Prediction of pulmonary gas exchange disorders in patients with long-term COVID-19 using machine learning methods

Abstract

Introduction. Hospital discharge after COVID-19 does not mean a complete recovery.Aim. To predict lung gas-exchange impairment in patients after COVID-19-associated pneumonia.Materials and methods. An observational retrospective cross-sectional study was conducted. 316 patients (78% men) with long-term COVID-19 and postCOVID computed tomography (CT) changes, without lung diseases in history were enrolled. Spirometry, body plethysmography, diffusion test were performed.Results. In whole group the medians of ventilation parameters were within the normal ranges. However, 78 (25%) patients had a restrictive type of ventilation disorders, 23 (7%) had airway obstruction, and 174 (55%) had a decrease in diffusion capacity of the lungs (DLCO). The general group was divided into two subgroups depending on the DLCO value: subgroup 1 – DLCO is within the normal range and subgroup 2 – DLCO is reduced. The DLCO analysis between the subgroups showed statistically significant differences in duration from the COVID19 onset (lower in subgroup 2) and in the computer tomography abnormalities in the acute period of COVID-19 (CTmax) (more in subgroup 2) whereas there were no differences in gender, age, body mass index (BMI). Analyzing the odds ratio showed that the chance of a decrease in DLCO after COVID-19 increased 6.5 times with CTmax of more than 45%, 4 times with a duration from the COVID-19 onset less than 225 days, 1.9 times if the age is younger than 63 years while male gender and BMI did not have an impact on DLCO in the post-COVID period. The logistic regression model with identified predictors demonstrated the accuracy, sensitivity and specificity of 81%, 82%, 80%, respectively.Conclusion. According to our model CTmax of more than 45%, the duration from the COVID-19 onset less than 225 days, age younger than 63 years are important predictors for reducing DLCO after COVID-19.