Novel approach to predict outcomes in COVID-19 patients

In a recent study published on the preprint server medRxiv*, researchers have used the predictive power of chest computed tomography (CT) scans and plasma cytokines to predict death and severity in coronavirus disease 2019 (COVID-19) patients.

Study: Quantitative chest CT combined with plasma cytokines predict outcomes in COVID-19 patients. Image Credit: QinJin /


Despite the COVID-19 pandemic persisting for almost two years since the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the virus responsible for COVID-19, originally emerged in December 2019, there are currently no reliable predictors that can assist physicians in predicting patient outcomes.

During COVID-19 progression, the cytokine assessment has been useful in the prediction of death. Furthermore, lung CT features have shown high predictive performance for COVID-19 severity.

In the current study, the researchers hypothesized that a combination of plasma cytokines and CT measurements would have a higher predictive power of COVID-19 outcomes than each method would predict independently. To test this, the authors of this study applied a data-driven machine learning approach.

About the study

The study cohort included 152 COVID-19 patients from the Mount Sinai Health System in New York admitted between March and September 2020. Each of these patients had their plasma cytokine levels assessed and a chest CT scan performed within 5 days of their admission into the hospital.

These patients were selected based on the following inclusion criteria for this study:

  • Hospitalized for COVID-19
  • Plasma cytokine assessment within 48 hours upon hospital admission
  • Chest CT scan performed up to 5 days apart from plasma cytokine assessment

In addition to the plasma cytokines of interleukin 6 (IL-6), IL-8, and tumor necrosis factor α (TNF-α) that were assessed and the chest CT, the researchers collected information on the patient demographics, as well as any clinical and laboratory variables. The researchers took meticulous care to exclude patients whose data may skew the objective of this study. For example, patients with acute conditions overlapping COVID-19 that may affect the cytokines were excluded.

In the study, radiologists calculated a CT qualitative score according to the percentage of lung parenchyma of each lobe affected by ground-glass opacification (GGO) and/or consolidations. This information was then summarized to yield an overall CT qualitative score that was between 0-20.

The CT quantitative assessment, which was performed using the open-source software three-dimensional (3D) slicer ( and the Chest Imaging Platform plug-in (, yielded five distinct variables. These variables included total lung volume, well-aerated lung volume, GGO volume, consolidation volume, and GGO to aerated lung ratio. In addition, the researchers also used the ELLA cytokine platform to measure IL-6, IL-8, and TNF-α.

To develop a robust tool for patient risk stratification to prioritize them for care, the researchers selected these two as appropriate outcomes, including maximum disease severity during hospitalization and hospital death. The World Health Organization (WHO) ordinal scale (0-7) was also used to assess disease severity prior to death.

All data were analyzed starting from 4 different scenarios, including cytokines, CT qualitative, CT quantitative, and combined scores.

The researchers also evaluated the probability of survival using Cox proportional hazard models to identify potential markers and used elastic net regression for predictive capabilities to separate patients that survive per scenario. Notably, the researchers indicated from the results that oxygen saturation and demographic variables had poor power in predicting death, even though these produce significant prognostic models in assessing the risk of death.

The researchers also increased their model robustness by performing a combination of random testing/training sets and cross-fold validation. A coefficient-based selection was also used to filter the significant models and select the variables relevant for predicting.

Study results

The study found that the hospital mortality rate for this cohort was 17.1%. While there were no significant differences in sex, race, ethnicity, or age between patients who died as compared to those who survived, patients who died had a higher WHO ordinal score and lower oxygen saturation at presentation as compared to those who survived.

The researchers also found that IL-6 and IL-8 levels were significantly higher in patients who died, while no significant difference for TNF-α. However, IL-6, TNF-α, and IL-8 were all correlated to disease severity.

On predicting COVID-19 death, the researchers showed that a combined model of CT scans based on additional information from cytokine assays increases the predictive power of death prediction. The optimized prediction scenario contained IL-6, IL-8, TNF-α, GGO to aerated lung ratio, and age.

Predicting the COVID-19 maximum severity score, the researchers found that the combined scenario performed better than the optimized scenario, CT quantitative, CT qualitative, and cytokines assessments alone. With the use of these variables, the researchers then build a risk prediction nomogram.  This nomogram uses selected variables of GGO to aerated lung ratio, age, TNF-α, IL-6, and IL-8 to provide a score for risk of death.

The researchers also provided a simple scoring system using plasma IL-6, IL-8, TNF-α, GGO to aerated lung ratio, and age as novel metrics, thereby suggesting these may be used as red flags when monitoring the patients. This would help physicians make critical decisions for patients at high risk of death for COVID-19.


Taken together, the combined approach of both chest CT scans with the assessment of plasma cytokines was found to be good predictors of death and maximum severity of COVID-19.

Notably, the CT quantitative was better at predicting severity, while the cytokine measurements better predicted death. The researchers also built a nomogram to predict the risk of COVID-19-related death using a combination of cytokines and CT variables.

Overall, the findings presented here can assist physicians in risk stratification and making critical decisions for individualized therapeutic strategies for patients.

*Important notice

medRxiv publishes preliminary scientific reports that are not peer-reviewed and, therefore, should not be regarded as conclusive, guide clinical practice/health-related behavior, or treated as established information.

Journal reference:
Dr. Ramya Dwivedi

Written by

Dr. Ramya Dwivedi

Ramya has a Ph.D. in Biotechnology from the National Chemical Laboratories (CSIR-NCL), in Pune. Her work consisted of functionalizing nanoparticles with different molecules of biological interest, studying the reaction system and establishing useful applications.


Please use one of the following formats to cite this article in your essay, paper or report:

  • APA

    Dwivedi, Ramya. (2021, October 19). Novel approach to predict outcomes in COVID-19 patients. News-Medical. Retrieved on December 01, 2021 from

  • MLA

    Dwivedi, Ramya. "Novel approach to predict outcomes in COVID-19 patients". News-Medical. 01 December 2021. <>.

  • Chicago

    Dwivedi, Ramya. "Novel approach to predict outcomes in COVID-19 patients". News-Medical. (accessed December 01, 2021).

  • Harvard

    Dwivedi, Ramya. 2021. Novel approach to predict outcomes in COVID-19 patients. News-Medical, viewed 01 December 2021,


The opinions expressed here are the views of the writer and do not necessarily reflect the views and opinions of News Medical.
You might also like... ×
How COVID-19 protection conferred by infection is modified by vaccination