Mega-trials and Meta-analysis: Large-scale Randomised Evidence

R Peto
Radcliffe Infirmary
Oxford, UK

There have been many large randomised trials of potential cancer preventive agents, most of which have shown no differences between treatment groups and controls. While this may seem disappointing, it is important that negative results are known. Large randomised trials of different hypotheses are needed to test laboratory and epidemiological concepts, since some cancer treatment trials have shown unexpected but important results.

In order to assess which cancer treatments have a real effect, strictly randomised evidence from large numbers of patients is required. Currently, many importantly incorrect answers and unjustified claims of improved survival are resulting from small non-randomised trials, as small trials can produce an impression of benefit that is simply due to chance. Meta-analyses of small trials are unreliable, and there is recurring emphasis on selected parts of the randomised evidence, which produces distortion of the treatment effects.

"The fundamental assumption in cancer treatment trials is the need either to detect or to refute moderate differences in 10-year survival", explained Dr Peto. To avoid bias, there has to be proper randomisation, analysis according to allocated treatments, emphasis on overall results, and systematic overview of all relevant randomised trials. The key requirement for accuracy is large numbers of patients.

Tamoxifen Trials

During the past 25 years, there have been more than 55 placebo-controlled trials of tamoxifen in early breast cancer (table 1). A total of 37,000 women with stage I or II breast cancer have been randomised to receive either tamoxifen or no tamoxifen.¹ These trials have implications both for the treatment and prevention of breast cancer.

The early results were wrongly assumed to indicate that tamoxifen was not effective. One reason for these mistaken assumptions was that the early trials did not include sufficient patients. Another fault was that these trials investigated tamoxifen for an inadequate duration (1 to 2 years), whereas it is now clear that tamoxifen must be taken for at least 5 years for protection from breast cancer. In addition, many patients with oestrogen receptor-negative (ER-) breast cancer were given tamoxifen since this agent acts on the oestrogen receptor it is probably only effective in women with ER+ breast cancer. It is now known that tamoxifen is of substantial value for patients who have ER+ disease, but the early experience led to a false negative result for the use of tamoxifen in breast cancer.

Table 1. Early breast cancer trials adjuvant therapy versus controls

Using the appropriate tamoxifen regimen for the appropriate disease (early stage ER+ breast cancer) results in a statistically significant 16% difference (p < 0.00001) in recurrence rate versus placebo, while 10-year mortality rates show an 8% difference in favour of tamoxifen over placebo (20% vs 28%, respectively). Importantly, this difference in survival was not shown in the early trials, and 15 years ago, the belief was that tamoxifen was clearly not effective. Today, among women with early breast cancer, a large proportion are still not offered tamoxifen, although the benefits could be increased by using this drug more widely.

Table 2. Reduction of new breast cancer incidence in trials of tamoxifen given for 5 years

An unexpected finding of the tamoxifen trials is that the incidence of contralateral breast cancer (disease incidence) is reduced by interference with hormonal stimulation of the tissues (p < 0.00001). In effect, tamoxifen prevents the emergence of a new cancer in the opposite breast. The same result has been obtained in women who are at high risk of breast cancer but who have never had the disease (table 2). Small trials of 5 years' tamoxifen treatment versus longer treatment have, perhaps wrongly, suggested that 5 years of treatment is sufficient. However, these trials are too small and do not have a sufficiently long duration of follow-up to provide reliable answers.

In Conclusion

Large scale randomised evidence with long-term follow-up is required for both treatment and preventive measures. "We really need to be much more serious about getting evidence that involves large numbers [of patients] and reliable results, and we need large-scale randomised evidence", concluded Dr Peto.

Reference

1. Tamoxifen for early breast cancer: an overview of the randomised trials. Early Breast Cancer Trialists' Collaborative Group. Lancet 1998;351:1451-1467.