Provider Spotlight: Faster Identification of Suspected Pancreatic Cancer Increases Access to Trials for Underrepresented Populations and Likelihood of Treatment

Byline:

Susan Beane醫學博士，第一保健執行醫療總監
Dr. Daniel King, Director of Research and Development, Northwell Health Cancer Institute's Center for Genomic Medicine

Susan Beane醫學博士導言

All types of cancer take their toll, but pancreatic cancer is especially aggressive.

According to the National Cancer Institute, the 5-year survival rate for pancreatic cancer is 13%, and just 3% if cancer has spread to distant parts of the body.
StatPearls describes the prognosis after one year of diagnosis as “dismal” — the majority of cases are advanced at the time of presentation, surgery is “complex and technically demanding,” and quality of life is significantly impacted.
Research from Ohio State University has found pancreatic cancer mortality rates are more than 6% higher in areas with a high social vulnerability index, suggesting language barriers, housing structures, and minority status impact the ability to receive necessary care.

There's a clear need for earlier identification and treatment of pancreatic cancer. We also need to remove the friction that typically comes with the many steps of the patient's journey - initial appointment, scan, biopsy, tumor board, oncology appointment, clinical trial, and ultimately treatment. As we know, the more steps required for treatment - the more visits to an unfamiliar facility on the other side of the city during the middle of the day, for example - the lower the odds we get patients the help they need.

Northwell Health found that only 1 in 4 patients with suspected pancreatic cancer would ultimately be diagnosed and treated for their disease. While many were diagnosed at such a late stage and in such poor health that treatment would be ineffective, others dropped off during the complex patient journey.

To address this issue, Northwell Health built an artificial intelligence model that could review imaging results and better identify patients at risk of a pancreatic cancer diagnosis. Critically, the health system didn't stop there. The team led by Dr. Daniel King, director of research and development at Northwell Health Cancer Institute's Center for Genomic Medicine, used the output of the analysis to reframe how Northwell reaches out to patients - and to think more broadly about AI's role in cancer care.

Look for patients who haven’t yet been seen
By Dr. Daniel King, director of research and development at Northwell Health Cancer Institute's Center for Genomic Medicine

Because pancreatic cancer is so aggressive, Northwell Health has dedicated resources to identify patients with the condition sooner. This increases the likelihood that patients will be able to receive treatment and experience a better quality of life following an early-stage diagnosis.

We started with the common radiology database in use at our health system's 21 hospitals and 10 outpatient cancer centers. We decided to look at the preliminary imaging scans that patients received prior to getting a biopsy to assess the extent of the cancer. We built a tool called iNav, leveraging an open-source natural language processing product from Google called BERT, to review the reports that accompany scans.

Our study looked at the 20,000 abdominal imaging scans done across our system in January 2023, which iNav was able to do in seconds. From there, iNav flagged 149 of the scans with suspected pancreatic cancer; for example, the system would identify terms such as "mass in the pancreas" as part of the radiology report. After clinical staff conducted chart review, we found 67 patients who had never seen an oncologist. That's more than two per day.

Then, we reviewed the data six months later to see whether the clinical workflows in place at the time had given these patients the care they needed. Of the 67 patients, only 11 got a biopsy, saw an oncologist, and started treatment. It took these patients an average of 56 days - almost two months - to get treated. Our health system had long suspected that pancreatic cancer care delivery was inconsistent and slow. The iNav analysis confirmed it.

Using data to improve care navigation

Instead of putting these findings on a shelf, our next step was to link iNav to clinical care. Every morning, iNav looks over every scan done across Northwell Health in the last 24 hours. When iNav flags someone with suspected pancreatic cancer, we connect them to a care navigator to ensure prompt follow-up with next steps. (As noted above, this typically consists of diagnostic biopsy, tumor board review, oncologist visit, and clinical trial screening.)

Since we've put this AI-assisted workflow in place, the time it takes for patients to receive a referral has been cut in half. Additionally, the number of patients with pancreatic cancer participating in clinical research has tripled, which has helped our health system increase access to trials for New York City's underrepresented populations.

A big part of the reason for this success, we feel, is that we worked to align the handoffs from iNav with our existing patterns of care. We integrated the alerts from iNav into existing workflows and job responsibilities to facilitate the care navigator's work, not conflict with it. We also publicly recognized the care navigators who were using results of iNav analysis to complement patient outreach, and this recognition has helped to encourage additional participation.

We are also proud that iNav was selected as one of the Best Inventions of 2024 in TIME.

Our goal with iNav is to ensure that no patient gets left behind. About 25% of the patients who receive imaging scans at Northwell don't have primary care physicians at Northwell. We still share the results with their physicians and encourage them to get care, even if they ultimately go somewhere else. We recognize the potential for our AI tool to help patients regardless of where they are - and we know patients are better off when they have access to the information that can guide them to the care they need.