2024 Latin America Conference on Lung Cancer (LALCA) - Posters-PP01.04: Analysis of the Lipi Score as a Prognostic Factor in Advanced Non-Small Cell Lung Cancer

Pdf Summary

Recent clinical trials have highlighted the effectiveness of targeted therapies for non-small cell lung cancer (NSCLC) patients with MET overexpression, determined through immunohistochemistry (IHC). However, variations among pathologists in interpreting MET IHC staining necessitate better standardization methods. This study assesses the consistency of MET IHC interpretation across 33 pathologists and compares it with an AI model to explore its potential for more standardized assessments.<br /><br />Moderate to substantial interobserver agreement was achieved among pathologists, with the level of agreement influenced by the pathologist's experience and the type of specimen. The AI model exhibited moderate to perfect agreement with pathologists, indicating its potential for aiding MET IHC interpretation standardization.<br /><br />A total of 30 lung adenocarcinoma cases were analyzed, comprising 15 resection and 15 biopsy specimens, using the Ventana BenchMark ULTRA IHC/ISH system and the SP44 antibody. MET protein expression was scored based on tumor cell staining percentages, with scores ranging from 0 (none) to 3 (strong). AI analyses employed a DeepBio model for MET scoring, and H-scores were calculated to further quantify MET expression.<br /><br />Statistical analyses showed an interobserver agreement using intraclass correlation coefficients (ICC) and kappa statistics. The AI model's involvement demonstrated promising results for establishing a consensus in MET IHC interpretation, overcoming pathologists' varying thresholds, especially in very weak staining scenarios.<br /><br />Overall, AI can provide meaningful collaboration with pathologists, potentially reducing variability and enhancing the reliability of MET overexpression assessments in NSCLC, ultimately impacting treatment decisions. Thus, integrating AI in interpreting IHC results for cancer diagnostics demonstrates a path forward for improving accuracy and uniformity in clinical settings.

Asset Subtitle

Eduardo Richardet

Keywords

NSCLC

MET overexpression

immunohistochemistry

pathologists

AI model

standardization

lung adenocarcinoma

Ventana BenchMark ULTRA

DeepBio model

treatment decisions