Computationally assisted patient finding for navigation to optimize pancreatic cancer care access

BACKGROUND: Patient navigators are increasingly utilized in cancer care but ensuring patients are properly identified and referred to navigators is a significant challenge. The primary objective was to compare time from radiographic report to biopsy, oncology visit, and treatment before versus after implementation of a computationally assisted navigation referral stream. Secondary objectives included evaluating care delivery across demographic groups and assessing survival outcomes. MATERIALS AND METHODS: A quality initiative at Northwell Health compared care delivery metrics between 2 cohorts of patients with suspected pancreatic cancer: those identified retrospectively using computational methods in January 2023 and those identified and navigated prospectively in June 2023. Radiology reports from a centralized health information exchange were analyzed by an ML-based natural language processing (NLP) model to detect findings suspicious of pancreatic cancer. Participants deemed eligible for navigation were contacted by a navigator to improve the likelihood and expediency of follow-up care. RESULTS: Seventy-one patients were included, with 38 patients in the retrospective cohort and 33 patients in the prospective cohort. The prospective cohort showed numeric reduction in time to biopsy (12-6 days, P = 0.173), oncology appointment (27-17 days, P = 0.192), and treatment (56-35 days, P = 0.136), though these results were not statistically significant. These metrics showed a significant reduction in standard deviation (P < 0.001), including among racial and ethnic minorities. The survival of patients in both cohorts was comparable (hazard ratio [HR] = 0.82, P = 0.66). CONCLUSION: This study provides promising evidence that an NLP-assisted identification workflow can improve care delivery and investigation in a larger study is warranted to validate these findings.