Identificação de preditores de tromboembolismo venoso em pacientes com Covid-19 por meio de técnicas de inteligência artificial: resultados do registro brasileiro de Covid-19

Abstract

Introduction: Previous studies that evaluated risk factors for venous thromboembolism (VTE) in patients with COVID-19 found inconsistent results. The aim of this study was to investigate VTE predictors by two analysis approaches - logistic regression (RL) and machine learning (ML) - due to their potential complementarity.

Methods: This is a retrospective cohort study based on a Brazilian COVID-19 registry, which included adult patients with confirmed COVID-19 admitted to 16 hospitals between March and September 2020. Pregnant women and patients who manifested COVID-19 during hospitalization were excluded. Symptomatic thromboembolic events were confirmed by imaging tests. All variables were collected at hospital admission, with the exception of the use of anticoagulants during hospitalization. The assessment of VTE predictors was analyzed using traditional statistics (RL) and ML, more specifically tree-based algorithm and bagging combined with Shapley Additive ExPlanation (SHAP) to investigate the association of variables with VTE.

Results: Among the 4,120 patients included (55.5% men, 39.3% critically ill patients), VTE was confirmed in 274 (6.7%). In the multivariate analysis of LR, obesity (Odds ratio (OR) 1.50, Confidence interval (CI) 95% 1.11–2.02); being an ex-smoker (OR 1.44, 95% CI 1.03–2.01); surgery in the last 90 days (OR 2.20, 95% CI 1.14–4.23); axillary temperature (OR 1.41, 95% CI 1.22–1.63); D-dimer 4 times or more above the upper limit of the reference value (OR 2.16, 95% CI 1.26–3.67), lactate (OR 1.10, 95% CI 1.02–1.19 ), C-reactive protein levels (CRP; OR 1.09, 95% CI 1.01–1.18); and elevated neutrophil count (OR 1.04, 95% CI 1.005–1.075) were independent predictors of VTE. Atrial fibrillation (OR 0.30, CI 95% 0.09 – 0.99), elevated peripheral oxygen saturation/fraction of inspired oxygen ratio (SF; OR 0.87, CI 95% 0.83 – 0.93) and prophylactic use of anticoagulants (OR 0.20, 95% CI 0.15 – 0.26) were protective. Temperature on admission, neutrophil count, D-dimer, CRP and lactate levels were also identified as predictors by ML methods.

Conclusion: Using ML and RL analyses, we show that D-dimer, axillary temperature, neutrophil count, CRP and lactate levels are risk factors for VTE in patients with COVID-19. Thus, until more definitive data are available, we suggest that patients with

11

alterations in these risk predictors be monitored more closely for the occurrence of VTE.