LQCL2515 - CMLE - Effect of Statistical Assumption Violations on Post Hoc Testing: A Methodological Case Study in Chronic Myeloid Leukemia-LQCL2515

LQCL2515 - Educational Acitivity

Pdf Summary

The document outlines a methodological case study centered on the importance of validating statistical assumptions within molecular pathology research, specifically studying lactate dehydrogenase (LDH) levels in chronic myeloid leukemia (CML) patients. Initially, a one-way ANOVA was employed to compare LDH levels across three groups: an accelerated CML phase group, a chronic CML phase group, and a control group. This analysis identified significant differences. However, the results were challenged as they neglected essential statistical assumptions, such as normality and homogeneity of variances.<br /><br />When reviewed, it emerged that data failed normality tests using the Shapiro-Wilk test and contained outliers, prompting the switch to a non-parametric Kruskal-Wallis test and Dunn’s post hoc analysis as more fitting alternatives. These tests also showed group differences but with different significant pairwise comparisons than initially observed. Further exploration using various post hoc tests revealed inconsistent p-values, underscoring the implications of selecting appropriate statistical methods.<br /><br />The document reinforces that failure to verify statistical assumptions can distort analyses, affecting both detection of true differences and masking genuine patterns. It emphasizes the importance of choosing statistical techniques based on data characteristics. Particularly when assumptions like normality and homogeneity of variance are not met, alternative methods, such as the Kruskal-Wallis test, or robust post hoc tests like Games-Howell or Tamhane’s T2, are recommended.<br /><br />The case study illuminates the broader methodological issue within laboratory statistics regarding the influence of assumption violations on subsequent analysis. It advocates for thorough verification of assumptions and careful selection of statistical methods aligned with data properties to yield reliable and interpretable results.

Keywords

molecular pathology

statistical assumptions

lactate dehydrogenase

chronic myeloid leukemia

ANOVA

Kruskal-Wallis test

normality test

homogeneity of variances

post hoc analysis

methodological case study