The BEMC talk in June 2020 titled “Disorderly World of Diagnostic and Prognostic Models for Covid-19” was presented by Laure Wynants. The project started at the beginning of the pandemic. The author read a tweet from somebody asking for advice on building a prediction model for COVID-19 patients. There is a clinical need for prediction models to improve care and reduce costs. If a good model is built it could help allocate core resources for clinician to answer these questions (1) who needs to undergo further diagnostic work-up, (2) to speed up CT interpretation, (3) who should be admitted to ICU?

Their first review named “Prediction models for diagnosis and prognosis of covid-19: systematic review of critical appraisal” took 18 days from building the idea to being accepted. This work involved 14 experienced risk modelling reviewers. Screening and data extraction was done independently by 2 reviewers. Laure Wynants and her research team did not only look at the published materials but also pre-prints. The number of pre-prints was much bigger than traditional scientific publishing. Although there are many issues around pre-prints, given the pandemic situation, the traditional scientific publishing is not keeping up the speed of scientific research that’s being done. The scope of the review is having prediction models for diagnosing coronavirus diseases 2019 (covid-19) in patients with suspected infection, for prognosis of patients with covid-19, and for detecting people in the general population at risk of being admitted to hospital for covid-19 pneumonia. Clinical prediction models will support clinical decision-making for individual patients by combining and giving appropriate weights to several inputs (e.g., CT image characteristics, signs and symptoms, lab test results, demographics …). Laure Wynants and her research team used PROBAST tools to assess the risk of bias and use CHARMS for data extraction. Then the PRISMA and TRIPOD were used for the author’s reporting. With collaboration with Cochrane prognosis methods group and the expanded group, the first round was run with an AI tool. The first round included 107 studies with 145 models which proposed diagnostic of prognostic model for covid-19. Among 107 papers, there were 87 preprints and 20 peer-reviewed and published. Among those models, 4 model for detecting people at risk in general population, 91 diagnosis model for COVID-19 or covid-19 pneumonia and 50 prognosis models for predicting mortality risk, progression to severe disease, or composite outcomes.

Contributed by: Pham Thi Thu