Current Models for COVID-19 Diagnosis, Prognosis Unreliable
Published models may be biased and unreliable and have overly optimistic performance estimates
FRIDAY, April 10, 2020 (HealthDay News) -- Novel coronavirus (COVID-19)-related models that are published in the literature are poorly reported and at high risk for bias, raising concern that their predictions could be unreliable when applied in daily practice, according to a review published online April 7 in The BMJ.
Laure Wynants, Ph.D., from Maastricht University in the Netherlands, and colleagues conducted a rapid systematic review and critically appraised studies that developed or validated a multivariable COVID-19-related prediction model.
The researchers identified 27 studies describing 31 prediction models, including three models for predicting hospital admission from pneumonia and other events (as proxy outcomes for COVID-19 pneumonia) in the general population; 18 models for diagnosing COVID-19 infection (13 of which used machine learning based on computed tomography scans); and 10 prognostic models for predicting mortality risk, progression to severe disease, or length of hospital stay. Only one study used patient data from outside of China. Age, body temperature, and signs and symptoms were the most reported predictors of the presence of COVID-19 in patients with suspected disease. For severe prognosis, the most reported predictors included age, sex, features derived from computed tomography scans, C-reactive protein, lactic dehydrogenase, and lymphocyte count. For prediction models for the general population, C index estimates ranged from 0.73 to 0.81; they ranged from 0.81 to more than 0.99 in diagnostic models (for 13 of 18 models) and from 0.85 to 0.98 in prognostic models (for six of 10 models). There was a high risk for bias in all studies, mostly because of nonrepresentative selection of control patients, exclusion of patients who had not experienced the event of interest by the end of the study, and a high risk for model overfitting. Most included studies did not include a description of the study population or the intended use of the model.
"Immediate sharing of well-documented individual participant data from COVID-19 studies is needed for collaborative efforts to develop more rigorous prediction models and validate existing ones," the authors write.