Beyond average-based medicine: HIE as a blueprint for data-informed care

Historically, outcome prediction in medicine has followed a familiar formula: run a clinical trial, publish the results, guide care based on averages. The model has served for decades, despite its limits.

In neonatal care, where decisions can carry lifelong consequences, averages are often insufficient. That’s why Ellen Grant, MD, MSc, director of the Fetal Neonatal Neuroimaging and Developmental Science Center at Boston Children’s Hospital, along with Yangming Ou, PhD, and their team, are challenging the model. They’re doing this by using big data to reimagine how outcomes are predicted, starting with hypoxic-ischemic encephalopathy (HIE), a brain injury caused by oxygen deprivation at birth.

“Published papers combine patients into groups and results are aggregated, so it’s difficult to extract answers to more granular questions,” Grant says. “We want to make the data answer the specific question in front of us.”

The need for more specifics

In practice, clinicians often rely on averages — even when sitting with a newborn whose clinical presentation involves dozens of interacting variables. A study may report overall outcomes for a group of infants with a similar condition, but it can’t isolate those who share a specific combination of variables, such as gestational age, MRI findings, and maternal history. The raw, patient-level detail exists in the original trials, but once findings are summarized for publication, much of that nuance is lost. To remedy this, Grant’s team decided to go back to the source.

Harmonizing data and building a platform

With funding from NIH, the group obtained the complete datasets — every recorded variable — from two major HIE clinical trials. The scale was immense: more than 1,000 clinical data elements per infant, across roughly 500 patients treated at 21 U.S. sites.

But gathering the data was only the beginning. Variables were coded differently, definitions varied, imaging descriptions didn’t always align. Before the team could analyze anything, they had to “harmonize” the datasets — standardize terminology, reconcile coding differences, and structure the data so each data point meant the same thing across sites.

The harmonized database then became a platform for precision prediction. The goal: rather than asking what happens “on average,” clinicians can find out what has happened in infants who most closely resemble their current patient. The system can incorporate evolving information: starting with early delivery-room data such as Apgar scores and blood gases, then update projections once MRI findings and additional clinical details become available.

Early analyses focused on 52 carefully selected variables spanning maternal health, physiologic measures, neuroimaging findings, and more. Using machine learning, the team determined which combinations of factors best predicted adverse and non-adverse outcomes. They developed an online calculator in which clinicians input approximately 10 variables to generate, within seconds, an individualized 0–100 percent predicted risk of neurocognitive deficits by 2 years of age. Overall predictive accuracy approached 90 percent.

“Our work aligns with a broader shift in medicine,” Grant says. “Using large, curated datasets to produce a ‘digital twin’ of each patient, we can find a group of prior patients whose profiles closely resemble the infant being treated.”

From trials to bedside tools

The team also developed a specialized chatbot built on the harmonized database that allows clinicians to ask patient-specific questions and receive evidence-based answers.

Grant, Ou, and their team are showing that when patient-level data is carefully curated and paired with AI, outcome prediction can move beyond averages to deliver precise, individualized insight and care.

“We’re treating clinical trial data as fluid, not static,” Grant says. “We’re turning trial information into tools that help clinicians better understand each individual patient in real time and have clearer, more informed discussions with families.”

Explore research taking place in the Fetal Neonatal Neuroimaging and Developmental Science Center.