AI-enhanced literature review: Comparing advanced pancreatic and gastric cancer trial outcomes by performance status.

Abstract

e16427 Background: Clinical trials have traditionally limited enrollment to patients with ECOG Status 0-1, representing patients with relatively better prognosis and ability to tolerate treatment (i.e., performance status). There has been debate about expanding enrollment to patients with ECOG Status 0-2 to better serve patients with poorer health status and more accurately reflect real world clinical practice. Methods: We conducted an AI-enhanced literature review in pancreatic and gastric cancer to test whether the primary clinical outcomes, median progression free survival (PFS), and median overall survival (OS), differed in trials with broader ECOG inclusion criteria. Completed trials in advanced/metastatic pancreatic and gastric cancer with inclusion criteria of ECOG 0-2 and reported PFS and OS outcomes were identified manually on Clinicaltrials.gov. This was compared to the PFS and OS outcomes that were reported for all advanced/metastatic pancreatic and gastric cancer regardless of ECOG inclusion criteria which were identified in PubMed using EvidAI, an AI literature search and data extraction tool. The EvidAI results were quality checked by a manual review of the top 3% of highest and lowest extracted outcome values, a unit test on all values, and a random spot check on an additional 3% of values. No correction for sample size, study characteristics, or study quality was made. Results: A total of 7 studies that included ECOG2 patients with 12 cohorts were identified using Clinicaltrials.gov. In comparison, a total of 734 studies and 1,138 cohorts that included patients with any ECOG status were identified on PubMed using EvidAI. The average median PFS and OS values are shown in the table below. There was no significant difference between mPFS and mOS values in studies including patients with an ECOG status of 2 and the corresponding values identified by AI extraction. Conclusions: This analysis suggests that there is unlikely to be a significant decrement in PFS and OS trial outcomes with the inclusion of patients with an ECOG status of 2, however there were a limited number of completed studies including ECOG2 patients and this comparison should be interpreted with care. AI literature review and data extraction can be a useful tool to efficiently filter and analyze large quantities of clinical data but must be carefully supervised to ensure reliability. Studies including ECOG2 (clinicaltrials.gov) AI-extracted trial literature (PubMed) P Value (ECOG2 vs all trials) mPFS - Pancreatic 3.5m 4.3m 0.29 mOS – Pancreatic 7.0m 8.0m 0.29 mPFS – Gastric 7.6m 5.4m 0.23 mOS – Gastric 13.9m 10.6m 0.28