These were composed of 19 residents (23%), 19 fellows (23%), 45 CNS (54%). Clinicians' rating on the probability of LNS yielded an AUC of 0.82 (95% CI 0.74 to 0.90). There was no significant difference of AUC among the groups of clinicians (P=0.8). The receiver-operating characteristic curve analysis showed there was no significant difference between our prediction score and the clinicians' own rating of probability of true LNS (P=0.3). However, the graphical plots of calibration are presented in Figures 1 and 2. This shows that the calibration of LNS score was good, (Figure 1) and yields a H–L test P-value of 0.62. In comparison to the LNS score, the calibration curve of clinician's prediction was considerably below the dotted ideal line. This suggests a trend to overestimate the probability of LNS (Figure 2), but H–L test P-value was 0.28. It appears from this analysis that the LNS score may be superior to the performance of the clinician rating.