Al-assisted navigation increases referrals for suspected pancreatic cancer
Click Here to Manage Email Alerts
Key takeaways:
- AI-guided navigation workflow shortened time to biopsy and start of treatment.
- The monthly average of patients being approached for biospecimen studies more than doubled.
CHICAGO — An AI-guided navigation workflow helped identify patients earlier in the diagnostic timeline than the traditional cancer referral process while reducing times to biopsy, follow-up and treatment initiation, study results showed.
Researchers, who presented their findings at ASCO Annual Meeting, also noted that implementation of the AI-guided navigation workflow increased patient referrals to pancreatic multidisciplinary clinics and tripled participation in biospecimen studies at the institution in which the study took place.
“[Traditional cancer care delivery] is a slow and sequential, reactive process that puts a tremendous amount of onus on patients to navigate our complex medical system, and that is not adequate for pancreatic cancer, a disease where we have no time to wait,” Daniel A. King, MD, PhD, a gastrointestinal medical oncologist at the R.J. Zuckerberg Cancer Center and director of research and development for the Northwell Health Cancer Institute’s Center for Genomic Medicine, said during a presentation.
“A quarter of our patients [who] are diagnosed with pancreatic cancer die within 1 month,” he added. “We know that time to treatment is an important endpoint because we can see that treatment within 6 weeks is already a favorable prognostic indicator of survival ... we spent a lot of time thinking about how to improve things and we hypothesized that if we could intercept patients early using radiology ... maybe we could make a difference in care outcomes for our patients.”
Background and methodology
Researchers trained a natural language processing classifier to identify radiology reports deemed suspicious for pancreatic cancer from patients who underwent imaging services at Northwell Health Cancer Institute.
The daily workflow included the natural language processing flagging a suspicious report, a coordinator validating the finding, a gastrointestinal oncologist and navigator coordinating specialty oncology and tumor board referral, and a clinical research coordinator prescreening for research studies.
Researchers excluded patients who chose to go to hospice or those with nonpancreatic ductal adenocarcinoma.
Before implementation of the AI-guided navigation workflow, patients suspected of having pancreatic cancer within the institution experienced a mean of 22 days to biopsy, 32 days to an oncology visit and 56 days to treatment initiation from a radiology report, with 17% of patients being referred to a pancreatic multidisciplinary clinic.
On average, 1.8 patients per month received interest regarding biospecimen studies, with 1.5 patients per month consenting and 0.9 patients per month being enrolled.
Results
In the month following implementation, the AI-assisted navigation pilot flagged 1,666 patient reports, resulting in identification of 38 patients with new suspicion of pancreatic cancer.
Among the 38 patients, 53% underwent a biopsy, 50% visited an outpatient oncologist, 29% received a referral to a pancreatic multidisciplinary clinic and 42% received treatment at Northwell Health Cancer Institute.
Starting from the date of radiology report, patients underwent biopsy at a mean of 7 days, received an outpatient oncology visit at 15 days and began treatment at 34 days.
Researchers noted a monthly average of four patients being approached for biospecimen studies, with four patients consenting and 3.2 patients being enrolled, all more than doubling monthly averages prior to implementation.
Next steps
Researchers acknowledged potential study limitations, including its small sample size and single-arm design, coupled with limited navigator resource .
Next steps, according to the researchers, are to increase the sample size in future studies with a new navigator and new support.
“We want to expand this ... [our next project] is about doing a randomized version of this, a much bigger version with about 200 patients over the next 2 years,” King said. “What we’re hoping to prove is that this type of technology and type of process allows us to find and pull patients much faster.”
Collapse
John KM, et al. Abstract 11002. Presented at: ASCO Annual Meeting 2024; May 30 – June 4, 2024; Chicago.
Click Here to Manage Email Alerts