Machine Learning Model Improves Mortality Risk Prediction for Cardiac Surgery Patients

In cardiac surgery, risk scores provided by The Society of Thoracic Surgeons (STS) are often used to evaluate a patient's procedural risk. While these scores remain vital for hospitals to assess and improve their performance, they are drawn from population-wide data, which can fall short of accurately predicting risk for specific patients with complex pathologies.

Now, cardiovascular surgeons and data science specialists at Mount Sinai (New York, NY, USA) have developed a machine learning-based model that predicts mortality risk for individual cardiac surgery patients, offering a considerable performance advantage over current population-based models. This data-driven algorithm, built on extensive electronic health records (EHR), is the first institution-specific model of its kind for pre-surgery cardiac patient risk assessment. It allows healthcare providers to determine the optimal treatment strategy for each patient.

The team theorized that models based on EHR data from their own institution, created via machine learning, could provide a useful solution. Using routinely gathered EHR data, they developed a robust machine learning framework to generate a risk prediction model for post-surgery mortality that is customized to both the patient and the hospital. This model incorporates vital data about Mount Sinai’s patient population, including demographic, socioeconomic, and health characteristics. This is in contrast to population-based models like STS, which rely on data from various health systems across the U.S. The effectiveness of this approach is further enhanced by an efficient open-source prediction algorithm called XGBoost, which assembles a group of decision trees by progressively focusing on harder-to-predict segments of training data.

The research team utilized XGBoost to model 6,392 cardiac surgeries conducted at The Mount Sinai Hospital from 2011 to 2016, encompassing heart valve procedures, coronary artery bypass grafts, aortic resections, replacements, or anastomoses, and reoperative cardiac surgeries, which significantly increase mortality risk. The team then compared the performance of their model to STS models for the same patient sets. The study found that the XGBoost model outshone STS risk scores for mortality in all frequently performed cardiac surgery categories for which STS scores were designed. The predictive performance of the XGBoost model across all types of surgeries was also high, indicating the potential of machine learning and EHR data for constructing effective institution-specific models.

“The standard-of-care risk models used today are limited by their applicability to specific types of surgeries, leaving out significant numbers of patients undergoing complex or combination procedures for which no models exist,” said senior author Ravi Iyengar, PhD, the Dorothy H. and Lewis Rosenstiel Professor of Pharmacological Sciences at the Icahn School of Medicine at Mount Sinai, and Director of the Mount Sinai Institute for Systems Biomedicine. “Our team rigorously combined electronic health record data and machine learning methods to demonstrate for the first time how individual institutions can build their own risk models for post-cardiac surgery mortality.”

Related Links:
Mount Sinai