1 - Epidemiology and Statistics Office, Research Unit, Centro Hospitalar de Lisboa Central, Lisbon, Portugal;
2 - NOVA Medical School, Universidade Nova de Lisboa, Lisbon, Portugal;
3 - Centre of Statistics and its Applications, Faculty of Sciences (CEAUL), University of Lisbon, Lisbon, Portugal.
Eur J Pediatr. 2018 Dec 5. doi: 10.1007/s00431-018-3296-6. [Epub ahead of print]
Letter to editor
The study by Doudin et al., exploring the association between serum 25(OH)D and hematological indices in adolescents, stands out for the wise use of cross-sectional data from a German cohort to explore a novel issue related to vitamin D. The authors’ interpretation of the findings is misled by the use of p value. Applying it to assess associations whose underlying plausibility is ignored and is not suggested by the correlation coefficients, easily leads to false-positive findings. In fact, p-values are strongly related to sample size, “sometimes attracting attention to very small effects that have little realworld importance”. The authors acknowledge that “the correlation coefficients and the effect sizes are very small”. Indeed, the correlation coefficient estimates range from r = −0.04 to r = 0.08 (very close to r = 0, i.e., absence of correlation), and the differences on hematological indices by levels of serum 25(OH)D are minimal (e.g., Hb [g/dl] 1st tertile: 13.7 ± 1.2; 2nd tertile: 13.5 ± 1.1; 3rd tertile 13.6 ± 1.1), in spite of the low estimated p values. None of the “statistically significant” differences found have a clear “clinical relevance”. This frequent methodological error prompted methodologists and editors to discourage the widespread use of p values and to promote the use of alternative measures, with scarce success. “False positive findings are detrimental to science and society, as once published, they accumulate persistent untrue evidence”.
Palavras Chave: Erythropoiesis; Hematological markers; p value; Serum 25(OH)D