Validity of administrative data for the diagnosis of primary sclerosing cholangitis: a population-based study

• Published in: Liver international Vol. 31; no. 5; pp. 712 - 720
• Main Authors: Molodecky, Natalie A., Myers, Robert P., Barkema, Herman W., Quan, Hude, Kaplan, Gilaad G.
• Format: Journal Article
• Language: English
• Published: Oxford, UK Blackwell Publishing Ltd 01.05.2011
• Subjects: Adult; Aged; Alberta - epidemiology; Algorithms; Biopsy; Cholangitis, Sclerosing - classification; Cholangitis, Sclerosing - diagnosis; Cholangitis, Sclerosing - epidemiology; Data Mining; Databases as Topic; Female; Humans; International Classification of Diseases; Logistic Models; Male; Middle Aged; positive predictive value; Predictive Value of Tests; primary sclerosing cholangitis; Reproducibility of Results; ROC Curve; sensitivity; Time Factors; validation; Alberta
• ISSN: 1478-3223; 1478-3231; 1478-3231
• DOI: 10.1111/j.1478-3231.2011.02484.x

Background/Aims: Administrative databases could be useful in studying the epidemiology of primary sclerosing cholangitis (PSC); however, there is no information regarding the validity of the diagnostic code in administrative databases. The aims of this study were to determine the validity of administrative data for a diagnosis of PSC and generate algorithms for the identification of PSC patients. Methods: The sensitivity (Se) and positive predictive value (PPV) of a PSC diagnosis based on administrative data from 2000 to 2003 were determined through chart review data. Algorithms were developed by considering variables associated with PSC and coding details. A logistic regression model was constructed using covariates associated with PSC. Based on this model, each subject was assigned a probability of having PSC. A cutoff value was selected that maximized the Se and specificity (Sp) of correctly predicting PSC cases. Results: In the administrative data, the initial Se and PPV were 83.7 and 7.2% respectively. The optimal algorithm included one PSC code and one inflammatory bowel disease code and had Se 56% and PPV 59%. Overall, the algorithms yielded inadequate PPV and Se estimates to identify a cohort of true PSC cases. The predictive model was constructed using six covariates. For this model, the area under the receiver operating characteristic curve was 93.5%. A cutoff of 0.0729 was used, which maximized the Se 81.9% and Sp 90.7%; however, the PPV was 41.0%. Conclusion: An algorithm for the identification of true PSC cases from administrative data was not possible. We recommend that PSC receives a distinct ICD code from ascending cholangitis.