ISSUE: The use of machine learning (ML) in biomedicine is increasing at a rapid rate. However, there is a substantial and well-documented division of opinion on its value; critics argue that the methodology is not transparent and too often leads to mistaken inference, calling it the new alchemy. Supporters argue that ML is a potential game-changer and enables powerful insights when traditional approaches fall short.
OVERVIEW: This debate has not yet fully addressed the area of market access and health economics and outcomes research. Whilst common in other industries and widely used in the drug discovery process, ML application to patient medical record data presents several specific challenges, particularly when it is used as a diagnostic tool. In 2018, the new European legislation on data protection (General Data Protection Regulations) will potentially prohibit some current uses of ML (e.g., automated individual decision making and profiling). Considering the current real-world data explosion, the benefit of ML is potentially substantial; however, we also need to acknowledge the potential for incorrect application and thus the subsequent likelihood of mistaken insights.The purpose of this panel will be to represent the spectrum of opinions on ML in our industry and will aim to explore opportunities and challenges with the participation of the audience. Examples of poor and successful applications of ML in the public domain will be used to make the debate accessible to a non-technical audience. Specifically, we plan to address
- hypothesis generation (e.g., informing drug development) vs decision making (e.g., regulatory, reimbursement, treatment choice by doctors)
- validation processes (how does ML compare to traditional statistical methods?)
- interpretation of ML findings (how often are they spurious and can we trust results even if we cannot explain them?)