Blood transcriptomics and machine learning can predict secondary respiratory bacterial infections in COVID-19 patients

Researchers from the University of Queensland have utilized machine learning, specifically the “least absolute shrinkage and selection operator” (LASSO), to predict the risk of secondary bacterial infections in hospitalized COVID-19 patients. The study, published in The Lancet Microbe, involved the analysis of blood samples from COVID-19 patients in six countries. The team identified the expression of seven host genes (DAPP1, CST3, FGL2, GCH1, CIITA, UPP1, and RN7SL1) in the blood at the time of study admission as predictive of the risk of developing a secondary bacterial infection of the respiratory tract more than 24 hours after admission. Combining the expression of these genes with a patient’s World Health Organization (WHO) severity score at the time of study enrollment resulted in a highly predictive model. The combination achieved an area under the curve of 0.98 in the test cohort, with a true positive rate (sensitivity) of 1.00 and a true negative rate (specificity) of 0.94. The researchers suggest that the predictive model can assist in making informed choices about antibiotic use, reducing the risk of over-treatment and potential antibiotic resistance.

Key Points:

1. Machine Learning for Risk Prediction: The study utilized machine learning, specifically the least absolute shrinkage and selection operator (LASSO), to predict the risk of secondary bacterial infections in hospitalized COVID-19 patients.
2. Prediction Model based on Gene Expression: Blood samples from COVID-19 patients were analyzed to identify the expression of seven host genes (DAPP1, CST3, FGL2, GCH1, CIITA, UPP1, and RN7SL1) predictive of the risk of developing a secondary bacterial infection of the respiratory tract more than 24 hours after admission.
3. Combination with WHO Severity Score: The expression of the identified genes was combined with a patient’s World Health Organization (WHO) severity score at the time of study enrollment to create a predictive model.
4. High Predictive Performance: The combination achieved an area under the curve of 0.98 in the test cohort, with a true positive rate (sensitivity) of 1.00 and a true negative rate (specificity) of 0.94.
5. Informing Antibiotic Use: The predictive model can assist clinicians in making informed choices about antibiotic use, potentially reducing the risk of over-treatment and antibiotic resistance.
6. Potential Impact on Antibiotic Prescription: The data raise the possibility that gene transcription and analysis, combined with machine learning, at the time of clinical presentation in a hospital can provide valuable information for guiding antibiotic prescription.
7. International Collaboration: The study involved an extensive international collaboration of clinicians, virologists, bioinformaticians, and experts, including the PREDICT Consortium, the Snow Foundation, and the Nepean Hospital.
8. Simplified Machine Learning Approach: LASSO, utilized in this study, is considered a less complex machine learning method compared to some other artificial intelligence methods, making it more accessible and transparent for scientists across different fields.
9. Importance of Data Science in Medicine: The study emphasizes the potential for data science, particularly machine learning, to revolutionize the medical industry, making it less of a black box and promoting better understanding among scientists.

More information: Meagan Carney et al, Host transcriptomics and machine learning for secondary bacterial infections in patients with COVID-19: a prospective, observational cohort study, The Lancet Microbe (2024). DOI: 10.1016/S2666-5247(23)00363-4