Researchers create model to predict COVID-19 mortality

By Sam HancockReviewed by Aimee MolineuxJan 31 2022

The coronavirus disease 2019 (COVID-19) pandemic has caused pandemonium for hospitals worldwide, with many reporting overflowing intensive care units (ICUs) and a lack of beds.

While mass vaccination schemes have reduced hospitalizations in developed nations, many countries are still struggling with the disease.

Researchers from the Georgia Institute of Technology have created a model that can help determine the outcome of disease in patients in the ICU.

This news article was a review of a preliminary scientific report that had not undergone peer-review at the time of publication. Since its initial publication, the scientific report has now been peer reviewed and accepted for publication in a Scientific Journal. Links to the preliminary and peer-reviewed reports are available in the Sources section at the bottom of this article. View Sources

A preprint version of the study is available on the bioRxiv* server while the article undergoes peer review.

The study

The researchers characterized and quantified serum antibodies against canonical antigens from blood drawn from 21 severe COVID-19 patients using their platform. These patients were diagnosed using a standard nasopharyngeal swab followed by PCR and admitted to the ICU ~6 days following symptom onset.

The serum samples were taken within 24 hours of ICU admittance. The scientists aimed to determine if the antibody profiles of these samples taken at the point of ICU admission could predict the outcome of the disease. After observing that both those who survived and those who did not have significantly higher levels of specific antibodies than health controls from before the pandemic, the researchers decided to pursue a multivariate machine learning approach, incorporating different aspects of the antibody response.

A two-step machine learning approach was initially used, with feature selection using the LASSO and L1 regularization to prevent overfitting, followed by classification using the down-selected features. The models they created predicted outcome measured in a k-fold cross-validation with permutation testing, and were based on anti-spike IgA, anti-spike RBD IgA2 and RBD-directed antibody galactosylation. Partial least squares discriminant analysis (PLS-DA) using these three down-selected features helped visualize the stratification and demonstrated successfully that they could stratify the survivors and non-survivors, with all three features higher in survivors.

To ensure the model was reliable, its performance was tested on a cohort of severely ill ICU patients. The model generated was still significantly predictive, although its performance was slightly worse. This was explained due to a reduction in the information available on antigen-specific antibody glycosylation measurements, one of the key factors the model relied upon. Despite this, the results were still precise enough to validate the effectiveness of the model.

Following this, the scientists attempted to examine the levels of antibody responses that are directed against the non-canonical antigens nsp3, nsp13, orf3a and orf8 in patients with severe COVID-19 to discover if they were independently predictive of outcome. They detected antibody responses to these in both survivors and non-survivors, but could not identify any significant differences between them, so constructed a model similar to the one described above.

Unexpectedly, this model was as good at predicting outcomes as the previous model. It selected four features to analyze: anti-orf8 IgA, anti-nsp13 IgG3, anti-M antibody FcR3A binding and anti-M antibody galactosylation. PLS-DA analysis revealed these four features could differentiate between survivors and non survivors – typically showing higher results in survivors.

The results indicate higher antibody titers for isotopes directed against both canonical and non-canonical antigens can be associated with survival. The model also helped discover that increased galactosylation of RBD- and M- specific antibodies are associated with favorable outcomes, and this was corroborated by the construction of a predictive model combining both. To examine if the ratios of antibodies directed against certain outcomes were predictive of survival, the researchers conducted a post-hoc analysis focusing on the ratios of IgA/IgG antibodies for features identified in the previous two models. Multivariate PLS-DA visualization showed that these could successfully discriminate between survivors and non-survivors.

During the prior analyses, the scientists noticed levels of reactivities to nsp13 and nsp3, amongst others, in health control patients, and that these were different to the antibodies directed against canonical antigens. As previous studies have shown that antibodies can be generated against SARS-CoV-2 antigens through infection with different species of coronaviruses, they decided to test if this was the case here. They analyzed the sequence similarity of the spike protein and nsp13 against corresponding antigens in coronaviruses that circulated before SARS-CoV-2, and discovered that nsp13 had high similarity, while the spike protein did not.

Following this, they created a 3-way multivariate machine learning model to respond to the possibility of pre-existing cross-reactive antibodies reactive against SARS-CoV-2. This performed significantly better than the previously constructed models and included five features taken from all previous analyses.

Conclusion

The authors highlight that they have successfully created a model that can predict with high reliability the outcome of severe COVID-19 infection in patients from samples acquired shortly after they enter the ICU.

This model could be extremely useful for healthcare workers, and could potentially be used in triage for serious outbreaks, as well as helping to direct attention where it is most necessary.

This news article was a review of a preliminary scientific report that had not undergone peer-review at the time of publication. Since its initial publication, the scientific report has now been peer reviewed and accepted for publication in a Scientific Journal. Links to the preliminary and peer-reviewed reports are available in the Sources section at the bottom of this article. View Sources