We have read a recent article [1] in J Cachexia Sarcopenia Muscle with great interest. This study aimed to determine the prevalence of low muscle mass within people living with HIV (PLWH) and to identify associated factors. By using multivariate logistic regression analysis, they identified antiretroviral medication types, specifically Zidovudine; BMI and NRI can be independent risk factors for low muscle mass in men with HIV. Despite these definite results, we would like to point out some statistical concerns in this study while the predictors of their research may not be accurate.

First, before the commentary, we want to reiterate the fundamental statistical rule; there should be 10 events (outcome of interest) for 1 variable to be tested for the predictor or risk factor logistic regression analysis [2-5]. Thus, 27 low muscle mass cases at most analysed three variables in this study. Surprisingly, there were 12 variables in Table 3 (quintile group) of this study when analysed the factors associated with risk of low muscle mass estimated by multivariate logistic regression analysis. Thus, a fourfold overfitted multivariable analysis could not obtain accurate factor results in this study; otherwise, 120 (12 × 10) low muscle mass cases are needed for the factors' statistical analysis. Additionally, a similar overfitted analysis was also found in the AWGS criteria group in Table 3 (27 low muscle mass cases were used to analyse 8 variables).

Second, we are curious about why not the author used the univariate logistic regression analysis to reduce the factors before the final multivariate logistic regression analysis in Table 3. It could obtain more reliable results.

Third, we are curious about the rationale for selecting these 8 variables (quintile group) or 12 variables (AWGS criteria) for the predictor analysis in Table 3, were they chosen at random or according to their clinical experience? According to the commonly accepted statistical rule, the author could use the significant variables in Table 2 (statistical analyses between the normal muscle and low muscle based on quintile group or AWGS criteria) for the factors analysis in Table 3. But the authors seem not to use these significant variables in Table 2 for the factors analysis in Table 3. Thus, selecting the variables at random could not obtain an accurate risk factor in this study.

Lastly, it is a great honour to comment on Xu et al.'s despite these comments.

我们怀着极大的兴趣阅读了 J Cachexia Sarcopenia Muscle 最近的一篇文章 [1]。本研究旨在确定 HIV 感染者 （PLWH） 中肌肉质量低的患病率并确定相关因素。通过使用多变量 logistic 回归分析，他们确定了抗逆转录病毒药物类型，特别是齐多夫定;BMI 和 NRI 可能是 HIV 感染者肌肉质量低的独立危险因素。尽管有这些明确的结果，但我们想指出本研究中的一些统计问题，而他们研究的预测因子可能不准确。

首先，在评论之前，我们想重申基本的统计规则;1 个变量应有 10 个事件（感兴趣的结果）进行预测因子或危险因素 logistic 回归分析 [2-5]。因此，在这项研究中，最多 27 例低肌肉质量病例分析了三个变量。令人惊讶的是，当分析通过多变量 logistic 回归分析估计的与低肌肉质量风险相关的因素时，本研究的表 3（五分位数组）中有 12 个变量。因此，在本研究中，四重过拟合多变量分析无法获得准确的因子结果;否则，需要 120 （12 × 10） 个低肌肉质量案例用于因子的统计分析。此外，在表 3 中的 AWGS 标准组中也发现了类似的过拟合分析（使用 27 个低肌肉质量病例分析 8 个变量）。

其次，我们很好奇为什么作者不在表 3 中最终的多变量 logistic 回归分析之前使用单变量 logistic 回归分析来减少因素。它可以获得更可靠的结果。

第三，我们很好奇选择这 8 个变量（五分位数组）或 12 个变量（AWGS 标准）进行表 3 中预测因子分析的基本原理，它们是随机选择的还是根据他们的临床经验选择的？根据普遍接受的统计规则，作者可以使用表 2 中的显著变量（基于五分位数组或 AWGS 标准在正常肌肉和低位肌肉之间的统计分析）进行表 3 中的因子分析。但作者似乎没有将表 2 中的这些显著变量用于表 3 中的因子分析。因此，在本研究中，随机选择变量无法获得准确的风险因素。

最后，尽管有这些评论，我还是很荣幸能对 Xu 等人发表评论。