Is S100-B better then rolling a die?
Muller et al report that S-100B had a specificity of 12.2% and a sensitivity of 86.4% and conclude that use of serum S-100B as a biomarker for CCT triage may improve patient screening and decrease the number of CCT scans performed. However, if we were to roll a 6-sided die and call the number 6 negative and the numbers 1 to 5 positive we would expect to have a test with similar parameters: 83.3% sensitivity (i.e. positive 5/6 times in those with the disease) and 16.6% specificity (i.e. negative 1/6 times in those without the disease). This illustrates how adopting a low threshold for positivity can produce and apparently high sensitivity at the expense of specificity for an essentially worthless test (like rolling a die). Reporting likelihood ratios overcomes this problem. The likelihood ratios for a positive and negative test based on the estimates of sensitivity and specificity reported by Muller are 0.98 and 1.11 respectively.
In fact, the performance of S-100B may not be quite as bad as their reported estimates suggest. Based on the data reported in Table 2 the sensitivity is indeed 86.4% (19/22) but the specificity is actually 31.8% (67/211), giving positive and negative likelihood ratios of 1.27 and 0.43 respectively. This is still not much help for decision-making but probably better than rolling a die.
Conflict of Interest: