An AI-assisted navigation approach for patients with radiographic suspicion of new pancreas cancer.

person Kristen M. John Donald and Barbara Zucker School of Medicine at Hofstra/Northwell, Hempstead, NY info_outline Kristen M. John, Joseph Tenner, Rolando Croocks, Cristina Valente, Bernadette Bingham, Amber N Habowski, Tara McEvoy, Tiffany Zavadsky, Kristen Beyer, Rita Mercieca, Sandeep Nadella, Anthony Carvino, Matthew Barish, Daniel King

Authors person Kristen M. John Donald and Barbara Zucker School of Medicine at Hofstra/Northwell, Hempstead, NY info_outline Kristen M. John, Joseph Tenner, Rolando Croocks, Cristina Valente, Bernadette Bingham, Amber N Habowski, Tara McEvoy, Tiffany Zavadsky, Kristen Beyer, Rita Mercieca, Sandeep Nadella, Anthony Carvino, Matthew Barish, Daniel King Organizations Donald and Barbara Zucker School of Medicine at Hofstra/Northwell, Hempstead, NY, Northwell Health Cancer Institute, New Hyde Park, NY, Northwell Health, New Hyde Park, NY, Cold Spring Harbor Laboratory, Cold Spring Harbor, NY, Division of Surgery, Northwell Health Cancer Institute, New Hyde Park, NY, Northwell Health Cancer Institute, Lake Success, NY Abstract Disclosures Research Funding No funding sources reported Background: The traditional cancer care paradigm relies on passive referral streams, resulting in suboptimal care delivery. Here, we assessed the results of an AI-assisted navigation approach to proactively coordinate specialty cancer care services for patients with radiographically suspected pancreas cancer. Methods: A natural language processing (NLP) classifier was trained to identify radiology reports suspicious for pancreas cancer from imaging reports generated within Northwell Health. The daily workflow consists of the following: the NLP flags reports with suspicion of pancreas cancer, a coordinator validates the finding, a GI oncologist and navigator coordinates specialty oncology and tumor board referral, and a clinical research coordinator pre-screens for research studies. Patients determined to be hospice-bound or with non-PDAC malignancy are excluded. Care delivery metrics were compared by Mann-Whitney U tests. Results: Prior to implementation, patients with new suspicion of pancreas cancer at our institution experienced a mean of 22 days to biopsy, 32 days to oncology visit, and 56 days to treatment initiation from radiology report with 17% referred to pancreatic multidisciplinary clinic (PMDC). Per monthly average, 1.8 patients were approached for biospecimen studies, 1.5 were consented, and 0.9 were enrolled. In the month following implementation of the AI-assisted navigation pilot, reports from 1666 patients were flagged, resulting in 38 patients with new suspicion of pancreas cancer identified (Table). Of these, 53% underwent biopsy, 50% were seen by an outpatient oncologist, 29% were referred to PMDC, and 42% received treatment at our institution. From date of radiology report, these patients underwent biopsy at a mean of 7 days (p<0.01), outpatient oncology visit at 15 days (p=0.285), and treatment initiation at 34 days (p=0.3479). A monthly average of 4 patients were approached for biospecimen studies, 4 were consented, and 3.2 were enrolled. Conclusions: An AI-guided navigation workflow identified patients earlier in the diagnostic timeline; reduced time to biopsy, follow-up, and treatment initiation; increased patient referrals to PMDC; and tripled participation in biospecimen studies at our institution. Demographics of patients identified through AI-guided workflow (n=38). Age 69 +/- 12 Gender Male 55% Race Asian 16% Black 24% White 47% Other 13% Scan type CT 87% MRI 13% Scan setting Inpatient 89% Outpatient 11%