摘要：

【目的】调查医学论文中成组t检验的P值错误,分析错误原因,并提出相应措施。【方法】选取236种医学期刊,每种期刊选取1项成组t检验,核验其P值,应用χ ²检验、Mantel-Haenszel法、二项式logistic回归分析P值错误。 【结果】236项成组t检验中,50项存在P值错误。单因素分析结果显示,核心期刊与非核心期刊相比,P值错误发生率差异具有统计学意义(χ ²=4.871,P=0.027);给出具体P值组与未给出具体P值组相比,P值错误发生率差异具有统计学意义(χ ²=15.440,P<0.0001)。将是否给出具体P值作为混杂因素,比较核心期刊与非核心期刊P值错误发生率,差异无统计学意义(χ ²=2.703,P=0.100)。多因素分析结果显示,是否方差齐(OR值为0.470,95%CI为0.230~0.961)、是否给出具体P值(OR值为5.459,95%CI为2.311~12.895)具有统计学意义。 【结论】医学论文成组t检验P值错误较多。为及时发现P值错误,期刊编辑应当重视对统计学方法应用条件的审查,要求作者给出统计描述以及统计推断的具体结果,能够利用简单易学的统计学软件核实P值。

关键词: 医学论文, 成组t检验, P值错误, 原因分析, 方差齐性

Abstract:

[Purposes] This paper investigates P-value errors in two-sample t-test of medical papers, analyzes the causes of errors, and puts forward the corresponding measures. [Methods] We reviewed 236 two-sample t-test cases from 236 medical journals. The P-value was verified and the P-value errors were analyzed by χ ² test, Mantel-Haenszel method and binomial logistic regression. [Findings] Among the 236 cases, 50 cases have P-value errors. The univariate analysis shows that there is a significant difference between core journals and non-core journals in incidence of P-value errors (χ ²=4.871, P=0.027), and there is a significant difference between cases with concrete P-value and cases without concrete P-value in incidence of P-value errors (χ ²=15.440, P<0.0001). When we regard whether the concrete P-value is described as the confounding factor, no significant difference in the incidence of P-value errors is found between core journals and non-core journals (χ ²=2.703, P=0.100). The multivariate analysis shows that whether the variance is equal(OR=0.470,95%CI:0.230-0.961)and whether the concrete P-value is described (OR=5.459,95%CI:2.311-12.895) are significant variables. [Conclusions] There is a high incidence of P-value errors in two-sample t-test of medical papers. In order to discover the P-value errors in time, editors should pay attention to the examination of the application conditions of statistical methods, ask the authors to give statistical description and statistical inference result, and verify P-value by simple statistical software.

Key words: Medical paper, Two-sample t-test, P-value error, Reason analysis, Homogeneity of variance