THE CREATION OF PREDICTIVE MODELS FOR ASSESSING THE SEVERITY OF COMMUNITY-ACQUIRED PNEUMONIA

Abstract

Community-acquired pneumonia (CAP) is a leading cause of mortality from lower respiratory tract infections and is associated with high incidence and unfavorable prognosis. In this regard, the timely assessment of the severity of CAP at the stage of hospitalization of the patient comes to the first place. The existing scales have a number of limitations, therefore they can’t always be better than the clinical solution. The aim of the research is to search for predictors of severe CAP and combine the most significant ones into a predictive model. There were examined 418 patients with CAP. The severity was determined according to IDSA/ATS criteria. Static analysis was performed in IBM SPSS Statistics. Logistic regression was used to identify and combine in a model the most significant criteria. The criteria were included in the predictive model with odds ratio (OR) >2. Demographic, laboratory, radiological and clinical indicators were analyzed in the course of the retrospective analysis. Significant differences between groups of the severity of pneumonia groups were revealed in 16 predictors. All predictors were included in the predictive model with odds ratios >2. As a result there were selected 7 criteria: age >40 years old, heart rate >93 bpm, the presence of HIV infection, liver disease, lesion >1 lung lobes, C-reactive protein >156 mg/L, creatinine >123 mmol/L. All predictors were combined using logistic regression. The resulting model was examined by ROC analysis. The area under the curve (AUC) was 0.88. Sensitivity and specificity were 87.5 and 73.5%, respectively. Thus, the article proposes a model for determining the severity of pneumonia (AUC=0.88), which includes the criteria used in the routine practice of pulmonologists in the Russian Federation. Further research is needed to create a scale based on the presented model.