A new software tool developed by Northwell Health will help the hospital network mine its data to forecast COVID-19 outbreaks weeks before they happen, hospital executives said.
The two-week advance warning system was created this summer by the customer insights group, in collaboration with information technology and clinical teams, in the wake of the COVID-19 surge that struck New York State’s largest health system last spring.
The new digital predictor collects 15 indicators and feeds them into a machine learning algorithm to recognize patterns in website traffic, which includes everything from emergency department wait time searches to physician page clicks. The result is a rolling two-week forecast that has closely mirrored caseload to date.
New Hyde Park-based Northwell said being able to anticipate a COVID-19 surge helps prepare staffing, supplies and patient handling.
"We have not had a tool that looks into the future," said Dr. Eric Cruzen, chief medical informatics officer of the Emergency Medicine Service Line at Northwell Health and chairman of the Emergency Department at Lenox Health Greenwich Village in Manhattan. "If alarm bells start to ring, we will pay attention."
Northwell, the largest health system in the state, treated about 85,000 patients, including 16,000 hospitalized patients, between March and Labor Day.
The encouraging news: The predictor "shows that at this point, everything looks stable as we look two weeks out," said Paul Lambson, corporate director of marketing and consumer insights, analytics, at Northwell Health.
Northwell has reported between 65 and 95 COVID-19 patients over the last six weeks at its hospitals.
Northwell’s predictive model examines the traffic volume of its website — more than 20 million page views since the first week of March — coupled with the geographic concentration of its 19 hospitals, 800 outpatient facilities and 52 urgent care centers across Long Island, New York City and Westchester. Its numbers have been fact-checked and correlate closely with the number of cases seen systemwide.
"We used all sorts of data and algorithms to see what had predictive value," Lambson said. "We found nuances. For example, we saw people shift from searching from doctor profiles to wait times. If they really needed care, we saw an increase in searches for hospital information."
Lambson said the data itself is proprietary, and will not be released publicly.
The predictor is one of many sets of data the health system uses to prepare for a COVID-19 increase, Cruzen said.
"I worry most about an uptick in viral syndrome presentation in our emergency departments, from fevers to low oxygen levels," Cruzen said. "We see those numbers in real time. I also pay attention to laboratory testing and because our health system is so large we can draw real conclusions from the data. But we have to be proactive instead of reactive."
Northwell said it plans to give away the source code for the predictive tool to other health systems.
"It's labor intensive, so it might not be easy for a health system to make it work with their website," Lambson said. "But we will show everyone what we did, and they can see if it'll work with their data."