When artificial intelligence enters health care, it doesn’t just meet complexity—it meets history: systems shaped by uneven access, legacy data, and urgent needs. That intersection was the focus of the third panel at Data Science Day 2025, hosted by the Data Science Institute.
The session, called the Science of Self Care: Innovations in Consumer Health, brought together researchers developing AI systems not to replace clinical judgment, but to support it—while keeping equity, context, and human experience at the center.
Read 5 Insights from Data Science Day 2025
Takeaways from the Discussion
1. Chronic Conditions Need Better Data—And New Definitions of Evidence
Noémie Elhadad, Associate Professor and Chair of Biomedical Informatics at the Vagelos College of Physicians and Surgeons, described her lab’s work on endometriosis—an underdiagnosed condition affecting 1 in 10 women. Traditional datasets, she said, don’t capture the daily realities of living with the disease.
Her team’s solution: the Phendo app, which gathers patient-reported data to support detection, care, and self-management.
“We’re not just modeling the disease,” she said. “We’re modeling how people live with it.”
2. Fairness Can Start with Staffing
Carri Chan, John A. Howard Professor of Business at Columbia Business School, focused on resource allocation in emergency departments—where wait times, staffing levels, and patient outcomes are tightly linked.
Her team builds predictive models that use real-time and historical data to adjust nurse staffing dynamically.
“Hospitals make trade-offs every day,” she said. “Data-driven tools help us make those trade-offs more visibly—and more fairly.”
3. Causality Reveals Structural Inequity
Drago Plecko, Postdoctoral Research Scientist in the Department of Computer Science at Columbia Engineering, presented a framework for causal fairness analysis. His work focuses on identifying not just disparities, but the mechanisms behind them.
In one study, his team found that Indigenous patients in Australia were more likely to be admitted to ICUs—not because they were sicker, but because they lacked access to primary care.
“Causal tools help us move from ‘Is this biased?’ to ‘What’s driving the disparity?’” he said.
His team’s IICE Radar system tracks these patterns geographically to support targeted public health responses.
4. Models Don’t Operate in Isolation
Lena Mamykina, Associate Professor of Biomedical Informatics and the panel’s moderator, closed with a reminder about context:
“These models don’t live in a vacuum,” she said. “They live in institutions—with policies, workflows, and histories. And those systems need care, too.”
This panel offered a look at how AI can support—not sidestep—the complexity of healthcare. What emerged was a vision of alignment: between models and real-world context, between prediction and care, between systems and the people they serve.
At the Columbia University Data Science Institute, that kind of alignment shapes the work from the beginning.
