Improved Accuracy of Long Covid-19 Prediction in Hospitalized Hypertensive Patients Compared to the General Cohort

Objective: The accuracy of prediction of Long COVID-19 development is currently suboptimal. Phenotyping a general post-COVID-19 population based on simple clinical and anamnestic criteria may be used to address this problem. The purpose of this study was to identify the early predictors of Long COVID-19 syndrome at 3 months after discharge in hospitalized patients with a history of hypertension (HT). Design and method: 221 hospitalized patients with COVID-19 (mean age 53.4+-13.6 years, 53% female, 40% with a history of hypertension) underwent comprehensive transthoracic echocardiography and a 6-minute walk test (6MWT) 1-2 days prior to discharge and after 1 month. A combination of logistic regression analysis and unsupervised machine learning was used to derive predictive models of self-assessed Long COVID-19 at 3 months after discharge (166 patients who have completed all visits comprised the final cohort). Results: Self-assessed Long COVID-19 at 3 months was detected in 50 (77%) of hypertensive and 74 (73%) of normotensive participants, p = 0,597. In the general cohort, a modest performance of 83% was observed in the validation subset for the derived pre-discharge predictors-based ML classification model, which only constituted to re-classification of 1/3 of otherwise false-positive cases compared to the assumption of 100% prevalence of Long COVID, with a total net reclassification improvement (NRI) of 8%. Phenotyping the general cohort based on the presence of hypertensive anamnesis significantly improved the predictive accuracy for the HT-positive subgroup: 100% accuracy was observed in the validation subset with an NRI of 23% vs baseline assumption, Chi-square test P < 0,001 vs the general cohort model confusion matrix. The derived models used sex, MRC dyspnea class, Charlson comorbidity index, and either (1) pre-discharge systolic BP and mitral E to left ventricular global longitudinal strain ratio, or (2) peak D-dimer, pre-discharge 6-minute walk distance, and peak capillary oxygen desaturation during the 6MWT. Conclusions: History of hypertension can be used as a phenotyping criterion to improve the accuracy of Long COVID-19 prediction among hospitalized patients: the derived ML-based classification models yielded a 100% predictive accuracy in the pre-specified validation subset of hypertensive patients hospitalized for acute COVID-19.