Short-term predictor for COVID-19 severity from a longitudinal multi-omics study for practical application in intensive care units
Talanta, 2023
Kugler S., Hahnefeld L., Kloka J., Ginzel S., Nürenberg-Goloub E., Zinn S., Vehreschild M., Zacharowski K., Lindau S., Ullrich E., Burmeister J., Kohlhammer J., Schwäble J., Gurke R., Dorochow E., Bennett A., Dauth S., Campe J., Knape T., Laux V., Kannt A., Köhm M., Geisslinger G., Resch E., Behrens F.
| Disease area | Application area | Sample type | Products |
|---|---|---|---|
Infectious Diseases | Patient Stratification | Plasma | O Olink Target 96 |
Abstract
Background
The COVID-19 pandemic challenged the management of technical and human resources in intensive care units (ICU) across the world. Several long-term predictors for COVID-19 disease progression have been discovered. However, predictors to support short-term planning of resources and medication that can be translated to future pandemics are still missing. A workflow was established to identify a predictor for short-term COVID-19 disease progression in the acute phase of intensive care patients to support clinical decision-making.
Methods
Thirty-two patients with SARS-CoV-2 infection were recruited on admission to the ICU and clinical data collected. During their hospitalization, plasma samples were acquired from each patient on multiple occasions, excepting one patient for which only one time point was possible, and the proteome (Inflammation, Immune Response and Organ Damage panels from Olink® Target 96), metabolome and lipidome (flow injection analysis and liquid chromatography-mass spectrometry) analyzed for each sample. Patient visits were grouped according to changes in disease severity based on their respiratory and organ function, and evaluated using a combination of statistical analysis and machine learning. The resulting short-term predictor from this multi-omics approach was compared to the human assessment of disease progression. Furthermore, the potential markers were compared to the baseline levels of 50 healthy subjects with no known SARS-CoV-2 or other viral infections.
Results
A total of 124 clinical parameters, 271 proteins and 782 unique metabolites and lipids were assessed. The dimensionality of the dataset was reduced, selecting 47 from the 1177 parameters available following down-selection, to build the machine learning model. Subsequently, two proteins (C–C motif chemokine 7 (CCL7) and carbonic anhydrase 14 (CA14)) and one lipid (hexosylceramide 18:2; O2/20:0) were linked to disease progression in the studied SARS-CoV-2 infections. Thus, a predictor delivering the prognosis of an upcoming worsening of the patient’s condition up to five days in advance with a reasonable accuracy (79 % three days prior to event, 84 % four to five days prior to event) was found. Interestingly, the predictor’s performance was complementary to the clinicians’ capabilities to foresee a worsening of a patient.
Conclusion
This study presents a workflow to identify omics-based biomarkers to support clinical decision-making and resource management in the ICU. This was successfully applied to develop a short-term predictor for aggravation of COVID-19 symptoms. The applied methods can be adapted for future small cohort studies.