French Hospital and Intel Predict ER Visits and Admissions


For hospital administrators, predicting the number of patient visits to emergency departments, along with their admission rates, is critical for optimizing resources at all levels of staff. Ultimately, this reduces wait times in emergency departments and improves the quality of patient care.

Intel and the Assistance Publique-Hôpitaux de Paris (AP-HP), the largest university hospital in Europe, worked together to build a cloud-based solution for predicting the expected number of patient visits and hospital admissions using advanced data science methodologies and the Trusted Analytics Platform (TAP). TAP is an open source platform that accelerates the creation of applications driven by big data analytics. Using data from four emergency departments within APHP, data scientists from Intel and medical experts from AP-HP evaluated three different approaches to time series analytics, optimizing model parameters and identifying the best predictive features to include in each. The team selected an Autoregressive Integrated Moving Average with Exogenous Input (ARIMAX) approach that proved to be simultaneously accurate, scalable, and easily adaptable to the needs of both data scientists and hospital staff.

The team moved into the model optimization phase of the project, using such metrics as the Akaike Information Criterion, or AIC, to explore which features to include in the model to balance accuracy and complexity. The team also developed an Apache* Spark-based implementation of the ARIMAX algorithm to take advantage of the speed and scalability of TAP’s distributed processing infrastructure.