Ai-Powered Study Prioritization

The Customer: 

Located thirty minutes north of New York City, Montefiore Nyack Hospital is a 391-bed community hospital providing emergency and acute care services to the residents of Rockland County and surrounding areas.

This Level II Trauma Center handles approximately 60,000 emergency room visits per year, and is 1 of 10 member hospitals in the nationally recognized Montefiore Health System. Since its founding in 1895, Montefiore Nyack has focused on practicing consistent, innovative medical care.

The Challenge: Moving Urgent Studies to the Front of the Queue

The Radiology Department at Montefiore Nyack Hospital is committed to continuous quality improvement. The team of radiologists reads more than 100,000 studies annually, and is always searching for ways to improve productivity without sacrificing quality.

A long-term customer of Change Healthcare, Montefiore Nyack first implemented Change Healthcare Radiology Solutions™ in 2009.

In 2015, the hospital served as the beta site for the Change Healthcare Workflow Intelligence™ solution, an imaging workflow rules engine designed to balance workload. The solution features a universal worklist that incorporates all of a radiologist’s tasks in one place for greater productivity.

“Our worklist prioritization was initially based on exam metadata,” explains Dr. Evan Kaminer, Director of Radiology at Montefiore Nyack Hospital. “How old was the patient? Where is the patient located? Inpatient or outpatient? When was the scan done?”

In partnership with Change Healthcare, the hospital built a detailed prioritization model that used data elements, such as referring doctor and exam type, to rank studies in the order in which they should be read. The new model allowed Montefiore Nyack to improve its Emergency Department (ED) turnaround times by 27% over the first three months.

The hospital still sought a reliable way to triage its ED cases. When the hospital’s leadership heard that Change Healthcare had partnered with Aidoc, a pioneer of medical artificial intelligence (AI), they were interested in how an AI-driven workflow could expedite care for critical patients.

The Solution: AI-powered Worklist Flags Positive Findings for Immediate Attention

In April 2019, Montefiore Nyack was the first hospital to implement Change Healthcare Workflow Intelligence 3.0, which offers AI decision support capabilities via Aidoc integration. The solution uses AI algorithms to scan diagnostic images for specific clinical findings, such as analyzing CT Head exams for anomalies related to intercranial hemorrhages.

After a scan is performed, the image is sent to an on-premises server, which anonymizes the data before transmitting it to Aidoc. The AI algorithms analyze the data for specified study types, sending the results back before Montefiore Nyack’s radiologists have read the study. If the platform detects any abnormalities, the exam is escalated to the top of the radiologists’ worklist.

“For the first time, with AI, we can use the results of what’s actually on the image to prioritize our studies,” says Dr. Kaminer. “If our goal is to read the most important study first, what’s more important than a study that has a positive finding?”

In keeping with its goal of triaging ED cases, Montefiore Nyack has implemented three AI algorithms designed to detect specific findings— intracranial hemorrhage, pulmonary embolism, and CT cervical spine fracture.

“We also have what’s called incidental pulmonary embolism, in which AI scans all of the chest CTs to see if there’s a pulmonary embolism that the radiologist may not have been looking for,” explains Dr. Kaminer.

At first, Montefiore Nyack’s leadership was concerned about the radiologists’ reaction to the new AI-driven workflow. Given the industry hype about AI’s potential impact, they didn’t want radiologists to feel undermined. Fortunately, the AI algorithms soon proved their worth.

Streamlined Workflow Speeds Results

Initially, Montefiore Nyack struggled with long wait times, as the studies were slow to be read by the AI algorithm. When the department realized their turnaround time benchmarks were at risk, they worked directly with Aidoc to fix the issue.

“If the AI takes 10 minutes or more to detect results, the radiologists’ productivity is negatively impacted, because they’re waiting for the result to come back,” explains Dr. Kaminer.

With a few adjustments to the server allocation and workflow, the system is now functioning much more smoothly. “We worked with Aidoc to reduce that time to four minutes. By the time the radiologist is ready to read it, the results are there or very close,” says Dr. Kaminer.

The Results: AI-Enabled Prioritization Improves Turnaround Times for Positive Findings

Now that the solution has been in place for a few months, Montefiore Nyack is beginning to quantify its results. As its main goal was to read positive studies first, the Radiology Department compared its turnaround time for reports marked positive by AI versus reports marked negative.

“We found that a study with an AI-positive finding had a turnaround time of 15 minutes, which includes getting the ER physician on the phone to deliver results,” says Dr. Kaminer. “That compares to 18 minutes for an AI-negative study.”

To an outsider, this 17% increase in speed may not seem impressive, but Dr. Kaminer believes the results are significant. Reading a positive brain study takes longer than reading a negative study, as the radiologist must describe the findings and interpret the results.

Since the 15-minute window also includes contacting the ER physician—an unnecessary step for negative studies—the 17% increase in turnaround time represents considerable time savings in the communication of urgent abnormalities. After all, for life-threatening conditions, every second counts. “While that three minutes may not seem like a lot, it could be critical for a patient who has a severe bleed,” says Dr. Kaminer.

As one of the Radiology Department’s biggest clients, the ED is pleased with the faster service. “The emergency room is thrilled that we’re using AI to scan their patients. They want the positive results the soonest, and that’s exactly what we achieve for them,” says Dr. Kaminer.

Automatic Double-Reading Improves Outcomes and Helps Reduce Physician Stress

The solution is also impacting clinical outcomes for Montefiore Nyack’s patient population. In one month alone, Aidoc flagged 77 patients with acute intracranial hemorrhage—some with subtle bleeds that might have been missed by radiologists.

“We’ve known for a long time that double-reading radiology images improves quality,” says Dr. Kaminer. “But it was never cost-effective. It was too expensive to have two radiologists look at each study.”

When one of those radiologists is a computer, however, double-reading becomes financially feasible. “Now, AI can augment the radiologist’s review to improve quality, increase accuracy, and ultimately produce better outcomes for our patients,” says Dr. Kaminer.

As Montefiore Nyack is a community hospital, sub-specialists are not always on the schedule. In many instances, radiologists read outside their specialty. “Sometimes, it causes a lot of stress for radiologists to read something that they’re not fellowship-trained in,” explains Dr. Kaminer. “The AI is a safety net. Having that backup over-read in real time has been a great benefit.”

Many of the doctors at Montefiore Nyack report that the solution has decreased their overall stress—a recognized component of physician burnout. While the financial impact of AI is still under review, Dr. Kaminer suggests that the new technology supports improved productivity by increasing diagnostic confidence.

“Some research shows that radiologists are more efficient with AI. They can read studies just a little bit faster, because the AI has their back,” says Dr. Kaminer. “That increased efficiency can translate into cost savings for the hospital.”