Abstract
In the September 2013 issue of Thyroid, Mazeh and Sippel expertly reviewed multiple aspects of familial nonmedullary thyroid carcinoma (FNMTC) (1). In their section on tumor aggressiveness, they pointed out that the results of different studies are conflicting. Four studies demonstrate that FNMTC is more aggressive than sporadic NMTC. In contrast, four other studies fail to find a difference. We suggest that this discrepancy may be accounted for by the use of statistical approaches that do not appropriately address two potentially confounding issues: (i) heterogeneity of two-member kindreds, and (ii) the small numbers of patients within some cells. Furthermore, the selected secondary subgroup analyses that some authors used to compare two-member kindreds to kindreds with three or more subjects may be misleading, since subgroup analyses are subject to type 1 and type 2 statistical errors (1). In contrast, McDonald et al. (2) investigated the same clinical question by applying a statistical approach that accounted for both of these confounding issues. Furthermore, this approach avoided potentially misleading selected secondary subgroup analyses. We discuss the clinical problem in relationship to the statistical approach and suggest that conclusions derived from the application of a more appropriate statistical analysis can be viewed with greater confidence.
Heterogeneity of two-member kindreds
As Mazeh and Sippel correctly point out, Charkes analysis indicates that two-member kindreds are likely to be a heterogeneous group (3). Two-member kindreds are enriched with the concordance of two sporadic papillary thyroid carcinomas (PTC) in the same kindred. Two-member kindreds may be caused by polymorphisms that confer less penetrant and aggressive disease than seen in kindreds with three or more family members. Alternatively, two-member kindreds may be caused by the same polymorphisms that occur in larger kindreds, but are restricted in size because fewer relatives are available for analysis. The clinical studies summarized by Mazeh and Sippel combined the two-member kindreds with larger kindreds in their initial analysis. This approach is attractive, since it increases the statistical power of the study. Not surprisingly, it was used in all eight studies reviewed by Mazeh and Sippel in their table 3 (1). However, as suggested by Charkes analysis, this approach may overlook a small effect of FNMTC on tumor aggressiveness, since the FNMTC group is enriched with sporadic NMTC. A statistical approach that accounts for the potential intermediate risk of two-member kindreds is preferred.
Small numbers of individuals
The outcome of FNMTC is generally good, and similar to sporadic NMTC. Therefore, there are small numbers of patients in the cells that describe aggressive characteristics. Many commonly used statistical tests overestimate the statistical significance, when there are small numbers of patients in some cells. An example of this is the commonly used chi-square test. The Fisher exact test is preferred over the chi-square test when the numbers of patients in some cells is small. A statistical approach that accounts for the small numbers of subjects in some cells is preferred.
A statistical approach for the clinical problem of FNMTC aggressiveness
The statistical approach that was used by McDonald et al. is more appropriate for the clinical problem of FNMTC aggressiveness (2). First, McDonald et al. reasoned that if FNMTC is more aggressive than sporadic NMTC, then the NMTC of two-member kindreds (a subset enriched with sporadic NMTC) should have an intermediate risk of aggressiveness. If this is the case, then the two-member kindreds should be analyzed as a distinct group and should not be combined with the larger FNMTC kindreds. McDonald et al. separated the two-member kindreds into a distinct group and employed a chi square by trend analysis to account for the potential intermediate risk of aggressiveness in two-member kindreds. Second, McDonald et al. used an exact test correction to ensure that the small numbers of patients in some cells did not inflate the statistical significance. Third, they refrained from performing selected secondary subgroup analyses that could be misleading. Using this approach, they observed that FNMTC was significantly more aggressive than sporadic NMTC at the following end points: distant metastases, need for additional surgery, need for additional treatment with 131-iodine, need for reoperation, and death due to disease (2).
In summary, we propose that the statistical approach that is selected to evaluate FNMTC tumor aggressiveness should account for the heterogeneity of two-member kindreds and for the small sample size of some cells. If these statistical modifications are employed in future studies, then a more consistent and accurate outcome might be observed. FNMTC was found to be more aggressive in a study that employs a statistical approach that accounts for the intermediate risk of two-member kindreds and for the small number of subjects in some cells (2).