Clinical Prediction Models For Inpatients With COVID-19: Do They Work?

Which prognostic models or indicators effectively predict critical care outcomes among patients hospitalized for COVID-19 pneumonia?

Models for predicting all-cause death and critical illness among hospitalized patients with acute COVID-19 show modest performance, according to study results published in Respiratory Medicine.

Researchers in Italy sought to validate and potentially improve clinical models for prognosis prediction in patients hospitalized with COVID-19. Toward that end, they conducted a study of numerous clinical models for predicting all-cause death or critical illness at 30 days of hospital admission for COVID-19 pneumonia.

The researchers enrolled 1044 consecutive patients with COVID-19 pneumonia at 5 nonintensive care unit centers during a 2020 outbreak in Italy. For each patient, prognostic scores for pneumonia and/or sepsis, along with COVID-19 specific scores, were calculated using 12 validated risk stratification tools (APACHE II, COVID-GRAM, CSS, CURB-65, 4C, HACOR, MEWS, NEWS II, qSOFA, REMS, ROX-index, and SOFA). Researchers compared the accuracy of scores for predicting in-hospital death at 30 days, the composite of death and orotracheal intubation.

The researchers found that 28.9% of patients presented critical illness 21.6% died during their hospital stay. Of 34 items included in the prognostic scores, the following were found to be independent predictors of in-hospital death at 30 days: (1) age of 60 years or older (hazard ratio (HR) 4.13; 95% CI, 1.49-11.43); (2) the presence of at least 2 comorbidities (HR 2.43; 95% CI, 1.57-3.76); (3) Glasgow coma scale lower than 15 (HR 1.94; 95% CI, 1.07-3.54); (4) mean blood pressure lower than 70 mmHg (HR 4.19; 95% CI, 1.50-11.71); (5) respiratory rate higher than 20 bpm (HR 1.58; 95% CI, 1.00-2.50); (6) ratio of partial pressure of oxygen in arterial blood (PaO2) to fraction of expired oxygen (FiO2) lower than 200 mmHg (HR 1.88; 95% CI, 1.22-2.89); (7) PaO2 lower than 60 mmHg (HR 1.60; 95% CI, 1.05-2.45); (8) oxygen saturation lower than 90% (HR 1.73; 95% CI, 1.12-2.69); and (10) respiratory index lower or equal to 3.8 (HR 1.82; 95% CI, 1.19-2.80).

The ability to discriminate categories of patients with varying levels of risk for in-hospital death at 30 days was deemed acceptable for APACHE II, COVID-GRAM, REMS, CURB-65, NEWS II, ROX-index, 4C, and SOFA, and poor for the rest. Both the REMS and 4C scores had high negative predictive values, 100.0% and 98.7%, respectively. Positive predictive value was poor overall and highest for ROX-index (75.0%).

This study is limited by its observational nature, the use of admission data as predictors, and the potential inclusion of patients with extremely poor prognosis. Also, results from this study population of hospitalized patients not in intensive care cannot be generalized to the whole COVID-19 population, and vaccination may now influence clinical outcomes, the researchers noted.

The researchers concluded that the overall performance of prognostic scores was modest, and that “simple predictors [such] as age and PaO2/FiO2 ratio can also be useful as standalone predictors to inform decision making.”


Vedovati MC, Barbieri G, Urbinin C, et al. Clinical prediction models in hospitalized patients with COVID-19: a multicenter cohort study. Respir Med. Published online August 21, 2022. doi:10.1016/j.rmed.2022.106954