To the Editor: We question the conclusions drawn by Canfell and colleagues1 from their analysis of trends in hormone replacement therapy (HRT) prevalence and breast cancer incidence for Australian women aged 50 years or older. Their ecological analysis lacks individual-level information on HRT use and information on tumour oestrogen receptor (ER) status, and captures only 2 years following the decline in HRT prevalence. This is an inadequate design within which to judge issues of causality; it is at best an hypothesis-generating exercise.2

The examination of only 2 years of breast cancer incidence after the decline in HRT use is unsound because of substantial unexplained annual variability in national breast cancer incidence. This is evidenced, for example, by the graph in Canfell et al’s Box 2, which identifies a fall in 1998–1999 that was not ascribed to changes in HRT prevalence. The lack of data on tumour ER status is another weakness. HRT use increases the risk of ER+ tumours, so any decline in HRT prevalence would be expected specifically to reduce ER+ tumour incidence.

We analysed Victorian Cancer Registry data for women aged 50 years or older for the period 2001–2005 — 2 years past the cut-off in Canfell et al’s analysis. Tumour ER status was available for 87% of breast cancer cases in 2001, rising to 90% in 2005. The demographic characteristics of the women for whom these data were available did not change between 2001 and 2005. Over this period, the proportion of all breast cancers that were ER+ increased from 65% to 71%, and the proportion of tumours with known ER status that were ER+ increased from 74% to 79%.

Poisson regression analysis of the age-specific incidence rates for both total and ER+ breast cancer for women aged 50 years or older estimated an average annual decline of 1.7% in the total incidence rate (P = 0.0009) and an annual increase of 0.2% in the ER+ incidence rate (P = 0.7). Our findings are illustrated in the Box. Although there was an apparent small decline in total breast cancer incidence in 2001–2003, this trend was reversed in 2004–2005. More importantly, the trend in ER+ tumour incidence was stable across the entire period.

Age-standardised incidence* of invasive breast cancer in women in Victoria aged ≥ 50 years, 2000–2005

Vertical bars represent 95% confidence intervals. ER+ = oestrogen receptor-positive. * Standardised to World Standard Population.

These Victorian trends, covering a longer time period and including information on the tumour type most likely to be affected by changes in HRT use, provide no support for the hypothesis of Canfell and colleagues.

In reply: In saying that our work on the effect of hormone replacement therapy (HRT) on the development of breast cancer is “at best an hypothesis-generating exercise”, Giles and colleagues ignore both the preceding literature and the fact that our analysis of Australian data explicitly tested the hypothesis raised by an analysis of United States data. Both the US and Australian analyses showed falling breast cancer incidence from 2001 in women aged ≥ 50 years, but not in younger women, following large reductions in HRT use.1,2

The logic, methodology and statistics presented by Giles et al are unsound. They present data on incident breast cancers for women in Victoria only, a subset of the national data included in our analysis. Not only is it invalid to use a subset of data to retest a hypothesis previously tested on the larger dataset (unless complex statistics are applied3-5), but Giles et al misrepresent Victorian trends. It is clear that in Victorian women aged ≥ 50 years, breast cancer incidence rates fell substantially and significantly by 11.5% between 2001 and 2003 (95% CI, 6.0%–16.6%; P < 0.0001), with no significant change among women aged 20–49 years (P = 0.3) (Box).

Giles et al present no data on HRT use. The largest drop in HRT use in Australia occurred between 2001 and 2003, and thereafter use began to plateau.6 It is therefore statistically inappropriate to calculate an “average annual decline” in breast cancer incidence from 2001 to 2005 because changes in HRT use over this time were not linear. In addition, their claim that the “trend was reversed in 2004–2005” is not supported by statistical testing, which shows no significant change in breast cancer incidence in women aged ≥ 50 years in Victoria from 2003 to 2005 (P = 0.4).

US data showed that the fall in breast cancer incidence following the drop in HRT use was seen particularly in oestrogen receptor-positive (ER+) tumours.1 As we pointed out,2 the lack of reliable Australian data on ER status means that no conclusions can be drawn about Australian trends in ER+ tumours. Giles et al acknowledge that the Victorian ER data are incomplete and that the completeness has increased over time. This, together with the possibility of differential ascertainment of ER status in women who used HRT over the period of interest, render the trends they present on ER+ breast cancers uninterpretable.

Overall, trends in breast cancer incidence in the subgroup of Victorian women cannot be said to differ materially from the national trends. A recent editorial in the Lancet draws attention to the fact that trends in HRT use and breast cancer incidence similar to those we reported have now been observed in many countries.7 The Australian findings add to the accumulating worldwide evidence that, in settings where HRT use was common and breast cancer screening rates relatively stable, a rapid decline in HRT use after 2001 was followed by a fall in breast cancer incidence.

Age-standardised incidence* of invasive breast cancer in women in Victoria, 1996–2005

Vertical bars represent 95% confidence intervals. Vertical dotted line indicates commencement of the period over which there was a hypothesised decrease in breast cancer incidence in women aged ≥ 50 years but not in women aged < 50 years. * Standardised to the Australian 2001 population.