Identification of Early Biomarkers of Mortality in COVID-19 Hospitalized Patients: A LASSO-Based Cox and Logistic Approach

This study aimed to identify possible early biomarkers of mortality among clinical and biochemical parameters, iron metabolism parameters, and cytokines detected within 24 h from admission in hospitalized COVID-19 patients. We enrolled 80 hospitalized patients (40 survivors and 40 non-survivors) with COVID-19 pneumonia and acute respiratory failure. The median time from the onset of COVID-19 symptoms to hospital admission was lower in non-survivors than survivors (p < 0.05). Respiratory failure, expressed as the ratio of arterial oxygen partial pressure to the fraction of inspired oxygen (P/F), was more severe in non-survivors than survivors (p < 0.0001). Comorbidities were similar in both groups. Among biochemical parameters and cytokines, eGFR and interleukin (IL)-1β were found to be significantly lower (p < 0.05), while LDH, IL-10, and IL-8 were significantly higher in non-survivors than in survivors (p < 0.0005, p < 0.05 and p < 0.005, respectively). Among other parameters, LDH values distribution showed the most significant difference between study groups (p < 0.0001). LASSO feature selection combined with Cox proportional hazards and logistic regression models was applied to identify features distinguishing between survivors and non-survivors. Both approaches highlighted LDH as the strongest predictor, with IL-22 and creatinine emerging in the Cox model, while IL-10, eGFR, and creatinine were influential in the logistic model (AUC = 0.744 for Cox, 0.723 for logistic regression). In a similar manner, we applied linear regression for predicting LDH levels, identifying the P/F ratio as the top predictor, followed by IL-10 and eGFR (NRMSE = 0.128). Collectively, these findings underscore LDH's critical role in mortality prediction, with P/F and IL-10 as key determinants of LDH increases in this Italian COVID-19 cohort.