COVID-19 mortality prediction models: Questions on Reliabili
Now open: Certificate Course in Management of Covid-19 by Govt. Of Gujarat and PlexusMDKnow more...Now open: Certificate Course in Management of Covid-19 by Govt. Of Gujarat and PlexusMDKnow more...
Get authentic, real-time news that helps you fight COVID-19 better.
Install PlexusMD App for doctors. It's free.
Research teams at Mount Sinai and Johns Hopkins have released two different prediction models to predict disease severity and chances of death in COVID-19 patients.

• MOUNT SINAI:

Used machine learning and clinical dataset (over 3,800 patients) including three major clinical features - age of the patient, minimum oxygen saturation and type of patient encounter (inpatient, outpatient or telehealth visits). The study, published in The Lancet, suggests the model is highly accurate.

• JOHN HOPKINS:

COVID-19 risk calculator (available online) includes various COVID-19 risk factors - patient age, BMI, presence of chronic disease and lung health along with the vital signs of the patient and the symptom severity at the time of admission to the hospital. The study, published in the journal Annals of Internal Medicine, was a retrospective cohort analysis that included 832 patients.

Link to John Hopkins Calculator: https://rsconnect.biostat.jhsph.edu/covid_predict/

• QUESTIONS ON RELIABILITY

A letter published in the European Respiratory Journal, a group of researchers from the UK and Netherlands mentioned 66 prediction models and suggested that all these models are prone to bias due to data quality, reporting and statistical analysis.

The authors of the study explained how missing data (patient information which is bound to be missing for all the factors needed to make a prediction) may affect the accuracy of the model and how artificially balancing data is not a good predictor of real outcomes.

Source: https://www.firstpost.com/health/covid-19-mortality-prediction-models-how-reliable-are-they-in-predicting-death-during-the-pandemic-8844001.html
Dr. T●●●●z H●●●●●●i and 3 others like this
Like
Comment
Share