We read the article by Esplin et al on the proteomic identification of serum peptides that predict subsequent spontaneous preterm birth with great interest because it is a very relevant topic. However, there appears to be biased sample selection and incomplete statistical analysis that lead to results that need additional interpretation. In study 1, 40 cases (spontaneous preterm birth) and 40 control subjects were selected from 2929 women who were eligible with serum samples at 24 weeks’ gestation. It is not clear whether the cases were matched with control subjects. In study 2, another 40 cases (spontaneous preterm birth) and another 40 control subjects were selected from 2929 women who were eligible with serum samples at 28 weeks gestation. Again, it does not appear that cases and control subjects were matched. Because control subjects were selected randomly, it is not clear how they were handled if the same control was selected for study 1 and study 2. Another question is the sample size of 40 cases and 40 control subjects, given such a large eligible cohort. In addition, it is not clear whether the samples that were used for model building were also used for validation, which would create additional bias. The rationale for the choice of markers is not provided in the article.

With respect to statistical analysis and results, because the nonparametric method was used to compare the 2 groups, median and range must be provided in addition to the mean and SE. In Table 3, with large SEs, it is not realistic to obtain significant probability values. For example, in study 1 for thrombin antithrombin, the means (and SE) are 274.8, 255.9 for cases and 293.8, 269.6 for control subjects. Based on these data, being able to obtain significance with a probability value of .02 seems unlikely. Figure 1 does not display other selected cutoff values of M/Z (mass to charge ratio) as stated in the statistical “Methods” section. In Figure 3, it appears that the correlation is very weak (or zero), but the authors have found it to be significant with a probability value of .001.