Estimation of Prevalence and Incremental Costs of Systemic Lupus Erythematosus in a Middle-Income Country Using Machine Learning on Administrative Health Data

Abstract

ObjectivesSystemic lupus erythematosus (SLE) is a chronic, autoimmune disease that may cause physical and functional disability. The objective of this study is to measure prevalence and estimate incremental cost of SLE treatment using information from administrative databases in Colombia. MethodsWe use data from the patients on the Colombian contributive health system with a period of study from 2015 to 2017. The incremental cost of SLE is estimated using a matched study by propensity score and multivariate balance of covariates. To reduce the effect of possible specification problems, we use Extreme Gradient Boosting, a flexible machine learning algorithm. We use paired t statistical comparison and Bootstrap to validate the robustness of the method. In addition, we use a machine learning regression approach on the cost of control patients to achieve double robustness and compare the results. ResultsSLE prevalence ranges between 41.65 and 54.47 (cases/100 000), which is lower than other Latin American countries. Using the operative definition of SLE, 5527 patients were selected. The potential control sample was composed of 1 942 253 patients. The total annual direct estimated cost per patient was US $2172. Adjusted incremental cost was US $1662. Considering 4 severity classes of SLE, the cost ranges from US $8823 for severe to US $447 for mild cases. ConclusionsIncremental costs of SLE in Colombia are similar to those from other middle-income countries. Compared with high-income countries, the cost is lower; nevertheless, if it is calculated proportional to the per capita health expenditure, it is comparable.