Statistical modeling to determine severity of respiratory sarcoidosis and parameters associated with cardiac sarcoidosis: as a way to stratify the risk of developing pulmonary hypertension

Aim. Using statistical modeling techniques, we aim to develop a model that optimizes the prediction of severity of sarcoidosis that affects the respiratory system (SRS) based on the identification and determination of signs (anamnestic, clinical, laboratory, instrumental examination data, etc.) associated with disease severity and subsequent stratification of the long-term risk for pulmonary hypertension (PH) development. Materials and methods. The 12-year observational cohort comparative study included 298 participants, both male and female, who had SRS. More than 200 different patient examination parameters were analyzed. The models were built using logistic regression and linear discriminant analysis. The quality of the models was assessed by constructing a classification matrix, calculating sensitivity and specificity as well as calculating the area under ROC curve. Results. As a result of the study, optimal classification models were developed for predicting SRS severity, constructed using various methods of statistical modelling. The models demonstrated that several characteristics, including parameters of echocardiography examination of patients (including indicators that allow for indirect diagnosis of PH), are associated with disease severity. A set of characteristics associated with particular sarcoidosis severity will allow prediction of it upon confirmation of diagnosis (individual prognosis), as well as patient management (observation or requiring the prescription of pathogenetic immunosuppressive therapy). Conclusion. Such a complex model for predicting disease severity in patients with a non-cardiac profile (SRS) is of great importance for risk stratification in terms of PH development in patients with severe sarcoidosis. Further analysis of the features identified during model construction can help clinicians to contribute to more accurate predictions of SRS severity in real-world clinical practice.