Table 2.

Receiver operating curve AUCs for prediction of breast cancer from untargeted NMR features.

Study group (cases and matched controls)	Number of cases and controlsa	Median case time to diagnosis (years)	k-fold cross-validation procedureb	Receiver operating curve prediction accuracy for test set (AUC, 95% CI)c
All participants	1572	4.82	10-fold	0.51 (0.45–0.57)
Menopausal status at blood collection
Pre-menopausal	354	5	5-fold, 5 times	0.61 (0.49–0.73)
Post-menopausal	1218	4.77	10-fold	0.53 (0.46–0.59)
Age at diagnosis
<55 years	265	2.7	5-fold, 5 times	0.63 (0.49–0.77)
≥55 years	1307	5.48	10-fold	0.49 (0.43–0.55)
Fasting status at blood collection
Yes	572	4.73	10-fold	0.52 (0.43–0.62)
No	1000	4.96	10-fold	0.53 (0.46–0.61)
Time from blood collection to cancer diagnosis
<2 years	317	0.97	5-fold, 5 times	0.59 (0.46–0.72)
<5 years	752	7.67	10-fold	0.51 (0.43–0.59)

AUC area under curve.

CIs were estimated by DeLong’s method.

^aPlasma samples of unknown waiting time to fractionation (n = 10) were excluded.

^bUsing a randomly selected 75% of the data.

^cUsing a randomly selected 25% of the data excluded from model training.