Table 2.

95% AUC estimate/confidence intervals (CI) for maximum and minimum cases (REFINE SPECT).

Method	Bootstrap validation	50/50 strat split validation	70/30 strat split validation	tenfold strat. cross val	10 × repeated tenfold strat cross val
Logistic regression: Max	0.792 [0.787–0.790]	0.816* [0.789–0.843]	0.834* [0.802–0.866]	0.797 [0.772–0.815]	0.795 [0.788–0.802]
Logistic regression: Min	0.788 [0.787–0.790]	0.767* [0.716–0.797]	0.756* [0.716–0.797]	0.787 [0.768–0.810]	0.792 [0.786–0.799]
Gaussian Naïve Bayes: Max	0.747 [0.694–0.698]	0.777** [0.746–0.808]	0.790** [0.752–0.828]	0.739 [0.716–0.762]	0.729 [0.722–0.736]
Gaussian Naïve Bayes: Min	0.741 [0.694–0.698]	0.708** [0.655–0.750]	0.703** [0.655–0.750]	0.705 [0.682–0.729]	0.712 [0.710–0.725]
Linear discriminant analysis: Max	0.788 [0.783–0.786]	0.817* [0.790–0.845]	0.827* [0.794–860]	0.798 [0.769–0.811]	0.795 [0.789–0.802]
Linear discriminant analysis: Min	0.785 [0.783–0.786]	0.756* [0.713–0.795]	0.754* [0.713–0.795]	0.790 [0.770–0.812]	0.792 [0.786–0.799]
Random forest: Max	0.771 [0.766–0.770]	0.801* [0.771–0.830]	0.817** [0.783–0.851]	0.784 [0.765–0.799]	0.777 [0.770–0.784]
Random forest: Min	0.768 [0.766–0.770]	0.741* [0.684–0.775]	0.730** [0.684–0.775]	0.766 [0.747–0.783]	0.772 [0.765–0.779]

95% CI calculated using Delong method with concatenated predictions. For bootstrap validation, AUC computed as mean from the AUC of the 500 iterations, sometimes falling out of range of 95% CI. CI calculated using Delong method in brackets.

*Signifies measured statistical significance p-value < 0.05 in independent ROC comparison to max–min counterpart for the same algorithm and validation technique.

**Indicates p-value < 0.01 in independent ROC comparison. ROC receiver operating characteristic curve, AUC area under the ROC curve.