In survival analysis, time-varying covariates are endogenous when their measurements are directly related to the event status and incomplete information occur at random points during the follow-up. Consequently, the time-dependent Cox model leads to biased estimates. Joint models (JM) allow to correctly estimate these associations combining a survival and longitudinal sub-models by means of a shared parameter (i.e., random effects of the longitudinal sub-model are inserted in the survival one). This study aims at showing the use of JM to evaluate the association between a set of inflammatory biomarkers and Covid-19 mortality. During Covid-19 pandemic, physicians at Istituto Clinico di Città Studi in Milan collected biomarkers (endogenous time-varying covariates) to understand what might be used as prognostic factors for mortality. Furthermore, in the first epidemic outbreak, physicians did not have standard clinical protocols for management of Covid-19 disease and measurements of biomarkers were highly incomplete especially at the baseline. Between February and March 2020, a total of 403 COVID-19 patients were admitted. Baseline characteristics included sex and age, whereas biomarkers measurements, during hospital stay, included log-ferritin, log-lymphocytes, log-neutrophil granulocytes, log-C-reactive protein, glucose and LDH. A Bayesian approach using Markov chain Monte Carlo algorithm were used for fitting JM. Independent and non-informative priors for the fixed effects (age and sex) and for shared parameters were used. Hazard ratios (HR) from a (biased) time-dependent Cox and joint models for log-ferritin levels were 2.10 (1.67-2.64) and 1.73 (1.38-2.20), respectively. In multivariable JM, doubling of biomarker levels resulted in a significantly increase of mortality risk for log-neutrophil granulocytes, HR=1.78 (1.16-2.69); for log-C-reactive protein, HR=1.44 (1.13-1.83); and for LDH, HR=1.28 (1.09-1.49). Increasing of 100 mg/dl of glucose resulted in a HR=2.44 (1.28-4.26). Age, however, showed the strongest effect with mortality risk starting to rise from 60 years.
University of Milan, Italy - ORCID: 0000-0002-6481-990X
University of Milan, Italy - ORCID: 0000-0002-4542-4996
University of Milan, Italy
University of Milan, Italy - ORCID: 0000-0002-3199-9369
University of Milan, Italy - ORCID: 0000-0001-9358-011X
Chapter Title
Longitudinal profile of a set of biomarkers in predicting Covid-19 mortality using joint models
Authors
Matteo Di Maso, Monica Ferraroni, Pasquale Ferrante, Serena Delbue, Federico Ambrogi
Language
English
DOI
10.36253/978-88-5518-461-8.36
Peer Reviewed
Publication Year
2021
Copyright Information
© 2021 Author(s)
Content License
Metadata License
Book Title
ASA 2021 Statistics and Information Systems for Policy Evaluation
Book Subtitle
BOOK OF SHORT PAPERS of the on-site conference
Editors
Bruno Bertaccini, Luigi Fabbris, Alessandra Petrucci
Peer Reviewed
Publication Year
2021
Copyright Information
© 2021 Author(s)
Content License
Metadata License
Publisher Name
Firenze University Press
DOI
10.36253/978-88-5518-461-8
eISBN (pdf)
978-88-5518-461-8
eISBN (xml)
978-88-5518-462-5
Series Title
Proceedings e report
Series ISSN
2704-601X
Series E-ISSN
2704-5846