At the height of the pediatric tripledemic in 2023, 12‑year‑old Molly left Children’s National Hospital after waiting four hours in the emergency department. Forty patients in front of her were still waiting to be seen.
Later that day, she was back in the ED, except this time she arrived by ambulance.
Confused, breathing fast, and critically ill, Molly had undiagnosed diabetes, dangerously high blood sugar, and complications that landed her in the ICU for days.
She recovered, but her case stayed with the team.
“Had she been seen on that first visit,” said Sarah Isbey, MD, emergency medicine physician, “she would have received care before she got as sick as she did.”
Children’s National sees more than 96,000 emergency visits each year. On the busiest days — when more than 240 patients checked in — nearly one in five children left without being seen.
“For providers, that was heartbreaking,” Isbey said. “It felt like we were failing our patients.”
After Molly’s case, the team asked themselves: How can we stop this from happening to other kids?
What resulted was a technological innovation that led to immediate improvement.
After three years, overall left-without-being seen (LWBS) rates had dropped 70%. On the busiest days, the rate fell 60%.
Prediction leads to prevention
Using years of emergency department data, the Children’s National team built a machine‑learning model that predicts when high‑acuity patients are likely to leave without being seen — hours before it happens.
Specifically, it predicts when two or more high‑acuity children (ESI 1–3) are likely to leave in the next eight hours, which enables the hospital to align forecasting with typical shift schedules and real staffing decisions.
“The model ingests real-time data to create actionable predictions that allow operations leaders to consistently bring in extra resources when they are needed without wasting scarce resources,” said Kenneth McKinley, MD, an emergency medicine physician who helped develop the model along with emergency medicine fellow Brandon Kappy, MD.
The model relies only on data hospitals already have: arrivals, current census, waiting times, boarding burden, time of day, day of week, and components of standard measures like the National Emergency Department Overcrowding Scale (NEDOCS).
McKinley tested several approaches and found advanced models like gradient boosting (XGBoost) consistently outperformed traditional crowding measures alone.
"This isn’t about algorithms. It’s about giving kids a real chance to be seen when they need us most."
In plain terms, the model learned how small changes in volume, crowding, and timing combine — often subtly — to signal when the ED is about to fail its sickest patients.
Predictions are generated hourly. When risk crosses a threshold, operational leaders get an email or text. Then a surge team, consisting of a nurse and physician, is activated and sent to a dedicated space near the waiting room.
Four on-call surge teams are built into the schedule: daytime, evening, overnight, and fast track.
“This shift is really gratifying to us physicians because we know we’re helping people be seen quicker,” Isbey said. “It gives an objective reason for being on call instead of someone’s gut feeling.”
When McKinley was called in for a surge shift based on the algorithm’s prediction, he encountered a 6‑year‑old boy who had been waiting a couple hours and previously would have been at risk of leaving. Instead, McKinley discovered a dangerous intracranial bleed and referred him immediately to surgery. He made a full recovery.
“I don’t know how long he would have waited otherwise,” McKinley said. “But I know we wouldn’t have found him when we did.”
Taking the work further
With LWBS rates down and early results showing fewer high‑acuity children leaving without care, the team is focused on what comes next: expanding predictive tools, refining triage, and exploring how data can help identify children at risk of deterioration even earlier.
The work is also moving beyond a single site. The team is already collaborating with other children’s hospitals to test whether the approach works in different environments, with different volumes and staffing models.
For the clinicians doing the work, the impact is personal.
“This isn’t about algorithms,” Isbey said. “It’s about giving kids a real chance to be seen when they need us most.”
