统计问题第106问：荟萃分析（8）

Question

Researchers undertook a meta-analysis of the effects of monitoring blood pressure at home. Randomised controlled trials were included if participants had essential hypertension and were attempting to achieve blood pressure targets. Home blood pressure monitoring was achieved with ambulatory monitors, whereas the control intervention was standard blood pressure monitoring in the healthcare system. An extensive search of electronic databases of reports and journal publications identified 13 randomised controlled trials. For these trials, the length of intervention varied between 2 and 36 months.

The meta-analysis reported that home blood pressure monitoring resulted in lower systolic blood pressure than standard monitoring, with an overall mean difference of 4.2 mm Hg (95% confidence interval 1.5 to 6.9). Potential publication bias was investigated by Egger’s test (P=0.038).

Which of the following statements, if any, are true?

·a) Publication bias would have occurred if trials with results that were not statistically significant were not published

·b) Publication bias would have occurred if only studies published in English were included in the meta-analysis

·c) The results of Egger’s test showed that publication bias existed for this meta-analysis

提示：这是一道多选题。

Answer

Statements a and c are true, whereas b is false.

The meta-analysis investigated the effects on blood pressure of home monitoring compared with standard monitoring. The validity of the meta-analysis depended on identifying all relevant randomised controlled trials ever undertaken. However, it is difficult to know if all such trials had been published and were easily accessible. The meta-analysis could have been prone to several types of bias influencing which studies were included.

Publication bias is probably the most common type of bias affecting the validity of a meta-analysis. It is the failure to include all relevant trials ever undertaken because they were not available in the databases searched. This can originate for a variety of reasons. Trials with results that are not statistically significant—typically those studies with small samples—are less likely to be published, either because they are not submitted for publication or not accepted for publication after submission (a is true). It is also possible that studies with results that favour the control group rather than the active treatment group may not be submitted for publication. Publication could have also been influenced by who funded the study or even the identity of the research group.

Another type of bias that may have influenced which studies were included in the meta-analysis is language bias, where studies published in any language other than English may not have been easily accessible (b is false).

Egger’s test is one of several statistical tests for publication bias. It is similar to traditional statistical hypothesis testing, there being a null hypothesis and an alternative hypothesis.2 The null hypothesis starts at the position of no publication bias, with the alternative hypothesis stating that bias exists. The P value for Egger’s test is equal to 0.038. Therefore, the null hypothesis is rejected in favour of the alternative at the 5% level of significance with the conclusion that publication bias exists (c is true). Methodology is available to adjust the results of a meta-analysis, so that the potential effects of publication bias can be assessed.

Meta-analyses have become integral to evidence based medicine, whereby the results of the most comprehensive evidence available, together with clinical expertise, are used in patient care. However, it is essential to be aware of potential publication bias to avoid making inappropriate treatment decisions on the basis of incomplete evidence.

所以答案是选择  a c

每天学习一点，你会更强大！