AI Milestones: AI-enabled EKG reads and alerts save lives: human-AI collaboration in practice
In this series, Parexel’s chief data officer Stephen Pyke considers the potential and challenges of artificial intelligence as AI comes of age in drug development. His conversations with research experts, in disciplines ranging from early development to clinical trial design, shed light on state-of-the-art applications and on AI’s dazzling possibilities to advance novel medicine, improve development efficiencies and enhance patient outcomes.
In April, Nature Medicine published results from a large, pragmatic randomized clinical trial that used AI-enabled EKG reads to identify patients at greatest risk of mortality, together with AI alerts to prompt treating physicians. The study, conducted by Chin-Sheng Lin and coworkers, enrolled 16,000 in-patient or emergency room patients in Taiwan; recent EKG traces of consenting patients were accessed from EHRs. For the intervention group, AI-enabled physician alerts included AI reports and warning messages delivered immediately. The control group received usual care and management. The AI alerts were associated with a 3.6% reduction in all-cause mortality, compared to a 4.3% reduction in controls. In the top 10% of AI-identified high-risk patients, risk of death was reduced by 31% over a 90-day observation period (https://pubmed.ncbi.nlm.nih.gov/38684860/).
Stephen Pyke. The potential value of AI in a healthcare setting has been recognized for a long time. In particular, AI may be used for the analysis of medical images (X-rays, CT scans, etc.) as well as EKG traces. And while an expert may be able to spot features indicative of underlying health problems, well-trained AI is able to match or even exceed issue detection rates, as well as reduce time taken for review and diagnosis. Given access to a large enough body of EKGs from which to learn, AI may actually be able to spot patterns which even an expert might miss.
While this is not news, what I think is important about this study is that it shows how AI can be integrated within a healthcare setting—in a practical, workable way—to deliver dramatic results. And a 31% reduction in mortality is certainly dramatic!
Nancy Kivel. Yes, the integration of artificial intelligence with a foundational practice like EKG assessments is a major story when it achieves such important benefits. I think the approach they used here could be a milestone in the evolution of EKG, which is clearly fundamental in not only clinical care but also crucial in clinical research.
What they did was create a two-step process. They combined a highly effective AI-enabled cardiac risk stratification scheme with timely AI alerts. They did this in a way that avoided the problem of physician desensitization. Desensitization—alert fatigue—can happen all too easily when physicians are inundated with alerts. They simply ignore the alerts and rely upon more familiar tools, combined with their skills and experience. In this study, they used AI reads to identify patients at greatest risk, targeting the top 10% of risk. Then they intervened with intensified monitoring or treatment, as appropriate. And this more select use of AI alerts seemed to work very well, judging by the study results.
Stephen Pyke. So they had better results with this use of risk stratification because the alerts were sufficiently few in number that physicians paid attention to them.
Nancy Kivel. It shows just how powerful AI-based tools can be when applied thoughtfully. AI gives us richer information, deeper analysis, faster reporting. We need more investigations like this to show us how to use it to best effect.
Starting in the 1960s, EKG machines had computer programs built in that produced a readout that appeared at the top of the paper tracing. They were notoriously prone to interference, there was no standardization across the programming of various models, and they did not take into account the condition of the patient. AI-enabled EKG reads are another step in the evolution of this foundational tool in cardiovascular care. It provides another layer of surveillance in the critical steps of identifying and intervening with medical care for the most at-risk patients.
Stephen Pyke. And that capability is vital in clinical research as well as in the patient care setting. The volume and complexity of data are growing exponentially in drug development processes. Novel AI integrations like this one can advance our ability to identify complex patterns, stratify risk, and alert researchers to take action.
We could use this two-step AI-EKG model in clinical trials in a number of ways. For example, it might be used for inclusion/exclusion determination to enrich a clinical trial population with patients in a targeted risk stratification, or to de-risk a trial so that you exclude patients likely to suffer a cardiovascular event. Or it could be used as a tool to support ongoing patient care and management. This may be particularly helpful when studying the effects of treatments designed to prolong life in those most seriously ill.
Nancy Kivel. AI can make today’s familiar, effective tools even sharper, more precise. Safety signal detection could be another trial application. Ongoing safety assessments in clinical trials involve some subjective evaluation by the investigators. AI-enabled reads and alerts could increase accuracy in AE data collection and speed response.
Stephen Pyke. The concept of cardiovascular health-related risk scoring and risk stratification is not a new one: the Framingham Heart Study, which began in 1948, led to the development of one of the best known stratification schemes. This study shows us how to take that concept to the next iteration in its development.
Nancy Kivel. If this AI-EKG model, or something similar, were to be validated and become standard of care, we could very well see it used in clinical trials. In cardiology, the concept of “AI-enabled” is being framed as “human-AI collaboration.” I think that is an apt description of where we are headed as AI technology advances clinical research.
Reference:
CS Lin, WT Liu, DJ Tsai, et al, 2024. AI-enabled electrocardiography alert intervention and all-cause mortality: a pragmatic randomized clinical trial. Nat Med. Apr 29; doi:10.1038/s419591-024-02961-4.