The absence of statistician involvement and low journal impact factor predict statistical mistakes in dermatology journal articles: A cross-sectional analysis

Background: Statistical mistakes can undermine research credibility. Identifying common errors may help researchers avoid them in future studies.
Objective: This study evaluated the frequency and types of statistical mistakes in dermatology journal articles and identified article characteristics that predict these errors.
Methods: A cross-sectional analysis was conducted on articles published in the 2023 volumes of eight dermatology journals. Articles were screened for statistical tests, with a target sample of 200 selected pseudorandomly. Multivariable logistic regressions assessed predictors of statistical mistakes, including journal impact factor, statistician involvement, funding source, first author highest degree, and statistical package.
Results: Of the 189 articles analyzed, 78% contained at least one statistical mistake. Reporting mistakes were found in 67%, and test selection errors in 46%. The absence of statistician involvement (aOR 2.49, p=.03) and low journal impact factor (aOR 3.82, p=.02) predicted the presence of at least one mistake.
Limitations: This sample from eight journals is not representative of all dermatology literature. Original data was not available for testing of test assumptions, so appropriate test selection was determined using statistical conventions.
Conclusion: Statistical mistakes are prevalent in dermatology literature. Researchers should review statistical best practices and consider involving a statistician in their work.

研究背景：统计学错误会损害研究成果的可信度，识别常见的统计错误有助于研究者在后续研究中规避此类问题。
研究目的：本研究评估了皮肤病学期刊论文中统计错误的发生频次与类型，并识别出可预测此类错误出现的论文特征。
研究方法：针对8种皮肤病学期刊2023年卷刊载的论文开展横断面分析（cross-sectional analysis）。首先对论文的统计检验方法进行筛查，采用伪随机抽样法选取200篇论文作为目标样本。随后通过多变量逻辑回归（multivariable logistic regressions）分析统计错误的预测因素，涵盖期刊影响因子（journal impact factor）、是否有统计学家参与、资助来源、第一作者最高学历以及所使用的统计软件包。
研究结果：在纳入分析的189篇论文中，78%的论文存在至少1处统计错误，其中67%存在报告性错误，46%存在检验选择错误。无统计学家参与（校正优势比[adjusted odds ratio, aOR] 2.49，p=0.03）以及低期刊影响因子（校正优势比3.82，p=0.02）是论文存在至少1处统计错误的显著预测因素。
研究局限：本研究样本仅来自8种期刊，无法代表所有皮肤病学研究文献；且由于无法获取原始数据以检验统计检验的前提假设，因此依据统计学惯例判断检验选择的恰当性。
研究结论：皮肤病学文献中普遍存在统计错误，研究者应参考统计最佳实践规范，并考虑在研究中邀请统计学家参与协作。