VALIDATION OF AN ALGORITHM TO IDENTIFY PROGRESSIVE PULMONARY FIBROSIS USING THE CANADIAN REGISTRY FOR PULMONARY FIBROSIS
Author(s)
Yanni Fan, ScD1, Catherine Allende, MS2, Danielle Sienko, MS2, Phil Schwab, PhD2, Natalie Nightingale, MSc3, Masha Eberg, MSc3, Vinicius C. Antao, MD, PhD1, Daniel Marinescu, MD, MHSc4, Alyson W. Wong, MD4, Christopher J. Ryerson, MD4.
1Boehringer Ingelheim Pharmaceuticals, Inc., Ridgefield, CT, USA, 2Genesis Research Group, Hoboken, NJ, USA, 3IQVIA Solutions Canada Inc, Montreal, QC, Canada, 4Department of Medicine, University of British Columbia, Vancouver, BC, Canada.
1Boehringer Ingelheim Pharmaceuticals, Inc., Ridgefield, CT, USA, 2Genesis Research Group, Hoboken, NJ, USA, 3IQVIA Solutions Canada Inc, Montreal, QC, Canada, 4Department of Medicine, University of British Columbia, Vancouver, BC, Canada.
OBJECTIVES: Progressive pulmonary fibrosis (PPF) is characterized by worsening fibrotic interstitial lung disease (ILD) and early mortality. As clinical progression relies on physiologic, radiologic, and symptomatic assessments not consistently available in claims data, PPF is challenging to identify in secondary healthcare databases. This study validated an algorithm for identifying PPF among patients with non-idiopathic pulmonary fibrosis (IPF) ILD using the Canadian Registry for Pulmonary Fibrosis (CARE-PF).
METHODS: This observational validation cohort study used CARE-PF data from two University of British Columbia sites, including adults with non-IPF ILD enrolled between January 1, 2017, and December 31, 2024, with a diagnosis on or before December 31, 2022. Confirmed PPF cases were defined using INBUILD trial criteria within 24 months of diagnosis. An algorithm using healthcare utilization and treatment proxies was applied to identify PPF cases from electronic medical records. Performance was assessed descriptively using sensitivity, negative predictive value (NPV), specificity, positive predictive value (PPV), accuracy, and area under the curve (AUC). Several sensitivity analyses were conducted, including exclusion of patients treated with antifibrotics and removal of medication-related proxies from the algorithm.
RESULTS: Among 1,389 patients with ILD, 384 (27.6%) met the INBUILD PPF definition, and 893 (64.3%) were classified as having PPF by the algorithm. Sensitivity was 86.7% and NPV 89.7%, with specificity 44.3% and PPV 37.3%. Accuracy was 56.0% and AUC 0.75. Performance estimates remained similar after excluding antifibrotic-treated patients (n=1,262). Without medication proxies, sensitivity was 65.1-73.4% and NPV was 82.9-85.2%, with specificity of 58.6-64.7%, PPV of 40.4-41.3%, accuracy of 62.7-64.8%, and AUC of 0.69-0.71.
CONCLUSIONS: The algorithm demonstrated high sensitivity and NPV, supporting utility of screening ILD populations for PPF in secondary healthcare databases. Given the low specificity and PPV, additional confirmatory criteria are being explored to refine the algorithm for more accurate identification of patients with PPF.
METHODS: This observational validation cohort study used CARE-PF data from two University of British Columbia sites, including adults with non-IPF ILD enrolled between January 1, 2017, and December 31, 2024, with a diagnosis on or before December 31, 2022. Confirmed PPF cases were defined using INBUILD trial criteria within 24 months of diagnosis. An algorithm using healthcare utilization and treatment proxies was applied to identify PPF cases from electronic medical records. Performance was assessed descriptively using sensitivity, negative predictive value (NPV), specificity, positive predictive value (PPV), accuracy, and area under the curve (AUC). Several sensitivity analyses were conducted, including exclusion of patients treated with antifibrotics and removal of medication-related proxies from the algorithm.
RESULTS: Among 1,389 patients with ILD, 384 (27.6%) met the INBUILD PPF definition, and 893 (64.3%) were classified as having PPF by the algorithm. Sensitivity was 86.7% and NPV 89.7%, with specificity 44.3% and PPV 37.3%. Accuracy was 56.0% and AUC 0.75. Performance estimates remained similar after excluding antifibrotic-treated patients (n=1,262). Without medication proxies, sensitivity was 65.1-73.4% and NPV was 82.9-85.2%, with specificity of 58.6-64.7%, PPV of 40.4-41.3%, accuracy of 62.7-64.8%, and AUC of 0.69-0.71.
CONCLUSIONS: The algorithm demonstrated high sensitivity and NPV, supporting utility of screening ILD populations for PPF in secondary healthcare databases. Given the low specificity and PPV, additional confirmatory criteria are being explored to refine the algorithm for more accurate identification of patients with PPF.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
EPH38
Topic
Epidemiology & Public Health
Topic Subcategory
Disease Classification & Coding
Disease
No Additional Disease & Conditions/Specialized Treatment Areas, Respiratory-Related Disorders (Allergy, Asthma, Smoking, Other Respiratory)