Artificial intelligence (AI) is usually applied to analyze a complex set of variables to make correlations that are not easily made by unsupported observation. But the branch of artificial intelligence, sometimes referred to as causal AI, includes not only causation, and appears to have the potential to change the cardiovascular (CV) event prevention paradigm.
Ference, Director of Research in Translational Therapies, University of Cambridge (England): “Causal AI is a new generation of AI algorithms that enables AI to go beyond expectations to help guide each individual’s clinical decision making.” .
In a new study testing this hypothesis, called CAUSAL AI, this approach is explored with two major risk factors, high LDL cholesterol (LDL-C) and high systolic blood pressure (SBP). Based on a deep learning algorithm that examined the effect of these risk factors on the biology of atherosclerosis, the causal effects of these risk factors were evaluated and then included in the risk estimate.
Causal AI can predict treatment effect
The study showed that the accuracy of risk prediction can be significantly improved using causal AI, but more importantly it indicates that causal AI can predict the effect of specific actions to reduce this risk in the context of a patient’s path toward autobiographical events.
“Risk estimation algorithms are used to select high-risk patients who might benefit from interventions to reduce risk, but they do not include causal effects of changes in LDL-C and SBP,” explained Dr. Ferenc.
As a result, they “may not accurately estimate the underlying risk of cardiovascular events from LDL-C or SBP level or the benefit of treating these risk factors,” he added.
In the CAUSAL AI study, presented at the annual conference of the European Society of Cardiology, risk prediction embedded with causal AI demonstrated the ability to match predicted events with actual events in several large patient data sets.
“Inclusion of causal effects in risk estimation algorithms accurately estimates baseline cardiovascular risk from LDL and SBP and the benefit of lowering LDL, SBP, or both starting at any age and extending for any duration,” said Dr. Ferenc.
AI in deep learning has evaluated more than 300 genetic variants
The AI for deep learning was based on randomized Mendelian studies evaluating 140 genetic variants associated with LDL-C and 202 variants associated with SBP.
In one test of the predictive impact of causal AI, risk prediction was first performed in 445,771 UK Biobank participants with the British Joint Societies Risk Calculator (JBS3). For actual events in this population, JBS3 alone “consistently reduced the increased risk of elevated LDL, blood pressure, or both” over the patient’s lifetime, according to Dr. Ferenc.
It also systematically overestimated the risk of cardiovascular events among participants with low LDL-C, blood pressure, or both.
However, after including the causal effect of LDL and blood pressure, “the same algorithm was able to accurately predict cardiovascular disease risk,” Dr. Ferenc said. The improved accuracy resulted in “observable and approximately predictable event curves that are synthesized over time.”
Built-in causal effects accurately predict outcomes
Causal AI, embedded in risk analyses, was also able to correct for inaccurate risks benefits derived from short-term clinical trials. These also “systematically reduce the benefit of lowering LDL, blood pressure, or both,” according to Dr. Ferenc.
By contrast, after including the causal effects of LDL and blood pressure in the algorithm, the same algorithm accurately predicted the benefit of lowering LDL, blood pressure, or both at each age, again producing the observed and predicted event curves.
In another evaluation by Dr. Ferenc and the co-investigators, the JBS3 algorithm was applied to several major trials, such as the Heart Protection trial and HOPE-3. By itself, the JBS3 algorithm expected less benefit than what was actually observed.
“After including the causal effects of LDL and blood pressure, the same algorithm was able to accurately predict the benefit of lowering LDL, blood pressure, or both that was observed in trials 3-5 years later,” Dr. Ferenc reported.
In sensitivity analysis, prediction accuracy remained largely similar across strata by risk factors, such as male sex, presence of diabetes, family history of cardiovascular disease, and other variables. It was also similar across participants’ age before the cardiovascular event and all follow-up periods.
The data provided by Dr Ferenc provide compelling evidence that JBS3, which is widely used in the UK for risk assessments, does not accurately estimate cardiovascular risk from elevated LDL or SBP. It also fails to appreciate the benefit of treating these risk factors.
“Therefore, they cannot be used to determine the optimal timing, intensity, and duration of treatment for the prevention of cardiovascular events,” said Dr. Ferenc.
By including the causal effects of LDL-C and blood pressure through an AI-based algorithm, treatment benefit can be accurately estimated “starting at any age and continuing for any duration, thus providing essential information to inform individual treatment decisions regarding timing, intensity, and duration,” according to Dr. Ference.
Routine application awaits further steps
Despite the promise of the concept, there are many steps that need to be taken before bringing it into the clinic, emphasized co-designate Volkert Asselbergs, MD, PhD. In addition to testing accuracy in multiple populations, “we have to do experiments as well,” meaning that future evaluations to validate the concept are useful for improving results.
However, he does not doubt that the concept of causal AI is promising and likely to have a measurable impact on heart disease after further validation.
“Causal AI is a critical step we must take for more efficient healthcare,” he said. One reason for his caution is that many of the AI-enhanced risk scores, while not necessarily AI causal, showed only “modest predictive value” in many of the studies he cited.
“We hope that the data provided by the CAUSAL AI study will really help us take a step into the discussion to see how we can really benefit by including genetic information in the AI framework to include causality in predicting risks and predicting benefits of treatment,” said Dr. Asselbergs, Professor of Precision Medicine, Utrecht University Medical Center (Netherlands).
Dr. Ferenc reported having financial relationships with more than 15 pharmaceutical companies. Dr. Asselbergs reported no potential conflict of interest.
This article originally appeared MDedge.compart of the Medscape Professional Network.