AI Has Great Potential in CV Pharmacotherapy, but Much Work Lies Ahead: Review

Artificial intelligence (AI) is expected to have a major impact on cardiovascular pharmacotherapy in numerous ways over the next several years, touching on everything from drug discovery to clinical decision-making, according to a state-of-the-art review.

AI models could contribute to streamlining the process of bringing new drugs to market, personalizing treatment regimens to maximize benefits and minimize adverse effects and drug-drug interactions, refining how clinical trials are run, and more, lead author Francesco Costa, MD, PhD (University Hospital Virgen de la Victoria, Málaga, and Instituto de Salud Carlos III, Madrid, Spain), and colleagues write in their paper published online last week in the European Heart Journal.

“Even if we want to be conservative, the impact of artificial intelligence is going to be tremendous,” Costa told TCTMD. “I personally don’t see any scenario in which, in the next 5 or 10 years, any aspect of what we do will not be impacted by these models because they are so diverse and they impact on so many different methods. They will have a transformative impact on every aspect of cardiovascular pharmacotherapy.”

Still, only a small number of AI models have gone through rigorous prospective evaluation followed by external validation, which will be needed before these approaches can be safely integrated into everyday practice.

“With respect to clinical medicine, I think that there are very few models that have been externally validated, so I think that none of them still is actually ready for prime time,” Costa said.

He said there has been “a lot of hype” around the use of AI in cardiology more generally. Indeed, AI has been evaluated in a variety of areas of cardiovascular medicine, including detection of certain conditions—like type 1 MI, low LVEF, and diabetes—using ECG readings and identification of decompensation in heart failure using patient voice recordings.

In the current review, Costa and his colleagues set out to summarize the evidence around AI specifically in the realm of cardiovascular pharmacotherapy, an area of interest to both interventional and general cardiologists. They examine studies across various therapeutic areas, including hypertension, diabetes, dyslipidemia, thrombosis, CAD, heart failure, and arrhythmias, and discuss the potential uses of AI in computer simulation models as well as drug discovery and repurposing.

It’s extremely important that we educate our peers on what is critical in interpreting the results of these models. Francesco Costa

“Artificial intelligence in a broader sense can impact cardiovascular pharmacotherapy, especially for giving precise treatment—to decide a specific treatment based on a patient’s phenotypic characteristics,” Costa said, adding that AI-based approaches may also prove useful for monitoring and improving adherence to treatment regimens.

Moreover, AI models may be able to improve clinical trials by adapting inclusion criteria to target those most likely to benefit, thereby reducing required sample sizes and costs, Costa et al say.

Currently, however, the greatest impact of AI for CV pharmacotherapy is being felt in drug discovery pathways, he said, noting that there are models that can predict how proteins will fold. “For this reason, it is going to be much easier to design new drugs, to develop new drugs, and to test them in clinical trials,” Costa said, adding that this will result in a greater number of effective medications that will be available to practicing clinicians.

Need for More Validation, Education

Moving forward, there is a great need to test AI models in randomized clinical trials, as has been done with drug and device therapies. “We do need to do the same for artificial intelligence models, to test them and demonstrate in a randomized setting that they are actually better than the standard of care, which is our current decision-making process,” Costa said.

In addition, there are key issues that need to be watched during the development of all AI models, regardless of whether they’re used in the field of CV pharmacotherapy or elsewhere, he said. He pointed to the need for high-quality data from diverse populations to train the models and external validation to ensure broad applicability.

Using data from a range of patient types will help ensure that biases within the healthcare system aren’t perpetuated with new AI models. “Otherwise, we’ll have models that apply to the mainstream population that do not apply to the specific patient that we have in front of us,” Costa said.

“On the other hand, what we have seen is that most of the articles that have been published and that have an impact on cardiovascular pharmacotherapy are not externally validated,” he added. “So basically the results apply very well to that specific population, but we don’t have enough data to confirm that these models actually work outside of that specific population, which is extremely important if we want to trust the results of these models.”

Within the medical community, scientific societies, and guidelines, there is still “very poor literacy” on these and other issues associated with AI, Costa said. Thus, “it’s extremely important that we educate our peers on what is critical in interpreting the results of these models, which is basically the quality of the data, the size of the training cohort, and the importance of validation and unbiased data,” he said. “That will give to our community the tools to understand if this model could apply or not into your clinical practice.”

And in future iterations of guidelines, Costa added, societies should give specific recommendations on which AI models are better than others to aid in clinical implementation.