Data were checked for normality using the Kolmogorov-Smirnov test. Group comparisons were conducted with the Wilcoxon rank-sum, Kruskal-Wallis, or Fisher exact test. The Spearman rank correlation coefficient was used to assess the association between 2 continuous variables. Variables that were significantly different between the LPD and PFPS groups were further analyzed in a multiple logistic regression model. A penalized Firth correction was employed in the model because of quasi-complete data separation. The significance level was set to a 2-sided alpha of 5% for all statistical tests. In cases involving multiple tests, raw P values were adjusted by the Holm-Bonferroni method.
A simulation-based power analysis was conducted for the correlation analysis to evaluate the power to detect significant Spearman correlations between 2 variables using 91 samples. The significance level was set to alpha = 5%/36 to maintain a family-wise error rate of 5% across 36 correlation tests (all pairwise comparisons between 9 variables). In 10,000 iterations, 2 Gaussian variables with a given correlation coefficient were simulated. In these settings, 91 samples were sufficient to uncover a correlation of 0.38, with a power of 80% considered significant. All analyses were performed with R statistics software (version 3.4.0;www.r-project.org ).
A simulation-based power analysis was conducted for the correlation analysis to evaluate the power to detect significant Spearman correlations between 2 variables using 91 samples. The significance level was set to alpha = 5%/36 to maintain a family-wise error rate of 5% across 36 correlation tests (all pairwise comparisons between 9 variables). In 10,000 iterations, 2 Gaussian variables with a given correlation coefficient were simulated. In these settings, 91 samples were sufficient to uncover a correlation of 0.38, with a power of 80% considered significant. All analyses were performed with R statistics software (version 3.4.0;