AI and healthcare: future opportunities and challenges

Data is the fuel that feeds the AI/ML engine

My father was an accountant, and in 1965, he decided to explore the emerging field of information technology (IT). He joined an IBM training program and went to work selling their System 360 mainframe computers. That year’s model would set you back $2-3 million, for which you got a room full of refrigerator-sized cabinets supporting a central processor with about a megabyte of memory. These machines were used by businesses to automate accounting, and for billing and tracking inventory. Looking back these tasks now seem simple, straightforward, even trivial, but at the time these capabilities were revolutionary, automating tasks that took people enormous amounts of time and effort.

Complex endeavour offers greater payoff

What does this reflection have to do with applying artificial intelligence and machine learning (AI/ML) to healthcare? It’s a reminder that over the course of these past six decades, the ability to apply IT to increasingly complex realms of human endeavour has hinged on the capability of computer hardware, the development of computer science to exploit these machines, and progress in data science to supply techniques for gathering, organising, and extracting knowledge and insights from mountains of data. The more complex the endeavour, the more sophisticated the IT required to address it and the greater the potential payoff.

Today we have computing capabilities barely imaginable to those early adopters. AI/ML enables these devices to learn and improve their performance as they consume the enormous datasets that have been organised, standardised, curated, and stored to facilitate their use. This combination of capabilities enables the application of IT to healthcare, among the most complex fields of human endeavour. The life sciences, medical sciences, therapeutic development, clinical practice, epidemiology, population health, and public administration are just a few of the interrelated disciplines that stand to benefit from the application of AI/ML.

Development provides opportunity

Although IT development provides this opportunity, the challenge of realising its promise for improving people’s healthcare is not simply the technical task of applying algorithms to data. Privacy, safety, equity, and how societies bear and distribute the costs of care are social imperatives that must be addressed in the process.

Developed societies have highly regulated healthcare systems. AI/ML can improve care only if they are integrated into these complex systems in a way that addresses these social imperatives and fits within a sound regulatory framework. But we must bring their capabilities to bear in a responsible and forward-looking manner.

Data is the fuel that feeds the AI/ML engine. With improved and wider access to a higher quantity of higher quality datasets, we can create better algorithms on which the engines are trained, leading to more useful discoveries and insights. We are working to identify and catalogue priority data sources, both public and private, and work with government and private stakeholders alike to develop best practices for data curation, management, and use. For example, The Alliance for Artificial Intelligence in Healthcare (AAIH) - an international advocacy organisation dedicated to the responsible adoption and application of artificial intelligence and machine learning (AI/ML) across the life science and healthcare spectrum - is working to demonstrate and further develop the use of federated data, a method of promoting data sharing through the virtual aggregation of multiple contributor’s datasets. This approach avoids the privacy, security, and intellectual property challenges to the physical aggregation of multiple datasets on a single server, offering tremendous potential for enhanced research collaboration.

FDA collaborate on standardisation

The advent of AI/ML in healthcare creates the need for a new regulatory framework to accommodate the technologies they enable, in two ways.

First, there is the regulation of products, including diagnostics and devices, with AI/ML software at the heart of their operation. To address this need, the Food and Drug Administration’s (FDA) Centre for Devices and Radiological Health (CDRH) has created a Digital Health Centre of Excellence to work with industry and other stakeholders to develop the standards and concepts to enable effective regulation. For example, the Alliance is engaging with the Centre of Excellence on the development of standardised methodologies for developing and evaluating AI/ML algorithms that enable medical products. Additionally, we are looking at concepts for what type of labelling will be required to effectively notify consumers and clinicians about the function, performance, and testing of AI/ML software.

The second category takes advantage of AI/ML’s ability to make the drug development process quicker and less expensive. As always, the FDA will approve drugs based on testing and assessments of safety and efficacy. But technology has enabled the use of real-world data (RWD) to inform those assessments and approvals like never before. For example, electronic health records, patient and disease registries, claims data, and patient data gathered from mobile devices and wearable sensors can all be used to inform the FDA’s processes and decisions. The challenge is how to turn this RWD into real-world evidence (RWE), that can be used to shape clinical trials, support pharmacovigilance, and efficacy assessments.

As excited as we are about the benefits we can reap from this revolution in future, it’s even more thrilling to think about the advances made in the last 60 years -- foreshadowing how the health and wellbeing of all can be improved in the decades ahead, through dedicated effort.

Author: Steven Wolfe is Executive Director at The Alliance for Artificial Intelligence in Healthcare (AAIH), theaaih.org