Abbas Zaidi, MD, a pediatric cardiologist at Wilmington-based Nemours Children’s Health Delaware Valley, has developed an AI model that successfully predicted more than three-quarters of care gaps for pediatric congenital heart disease. He presented the research at the American Heart Association’s annual meeting in November.
Dr. Zaidi shared with Becker’s how the tool could be integrated into clinical workflows and other opportunities for the use of AI in cardiovascular care.
Editor’s note: Responses have been lightly edited for clarity and length.
Question: You presented research at the American Heart Association’s annual meeting on an AI-based model designed to predict which pediatric congenital heart disease patients are most at risk of experiencing gaps in care. What inspired this work?
Dr. Abbas Zaidi: This project was inspired by a simple yet critical question: Can we utilize AI to help ensure that no child with congenital heart disease falls through the cracks? Children with congenital heart disease require lifelong follow-up, yet many unintentionally fall out of regular cardiology care. We see this across the country for a multitude of reasons, for example, families move, their insurance coverage changes, transportation becomes difficult, or adolescents disengage during the transition to adulthood. These lapses can delay critical imaging, medication adjustments, or timely interventions.
Our team wanted to shift from reactive to proactive care by identifying which patients were most at risk before a gap occurred. We focused specifically on clinically meaningful lapses – situations where a child missed recommended cardiology follow-up beyond the safe interval. These “true gaps in care” can result in missed diagnoses, worsening symptoms, and avoidable emergency care.
Q: Can you walk us through how the machine learning model was developed? How might the model be used in a real-world pediatric cardiology setting?
AZ: We built a dataset that combined clinical complexity, demographic information, geographic distance to care, and insurance patterns. After cleaning and preparing the data, we tested several machine learning approaches. One approach to AI modeling, called the Balanced Random Forest model, performed the best. It is designed to handle the imbalance between common “no-gap” cases and less frequent true lapses.
The model achieved strong performance, and correctly identified 76% of patients who ultimately experienced a true gap in care. The most important predictors were age, distance from a Nemours Children’s Health facility, and anatomic complexity, highlighting how both medical and social factors shape and impact care continuity. In practice, this tool could function behind the scenes in the electronic health record. Care teams might receive a weekly or monthly list of patients flagged as “high risk for a future gap.” This would allow earlier outreach, telehealth scheduling, help with insurance renewal, or navigation support for families who face transportation or socioeconomic barriers. The goal is not to replace clinical judgment, but to give health systems and providers an early-warning signal so every child stays connected to the care they need.
Q: Are there any emerging trends or challenges in cardiology that you think deserve more attention from hospital and health system leaders?
AZ: Yes. Three major themes emerged from this work:
- Adolescents and young adults are the most vulnerable to falling out of care. Transition-age patients consistently show the highest risk. Health systems should invest in structured transition programs, navigation support, and care pathways that take into account young adult preferences.
- Social drivers of health have a major impact on cardiology outcomes. Geographic distance, transportation, and insurance instability are powerful drivers of care gaps. Predictive analytics make the impact of these inequities more visible. Resources should be dedicated to addressing them, through services such as mobile care, telehealth access, and enhanced patient navigation.
- AI is becoming essential, but must be used ethically and strategically. Predictive tools can transform care, but only when paired with human support systems. The purpose is not to label families; it is to identify needs early and intervene compassionately.
Leave a Reply