Skip to content

Q&A with THIS Space panel members 

During THIS Space 2024 delegates had the chance to put their questions to our stellar line-up of speakers. After the event, we caught up with a few of them, and they were delighted to answer some more of your questions. 

Over-testing and over-diagnosis  

The first session of the day looked at some of challenges that over-testing and over-diagnosis pose for healthcare, and the potential impact of AI. Speakers discussed some of the issues, which often stem from a well-intentioned desire to catch diseases early but can have unintended consequences. 

Natalie Armstrong is Executive Dean and Professor of Health Services Research of the School of Health and Medical Sciences at City St George’s, University of London. 

Natalie was asked how we can create systems that support better decisions and reduce defensive practices such as over-testing and overdiagnosis, which might be “driven by fear and blame.” While she admitted that she didn’t have all the answers, Natalie replied that it was important to consider, “how we signal that interpretation of guidelines and protocols is sometimes possible, how more senior or experienced staff role-model doing less, and how we help staff to clearly document what they decided not to do.”  

Natalie agreed that a lot of the concern around over-testing and diagnosis might be linked to avoiding risk, and time pressure. She also agreed with a delegate who suggested that in many cases, discharging people can take longer than admitting them. When asked how medical professionals could be supported to take necessary risks, she said, “I think more senior and experienced staff need to role model this, and organisations need to show that they will support their staff in making these kinds of decisions.” 

Tackling outdated diagnoses in medical records 

Florence Doo, Director of Innovation, University of Maryland talked about how outdated or incorrect diagnoses in electronic health records might contribute to unnecessary testing and treatment. She was asked, “How can healthcare systems tackle inaccuracies while making sure patients get the right care and clinicians can still trust the records?” Florence said that old entries saying things like “possible XYZ” on patient records might well encourage clinicians to order follow-up tests that may no longer be needed, and that “one way to minimise this cascade is with AI. We can use AI algorithms to scan for mismatches between active diagnoses and recent imaging or lab results. If a suspected issue never got confirmed by objective data, that diagnosis can be flagged for follow-up or removal — using AI to highlight inconsistencies and empower clinicians to provide excellent care.” 

Florence also believes in the power of patient involvement, saying that, “electronic patient portals can help patients spot inaccuracies. Patients often catch errors because they remember precisely when something was — or wasn’t — diagnosed.” 

She also stressed the importance of creating systems that reduce pressure to over-test, explaining that shared decision-making tools, integrated into electronic systems, could help clinicians weigh up the risks and benefits of additional tests in real time. Additionally, clear “safe harbours” that protect clinicians following evidence-based pathways could encourage more confident, conservative decision-making. 

Environmental stewardship 

Beyond individual patient care, the environmental impact of healthcare is an emerging concern. When asked whether environmental impact should be a key pillar of healthcare improvement, Florence argued that yes, environmental stewardship should become a core consideration. “This is something I’m personally passionate about as a radiologist who researches both AI and sustainability.” Radiology, with its energy-intensive equipment, offers opportunities to adopt greener practices, such as optimising machine usage and exploring telemedicine, “it’s not just a matter of turning off the lights,” she explained, “We need to think about the entire lifecycle of equipment, from manufacturing to disposal, and explore innovations that reduce waste.” 

At the same time, the integration of AI presents a dual challenge – Florence describes it as a “double-edged tool,” explaining that “AI tools might help to triage the patients who truly need imaging, cutting down on unnecessary scans and by extension, resource consumption. But the process of training large AI models consumes its own share of energy, which means we need to identify lean AI solutions that fit practical clinical needs without ballooning our carbon footprint.” 

Today’s challenges in responsible data science and healthcare innovation 

The final session of the day focused on addressing equitable access, data privacy, patient voice and other ethical considerations in data science and healthcare innovation.

Jessica Morley, a postdoctoral research associate at Yale Digital Ethics Center, highlighted the complexity of implementing AI in healthcare. When Jessica was asked whether we should push back against AI for healthcare delivery, she answered that while implementing AI is complex, “when designed well, supported by suitable information infrastructure, and robustly governed – it can also deliver enormous benefits. To push back against its use entirely would, therefore, be a mistake that would result in significant opportunity costs.” 

She suggested that instead, we should reject the narrative that AI can solve all of healthcare’s myriad problems and that it’s somehow more objective and more accurate than humans by default. “This is a false and harmful narrative that will prevent the development of the appropriate safeguards we need to put in place to make sure healthcare systems are able to capitalise on the opportunities presented by AI while proactively mitigating the harms.”

Finally, a delegate wanted to know how the quality and scope of data could be improved, without significantly adding to the ‘burden’ and assurance processes required of clinicians?  

Jessica replied that the most important thing would be making sure that collecting high quality data at the point of care was made as easy and as frictionless as possible. “To do this, we need well designed electronic health record systems that are specifically designed to make data entry seamless and easy and encourage less mistakes. We then need to invest more in research software engineering and the development of data platforms that automate ‘data curation’ – automatically cleaning and improving the quality of the data before researchers use it, in a standardised and transparent way. 

You can rewatch all the sessions on the THIS Space on-demand YouTube channel.  

Sign up to receive the latest news, reports and articles from THIS Institute.