PURPOSE: Machine learning is gaining in momentum among stakeholders in the healthcare ecosystem with the advent of Big Data and new possibilities for evidence generation to support patient access to novel therapeutics. However, machine learning remains a mysterious term for many stakeholders. The workshop will describe common machine learning methodologies and will focus on real applications of these techniques in the health economics and outcomes research (HEOR) area.

DESCRIPTION: ‘Machine learning’ is a generic term to describe algorithms and methodologies, which includes cluster analysis, decision trees, artificial neural networks, Bayesian networks, and others. These techniques, while used before in other fields, are becoming relevant in healthcare with the influx of new types and/or volume of data, as well as the new possibilities to consolidate datasets from several origins.

Helene Karcher will start by presenting the principles of machine learning, the main techniques, and the data needs for a few staple techniques with examples.

Christoph Gerlinger will outline how machine learning techniques, ie, cluster analysis, time series analysis, and random forests, were used to develop a patient-centered prediction model for the menstrual bleeding for women who start a new hormonal intrauterine device for contraception.

Gorana Capkun will discuss the use of machine learning for patient identification, its role in optimizing chart reviews and improving referrals for multispecialty diseases with difficult diagnostic patterns and no existing therapy.

Jackie Vanderpuye-Orgle will present applications and case studies of machine learning in oncology and pulmonary hypertension for patient identification, clinical decision support, and prognostic tools.

Finally, the presenters will discuss with the audience the proper place of machine learning techniques in the HEOR area.