1366P - Natural Language Processing (NLP) as promising artificial intelligence (AI) tool to improve patients (pts) enrollment in clinical trials (CT): Analysis in real-world conditions on a lung cancer cohort

Background

Oncology biomarker-driven CT are opportunities for pts with limited therapeutic options but screening is complex and time consuming. The improvement in AI methods, such as NLP algorithms could alleviate this burden. Intelligence for Health® (i4H) software allows automated pts cohorts generation by requesting selected criteria among structured and unstructured pts data hosted in hospital Electronic Medical Records (EMR), including hospital reports, scanned external documents and laboratory reports.

Methods

We conducted a retrospective study on real world data from 7400 pts’ EMR managed within an onco-pulmonology department between 01/01/18 and 04/30/21 which reflected real conditions of CT recruitment. Our objective was to assess capacity of i4H to retrieve pts matching with eligibility criteria from a Phase III CT (CodeBreaK 200, NCT04303780) in KRAS G12C mutated non-small cell lung cancer (NSCLC) pts. Sub-cohorts using different semantic approaches of main inclusion and exclusion criteria (age, histology, stage, molecular alteration, previous treatment) were set up. Stage I aimed at selecting sub-cohorts used for further analysis in up-coming stage 2. We reported here interim analysis on 987 EMR.

Results

7 pts were manually screened for the CT; 6 were enrolled and 1 was excluded (KRAS G12C negative).

Sub-cohorts 2 and 3 retrieved 100% of manually screened pts and excluded correctly negative KRAS G12C pt, with an acceptable size. Table: 1366P

Sub-cohorts results

Conclusions

NLP algorithms might help clinicians in providing a reasonable sized list of pts matching with main CT inclusion and exclusion criteria. It might be useful making a first funnel of screening and decreasing workload and time spent on this activity. Additional results on sensitivity and specificity will be presented at the congress.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

Collective Thinking.

Funding

Collective Thinking.

Disclosure

J. Mazieres: Financial Interests, Personal, Principal Investigator, personal fees: Roche, AstraZeneca, Pierre Fabre, Takeda, BMS, MSD, Pfizer, Jiangsu Hengruii, Blueprint, Daiichi Sankyo, Novartis, Amgen; Financial Interests, Personal, Other, grants: Roche, AstraZeneca, Pierre Fabre, BMS, Illumina. All other authors have declared no conflicts of interest.