Factors Associated With Lower Extremity Dysmorphia Caused by Lower Extremity Lymphoedema: Comment on Data Sparsity

We read the article by Yamamoto et al. with great interest. The authors aimed to examine independent factors associated with lower extremity dysmorphia (LED) in patients with secondary lower extremity lymphoedema. They found that the odds of LED were considerably higher among patients with stage IV leg dermal backflow (LDB) compared with those with stage 0 LDB (crude odds ratio [OR] 43.00, 95% confidence interval [CI] 4.92–375.76; adjusted OR 76.79, 95% CI 8.13–725.20), which is questionable. The authors believe that the very large effect size estimate, such as OR with considerably wide CI, can be yielded from sparse data.^,  Sparse data means that there are not enough observations in the different combinations of the exposure and outcome variables in the studied data. We tabulated the associations between LDB stage IV and LED (n_{Stage IV, LED+} = 43; n_{Stage IV, LED–} = 10; n_{Stage 0, LED+} = 1; n_{Stage 0, LED–} = 10) and found that data sparsity is expected. Hence, the authors' crude and adjusted ORs could be biased owing to the sparse data bias and should be decreased using the appropriate statistical methods.^{,  ,  ,}  We used one of the effective methods to decrease this aforementioned bias, namely penalisation via data augmentation using the log-f (3.9, 3.9) prior distributions. It was found that the penalised crude OR and its CI shrank and narrowed, respectively (OR 11.30; 95% CI 2.76–46.29). Hence, we respectfully suggest that the authors re-estimate their adjusted ORs on the association between LDB stage IV and LED in order to obtained unbiased ORs.