The figure you provide shows the results of a random effects meta-analysis, which is not an advisable method unless the population parameters are indeed a random sample from a super-population of populations, and the populations investigated in the meta-analysis would randomly change from one repeated meta-study to another according to this sampling from the super-population.

The circled p-value on the right is testing whether the mean of this presumably fictitious and indefensible super-population of population ORs is different from 1.

I'm not entirely sure what the p-value is for "the test for effect in subgroup." Perhaps it is a p-value from a fixed-effect model adjusting for study, testing the global null hypothesis that the population OR investigated in each of the 5 studies are all equal to 1. Perhaps instead it is a fixed-effect meta-analytic p-value testing the hypothesis that the population OR in American equals 1 by pooling studies 1 through 5.

Hello Barbora,

I couldn't tell what software you were using for your analysis. What I can say is that both CMA V2 and SPSS V29 yielded p = .35, which concurs with the circled value at the bottom left of your screen shot. (True using either OR or log-OR as the ES metric.)

For OR: p = .35 is the result of testing Ho: Pooled OR = 1

For log-OR: p = .35 is the result of testing Ho: Pooled log-OR = 0

I can't deduce what the other p-value (.56) is supposed to be representing without additional information.

Good luck with your work.

In a forest plot, the p-values you're referring to might be related, but they could also differ for various reasons. Let me break down the possibilities:

Here's why these p-values might be different:

• Statistical Noise: Subgroup analyses involve smaller sample sizes, which can lead to increased variability and "noise" in the data. As a result, the p-values for subgroup effects might be more variable and potentially less significant than the overall test p-value.
• Multiplicity Issues: When conducting multiple subgroup analyses, there's a concern about inflating the overall Type I error rate (the chance of finding a significant effect when there isn't one). This is often addressed through various statistical adjustments, such as the Bonferroni correction. The overall test p-value may account for these adjustments, while subgroup p-values might not.
• Interaction Effects: The treatment effect might vary across subgroups. For example, a treatment might be more effective in one subgroup and less effective in another. These interaction effects can result in different p-values for each subset.
• Random Chance: Due to random variation, even if the treatment has no actual effect in any subgroup, there's still a chance of observing small p-values in some subgroups just by chance. This can lead to inconsistencies between subgroup p-values and the overall p-value.

In an ideal scenario, if the treatment effect is consistent across all subgroups, you might expect the subgroup p-values to be similar or close to the overall test p-value. However, due to the factors mentioned above, differences can occur. It's crucial to interpret these p-values cautiously, considering the context, the sample sizes, and the statistical methodologies used. Additionally, if you're concerned about the differences, consulting with a statistician or conducting more in-depth statistical analyses might provide a clearer understanding of the results.

Header and table isn't matching I guess...

p-value in the rightmost column of a forest plot is the p-value for the overall test of the treatment effect across all subgroups. It is calculated by combining the results of the individual studies in the meta-analysis.

The p-value for the test for effect in the subgroup is the p-value for the test of the null hypothesis that the treatment effect in the subgroup is equal to zero. It is calculated using only the data from the studies in the subgroup.

The two p-values can be different for a number of reasons, including:

Heterogeneity: If there is a lot of heterogeneity in the treatment effects across subgroups, then the p-value for the overall test of the treatment effect will be lower than the p-value for the test for effect in any individual subgroup.

Sample size: If the sample size is small in a subgroup, then the p-value for the test for effect in the subgroup will be higher than the p-value for the overall test of the treatment effect.

Chance: It is also possible that the two p-values are different simply due to chance.

It is important to note that a statistically significant p-value for the test for effect in a subgroup does not necessarily mean that the treatment effect is different in that subgroup from the overall treatment effect. It is possible that the difference is due to chance or to other factors, such as heterogeneity or small sample size.

To determine whether there is a true difference in the treatment effect between subgroups, it is important to consider all of the evidence, including the p-values for the individual studies, the p-values for the tests for effect in the subgroups, and the p-value for the overall test of the treatment effect. It is also important to consider the magnitude of the differences in the treatment effects between subgroups and the clinical implications of those differences.

In general, it is more reliable to draw conclusions about subgroup differences based on the results of multiple studies than on the results of a single study.

Now coming to your table p-value in the right column of the forest plot is the p-value for the overall test of the treatment effect across all subgroups. It is calculated by combining the results of the individual studies in the meta-analysis. In this case, the p-value is 0.56, which is not statistically significant.

The p-value for the test for effect in the subgroup is the p-value for the test of the null hypothesis that the treatment effect in the subgroup is equal to zero. It is calculated using only the data from the studies in the subgroup. In this case, the p-value for the test for effect in the subgroup is 0.094035, which is statistically significant.

The two p-values are different because of the heterogeneity between the studies in the meta-analysis. The heterogeneity statistic (0.5) is very high, which indicates that there is a lot of variability in the treatment effects across studies. This variability could be due to a number of factors, such as different study designs, different populations of patients, and different treatment regimens.

When there is heterogeneity in the treatment effects across studies, it is more difficult to detect a significant overall treatment effect. This is because the variability in the treatment effects across studies can mask the true effect of the treatment.

In this case, the p-value for the overall test of the treatment effect is not statistically significant, but the p-value for the test for effect in the subgroup is statistically significant. This suggests that the treatment may be effective in the subgroup, but it is not possible to draw a definitive conclusion without further research.

It is important to note that a statistically significant p-value for the test for effect in a subgroup does not necessarily mean that the treatment is clinically effective in that subgroup. It is possible that the difference in the treatment effect is small or that it is not clinically meaningful.

To determine whether the treatment is clinically effective in a subgroup, it is important to consider the magnitude of the difference in the treatment effect and the clinical implications of that difference