Evaluating Pain Scores and Detecting NSCLC Using Named Entity Recognition and Agentic AI in Clinical Texts
Author(s)
Abhimanyu Roy, MBA1, Abhinav Nayyar, MBBA, MBA1, Mahainn Somani, B.Tech1, Vikash Kumar Verma, MBA, PharmD1, Ina Kukreja, MBA, PT1, Arunima Sachdev, MA1, Marissa Seligman, BS Pharma2, Rahul Goyal, BS Tech3, Louis Brooks Jr, MA4.
1Optum, Gurgaon, India, 2Optum, Boston, MA, USA, 3Optum, Phoenix, AZ, USA, 4Optum, Bloomsbury, NJ, USA.
1Optum, Gurgaon, India, 2Optum, Boston, MA, USA, 3Optum, Phoenix, AZ, USA, 4Optum, Bloomsbury, NJ, USA.
Presentation Documents
OBJECTIVES: This study aims to develop a Named Entity Recognition (NER) model to accurately identify Non-Small Cell Lung Cancer (NSCLC) and evaluate pain scores from unstructured clinical notes. The goal is to enhance patient management and outcomes through early and accurate assessment of pain scores.
METHODS: We analyzed Optum’s de-identified clinical notes from 2016 to 2023. NSCLC diagnoses were confirmed using ICD-10 codes and textual mentions. Pain scores were identified through manual review of 10% of clinical notes and subsequent NLP model application on the remaining notes. The NER model employed a Char CNNs - BiLSTM - CRF architecture, trained to detect medical entities specific to NSCLC and various pain scales. Agentic AI components were integrated to autonomously refine entity recognition capabilities. Model performance was evaluated using precision, recall, and F1-score metrics.
RESULTS: The NER model demonstrated high accuracy in identifying NSCLC stages and associated pain scores, with precision, recall, and F1-scores exceeding 90%. The integration of Agentic AI improved the model's adaptability and performance over time, ensuring precise and reliable data extraction. Early identification of pain scores facilitated timely interventions, potentially enhancing patient management and outcomes.
CONCLUSIONS: The AI-driven NER model effectively identified NSCLC and evaluated pain scores from clinical texts, supporting personalized and effective care strategies. The integration of Agentic AI enhanced the model's performance, demonstrating the potential of combining AI with human intelligence to improve clinical outcomes and patient quality of life. This approach underscores the transformative potential of AI in healthcare analytics.
METHODS: We analyzed Optum’s de-identified clinical notes from 2016 to 2023. NSCLC diagnoses were confirmed using ICD-10 codes and textual mentions. Pain scores were identified through manual review of 10% of clinical notes and subsequent NLP model application on the remaining notes. The NER model employed a Char CNNs - BiLSTM - CRF architecture, trained to detect medical entities specific to NSCLC and various pain scales. Agentic AI components were integrated to autonomously refine entity recognition capabilities. Model performance was evaluated using precision, recall, and F1-score metrics.
RESULTS: The NER model demonstrated high accuracy in identifying NSCLC stages and associated pain scores, with precision, recall, and F1-scores exceeding 90%. The integration of Agentic AI improved the model's adaptability and performance over time, ensuring precise and reliable data extraction. Early identification of pain scores facilitated timely interventions, potentially enhancing patient management and outcomes.
CONCLUSIONS: The AI-driven NER model effectively identified NSCLC and evaluated pain scores from clinical texts, supporting personalized and effective care strategies. The integration of Agentic AI enhanced the model's performance, demonstrating the potential of combining AI with human intelligence to improve clinical outcomes and patient quality of life. This approach underscores the transformative potential of AI in healthcare analytics.
Conference/Value in Health Info
2025-05, ISPOR 2025, Montréal, Quebec, CA
Value in Health, Volume 28, Issue S1
Code
MSR101
Topic
Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
SDC: Oncology