Cohort Definition Name : Earliest event of Ankylosing Spondylitis
Contributor name : Joel Swerdel’
Contributor OrcId :
Logic Description : Earliest occurrence of Ankylosing Spondylitis indexed on diagnosis date, for the first time in history cohort exit is the end of continuous observation
Recommended study application : target
Assertion statement : This cohort definition was executed on at least one real person-level observational health data source and resulted in a cohort with at least 1 person.
Target Clinical Description : Ankylosing spondylitis (AS) is a complex, chronic inflammatory disease of the axial spine that is insidious in onset, progressing to radiological sacroiliitis over several years. It is also known as radiographic axial-spondylarthritis (axSpA). Pathogenesis is poorly understood, however immune-mediated mechanisms and genetic and environmental factors are thought to have key roles. Whereas AS does not follow a predictable pattern of progression, early signs such as hip involvement may be prognostic of later disease severity. AS commonly starts in the second or third decade of life; the clinical picture of early (juvenile onset) AS differs from that of adult onset by the more frequent involvement of peripheral joints. Men are afflicted with AS approximately 2-3 times more frequently than women, and disease tends to be more severe in men. AS is one disease in a family of inter-related disorders (spondyloarthropathies) that share clinical and genetic characteristics distinct from rheumatoid arthritis. The current subcategories of spondyloarthropathies according to the European Spondylarthropathy Study Group are AS, reactive arthritis, psoriatic arthritis, inflammatory bowel disease (IBD) associated arthritis, and undifferentiated spondyloarthropathy. Approximately 1 million Americans and 20 million people globally are affected by AS.
"Evaluation conclusion : We developed a prevalent cohort definition for ankylosing spondylitis (AS) using a concept set of 4 concepts which incorporated all those found from the literature review, the analysis of PHOEBE, and orphan concepts in cohort diagnostics. The concept set removed juvenile forms of the disease. It also excludes those with Non-radiographic axial spondyloarthritis. We performed the evaluation across a network of claim data sources and 1 EHR US data source. The data sources are: IBM® MarketScan® Commercial Database (CCAE), Optum’s longitudinal EHR repository (Optum EHR), Optum’s Clinformatics® Data Mart (DOD), IBM® MarketScan® Multi-State Medicaid Database (MDCD), IBM® MarketScan® Medicare Supplemental Database (MDCR), Japan Claims Database (JMDC), Clinical Practice Research Datalink (CPRD) , IQVIA® Australia Longitudinal Patient Data (LPD) database (Australia), IQVIA® Disease Analyzer (DA) France database (France), QVIA® Disease Analyzer (DA) Germany database (Germany), IQVIA® Adjudicated Health Plan Claims Data (formerly PharMetrics Plus) - US database (PharMetrics), IQVIA® Ambulatory EMR (EMR) . The algorithm retrieves subjects from 12 databases tested. While there were indications of AS in the 2 months prior to the first code for AS, e.g., diagnoses of rheumatoid arthritis, the condition may be undiagnosed for a year or more so correcting by a few months was not determined to be of value. We developed a more specific cohort requiring a second diagnosis or observation code for AS in the time period 31-365 days after index. This cohort improves the specificity of the algorithm albeit at the expense of sensitivity as determined by PheValuator. This is one of the cases where there is a real trade-off in performance between the broad (one code) and narrow (two codes) definitions. The near perfect PPV for definition with 2 codes might be considered for use for some analyses where maximization of PPV is preferred. In addition, caution should be used when using the Australia and France datasets as the rates of AS decreased precipitously during the past five years which may be due to a dataset artifact.
Performance characteristics were determined for 7 of the 12 databases. The remaining databases did not contain enough subjects to produce an accurate diagnostic model. Using one code (the current definition), sensitivity ranged from about 77% in Optum EHR to about 97% in PharMetrics while positive predictive value ranged from about 56% in MDCD to about 85% in Amb EHR. Using two codes, sensitivity decreased, ranging from about 39% in MDCR to about 74% in PharMetrics while positive predictive value increased, ranging from 98% in MDCR and MDCD to >= 99% in all other databases. Overall, the mean sensitivity for Aone code was about 90% and the PPV was about 71%; for two codes the mean sensitivity was 58% while the PPV was about 99%.