How far can we trust AI in clinical trials?- INDUSTRY VOICES

The use of AI in clinical trials is beginning to find its place in the industry around the globe, particularly in Asia, which currently marches ahead in its plans to integrate artificial intelligence. With the promise of many benefits such as speeding up analytical processes and patient recruitment, it is clear to see the attraction in joining the race in integrating AI and machine learning into clinical trials. However, there are growing concerns over the current limitations, from handling data standards and safety to achieving acceptance from patients and regulatory bodies.

We asked seven clinical industry experts what they saw as the biggest challenges in using AI in clinical trials, and the ways in which the industry can overcome them.

Michael Song - Senior Manager R&D, MedImmune

Biggest challenge is robustness of AI and the number of data needed to improve AI accuracy. In addition, data can be dynamic and interlinked, AI will need to be able to handle this changing interlinkage between data as well as dynamic nature and external factors that influence the data. Best way to address this is to take it one step at a time. I see a lot of great aspirations, but to achieve them it requires multiple careful steps. Those initial steps is what we need to focus on as what we learn from each step will help shape how the next step will look like, and in time the aspiration may change due to those learnings and adjustments.

Micheal will be delivering a presentation on ‘Building models to engage more and ‘better’ patients in clinical trials’ at Clinical Trials Europe on 20 November in Barcelona.

Ruby Saharan, Senior Medical Advisor- RWE, Novartis Oncology UK and Ireland

There is a real lack of Data Translators and Architects within the industry at present. This can be potentially addressed through alternative hiring- such as increased placements of post-graduate tech students into the healthcare sector.

Ruby will be speaking Clinical Trials Europe in Barcelona, 20 November, on using clinical EHRs, including Artificial Intelligence and Machine Learning to assist the drug commercialization process.

Renee Deehan-Kenney PhD, Vice President of Computational Biology, QuartzBio

Artificial intelligence (AI) in clinical trials suffers from an imbalance in patient sample size (n) relative to data generated by high-throughput measurement assays (p), often on the order of tens of thousands of measurements per sample. This discrepancy results in overfitting and limits the applications of AI for learning-based approaches. Feature reduction through prior knowledge and/or data augmentation through publicly available data sets can be deployed to balance p with n. Prior knowledge and public data approaches should be strategically employed to enhance learning-based approaches and ensure optimal insight generation from your clinical trial data ecosystem.

Bruce Hellman, CEO, uMotif, @umotif

The biggest obstacle is finding the right use case for AI to help improve clinical trials of today.

There is a misconception that AI is only for data analysis. However, its power can be applied to engaging patients in clinical trials. Platforms such as Netflix and Amazon use AI to encourage people to consume more based on previous actions. How might we use AI to engage more patients and capture better quality data?

A common challenge with AI is the cleanliness of data meaning that accurate conclusions cannot be drawn. By using AI to better engage patients and capture more data at a higher quality the analytical power of AI can be unlocked further downstream.

Bruce will be taking part in a panel discussion on the dilemma of collecting massive amounts of data at Clinical Trials Europe on 20 November in Barcelona.

Jenny Royle, Patient-centricity Senior Program leader, digitalECMT, CRUK Manchester Institute, and The Christie NHS Foundation Trust

The reproducibility and trust required for synergistic working between computer and human. Computers and humans are good at different things – would you want a computer to interpret a rare medical complication? Or a highly experienced doctor stuck/bored monitoring a medical scans for abnormalities 24/7? We need to use this expertise and resource synergistically and this requires that AI is focussed on areas of human weakness first, and that people understand how an output is reached and how reliable that output is (even if it’s reached in a way a human wouldn’t have tackled it).

Siobhan Southam, Strategic Engagement Leader, digitalECMT at CRUK Manchester Institute, and The Christie NHS Foundation Trust

My views are similar to Jenny’s here. It also depends what you mean by AI as it’s a rather hyped term and I have sat in a meeting and heard someone describe it as ‘superior intelligence’, which it is not. Machine learning and algorithms might be more appropriate descriptions and here the challenge is transparency and trust. We need to be confident that ultimately there is a clinician making an informed judgement in a patients best interest. That requires oversight as well as understanding how the algorithm works, the decision tree and the dataset used to develop/train the algorithms. The other element is data security and this is something that patients themselves have raised.

Siobhan will be speaking about innovation in care pathways for clinical development at Clinical Trials Europe on 20 November in Barcelona.

James O’Loan, Consultant Pharmacist, Doctor4U

There’s no doubt that the advancements in Artificial Intelligence over the past decades have significantly improved clinical trials. AI has made the process of clinical trials a lot faster without the need for extensive manual labour, which in turn saves the cost and time that manual labour brings to clinical trials. AI can be used to discover new medicines, detect diseases, and deliver healthcare treatment and services a lot faster than humans can. It can make trials more successful and reliable by being able to continuously monitor patients throughout the trial and collate large amounts of data when humans can’t, plus it has the ability to make decisions the same way humans can.

However, the ability of AI to be faster and smarter than humans has posed some concern. If Artificial Intelligence can do everything humans can in clinical trials, will it replace human workers? This is a much-debated topic that raises concerns that AI may be unethical. This is one of the biggest obstacles to using Artificial Intelligence. The demands and complexities of clinical trials require the sophistication, stamina, intelligence, and speed of AI, as opposed to doctors and scientists who have the burden of human problems such as being burnt out, needing time off, and sometimes human error.

AI has many benefits and has significantly advanced our healthcare, however, the idea that AI is taking over the role of doctors and scientists and is unethical is in some way holding back what we can achieve with this technology. Although the edge that medical professionals have over AI is human emotion which is always needed for patient care. Human interaction with patients in trials is necessary so that patients can understand what they are taking part in, and decisions can be made based on their physical and mental response to the trial which may only be recognised by humans.

To get the benefits of both, doctors and AI should work together to create ethical, safe, and efficient clinical trials. We should be able to take advantage of these advancements in technology and provide the best healthcare we possibly can. Instead of fighting these advancements and seeing it as a threat we should be using AI to our advantage providing it has proved safe and effective, and not deny the public of advanced healthcare.

However, there is a compromise. In terms of healthcare, human intelligence will always be needed alongside AI. While Artificial Intelligence takes care of some of the administrative tasks in a clinical trial such as collating large volumes of data precisely and accurately and monitoring patients continuously, this allows human physicians to turn their attention to other aspects of the trial such as ensuring it is safe and ethical, and making decisions about the best treatment for their patients.

RELATED ARTICLE: Five AI Healthcare Startups Bringing Us Closer to Cures