ARS Ile-com-France takes stock of its datathon

[ad_1]

The ARS Ile-com-France had announced in March 2019 its ambition to develop a tool for predicting the activity of non-scheduled emergency care using artificial intelligence (AI), we recall.

It organized a "datathon" from 25 to 27 November 2019 to advance its work. The event brought together "emergency professionals, directors of healthcare establishments, coders, designers and data scientists"she said in a press release.

Solicited by TICsanté, Axelle Menu said it was "very satisfied" with the first results.

The data used came from "medical regulation files of the Samu, firefighters and SOS Doctors", as well as "from the activities of the emergency services of a hundred establishments in the region with a history of 10 years", she said.

"Knowing that there are 3 to 4 million emergency visits per year in the region, that represented the largest volume of data," she said.

Were also used "data relating to the sales of medicines available without prescription in Ile-com-France from Ospharm, meteorological data and data from the prefecture relating to events gathering more than 5,000 people" such as concerts and sports meetings .

The total data volume was 1.3 GB.

It "is not extremely important because we had previously aggregated the source granular data to save processing time during the datathon and to make data available in an open data format," said the ARS.

Participants were asked to respond to six challenges representing pre-identified needs, such as predicting the flow of emergency room visits correlated with "regional events", "external environmental factors and auditing relevant data" from Google trends. , the tool for knowing the frequency of searches on Google, or "based on epidemiological data" in the short and long term.

"We note that the mobilization of only emergency passage data and emergency medical services, without crossing other data, makes it possible to predict the rush to emergency departments with margins of less than 10%. This has already been demonstrated, but we wanted to 'objectify', underlined Axelle Menu.

The agency also found that an increase in emergency department calls affected emergency room visits in the days that followed. "It is known from SOS Doctors and emergency data, but less also by using data from Sdis (departmental fire and rescue service, editor's note) 91 and Samu. This encourages us to make a multi-tool -sources, "said the manager.

Unsatisfactory use of Google Trends data

On the other hand, the use of data from Google trends has not been satisfactory. The datathon "wanted to verify that the increased search for keywords linked to symptoms made it possible to observe a few days earlier the onset of a phenomenon such as a flu epidemic over several years".

"We have not observed this phenomenon," said Axelle Menu. "We should have: the region brings together a fifth of the French population, it is fairly smart health connected." The problem came "from a technical limitation: Google aggregates data by the week into open data, but we would have needed data by the day. But we will continue to dig."

Concerning the "very numerous" climatology data, "we underestimated the time necessary to build the indicators", she continued.

"We know that black ice increases emergency room visits, but black ice is not a given, it is a risk linked to humidity and a sudden drop in temperatures. We ran out of time to build this type indicators, but the data could still be useful, so we're not going to give up. "

The ARS does not only want "to anticipate the activity of emergency services, and in particular very concentrated peaks", but "to predict the number of emergency visits by type of patient differentiated by age category", underlined Axelle Menu.

This tool should be able to "be used by health professionals in health establishments, the medical regulators of the Samu or Sdis and the ARS", she said.

The results of the datathon are still "in the digestion phase" and should be presented to the steering committee at the end of February. "If the development of the tool is prioritized in terms of resources, we could have a Review version at the end of the summer," she said.

[ad_2]