First Datathon to predict unscheduled care

[ad_1]

ARS Île-com-France launched the first Datathon to predict unscheduled care in Ile-com-France. Presentation.

For three days, the main Ile-com-France health players worked on the development of a digital tool for predicting urgent and unscheduled care in the short, medium and long term in the region.

The goal? Equip health professionals, health establishments and the ARS with a tool to anticipate activity flows in order to optimize the organization of the health system and the mobilization of resources.

From November 25 to 27, non-scheduled health and care experts (emergency professionals, directors of health establishments), coders, designers and data scientists met at Freedom Living Lab to develop the first bricks of this digital tool. An unprecedented approach on a regional scale, which also aims to coordinate healthcare professionals and strengthen the supply of care. The experiment and its promising results are anchored in the national strategy "My Health 2022" of digital transformation of the health system and the Regional Health Plan of Île-com-France.

Predict unscheduled care using artificial intelligence

This project mobilized large health databases (activity of emergency services, medical regulation files for emergency medical services, firefighters data and SOS Doctors data), enriched by external climatology and air quality data ( Airparif), cultural or social events (Prefecture), data from pharmacies (OSPHARM) or Google requests from users on symptoms linked to specific epidemics and pathologies.

Each multi-professional team focused on one of the following six issues:

• Challenge 1 – How to predict the flow of non-scheduled short, medium and long term care activities in correlation with regional events?

• Challenge 2 – How to predict epidemiology in the short and medium terms, in correlation with external environmental factors and the audit of relevant data (Google trends)?

• Challenge 3 – How to predict non-scheduled care activity in the short term based on epidemiological data?

• Challenge 4 – How to predict long-term unplanned care activity based on epidemiological data?

• Challenge 5 – How to predict the flows of non-scheduled short-term care activities based on medical regulation files from SAMUs and external data made available (SDIS91, OSPHARM, etc.)?

• Challenge 6 – Focus: how to predict the flows of short-term non-scheduled care activities in a specific territory (91) with all the available data?

With the objective of proposing, for each challenge, an interactive tool making it possible to model on several geographical scales the prediction of unscheduled care on activity estimates from 7 days to 6 months and thus, in the long term, to anticipate the impacts possible (human resource needs, hospital bed needs, etc.).

Encouraging results

In three days, interesting results could be produced from data, some of which were never crossed with each other and which presented difficulties of appropriation due to their different natures. The predictions obtained constitute a first point of reference, allowing further research and improved results in the coming months.

In general, strong predictive axes emerge from the only study of the history of activity data with small margins of error (between 6 and 10%). Certain avenues and the intersection of the different results have also opened the way to targeted explorations which could improve predictions, in particular the effects of social and meteorological events, provided that for each challenge a finer data mesh is available.

At the end of the datathon, all the players were able to fulfill two original objectives:

1. Use modern approaches to data analysis and machine learning in order to assess the potential of all the data sets mobilized and crossed

2. Laying the foundations of an ergonomic decision-making aid tool, intended for all emergency professionals, the directors of health establishments and the ARS Île-com-France to anticipate peaks of non-scheduled care services.

Source: Île-com-France Regional Health Agency

Similar articles

[ad_2]