data-experts to well-oiled data teams of a few known names from the pharma world.

We smelled blood and started to work. We had agreed a number of different ways in the exploration of the data beforehand, thinking out-of-the-box and taking into account the interests of our team: prescribing behavior of anxiety and depression medication, the relationship between the consumption of diabetes medication and the demographic development in Brussels and finally the relationship between asthma medication and the environment.

The hours pass and we can not get away from it: the data are complex and not flawless. In contrast to the teams from the pharma world, we lacked the domain knowledge to immediately get to the heart of the problems. The project staff and employees of the NIHDI were very helpful, but our questions were still too general to make big leaps ahead. That's how we head into the night: tossing, turning, searching, surfing… until Jolien and Sonja leave a tireless David for a few hours at 1:30 am. It will be a short night, in which David finds little rest, entangled in a web of data dreams.

The next morning at 10 o'clock, the team has reached its full strength again. Meanwhile, several tables are linked, including the demographic data. Python, R, Tableau and Tableau Prep, all means are good and are used intensively.

With the arrival of the RIZIV staff, we achieved a small breakthrough: we learned from which codes we could find medication for specific diseases: the website of the World Health Organization (WHO) being the key here. For example, diabetes medication could be found under the code A10. This contains three subclasses, the most important group of which was the insulin.

Spectacular decrease in consumption of Alzheimer medication in hospitals for all provinces.

At the same time, new challenges lurked around the corner: the first graphs showed erratic patterns in the reimbursement of medications. For the correct interpretation of this, a dialogue with specialists was required. But time flew by, and we were not able to get to a predictive model. An important stumbling block here was that the data only concerned the hospitals. In other words, we could not deduce a decrease in consumption of a certain type of medicine from these data. For example: Consumption of Alzheimer medication is drastically decreasing (see graph). Does this mean that less of this medication is being administered, or is this medication provided via the GP or a rest home?

Consumption of Diabetes medication (red) together with demographic evolution between 2005 and 2017

Likewise, the capricious pattern in diabetic consumption raises questions. The peaks in the red curves below can not immediately be explained without domain knowledge. These are due to a policy decision, which are the result of problems with the data quality, etc.


Yet a story started to unfold for us. Like any data scientist, after a first data exploration, we had a lot of questions that we could turn to a domain expert, and this dialogue was essential to arrive at deeper insights. At the closing presentation, our story appeared to show many similarities with that of other teams: struggles with the data complexity, the intensity and the lack of time… but what an experience! What was striking, however, was that had we not limited ourselves to data visualization alone, we could have potentially looked for evidence-based insights on a case-by-case basis.

And so we concluded, happy with the commendation that we received about our inventive approach to demographic data, ready to go a further night: the New Year party of Ordina started right afterwards....

The VisionWorks data science team attended:

Dieter De Witte
David De Wachter
Jasper Lauwers
Jolien Vanaelst
Sonja Peters