Dear Editor,

Dolatshahi et al. carried out an interesting and clinically relevant meta-analysis study [1] focusing on the thyroid hormone levels in patients with Alzheimer Disease (AD). In the study, the authors performed a number of meta-analyses by including thirty-two observational studies. Accordingly, they identified that serum total T3 and free T3 and cerebrospinal fluid (CSF) total T3 levels were significantly lower in AD patients compared to controls. However, the certainty of these statistically significant findings could be adversely affected by the following issues.

First, the multiple testing issue was not taken into consideration in Dolatshahi et al.’s study [1]. In the study, the authors in total performed 13 important statistical tests (i.e., 13 meta-analyses): Figure 2 (3 tests), Figure 3 (4 tests), Figure 4 (3 tests), and Figure 5 (3 tests). According to the Bonferroni method [2], the threshold for multiple testing significance should be calculated by 0.05÷13 = 0.0038. According to this corrected threshold for significance, in the article all of the statistically significant findings regarding serum total T3 (P = 0.008, Figure 4), serum free T3 (P = 0.03, Figure 4), and CSF total T3 (P = 0.05, Figure 5) would not reach statistical significance.

Second, there was severe heterogeneity found in the meta-analyses of serum total T3 (I2 = 83%, Figure 4), serum free T3 (I2 = 93%, Figure 4), and CSF total T3 (I2 = 97%, Figure 5). In this case, authors are encouraged to conduct further subgroup analyses or meta-regression analyses to explore the sources of heterogeneity, or to conduct sensitivity analyses (such as using the leave-one-out method) to examine the robustness of pooled results. However, authors did none of them. Therefore, the credibility of these meta-analyses results is questionable.

Last, in this meta-analysis [1] authors included a total of 32 original studies, but they did not perform detection of publication bias. As for those meta-analyses involving ten or more than ten original studies, such as the meta-analyses of serum total T3 (involving 11 original studies, Figure 4) and serum free T3 (involving 15 original studies, Figure 4), authors should examine the presence of publication bias or not in accordance with common practice [3].