Using Natural Language Processing to Identify Sequencing Patterns in Advanced Urothelial Carcinoma: A Real-World Observational Study
Author(s)
Arthur PEREZ, PharmD, Lorraine HOUVET, MSc, Titouan Lacombe, MSc, Emmanuel Gilson, MSc, ISABELLE DELAROZIERE, PhD.
OSPI, Paris, France.
OSPI, Paris, France.
OBJECTIVES: Patient-centered decision-making in the care protocol and life trajectory could ensure optimal clinical and organizational management of pathologies. However these decisions need to be supported by high levels of evidence which can be costly and time consuming in real life settings due to scarce resources and heterogeneous data collection across settings. Using recent advances in AI knowledge could help accelerate the production of such evidence.This non-interventional study is an opportunity to collect updated long-term efficacy and follow-up data in the care of urothelial carcinoma. It aims to identify treatment sequences and describe clinical and demographic characteristics of locally advanced or metastatic urothelial carcinoma patients.
METHODS: For all patients meeting eligibility criteria in the 12 participating sites, the data of interest was collected from structured and unstructured data available in electronic health records over the study observation period defined as earliest identified chemotherapy date (or 01/01/2020) and 01/02/2025.AI technologies and NLP methods allowed semi-automated data collection among unstructured data.
RESULTS: Using a semi-automated collection method proved to be an effective way of collecting and treating vast amounts of data in a limited timeframe: data from 12 participating sites were collected in a four month period while guaranteeing a high level of homogeneity across more than 60 variables. It helped capture precise data on patients characteristics (ECOG status, adverse events of interest, biological abnormalities throughout the whole study timeframe) and link it to clinical pathways and outcomes (treatment lines changes, OS and DOT).
CONCLUSIONS: Using AI has a strong potential for accelerating real world clinical research by structuring data from EHRs, yielding consistency superior to manual entry by physicians and reduced amount of time. This may enhance the efficiency, accuracy, and scalability of EHR-to-database conversions to enhance patient-centered innovation.
METHODS: For all patients meeting eligibility criteria in the 12 participating sites, the data of interest was collected from structured and unstructured data available in electronic health records over the study observation period defined as earliest identified chemotherapy date (or 01/01/2020) and 01/02/2025.AI technologies and NLP methods allowed semi-automated data collection among unstructured data.
RESULTS: Using a semi-automated collection method proved to be an effective way of collecting and treating vast amounts of data in a limited timeframe: data from 12 participating sites were collected in a four month period while guaranteeing a high level of homogeneity across more than 60 variables. It helped capture precise data on patients characteristics (ECOG status, adverse events of interest, biological abnormalities throughout the whole study timeframe) and link it to clinical pathways and outcomes (treatment lines changes, OS and DOT).
CONCLUSIONS: Using AI has a strong potential for accelerating real world clinical research by structuring data from EHRs, yielding consistency superior to manual entry by physicians and reduced amount of time. This may enhance the efficiency, accuracy, and scalability of EHR-to-database conversions to enhance patient-centered innovation.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
CO260
Topic
Clinical Outcomes, Methodological & Statistical Research, Study Approaches
Topic Subcategory
Clinical Outcomes Assessment
Disease
No Additional Disease & Conditions/Specialized Treatment Areas, Oncology, Personalized & Precision Medicine, Urinary/Kidney Disorders