Hormone replacement therapy and false positive recall in the Million Women Study: patterns of use, hormonal constituents and consistency of effect

The variables for these analyses, including use of HRT, were defined according to what was reported on the recruitment questionnaire, that is, at baseline. Women were asked which specific proprietary preparation of HRT they had used most recently, and for these analyses the preparations are grouped according to the hormonal constituents of each preparation listed in the British National Formulary [6]. Women whose periods had ceased either naturally or as the result of a bilateral oophorectomy are defined as postmenopausal. Women aged 55 and over who had had a hysterectomy without oophorectomy were also defined as postmenopausal, as were women aged 55 and over who began use of HRT before their natural menopause [7].

These analyses include 87,967 postmenopausal women aged 50 to 64 years who did not report a past history of cancer (except non-melanoma skin cancer) at recruitment. (This includes a small number (n = 2,527, 2.9%) who were aged 49 or 65 years when screened, but were close to their 50th or 65th birthday, since women are invited to screening according to their year of birth, rather than their exact age.)

Relative risks (RR) are estimated as odds ratios and 95% confidence intervals (CI) calculated by conditional logistic regression, stratifying by screening centre, age at screening (50 to 52, 53 to 55, 56 to 58, 59 to 61, 62 to 64), time since menopause (<5 years, 5 to 9 years, ≥10 years), reported previous breast screening (no, yes), body mass index (<25 kg/m², ≥25 kg/m²) and history of previous breast surgery for a condition other than breast cancer (no, yes). All adjustment factors include an 'unknown' category. Adjustment for other potential confounding factors, such as deprivation index, educational attainment, family history of breast cancer, age at first birth and parity, had no material effect on the findings and are, therefore, not included in the analyses. All tests for trend by duration and time since last use of HRT were calculated using the appropriate median duration or time since last use in each category. Where results are presented graphically, relative risks are shown as black squares with areas inversely proportional to the variance of the log relative risk, indicating the amount of statistical information available for that particular estimate. The corresponding confidence intervals are shown as horizontal lines.

Articles presenting original epidemiological data on the effect of current and past use of HRT on the risk of false positive recall were located through searches of PubMed, the Science Citation Index and through searching reference lists of located references. Information on the relative risk of false positive recall in current and past users of HRT compared to never users was extracted and presented graphically. The summary relative risk was calculated as the exponent of the average of the log relative risks for the relevant studies, weighted according to the inverse of the variance of the individual log relative risks.

Table 1 shows the proportion of women experiencing false positive recall according to use of HRT, along with the associated relative risk estimates and their confidence intervals. Stratifying by screening centre, age, time since menopause, previous breast screening, body mass index and previous breast surgery, women who had ever used HRT had a significantly higher risk of false positive recall, compared to women who had never used HRT (RR 1.47, 95% CI 1.36–1.60), and this relative risk was particularly elevated among current users (RR 1.64, 95% CI 1.50–1.80) (Table 1). Women who had used HRT in the past were also at a significantly increased risk of false positive recall compared to never users (RR 1.21, 95% CI 1.06–1.38) and this relative risk differed significantly from that in current versus never users of HRT (χ² ₁(heterogeneity) = 13.7, p < 0.001). There was a significant trend in recall according to time since stopping use of HRT (χ² ₁(trend) = 14.0, p < 0.001), suggesting the wearing off of the effect of HRT over time in past users [3].

Among current users of HRT, the relative risk was significantly elevated in all duration groups, including those women who had been using HRT for less than one year, and there was no significant trend in the relative risk according to duration of use (Table 1). In contrast, there was a significant trend of increasing risk of false positive recall with increasing duration of use in past users (Table 1). Recent users of HRT are more likely to have used it for long durations, and there were insufficient data to separate out reliably the effects of duration of use and time since last use among past users of HRT.

The relative risk of false positive recall differed significantly according to whether a woman was currently using oestrogen-only, oestrogen-progestagen or other types of HRT: RR (95% CI) 1.62 (1.43–1.83), 1.80 (1.62–2.01) and 0.95 (0.73–1.23), respectively (χ² ₂(heterogeneity) = 22.2, p < 0.0001) (Fig. 1). This was mainly because the relative risk (95% CI) of false positive recall among current users of tibolone was 0.76 (0.52–1.10) compared to women who had never used HRT, and the difference in the relative risk of false positive recall between women currently using oestrogen-only, oestrogen-progestagen and tibolone was highly significant (χ² ₂(heterogeneity) = 21.0, p < 0.0001). There was no significant difference in the relative risk of false positive recall between women using oestrogen-only and oestrogen-progestagen HRT (χ² ₁(heterogeneity) = 2.3, p = 0.1) [3].

Among women currently using oestrogen-only or oestrogen-progestagen HRT, there was no significant difference observed according to the type of oestrogen (for conjugated equine oestrogen versus oestradiol, χ² ₁(heterogeneity) = 0.5, p = 0.5) or type of progestogen (for medroxyprogesterone acetate versus norethisterone versus norgesterel/levonorgesterel, χ² ₂(heterogeneity) = 4.3, p = 0.1) in use, according to the dose of oestrogens currently being used (for conjugated equine oestrogen ≤0.625 mg versus >0.625 mg, χ² ₁(heterogeneity) = 3.0, p = 0.08; for oestradiol ≤1 mg versus >1 mg, χ² ₁(heterogeneity) = 0.5, p = 0.5; for oestradiol ≤50 μg versus >50 μg, χ² ₁(heterogeneity) = 3.4, p = 0.07) or according to whether the oestrogen was taken orally, via a patch or an implant (χ² ₂(heterogeneity) = 2.8, p = 0.2) (Fig. 1). There was no significant difference in the relative risk of false positive recall between those using sequential and continuous oestrogen-progestagen HRT (χ² ₁(heterogeneity) = 0.3, p = 0.6).

To examine how consistent the effects of current and past use of HRT are across other possibly relevant factors, the adjusted relative risks of false positive recall and 95% confidence intervals in current and past users of HRT compared to never users are presented (Fig. 2). Significant heterogeneity was observed in the effect of current use of HRT (global χ² ₂₀(heterogeneity) = 38.2, p = 0.01), which was chiefly the result of variation according to the closely related factors of age within screening round and time since menopause. Although this heterogeneity appeared to be greater for time since menopause than for age, there were insufficient data to investigate their dual effects. There was no significant heterogeneity in the effect of HRT according to whether or not women had had a previous screen, after accounting for age. No significant heterogeneity was present for current use of HRT for any other factors examined, nor was significant heterogeneity found for the effect of past use of HRT (global χ² ₂₀(heterogeneity for past use) = 18.1, p = 0.6). These included comparisons between the different screening centres (Fig. 2), in that there was no significant difference between the nine centres using single-view mammography among previously screened women (centres A to E and G to J) and the centre using two-view mammography among previously screened women (centre F) (χ² ₁(heterogeneity for current use) = 1.1, p = 0.3; χ² ₁(heterogeneity for past use) = 0.2, p = 0.7).

Figure 3 summarises the results of this study, along with previous studies that have compared the risk of false positive recall among current and past users of HRT separately with that for never users [8‐10]. The overall results show a significantly elevated risk of false positive recall among current users of HRT compared to never users, and a lesser, but still significant, increase in risk among former users. There is significant heterogeneity in the results for current use of HRT but not for past use.