Problems with high-sensitivity C-reactive protein.

To the Editor: Last year, Ockene et al. (1) published a report in this journal that made the claim that high-sensitivity C-reactive protein (hs-CRP) has a degree of measurement stability that is similar to that of total cholesterol and that this provides further evidence of the potential clinical utility of hs-CRP screening as a novel tool for vascular risk prediction. The key evidence that Ockene et al. (1) present to justify their claim is a histogram showing an almost identical agreement in terms of group classification between first and second measurements for hs-CRP and total cholesterol. This apparent agreement is spurious and is attributable to the way in which Ockene et al. partitioned the hs-CRP data. Although the total cholesterol data in the histogram are divided into quartiles, the hs-CRP data are partitioned into arbitrary intervals that contain ϳ15%, 20%, 30%, and 35%, respectively, of the sample. Ockene et al. (1) provide two graphs showing the data for all 113 patients for serial cholesterol and CRP values ranked by mean concentration. These values are different for the two analytes. For cholesterol, the average intraindividual variation is 18.2%, and the intraindividual variation is roughly constant across all the range of data. For CRP, the average intraindividual variation is higher, at 44.2%. It is lowest at low CRP concentrations and then increases as the mean CRP concentration increases. Ockene et al. (1) also provide graphs showing the distributions of the total cholesterol and hs-CRP results. As expected, the total cholesterol distribution is approximately gaussian and the hs-CRP distribution is skewed. A result of the skewed distribution of CRP concentrations is that even if true quartiles had been used, the interquar-tile spacing would increase markedly as the mean CRP concentration increased. The arbitrary intergroup spac-ings used by Ockene et al. amplify this effect. The third group concentration range is twice as wide, and the fourth group concentration range is 16 times as wide as each of the first two groups. The group cutoffs used by Ockene et al. also allow the upper 65% of the study participants, who among them encompass the majority of the intrain-dividual variation in hs-CRP, to fall within these two wider intervals. It is readily apparent that the probability that a second or subsequent CRP value will fall outside the original group is lower if the original value was in the wide third or fourth group. The net result …