A NOVEL REGISTRY-DATABASE LINKAGE TO ENABLE REAL-WORLD EVIDENCE GENERATION IN RARE DISEASES: ALGORITHM DEVELOPMENT AND VALIDATION FOR MYOTONIC DYSTROPHY TYPE 1 IN THE FRENCH SNDS
Author(s)
Nadia Quignot, PhD1, Stephanie Read, PhD2, Gaelle Gusto, PhD1, Melinda Gyenge, PhD3, Artak Khachatryan, MPH, PhD, MD4, Guillaume Bassez, MD, PhD, MCU-PH3.
1Certara, Paris, France, 2Certara, Edinburgh, Midlothian, United Kingdom, 3Neuromuscular reference center, Pitié-Salpêtrière Hospital, Paris, France, 4Certara, Sutton, United Kingdom.
1Certara, Paris, France, 2Certara, Edinburgh, Midlothian, United Kingdom, 3Neuromuscular reference center, Pitié-Salpêtrière Hospital, Paris, France, 4Certara, Sutton, United Kingdom.
OBJECTIVES: Myotonic Dystrophy Type 1 (DM1) is a rare, progressive disorder with multisystem involvement and significant clinical and economic burden. Accurate patient identification in administrative databases is a prerequisite for real-world evidence generation. In the French Système National des Données de Santé (SNDS), only a broader code encompassing DM1 and DM2 is available, making case identification challenging. This study aimed to establish a novel registry-database linkage and a predictive algorithm for DM1 identification in the SNDS.
METHODS: DM-Scope, established in 2008, contains data for over 3,800 genetically-confirmed DM1 and DM2 patients. Registry data were linked to the SNDS using sex, month and year of birth, outpatient visit dates and centres, supplemented by clinical variables to improve specificity. Linked records were used to develop and validate a patient identification algorithm, with the cohort split into training and validation sets (2:1). Variables were selected based on statistical significance and clinical relevance; LASSO regression with cross-validation was applied in the training set and performance assessed using bootstrap resampling. The algorithm was subsequently applied to identify a nationwide cohort of probable DM1 patients.
RESULTS: Of 1,475 DM-Scope patients, 1,252 were successfully matched in the SNDS, of whom 1,194 contributed to algorithm development. The final algorithm comprised 12 variables, including cardiac conduction disorders, diabetes, respiratory disease, fractures, foot orthoses, treatment use, and age at first evidence of DM, with age emerging as the strongest predictor. The algorithm achieved an AUC of 0.86 in the training set and an optimism-corrected AUC of 0.84. At a 0.9 probability threshold, optimism-corrected sensitivity was 88% and specificity was 64%.
CONCLUSIONS: This study provides the first validated algorithm for DM1 identification in administrative healthcare data, demonstrating good discriminatory performance and supporting nationwide DM1 identification, although further refinement is needed to improve specificity. This methodology may be transferable to other rare diseases lacking disease-specific coding.
METHODS: DM-Scope, established in 2008, contains data for over 3,800 genetically-confirmed DM1 and DM2 patients. Registry data were linked to the SNDS using sex, month and year of birth, outpatient visit dates and centres, supplemented by clinical variables to improve specificity. Linked records were used to develop and validate a patient identification algorithm, with the cohort split into training and validation sets (2:1). Variables were selected based on statistical significance and clinical relevance; LASSO regression with cross-validation was applied in the training set and performance assessed using bootstrap resampling. The algorithm was subsequently applied to identify a nationwide cohort of probable DM1 patients.
RESULTS: Of 1,475 DM-Scope patients, 1,252 were successfully matched in the SNDS, of whom 1,194 contributed to algorithm development. The final algorithm comprised 12 variables, including cardiac conduction disorders, diabetes, respiratory disease, fractures, foot orthoses, treatment use, and age at first evidence of DM, with age emerging as the strongest predictor. The algorithm achieved an AUC of 0.86 in the training set and an optimism-corrected AUC of 0.84. At a 0.9 probability threshold, optimism-corrected sensitivity was 88% and specificity was 64%.
CONCLUSIONS: This study provides the first validated algorithm for DM1 identification in administrative healthcare data, demonstrating good discriminatory performance and supporting nationwide DM1 identification, although further refinement is needed to improve specificity. This methodology may be transferable to other rare diseases lacking disease-specific coding.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
RWD182
Topic
Epidemiology & Public Health, Methodological & Statistical Research, Real World Data & Information Systems
Topic Subcategory
Health & Insurance Records Systems
Disease
Cardiovascular Disorders (including MI, Stroke, Circulatory), Musculoskeletal Disorders (Arthritis, Bone Disorders, Osteoporosis, Other Musculoskeletal), Neurological Disorders, Rare & Orphan Diseases, Respiratory-Related Disorders (Allergy, Asthma, Smoking, Other Respiratory)