There is inevitably a certain amount of clinical heterogeneity between trials, because of differing protocols, treatments, and patient characteristics and this will be true irrespective of whether any statistical test for heterogeneity is significant. If statistical heterogeneity is observed, then the possible reasons for this should be explored [54]. Perhaps the treatments are sufficiently different and can be split into less heterogeneous subsets of trials. If such a source of heterogeneity is found and trials can be subdivided according to the appropriate characteristic, separate analyses can be performed. In fact, this may enable us to address questions such as which particular treatments perform best or which types of patient will benefit most. Indeed, it could be argued that modeling such heterogeneity using the random effect approach (see Section 11.6.1) is effectively throwing away valuable information. The example of chemotherapy in non-small cell lung cancer described in Section 11.7 provides an example of this. There is statistical heterogeneity if all the trials are grouped together irrespective of chemotherapy type, but because the trials were split into subsets by type of chemotherapy, the heterogeneity was explained mostly by the differences between subsets. Splitting the trials in this way, according to a good clinical rationale minimized the heterogeneity within the subsets and maximized the interaction term between subsets. It suggested a different effect according to the type of chemotherapy used and in this particular example it would not be informative to combine all trials in a random effect model. As in this example, it is useful to consider possible sources of heterogeneity at the outset rather than trying to explain it after the fact. Otherwise, in the search for the clinical source of statistical heterogeneity, there is likely to be some post hoc reasoning in the explanation, so that only cautious conclusions should be drawn.

Heterogeneity can be investigated fully only when IPD are available, particularly when the endpoint is the time to an event. Although reported trial characteristics may give some insight, for example by employing meta-regression techniques, it is unlikely that all reasonable possible sources of heterogeneity can be investigated in a meta-analysis based on data extracted from publications.

Quit Smoking Today

Quit Smoking Today

Quit smoking for good! Stop your bad habits for good, learn to cope with the addiction of cigarettes and how to curb cravings and begin a new life. You will never again have to leave a meeting and find a place outside to smoke, losing valuable time. This is the key to your freedom from addiction, take the first step!

Get My Free Ebook

Post a comment