West Coast Ballroom
Effect of Changes in International Classification of Diseases (ICD) Versions on Measurement of Time-Varying Covariates in Disease Risk Prediction Models (307886)
*Lisa M Lix, University of ManitobaStephanie Monkman, University of Manitoba
Lin Yan, George and Fay Yee Centre for Healthcare Innovation
Keywords: comorbidity; longitudinal; chronic disease; data quality
The World Health Organization's International Classification of Diseases (ICD) is the recognized standard for reporting diseases and health conditions worldwide. It is used to ascertain comorbid conditions in electronic health databases, such as outpatient records. However, changes in ICD versions can result in abrupt changes in longitudinal trends of comorbid conditions in specific populations. This can adversely affect the accurate estimation of time-varying comorbidity effects in disease risk prediction models. We propose to address this problem by: (1) testing for inflection points in years of ICD version changes, and (2) including inflection indicators in the prediction model. For the former, piecewise regression models based on a negative binomial regression were adopted. For the latter, binary measures for inflection points are included in the risk prediction model. The methodology is applied in a cohort study of the effect of parental and offspring comorbidity on offspring fracture risk prediction using administrative health data from 1970 to 2017 in one Canadian province.