Chapters Index (French) French Version Welcome page Contact

Unfortunately, the results of our trials are often of no real significant due to the low number of patients included in studies.

By grouping all of these non-significant studies, it is possible to build a meta-analysis taking into account all of the data from the original trials. This meta-analysis can, however, only reply to a simple question.

For instance, by grouping together all breast carcinoma adjuvant chemotherapy trials, including, in one arm anthracycline (independently of the type of anthracycline and of its dosage) and, in another arm, a standard treatment without anthracycline, it was possible to demonstrate that the presence of anthracycline in the chemotherapy regimen improves results.

Thus, meta-analyses offer new statistically interesting results.

Example of a meta-analysis on hormonotherapy in breast cancer. On the left, a positive effect of hormonotherapy. On the right, a negative effect. In the middle, no effect is observed. In this example, 5 trials are grouped together and pooled. None of them has included enough patients to be conclusive: the standard deviation is too high. Different types of hormonotherapy have been given to the patients (the physicians could not agree on a single prescription), however all studies have one arm with hormonotherapy and one arm without. The meta-analysis then constitutes an important number of patients and allows a positive conclusion.

However, meta-analysis should be analysed with caution.

No patient should have been excluded (especially due to toxicity).

Moreover, most authors generally only publish positive results and negative results are of little interest to them. The same can be said for many international publications, which, just like good journalists, like to create an event and do not always consider publishing “non-information”. Finally, the results of a study may not have been published because the study was too difficult to perform (for instance, many patients excluded for toxicity). The absence of all of these negative results will falsify the meta-analysis result.

Nevertheless, with simple reasoning, it is clear that studies which require a great number of patients (sometimes over a thousand!) in order to prove a significant difference between two therapeutic options, firstly demonstrate that these two options, in reality, give very close results (statistically speaking, there is even a 1% to 5% chance of obtaining identical results!).

Thus, the physician should not take any risks for his/her patient by prescribing a slightly better treatment.

As common sense would say, meta-analyses were not necessary to prove the efficiency of tuberculosis antibiotics or penicillin.