COMMENTARY

Title, Date of Publication & Journal

Does hormone replacement therapy (HRT) cause breast cancer? An application of causal principles to three studies. Part 5. Trends in breast cancer incidence in relation to the use of HRT.

Shapiro S, et al. J Fam Plann Reprod Health Care 2013;0:1–9. doi:10.1136/jfprhc-2012-100508

Claim supported by evidence?

The paper reviews the ecological evidence presented by three previous studies that the decline in HRT use ‘led to a reduction in breast cancer incidence’.

Through a thorough examination of several well-established principles of causal inference, this paper concludes that there is not enough strength or consistency in the ecological data to support this causal link. The paper does not rule out a link; but finds this particular evidence too weak to support it in the way these previous studies have argued.

This paper does not address all of the evidence linking HRT with breast cancer.

A principal part is that the decline in breast cancer started in 1999, whereas the decline in HRT followed the WHI announcement in July 2002.

Study Conclusions

• The study conclusions are carefully worded – they do not refute the entire evidence of a link between HRT and breast cancer.
• The authors conclude that the decline in breast cancer MIGHT NOT be causally related to the decline in HRT.
• The authors calculations for dose-response show an increasing risk for HRT:

o   As the authors state – confounders might explain this.

o   However, HRT could also explain this!!

• The commentary states:

o   Preparations in use nowadays are lower dose, often transdermal and identical to endogenous hormones, and as such could have completely different risk profiles. Recent observational data have suggested that the risk of breast cancer is neutral if natural progesterone is used; these findings require confirmation from prospective trials.

Strengths/Limitations

It uses well-established epidemiological principles of causality to pick apart claims made in the previous studies

It comes to a balanced conclusion that this data can neither support nor refute a causal link

This paper deals with the studies’ findings by using each of the main causal principles in turn that have been laid out in the epidemiological literature. They highlight areas and particular statements / results in each of the papers that can be refuted in light of each these principles. They cross reference where one finding / implication in the studies might fall foul of more than one causal principle.

This paper deals only with the ecological data presented and so (cannot) does not claim to make definitive statements about the possible link between HRT and breast cancer incidence (although it is clear the authors are generally sceptical and have written about this previously).

The authors do not estimate the relative risk for breast cancer due to HRT.  They do not show that the relative risk is within acceptable limits.  In epidemiology, it is difficult to gather conclusive evidence with a small effect – as the authors acknowledge.

Lack of evidence is not evidence of lack of effect!

Glossary

Ecological studies do not collect or analyse individual data, but rather take (often) routinely collected data and try to find correlations that might exist. They are mainly only used (and recommended) for hypothesis generation that may then lead on to other analytical studies in which these hypotheses can be tested by collecting individual-level data.

An example would be that high levels of alcohol consumption might be found to correlate with high rates of cirrhosis of the liver in a number of countries. This does not tell us anything about the risk faced by any one individual in that country, but the finding could prompt e.g. a case-control study in which individuals with cirrhosis are asked about their alcohol consumption, or a cohort study in which people’s drinking is assessed on a regular basis and they are then followed to determine who gets cirrhosis. More complicated and costly, but more definitive. It could be that the correlations apparent in the ecological study are not borne out by the analytical studies.

One particular problem with ecological studies (as explained in the paper) is that without individual-level data you cannot correct for other possible explanations of the correlation (i.e. confounders or biases).