统计问题第103问：荟萃分析问题4

Question

Researchers undertook a meta-analysis of the long term efficacy of approved anti-obesity drugs in reducing weight and improving health status. Only double blind randomised placebo controlled trials of drugs used in adults (age over 18 years) for one year or longer were included.

The primary outcome was change in weight from baseline. A separate meta-analysis was performed for each drug identified. Fourteen trials of orlistat were found and results of the meta-analysis were presented in a forest plot (figure⇓). Orlistat was more effective than placebo in reducing weight; with a mean difference in weight loss of 2.87 kg (95% confidence interval 2.53 to 3.21) greater than placebo.

• Forest plot of the long term effect of orlistat compared with placebo on weight change. For each trial and the total overall effect, the mean difference in weight change was derived as orlistat minus placebo.

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

·a) The line of no effect would be a vertical line going through unity on the forest plot.

·b) A negative weighted mean difference indicates placebo was more effective than orlistat in reducing weight.

·c) There was statistical heterogeneity between the sample estimates.

·d) The overall estimated change in weight with orlistat was significantly different from placebo at the 5% level of significance.

·e) It can be concluded that placebo was ineffective at reducing weight.

提示：正确答案只有一个。

Answer

Answer d is true, whereas a, b, c, and e are false.

Orlistat was compared with placebo by calculating the difference between group means in the primary outcome of weight change from baseline. A zero difference would indicate that the mean weight change was equal for orlistat and placebo. Therefore, the line of no effect would go vertically through zero on the forest plot (a is false). Previous questions described a meta-analysis of randomised controlled trials where the primary outcome was binary.Relative risk was used to compare the risk of the outcome occurring between groups. In that situation a relative risk of 1.0 would indicate equal risk in both groups and so the line of no effect would go vertically through 1.0 on the forest plot.

For the orlistat and placebo groups in each trial, the sample size, plus mean and standard deviation of weight change from baseline are shown on the left of the forest plot. The difference in sample means was calculated as orlistat minus placebo and presented as “weighted mean difference.” The percentage weight given to each difference—that is, the influence each trial had on the overall effect, is indicated in the column “weight (%).” The percentage weight for each trial was determined by the precision of its sample estimate; those trials with more accurate estimates of the population mean difference had greater weight. The sample mean was negative for both orlistat and placebo in all trials; therefore, participants in each group on average lost weight from baseline. For all of the trials weight loss was greater for orlistat, so the difference in means between orlistat and placebo was negative. A negative difference in sample means favours orlistat and is presented on the forest plot to the left of the line of no effect (b is false).

Last week’s question described how a meta-analysis incorporates a statistical test of heterogeneity to assess the extent of variation between sample estimates. The P value for the statistical test of heterogeneity for the meta-analysis of orlistat was 0.61, larger than 0.05—the traditional critical level of significance—and the Higgins I²statistic was zero. Therefore, no evidence of heterogeneity existed between the sample estimates (c is false). Because heterogeneity did not exist between sample estimates then a fixed effects meta-analysis could have been used to derive the total overall estimate and not a random effects model, as indicated at the top of the forest plot. However, if heterogeneity is not present then a random effects meta-analysis will produce the same total overall estimate as a fixed effects model.

The overall estimate of the mean difference in weight change between orlistat and placebo (orlistat minus placebo) was −2.87 kg (95% confidence interval −3.21 to −2.53). Therefore, taking orlistat reduced a person’s weight on average by 2.87 kg more than taking placebo. The 95% confidence interval did not include the mean difference of zero, which represents no difference between orlistat and placebo in mean weight change. Therefore, a statistically significant difference at the 5% level of significance existed between orlistat and placebo in mean weight change (d is true). This finding is corroborated by the result of the statistical hypothesis test for the overall effect—P<0.001.

Although the mean difference in weight change between orlistat and placebo was significant at the 5% level of significance and favoured orlistat, it does not mean that placebo did not produce weight loss (e is false). For each placebo group in each of the trials, the mean weight change was negative, which indicates that on average weight loss occurred.

所以答案是选择  d

每天学习一点，你会更强大！