Predictive Modeling of Poor Outcome in Severe COVID-19: A Single-Center Observational Study Based on Clinical, Cytokine and Laboratory Profiles

Abstract

Pneumonia is the main cause of hospital admission in COVID-19 patients. We aimed to perform an extensive characterization of clinical, laboratory, and cytokine profiles in order to identify poor outcomes in COVID-19 patients. Methods: A prospective and consecutive study involving 108 COVID-19 patients was conducted between March and April 2020 at Hospital Clínico Universitario de Valladolid (Spain). Plasma samples from each patient were collected after emergency room admission. Forty-five serum cytokines were measured in duplicate, and clinical data were analyzed using SPPS version 25.0. Results: A multivariate predictive model showed high hepatocyte growth factor (HGF) plasma levels as the only cytokine related to intubation or death risk at hospital admission (OR = 7.38, 95%CI—(1.28–42.4), p = 0.025). There were no comorbidities included in the model except for the ABO blood group, in which the O blood group was associated with a 14-fold lower risk of a poor outcome. Other clinical variables were also included in the predictive model. The predictive model was internally validated by the receiver operating characteristic (ROC) curve with an area under the curve (AUC) of 0.94, a sensitivity of 91.7% and a specificity of 95%. The use of a bootstrapping method confirmed these results. Conclusions: A simple, robust, and quick predictive model, based on the ABO blood group, four common laboratory values, and one specific cytokine (HGF), could be used in order to predict poor outcomes in COVID-19 patients.

Early Routine Biomarkers of SARS-CoV-2 Morbidity and Mortality: Outcomes from an Emergency Section

“…Indeed, the present retrospective study is focused at the level of the emergency section only to predict severe COVID-19 outcomes early by comparing three different groups of patients (those attending only the emergency section; those entering an emergency section and attending the ICU for COVID-19 after; those entering an emergency section and attending the ICU after but with a fatal outcome). In order to identify poor outcomes, similar studies were certainly performed but with different schedules and groups of patients [ 28 , 29 ]. Usually, the main criteria used are age, fever, respiratory rate, respiratory distress, oxygen saturation levels, arterial blood oxygen partial pressure, and the presence of bilateral and peripheral ground-glass opacities [ 30 ].…”

“…Few studies have been conducted in the emergency unit regarding the study of early routine biomarkers for the prediction of morbidity and mortality of COVID-19 disease [ 28 , 29 , 32 , 33 ]. D-dimer is a product of fibrinolysis widely used as a marker of activation of the coagulation and fibrinolytic systems and for the diagnosis of thromboembolism [ 34 ].…”

Clinical, laboratory data and inflammatory biomarkers at baseline as early discharge predictors in hospitalized SARS-CoV-2 infected patients

“…Having a diagnostic tool to predict the evolution of these patients could be very useful. To date, many studies have evaluated the ability of different parameters to predict a poor prognosis [ 11 , 18 , 20 – 26 ]. However, none have focused on studying which parameters could reliably predict a favorable outcome for a safe early discharge in hospitalized patients.…”

“…In addition, clinical data and laboratory parameters such as lymphopenia, high plasma levels of lactate dehydrogenase (LDH), C-reactive protein (CRP), ferritin and D-dimer have been associated with a worse evolution and/or mortality. However, most of these studies do not provide the positive and negative predictive value of their model, making it difficult to know the true predictive capacity of these parameters [ 5 – 11 ].…”

Prognostic models in COVID-19 infection that predict severity: a systematic review

“…AUC assessed in all 44 studies ranged from 0.53 to 0.94, and sensitivity and specificity ranging from 21.5% to 98.6% and 13.7% to 89.2%, respectively. A combination of model development with internal and/or external validation was provided by 24 (54.5%) out of 44 studies with an AUC of 0.94 for the two models with the best predictive performance [ 294 , 316 ] . Additional information on prediction time, model performance and predictors are presented in Supplementary material Table 4 .…”