Artificial intelligence (AI) is complicated. Now imagine applying it in the health care setting.
To address the challenges and complexity of applying AI in health, CTA brought together nearly 50 companies – from large tech giants like Google to non-profit leaders like American Medical Association – to provide standards and recommend best practices. Just a few months and many discussions later, the group of experts are now nearing completion of its first standard. This document will outline definitions and characteristics of AI in health care, building a foundation that doesn’t exist yet.
In speaking with CTA, the co-chairs of the working group, Pat Baird, Regulatory Head of Global Software Standards, Philips and Jerry Wilmink, Chief Business Officer, CarePredict, talk about the significance of this initiative and next steps in the standardization effort.
Pat Baird, Philips: Standards can help set the expectation of what “good” looks like, and there is so much hype and questionable claims about AI products and applications right now that we need standards to help differentiate between the good and the bad.
Jerry Wilmink, CarePredict: Standards are important in the fast-paced growth of AI to ensure technologies are effective and scalable across the entire health care system.
AI technologies are very new and introduce an entirely new set of challenges that have not yet been considered in the traditional health care ecosystem. Standards that cover the definition, development, deployment and maintenance of AI in health care do not exist yet, and we need them to ensure terminology and categorization is accurate so all parties can clearly communicate.
Baird: The main challenges have been around common terminology. Data science as well as regulations and standards for medical devices have been around for decades, and each have developed their own definitions for certain key words.
Wilmink: The rapidly evolving field of AI presents challenges for the members of the standards committee to keep up. This includes the impact of new types of input data collected from novel data collection systems and implementation of AI technologies in clinical workflows are current challenges.
The benefits, however, are revolutionary. For example, deep learning networks are being tested in a several heath care applications — imaging diagnostics on the frontline, clinical decision-making and “machine-augmented” preliminary diagnoses. Over the past year, studies have illustrated that such models can identify cancer up to 50% faster and with performance on par with leading radiologists and dermatologists. These algorithms have the potential to save our health care system billions of dollars by providing a preliminary diagnosis before a patient sees a specialist or visits an emergency room.
Baird: One of the benefits of working with a very engaged team is that someone will phrase something in a certain way, and suddenly a complicated and amorphous topic will suddenly be simplified in your mind.
An important concept that emerged during our discussions was the idea of “assistive intelligence,” where the software helps inform a clinician, but the ultimate decision lies within the caregiver. I think we all appreciate the value that an assistant brings to any organization – they help improve productivity, let us concentrate on the most important things on our plate, can handle the details of a problem, etc., but we maintain ultimate ownership of an issue. I think this simple term will reduce confusion.
Wilmink: This standard intends to provide the framework for better understanding AI health care technologies. Specifically, we intend to provide common terminology that users, technology companies and care providers can use to better communicate, develop and use AI health care technologies.
The standard also explores some of the unique challenges with AI that are specific to the application of health care including data availability and data interoperability, incorrect predictions, data distributed shift, clinician trust and implications with regulators. These challenges provide the framework for some of the future work CTA standards will explore.
Baird: At Philips, we believe a person’s health should be a connected journey that offers a seamless, integrated and highly personalized experience. We think that AI can help turn that vision into reality. The needs of the health care provider and the patient or consumer should always be at the forefront. Technology should adapt to their needs, extend their abilities, and help them achieve better outcomes.
Also, AI is cool and I wanted to play with it. I’m a standards geek!
Wilmink: CarePredict developed and commercialized one of the first AI platforms for predictive senior care, and fortunately, our organization has had years of experience working with leading senior care providers to develop, test and integrate our solutions into the senior care workflow. I previously led a senior wearable technology company and have been an active participant in CES and other CTA working groups. I am honored to have the opportunity with CTA on the AI in health care standards initiative.
AI in health care will be on display at CES 2020 and during the Disruptive Innovations in Health Care conference. The program, presented by CTA and the American College of Cardiology, will offer continuing medical education credits and cover topics such as AI in health care, value-based care and reimbursements. Register today.
For more information or to get involved with CTA’s Artificial Intelligence in Health Care Working Group, please contact Kerri Haresign.