Computationally Assisted Patient Finding for Navigation to Optimize Pancreatic Cancer Care Access

Comprehensive Summary

An AI-assisted model was implemented to improve the identification and care coordination of patients who were suspected to have pancreatic cancer. In this study, a machine learning natural language processing (NLP) model was used to analyze radiology reports from abdominal CT and MRI scans. The model flagged reports containing language suspicious for pancreatic cancer, allowing clinicians to identify patients earlier on and refer them to patient navigators for timely diagnostic work. Researchers compared a retrospective cohort identified through computational review with a prospective cohort identified in real time using the NLP referrals. Patients with newly suspected pancreatic cancer were also included in the analysis. The prospective cohort showed reductions in time to biopsy, oncology visit, and treatment initiation, although these differences were not statistically significant. However, the intervention reduced variability in these timelines, suggesting more consistent care delivery.

Outcomes and Implications

The integration of AI-based NLP tools into clinical spaces demonstrates the potential to improve cancer care delivery by identifying high-risk patients earlier on. The model helped with making streamlined referrals that connected patients with care navigators earlier in the diagnostic process. Improvements were noticeable among racial and ethnic minority patients as well, suggesting progress toward more equitable care. Although the study showed promising trends, the small sample size limited statistical power for some of the comparisons. Additional limitations included the need for manual review of cases. However, significant improvements in the consistency of care efficiency metrics were seen.