AI Predicts Stroke Risk in CF-LVAD Patients

(Image credit: Nnibedita)

Harnessing AI to Predict Stroke Risk in Heart Failure Patients with CF-LVAD Implants

In recent years, artificial intelligence (AI) has made remarkable inroads in healthcare, from diagnostics to personalized treatment planning. One particularly groundbreaking development lies in its potential to predict stroke risk in patients with advanced heart failure who rely on a continuous-flow left ventricular assist device (CF-LVAD). While CF-LVADs have transformed heart failure treatment, improving patients’ quality of life and longevity, they come with risks — chief among them is stroke, a persistent complication despite continuous technological improvements. Researchers at Baylor College of Medicine are exploring how AI, combined with biomolecular research, might change that.

“Early detection and management of post-CF-LVAD strokes are crucial for improving patient outcomes. Our study aimed to identify a predictive model for stroke risk following CF-LVAD implantation,” said Dr. Nandan Mondal, co-author and assistant professor in the Michael E. DeBakey Department of Surgery at Baylor College of Medicine.

A Deeper Dive into CF-LVADs and Stroke Complications

CF-LVADs serve as a lifeline for many individuals with end-stage heart failure, functioning by mechanically pumping blood from the left ventricle to the rest of the body. Unlike previous generations of assistive devices, CF-LVADs operate continuously, providing a steady blood flow that supports heart function. For patients who are ineligible for heart transplants, this implant has often been the difference between life and death. However, with this benefit comes the trade-off: an increased risk of thromboembolic events, primarily stroke, stemming from blood clot formation.

Despite enhancements in anticoagulation therapy, stroke remains a significant risk for CF-LVAD patients. The challenge has been that predicting which patients are at a higher risk is difficult, as strokes can result from a complex interplay of individual patient factors, device function, and biochemical responses. To tackle this, Dr. Mondal’s team explored a novel approach by analyzing mitochondrial protein levels in blood cells to identify indicators of heightened stroke risk.

The Role of Mitochondria and OxPhos Proteins in Stroke Risk

Mitochondria, often called the “powerhouses” of cells, play a critical role in maintaining cellular function by producing energy through oxidative phosphorylation, or OxPhos. Interestingly, previous research has suggested a link between mitochondrial dysfunction and stroke. Specifically, a reduction in OxPhos protein levels in white blood cells has been associated with stroke severity in patients with various cardiovascular conditions.

Given this background, Dr. Mondal and his team hypothesized that OxPhos protein levels in white blood cells might hold predictive value for stroke risk in CF-LVAD patients. Their study, published in the ASAIO Journal, examined whether fluctuations in these protein levels could indicate an increased risk of stroke following the implantation of the device.

Developing the Study: A Machine Learning Approach

To test this hypothesis, the research team recruited a group of 50 patients who had received CF-LVADs. They divided these patients into two groups: those who had experienced a stroke before the device was implanted and those who had not. Blood samples were collected from all participants before and after the CF-LVAD implantation, and OxPhos protein levels in white blood cells were measured.

This study generated an extensive dataset — analyzing mitochondrial proteins can produce hundreds of data points per patient. To manage and interpret this data, Dr. Mondal’s team incorporated machine learning. Jacob P. Scioscia, a medical student in Dr. Mondal’s lab, played a key role in this phase of the study, using machine learning algorithms to sift through the data and identify patterns associated with stroke risk.

“Dr. Mondal was seeking students with experience in machine learning, and I was particularly drawn to the cardiothoracic surgery field,” Scioscia said. “I saw this as an opportunity to gain hands-on experience with data analysis in a high-stakes medical research environment.” Scioscia joined the project through Baylor’s SOAR (Student Opportunities for Advancement in Research) database, which connects students with faculty research opportunities.

Discovering New Predictive Indicators for Stroke

The machine learning models identified six prognostic factors within the dataset that were associated with stroke risk. Notably, they found that patients who had experienced a stroke prior to CF-LVAD implantation had lower levels of OxPhos proteins, both before and after the device was implanted, compared to patients who had no history of stroke. Furthermore, the data indicated a notable drop in OxPhos protein levels in patients who suffered a stroke post-implantation, suggesting that mitochondrial health, as reflected by OxPhos levels, could be an important predictive factor for stroke risk in these patients.

“We discovered that patients who had a stroke both before and after CF-LVAD implantation exhibited consistently lower OxPhos protein levels,” explained Dr. Mondal. “This suggests that reduced mitochondrial function could be an underlying factor contributing to stroke vulnerability in CF-LVAD patients.”

A New Frontier in Predictive Healthcare for Heart Failure Patients

The Baylor study provides a new perspective on the potential of mitochondrial dysfunction as an indicator of stroke risk, opening the door to early, personalized interventions for patients with CF-LVADs. Unlike other forms of stroke risk assessment, which might focus on patient history or device performance, this research proposes a biomolecular approach that could give clinicians a powerful new tool. By measuring OxPhos levels in white blood cells, clinicians may be able to identify patients who are at a heightened risk for stroke, allowing them to tailor postoperative care accordingly.

“This combination of laboratory-based results with machine learning allows for a level of predictability that we haven’t previously seen in this area of cardiology,” noted Scioscia. “Working with Dr. Mondal and the team to analyze the data and witness how we can apply these findings has been incredibly rewarding.”

While the team’s findings are promising, further research is essential to refine the predictive power of OxPhos proteins as stroke indicators.