Early prediction of in-hospital death of COVID-19 patients: a machine-learning model based on age, blood analyses, and chest x-ray score

Abstract

An early-warning model to predict in-hospital mortality on admission of COVID-19 patients at an emergency department (ED) was developed and validate using a Machine-Learning model. In total, 2782 patients were enrolled between March 2020 and December 2020, including 2106 patients (first wave) and 676 patients (second wave) in the COVID-19 outbreak in Italy. The first-wave patients were divided into two groups with 1474 patients used to train the model, and 632 to validate it. The 676 patients in the second wave were used to test the model. Age, 17 blood analytes and Brescia chest X-ray score were the variables processed using a Random Forests classification algorithm to build and validate the model. ROC analysis was used to assess the model performances. A web-based death-risk calculator was implemented and integrated within the Laboratory Information System of the hospital. The final score was constructed by age (the most powerful predictor), blood analytes (the strongest predictors were lactate dehydrogenase, D-dimer, Neutrophil/Lymphocyte ratio, C-reactive protein, Lymphocyte %, Ferritin std and Monocyte %), and Brescia chest X-ray score. The areas under the receiver operating characteristic curve obtained for the three groups (training, validating and testing) were 0.98, 0.83 and 0.78, respectively. The model predicts in-hospital mortality on the basis of data that can be obtained in a short time, directly at the ED on admission. It functions as a web-based calculator, providing a risk score which is easy to interpret. It can be used in the triage process to support the decision on patient allocation.

Data availability

We are unable to share the dataset as it contains sensitive personal data collected during the pandemic at Spedali Civili di Brescia. We cannot share the full data since are data from patients. Interested researchers should contact the authors for any query related to data sharing and submit a project proposal Once defined the goal of the study, and the data needed authors will submit the potential project of collaboration to the IRB of Spedali Civili di Brescia to receive approval to access a deidentified dataset. Please note that other informations related to patients can be acquired, always following approoval of IRB of Spedali Civili di Brescia, not only the ones studied in the paper.Anyway, following request to the authors, it will be possible to share processed version of the dataset ( e.g. an Excel sheet with numbers used to plot the graphs and charts of the manuscript).All code used to analyse the data can be found on GitHub at https://github.com/biostatUniBS/BS_EWS

Article and author information

Author details

  1. Emirena Garrafa

    University of Brescia, Brescia, Italy
    For correspondence
    emirena.garrafa@unibs.it
    Competing interests
    The authors declare that no competing interests exist.
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-4761-6892
  2. Marika Vezzoli

    University of Brescia, Brescia, Italy
    Competing interests
    The authors declare that no competing interests exist.
  3. Marco Ravanelli

    University of Brescia, Brescia, Italy
    Competing interests
    The authors declare that no competing interests exist.
  4. Davide Farina

    University of Brescia, Brescia, Italy
    Competing interests
    The authors declare that no competing interests exist.
  5. Andrea Borghesi

    University of Brescia, Brescia, Italy
    Competing interests
    The authors declare that no competing interests exist.
  6. Stefano Calza

    University of Brescia, Brescia, Italy
    Competing interests
    The authors declare that no competing interests exist.
  7. Roberto Maroldi

    University of Brescia, Brescia, Italy
    Competing interests
    The authors declare that no competing interests exist.

Funding

Stefano Calza (PRIN 20178S4EK9)

  • Stefano Calza

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Reviewing Editor

  1. Evangelos J Giamarellos-Bourboulis, National and Kapodistrian University of Athens, Medical School, Greece

Ethics

Human subjects: The Institutional review board aprpoved the study with the entry code NP4000.

Version history

  1. Received: May 24, 2021
  2. Preprint posted: June 13, 2021 (view preprint)
  3. Accepted: October 17, 2021
  4. Accepted Manuscript published: October 18, 2021 (version 1)
  5. Accepted Manuscript updated: October 22, 2021 (version 2)
  6. Version of Record published: October 27, 2021 (version 3)

Copyright

© 2021, Garrafa et al.

This article is distributed under the terms of the Creative Commons Attribution License permitting unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 991
    views
  • 132
    downloads
  • 22
    citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Emirena Garrafa
  2. Marika Vezzoli
  3. Marco Ravanelli
  4. Davide Farina
  5. Andrea Borghesi
  6. Stefano Calza
  7. Roberto Maroldi
(2021)
Early prediction of in-hospital death of COVID-19 patients: a machine-learning model based on age, blood analyses, and chest x-ray score
eLife 10:e70640.
https://doi.org/10.7554/eLife.70640

Share this article

https://doi.org/10.7554/eLife.70640

Further reading

    1. Biochemistry and Chemical Biology
    2. Cell Biology
    Christopher TA Lewis, Elise G Melhedegaard ... Julien Ochala
    Research Article

    Hibernation is a period of metabolic suppression utilized by many small and large mammal species to survive during winter periods. As the underlying cellular and molecular mechanisms remain incompletely understood, our study aimed to determine whether skeletal muscle myosin and its metabolic efficiency undergo alterations during hibernation to optimize energy utilization. We isolated muscle fibers from small hibernators, Ictidomys tridecemlineatus and Eliomys quercinus and larger hibernators, Ursus arctos and Ursus americanus. We then conducted loaded Mant-ATP chase experiments alongside X-ray diffraction to measure resting myosin dynamics and its ATP demand. In parallel, we performed multiple proteomics analyses. Our results showed a preservation of myosin structure in U. arctos and U. americanus during hibernation, whilst in I. tridecemlineatus and E. quercinus, changes in myosin metabolic states during torpor unexpectedly led to higher levels in energy expenditure of type II, fast-twitch muscle fibers at ambient lab temperatures (20 °C). Upon repeating loaded Mant-ATP chase experiments at 8 °C (near the body temperature of torpid animals), we found that myosin ATP consumption in type II muscle fibers was reduced by 77–107% during torpor compared to active periods. Additionally, we observed Myh2 hyper-phosphorylation during torpor in I. tridecemilineatus, which was predicted to stabilize the myosin molecule. This may act as a potential molecular mechanism mitigating myosin-associated increases in skeletal muscle energy expenditure during periods of torpor in response to cold exposure. Altogether, we demonstrate that resting myosin is altered in hibernating mammals, contributing to significant changes to the ATP consumption of skeletal muscle. Additionally, we observe that it is further altered in response to cold exposure and highlight myosin as a potentially contributor to skeletal muscle non-shivering thermogenesis.

    1. Biochemistry and Chemical Biology
    2. Neuroscience
    Maximilian Nagel, Marco Niestroj ... Marc Spehr
    Research Article

    In most mammals, conspecific chemosensory communication relies on semiochemical release within complex bodily secretions and subsequent stimulus detection by the vomeronasal organ (VNO). Urine, a rich source of ethologically relevant chemosignals, conveys detailed information about sex, social hierarchy, health, and reproductive state, which becomes accessible to a conspecific via vomeronasal sampling. So far, however, numerous aspects of social chemosignaling along the vomeronasal pathway remain unclear. Moreover, since virtually all research on vomeronasal physiology is based on secretions derived from inbred laboratory mice, it remains uncertain whether such stimuli provide a true representation of potentially more relevant cues found in the wild. Here, we combine a robust low-noise VNO activity assay with comparative molecular profiling of sex- and strain-specific mouse urine samples from two inbred laboratory strains as well as from wild mice. With comprehensive molecular portraits of these secretions, VNO activity analysis now enables us to (i) assess whether and, if so, how much sex/strain-selective ‘raw’ chemical information in urine is accessible via vomeronasal sampling; (ii) identify which chemicals exhibit sufficient discriminatory power to signal an animal’s sex, strain, or both; (iii) determine the extent to which wild mouse secretions are unique; and (iv) analyze whether vomeronasal response profiles differ between strains. We report both sex- and, in particular, strain-selective VNO representations of chemical information. Within the urinary ‘secretome’, both volatile compounds and proteins exhibit sufficient discriminative power to provide sex- and strain-specific molecular fingerprints. While total protein amount is substantially enriched in male urine, females secrete a larger variety at overall comparatively low concentrations. Surprisingly, the molecular spectrum of wild mouse urine does not dramatically exceed that of inbred strains. Finally, vomeronasal response profiles differ between C57BL/6 and BALB/c animals, with particularly disparate representations of female semiochemicals.