Geneticist warns BMP2 wisdom tooth findings may not survive multiple-testing correction

A published letter to the editor argues that a recent case-control study linking a BMP2 variant to impacted wisdom teeth may have overstated its statistical case, because the authors tested far more genetic models than their significance threshold could support. The comment, written by Mostafa Saadat and published as a Correspondence item in the Egyptian Journal of Medical Human Genetics, takes aim not at the study’s subject matter, which he calls relevant and important, but at its analytical strategy.

The original study by Sabeti Akbar-Abad and colleagues examined whether a variant in the BMP2 gene is associated with impacted wisdom teeth, using a case-control design. Saadat acknowledges several strengths of that work, including appropriate adjustment for ethnicity and gender, and a control group that was in Hardy-Weinberg equilibrium, a standard check that genotype frequencies behave as population genetics predicts.

The concern centers on a practice that has become increasingly common in genetic association studies: applying multiple inherited models of how a variant acts, such as codominant, dominant, recessive, overdominant and allelic, to the same dataset and presenting all of the results side by side. Saadat argues this apparent comprehensiveness creates two serious problems, one statistical and one interpretive.

The statistical problem is one of multiplicity. Every additional test performed on the same data raises the total number of chances for a false positive. To keep the overall type I error rate, the probability of wrongly declaring an association, under control, analysts must adjust the significance threshold, for example with a Bonferroni correction that divides the nominal threshold by the number of tests. Without such adjustment, the chance of reporting at least one spurious finding becomes substantially higher than the nominal 5 percent.

Applied to the original study, the arithmetic is unforgiving. The authors tested five genetic models in both crude and adjusted forms, producing ten separate statistical tests reported in Table 3 of the article and interpreted together in its discussion. A Bonferroni correction for ten comparisons sets the adjusted significance threshold at 0.05 divided by 10, or 0.005. The smallest p-values reported in the original study were 0.006 for the dominant model and 0.011 for the codominant model. Both exceed the corrected threshold, meaning that none of the reported associations would remain statistically significant after proper adjustment for multiple comparisons.

The interpretive problem follows from the first. When several models yield different levels of significance, it becomes unclear which model truly reflects the underlying biological relationship between the variant and the condition. Saadat notes that testing multiple models may be acceptable in exploratory analyses intended to generate new hypotheses, provided the findings are clearly labeled as hypothesis-generating. In confirmatory analyses, however, a single pre-specified approach with appropriate correction is essential. He points out that a careful reader might reasonably ask why the dominant model was chosen over the recessive model, or why the allelic model was reported at all, an ambiguity that can lead to overinterpretation of results that may not be robust.

The letter is not purely critical. Turning to the genotype counts in the original Table 3, Saadat calculates crude odds ratios of 0.36 for the TC genotype and 0.35 for the CC genotype. Using a data-driven framework for genetic model selection described elsewhere, he concludes that the dominant model appears to be the most appropriate choice for this particular study. Under that approach, presenting the recessive, overdominant and allelic models would add little meaningful information while multiplying the number of statistical comparisons without clear justification. He adds that model selection in other contexts may also depend on biological hypotheses and analytical frameworks.

Beyond the specific study, the correspondent offers constructive recommendations for future genetic association research. First, the appropriate genetic model should be selected based on the observed pattern of effect sizes across genotypes, rather than testing all possible models simultaneously without justification. Second, when multiple testing is unavoidable, correction methods such as Bonferroni or false discovery rate control should be applied and the adjusted significance thresholds clearly reported. Third, adherence to established reporting guidelines such as STREGA, the STrengthening the REporting of Genetic Association studies guideline, can improve transparency and completeness.

For readers following the BMP2 and impacted wisdom tooth question, the practical takeaway is caution rather than dismissal. The original study’s design strengths stand, and its data-driven case for a dominant model offers a path toward a cleaner reanalysis. But as the letter makes plain, under a corrected significance threshold the currently reported associations do not hold, and any conclusion about the variant’s role will need to rest on analyses that survive the multiplicity problem the correspondent describes.

Subject of Research: Statistical critique of a case-control study of a BMP2 variant and impacted wisdom teeth

Article Title: Comment to: Unraveling the role of BMP2 variant in impacted wisdom teeth: insights from a case-control study

Article References: Saadat, M. (2026). Comment to: Unraveling the role of BMP2 variant in impacted wisdom teeth: insights from a case-control study. Egyptian Journal of Medical Human Genetics, 27(1), Article 82. https://doi.org/10.1186/s43042-026-00911-y

Image Credits: AI Generated

DOI: 10.1186/s43042-026-00911-y

Keywords: BMP2, impacted wisdom teeth, genetic association study, multiple comparisons, Bonferroni correction, genetic models, odds ratio, STREGA, case-control study, type I error, correspondence, statistical methods