AI vs. Expert Models in Predicting Heart Disease Deaths

Understanding Prognosis in Medical Treatment

When doctors plan treatment for patients, they must assess likely outcomes, often referred to as their “prognosis.” Statistical models assist healthcare professionals in evaluating a patient’s risk level and prognosis, enabling them to tailor the most suitable treatment options. Traditionally, these prognostic models have relied on expert medical research to pinpoint key risk factors associated with specific illnesses, a process that demands significant research effort and time.

Leveraging Electronic Health Records

With the rise in electronic health records, a vast amount of patient data is now accessible. Researchers at the Francis Crick Institute in London, UK, aimed to utilize artificial intelligence (AI) to analyze this data and create a “machine-generated” model for predicting the risk of heart disease-related deaths. Their findings were published in the journal PLOS ONE.

Research Methodology and Data Sources

The researchers accessed data from 80,000 patients via the CALIBER platform, which integrates four sources of routinely collected electronic health records in England: primary healthcare records, hospital discharge records, a national audit of myocardial ischemia, and death registrations. By employing a machine-learning approach, they developed an AI prognostic model for heart disease based on approximately 600 risk predictors.

Comparative Analysis of AI and Traditional Models

The accuracy of the AI model was compared with that of a traditional prognostic model developed by heart experts, which was based on 27 selected risk factors. The results indicated that the AI-derived model outperformed the conventionally derived model in predicting patient outcomes.

Novel Predictors Identified by AI

Interestingly, the AI model also identified new risk predictors, such as the frequency of home visits by doctors, which had not been recognized in traditional prognostic models. This factor could serve as a significant indicator of a patient’s health status.

Implications for Future Patient Care

This study highlights the feasibility of using machine learning to create prognostic tools. AI models can match the accuracy of conventional models while being less time-intensive to develop. This innovative approach holds the potential to enhance the development of models for various diseases and may significantly impact the management of patient care in the future.

References

Steele AJ, Denaxas SC, Shah AD, et al. Machine learning models in electronic health records can outperform conventional survival models for predicting patient mortality in coronary artery disease. PLOS One 13(8):e0202344. Doi:10.1371/journal.pone.0202344.
AI beats doctors at predicting heart disease deaths. The Francis Crick Institute EurekAlert https://www.eurekalert.org/pub_releases/2018-09/tfci-abd090418.php