Predictors of mortality in hospitalised patients with COVID-19: a 1-year case-control study

Abstract

Background: A paucity of predictive models assessing risk factors for COVID-19 mortality that extend beyond age and gender in Latino population is evident in the current academic literature.

Objectives: To determine the associated factors with mortality, in addition to age and sex during the first year of the pandemic.

Design: A case-control study with retrospective revision of clinical and paraclinical variables by systematic revision of clinical records was conducted. Multiple imputations by chained equation were implemented to account for missing variables. Classification and regression trees (CART) were estimated to evaluate the interaction of associated factors on admission and their role in predicting mortality during hospitalisation. No intervention was performed.

Setting: High-complexity centre above 2640 m above sea level (masl) in Colombia.

Participants: A population sample of 564 patients admitted to the hospital with confirmed COVID-19 by PCR. Deceased patients (n=282) and a control group (n=282), matched by age, sex and month of admission, were included.

Main outcome measure: Mortality during hospitalisation.

Main results: After the imputation of datasets, CART analysis estimated 11 clinical profiles based on respiratory distress, haemoglobin, lactate dehydrogenase, partial pressure of oxygen to inspired partial pressure of oxygen ratio, chronic kidney disease, ferritin, creatinine and leucocytes on admission. The accuracy model for prediction was 80.4% (95% CI 71.8% to 87.3%), with an area under the curve of 78.8% (95% CI 69.63% to 87.93%).

Conclusions: This study discloses new interactions between clinical and paraclinical features beyond age and sex influencing mortality in COVID-19 patients. Furthermore, the predictive model could offer new clues for the personalised management of this condition in clinical settings.

Keywords: COVID-19; INFECTIOUS DISEASES; Risk Factors.

Figure 1

Missing data and imputation. (A) Histogram of frequency of missing variables in the total of patients included (n: 584). (B) Heatmap for the distribution of missing data. (C) Distribution of imputed variables by MICE. The red lines correspond to the five imputed datasets, whereas the blue line corresponds to the original dataset. (D) Heatmap for levels of laboratories during the year of follow-up. MICE, multiple imputations by chained equation; PaO₂/FiO₂, partial pressure of oxygen to inspired partial pressure of oxygen ratio.

Figure 2

Classification and regression trees (CART). This strategy estimated a predictive model and 11 clinical profiles based on respiratory distress, haemoglobin, LDH, PaO₂/FiO₂, chronic kidney disease, ferritin, creatinine and leucocytes on admission. The accuracy model for prediction was 80.4% (95% CI 71.8% to 87.3%), with an area under the curve of 78.8% (95% CI 69.63% to 87.93%) on testing sets. AUC, area under the curve; LDH, lactate dehydrogenase; PaO₂/FiO₂, partial pressure of oxygen to inspired partial pressure of oxygen ratio; ROC, receiving operating characteristics curve.