AUTHOR=Bian Yi , Le Yue , Du Han , Chen Junfang , Zhang Ping , He Zhigang , Wang Ye , Yu Shanshan , Fang Yu , Yu Gang , Ling Jianmin , Feng Yikuan , Wei Sheng , Huang Jiao , Xiao Liuniu , Zheng Yingfang , Yu Zhen , Li Shusheng 

TITLE=Efficacy and Safety of Anticoagulation Treatment in COVID-19 Patient Subgroups Identified by Clinical-Based Stratification and Unsupervised Machine Learning: A Matched Cohort Study

JOURNAL=Frontiers in Medicine

VOLUME=Volume 8 - 2021

YEAR=2021

URL=https://www.frontiersin.org/journals/medicine/articles/10.3389/fmed.2021.786414

DOI=10.3389/fmed.2021.786414

ISSN=2296-858X

ABSTRACT=Objective: Explore efficacy of anticoagulation in improving outcomes and safety of COVID-19 patients within predefined clinical subgroups and in subgroups determined by unsupervised machine learning.
Methods: This single-center, retrospective and propensity score-matched cohort study unselectively reviewed 2272 patients with COVID-19 admitted to the Tongji Hospital between Jan 25 and Mar 23, 2020. Propensity score-matching between patients adjusted for potential covariates was carried out with the patients divided into two groups depending on whether or not they had received anticoagulation (AC) (AC group, ≥ 7 days of treatment; non-AC group, no treatment). This yielded 164 patients in each group. Predefined subgroup analysis and unsupervised learning approach were applied to identify clinically meaningful features of patient subgroups, detect patients benefiting most from AC and verify the safety.
Results: In-hospital mortality of the AC group was significantly lower than that of the non-AC group. There was a significantly higher incidence of clinically relevant non-major bleeding but not major bleeding in the AC group. Predefined subgroup analysis showed that, at admission, patients of severe or critical cases, moderate or severe acute respiratory distress syndrome (ARDS) cases, as well as patients with a D-dimer level ≥0.5 μg/mL, can benefit from AC. During the hospital stay, in critical cases and severe ARDS cases, patients who received AC had significantly lower in-hospital mortality. Unsupervised machine learning analysis established a six-class clustering model. Cluster 1, 2 and 6 were non-critical cases and had no response to AC, while cluster 3, 4 and 5 were critical patients. Among critical groups, cluster 4 could benefit from AC with no increase in bleeding events, whereas cluster 3 and 5, which had disturbance of consciousness, needed vasopressor support, and had higher white blood cells and neutrophils counts, cannot benefit from AC.
Conclusions: AC treatment decreased the risk of in-hospital mortality and did not increase major bleeding, especially in predefined critically ill COVDI-19 patients. Specifically, unsupervised learning analysis revealed that the most critically ill patients with disturbance of consciousness, hemodynamic unstable and elevated inflammatory biomarkers cannot benefit from AC and AC treatment can increase the incidence of bleeding events.