Table 2 Features selected by recursive feature optimisation (RFE) employing logistic regression (LR) and decision trees (DT).

From: Early outcome detection for COVID-19 patients

Coverage threshold

DT-RFE features

LR-RFE features

90% (171 patients, 59 variables)

Age(14), creatinine(14), platelets(18), lymphocytes(19), P/F(20)

Anticoagulants(12), diarrhea(14), chronic liver disease(15), asthma(17), age(20), COPD(20), P/F(20)

75% (74 patients, 68 variables)

Troponin(11), sodium(14), glucose(15), creatinine(19), BUN(20), PCT(20), P/F(20)

PCT(13), immunosuppressives (14), diarrhea(16), asthma(17), age(19), COPD(20), fever(20)

  1. We show two different experiments, with different initial selection of starting clinical variables (coverage of 90% and 75% of patients). In parentheses we show the frequency with which the feature was selected, out of 20 validation runs (20-fold cross validation). A frequency of 20 means features were selected in all validation runs. We only include features selected in over 50% of the runs.
Back to article page