Using artificial intelligence and electronic health records to predict cardiovascular outcomes
Scientists from the University of Utah (UT, USA) used Poisson Binomial-based comorbidity (PBC) to search electronic h...
Scientists from the University of Utah (UT, USA) used Poisson Binomial-based comorbidity (PBC) to search electronic health records (EHRs) for any comorbid diagnoses, medications and procedures underlying cardiovascular health outcomes.
The research, published in PLOS Digital Health, utilized the EHRs of over 1.6 million patients with diverse cardiovascular outcomes, focusing on the following three: congenital heart disease, sinoatrial node dysfunction and heart transplant. The results highlighted that PBC was effective in understanding the demographic factors and comorbidity of these key cardiovascular outcomes.
The researchers believe their approach of using artificial intelligence (AI) and EHRs can demonstrate healthcare disparities present in a specific health care system.
They sought to understand the relationships between outcomes and multiple clinical variables, which is why the AI solution approach was used. Since multimorbidity networks can only model up to 30 variables at once, the team had a unique advantage of being able to pre-identify any high impact variables through PBC.
The authors agree that their approach has its limitations as they are only able to model 30 health conditions at a time. The researchers suggested relaxing this limiting factor by allowing approximate solutions in the future. This will enable the multimorbidity network to scale up its complexity to thousands of other health conditions.
Recommended for you:
Another area to improve upon is the incorporation of continuous variables. Despite current software packages not permitting this, the scientists explained that theoretically there is no limitation preventing the use of continuous variables in a probabilistic graphic model framework.
The results of this study elucidate how multimorbidity networks provide a solution for understanding the impacts of medical procedures, diagnosis and medications on cardiovascular outcomes.
The analyses presented can provide the first steps in generating a global description of heart disease and associated comorbidities across the USA. The scientists have shown that given the right datasets, this approach can provide new insights, such as a mother-child cross-generational cardiovascular multimorbidity.
The team have made the multimorbidity network available online to help the scientific community in producing high-quality outcomes research.