AN NLP-BASED SOLUTION (REALLI) FOR RSV CASE IDENTIFICATION AND CLINICAL CHARACTERIZATION: A RETROSPECTIVE STUDY IN 13 FRENCH HOSPITALS (OMIP)
Author(s)
Justine Wenta, MD, MSc1, Laura MARTIN, MD, MSc1, Ayoub Boukhlal, PharmD1, Marc Jouve, MBA, MSc2, Barbara Lebas, MSc2, Florence Kpade, MSc2, Yasmine Baghdadi, PhD2, Elodie Blanchard, MD, PhD3, Paul Loubet, MD, PhD4, Jean Marie COHEN, MD5.
1PFIZER, Paris, France, 2Sancare, Paris, France, 3CHU Bordeaux, Bordeaux, France, 4CHU Nimes, Nimes, France, 5Medical Office, Open Rome, Paris, France.
1PFIZER, Paris, France, 2Sancare, Paris, France, 3CHU Bordeaux, Bordeaux, France, 4CHU Nimes, Nimes, France, 5Medical Office, Open Rome, Paris, France.
OBJECTIVES: French hospital discharge database (PMSI) is widely used for real-world research but incompletely captures clinical information often recorded as unstructured text (e.g., symptoms, comorbidities, test results, treatments). We aimed to describe the added value of a Natural Language Processing (NLP)-powered solution (REALLI) for extracting data from inpatient electronic health records (EHRs) in RSV hospitalizations not captured in PMSI.
METHODS: OMIP (Observatory of Infectious and Pulmonary Diseases) is a retrospective observational study in 13 French hospitals including patients aged ≥60 years hospitalized in medicine-surgery-obstetrics units over two seasons (May 2022-April 2023; May 2023-April 2024). REALLI combined of ICD-10 code-based algorithms, NLP methods from semantic rules-based research and Large Language Model to identify RSV hospitalizations. As REALLI accesses all hospital documents (discharge summaries, clinical notes, laboratory and imaging reports, prescriptions/administration records), it extracts data not usually coded in PMSI: symptoms, medical history, comorbidities, diagnostic tests, treatments, and diagnosis, thereby supplementing ICD-10 codes commonly used for RSV detection.
RESULTS: In May 2022-April 2023, 344 hospitalized patients for RSV were identified, 9.3% through NLP only. Among clinical variables not available in PMSI, NLP captured signs of respiratory distress in 14% patients, fever in 54%, COVID-19 vaccination in 48% of patients, influenza vaccination in 19%, pneumococcal vaccination in 7.0%. Median delay to inclusion was 25 days prior to the RSV epidemic peak. In May 2023-April 2024, 213 patients were identified, 16% through NLP only; respiratory distress was observed in 17%, fever in 54%, COVID-19 vaccination in 33%, influenza vaccination in 15%, pneumococcal vaccination in 10%; median delay to inclusion was 30 days prior to the RSV epidemic peak.
CONCLUSIONS: NLP-powered solution such as REALLI substantially enriches PMSI-based analyses by capturing available granular clinical information from EHR structured and unstructured data and contributes in identification of RSV patients.
METHODS: OMIP (Observatory of Infectious and Pulmonary Diseases) is a retrospective observational study in 13 French hospitals including patients aged ≥60 years hospitalized in medicine-surgery-obstetrics units over two seasons (May 2022-April 2023; May 2023-April 2024). REALLI combined of ICD-10 code-based algorithms, NLP methods from semantic rules-based research and Large Language Model to identify RSV hospitalizations. As REALLI accesses all hospital documents (discharge summaries, clinical notes, laboratory and imaging reports, prescriptions/administration records), it extracts data not usually coded in PMSI: symptoms, medical history, comorbidities, diagnostic tests, treatments, and diagnosis, thereby supplementing ICD-10 codes commonly used for RSV detection.
RESULTS: In May 2022-April 2023, 344 hospitalized patients for RSV were identified, 9.3% through NLP only. Among clinical variables not available in PMSI, NLP captured signs of respiratory distress in 14% patients, fever in 54%, COVID-19 vaccination in 48% of patients, influenza vaccination in 19%, pneumococcal vaccination in 7.0%. Median delay to inclusion was 25 days prior to the RSV epidemic peak. In May 2023-April 2024, 213 patients were identified, 16% through NLP only; respiratory distress was observed in 17%, fever in 54%, COVID-19 vaccination in 33%, influenza vaccination in 15%, pneumococcal vaccination in 10%; median delay to inclusion was 30 days prior to the RSV epidemic peak.
CONCLUSIONS: NLP-powered solution such as REALLI substantially enriches PMSI-based analyses by capturing available granular clinical information from EHR structured and unstructured data and contributes in identification of RSV patients.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
EPH222
Topic
Epidemiology & Public Health, Methodological & Statistical Research
Topic Subcategory
Disease Classification & Coding
Disease
Infectious Disease (non-vaccine)