Appendix Table 5.

Estimated AUROCC for All Candidate Models

Derivation groupValidation group
A priori symptom score0.770.790.750.56
Cough/fever heuristic0.710.750.670.55
Cough/fever symptom score0.710.760.670.57
Cough/fever/acute onset heuristic0.640.620.610.57
Cough/fever/acute onset symptom score0.730.770.670.55
Cough/fever/myalgia heuristic0.720.720.750.58
Cough/fever/myalgia symptom score0.750.760.740.55
Re-fit FluScore model (Ebell 2012)0.770.790.740.57
LASSO score0.860.780.710.60
LASSO heuristic0.500.740.57
CART (manual)0.810.820.670.55
C5.0 tree (manual)0.730.790.650.55
Conditional inference tree (manual)0.790.800.630.57
Bayesian Additive Regression Trees (BART)0.860.810.700.64
C5.0 tree (tuned)0.850.750.600.57
CART (tuned)0.810.820.670.55
Conditional inference tree (tuned)0.790.790.630.55
Elastic net logistic regression0.870.830.700.65
Unpenalized logistic regression0.880.840.670.62
k-Nearest Neighbors classifier0.890.920.660.65
LASSO logistic regression0.870.830.700.65
Naive Bayes classifier0.830.790.740.68
Random forest0.890.870.730.59
SVM (linear kernel)0.850.820.750.65
SVM (polynomial kernel)0.830.790.740.67
SVM (RBF kernel)0.830.820.740.69
Gradient-boosted tree0.850.870.710.61
  • Abbreviations: AUROCC, Area Under the Receiver Operating Characteristic Curve; LASSO, Least Absolute Shrinkage and Selection Operator; SVM, Support vector machines; CART, Classification and Regression Tree Algorithm.

  • Notes: The AUROCC was not estimable for the LASSO heuristic model on the validation set of clinician-reported symptom data, as all patients were assigned the same score in this set.