| >A unique two biomarker profile consisting of ANG-1/P-SEL was developed with machine learning, providing a classification accuracy for Long-COVID status of 96%. The first paper used a random forest-based decision tree classifier built on markers in blood assays. Neat. However, this study has a major flaw. They rated their classifier's accuracy on classifying blood marker profiles of acutely ill COVID patients, long COVID patients experiencing "diffuse symptoms" referred with "no selection process", and a healthy control group. The control group consisted of healthy patients whose blood samples had been banked prior to the COVID-19 pandemic. It's not clear whether they compared their classifier's results against people who've had COVID and recovered without issue, versus those who had COVID and continued to experience symptoms long after recovery. That is the entire point of developing such a classifier. This paper is worthless without that comparison. Second paper has the same problem, and is honest about it: >The healthy control subjects were individuals without disease, acute illness or prescription medications and were previously banked in the Translational Research Centre, London, ON (Directed by Dr. D.D. Fraser; https://translationalresearchcentre.com/). These latter samples were obtained prior to the emergence of SARS-CoV-2 in our region and therefore, were considered not to have been exposed to the virus. |
I only added these to say there's enough evidence to be questionable of the null hypothesis with regard to long COVID physiopathology.