Breast cancer screening

Abstract

Objective To assess the impact of the NHS breast screening programme on mortality from breast cancer in women aged 55-69 years over the period 1990-8.

Design Age cohort model with data for 1971-89 used to predict mortality for 1990-8 with assumption of no major effect from screening or improvements in treatment until after 1989. Effect of screening and other factors on mortality estimated by comparing three year moving averages of observed mortality with those predicted (by five year age groups from 50-54 to 75-79), the effect of screening being restricted to certain age groups.

Setting England and Wales.

Subjects Women aged 40 to 79 years.

Results Compared with predicted mortality in the absence of screening or other effects the total reduction in mortality from breast cancer in 1998 in women aged 55-69 was estimated as 21.3%. Direct effect of screening was estimated as 6.4% (range of estimates from 5.4-11.8%). Effect of all other factors (improved treatment with tamoxifen and chemotherapy, and earlier presentation outside the screening programme) was estimated as 14.9% (range 12.2-14.9%).

Conclusions By 1998 both screening and other factors, including improvements in treatment, had resulted in substantial reductions in mortality from breast cancer. Many deaths in the 1990s will be of women diagnosed in the 1980s and early 1990s, before invitation to screening. Further major effects from screening and treatment are expected, which together with cohort effects should result in further substantial reductions in mortality from breast cancer, particularly for women aged 55-69, over the next 10 years.

Introduction

The NHS breast screening programme was introduced in England and Wales in 1988, on the recommendations of the Forrest committee.[1] These recommendations were based on the findings of randomised controlled trials, which showed that mammography could reduce mortality from breast cancer in women aged 50 years and over by 25-30% over a period of about 10 years. In England and Wales women aged 50-64 are invited for screening every three years. Evidence from the programme itself indicates, and national incidence rates confirm, that the build up of activity was gradual and that the "prevalent" round of screening was not completed until 1995.

In 1992 the Department of Health set a target for breast cancer of a reduction in mortality of 25% in the age group invited for screening by the year 2000.[2] It was subsequently acknowledged that this target should be applied to the age group 55-69 years. Screening would not be expected to affect mortality in the 50-54 years age group because the average age at first screening for women in the programme is 51.5 years (as women are first invited between the ages of 50 and 52), and in the randomised controlled trials there was little or no effect of screening in the first four years. Also survival from breast cancer in the late 1980s was good (five year relative survival rate was almost 70% and 10 year survival over 50%)[3] and will have been higher in women detected by screening because of earlier detection during the preclinical phase. The programme has set targets relating to uptake and rates of cancer detection, which, if achieved, should eventually lead to the target reduction in mortality of 25%.[4]

The targets for cancer detection are based on the detection rates observed in the Swedish two county randomised controlled trial.[5] In the analysis of this trial, all deaths from breast cancer in women with date of diagnosis before the date of entry to the trial in both the study and control arms were excluded. In 1989, Day suggested that the reduction in mortality from breast cancer in the target population should be at least 25% after 10 years from the start of screening, but only in those women free from breast cancer when first invited to screening.[6] National mortality statistics will include both women who were and were not free from breast cancer at the time of their first invitation. The further away in time from the start of screening, the greater will be the proportion of women in the former category. Many deaths from breast cancer in the 1990s will be in women diagnosed with breast cancer before any invitation to screening, as full coverage of the population of England and Wales did not occur until 1995. Consequently the impact of screening on breast cancer on the national mortality statistics by the year 2000 is likely to be much less than 25%.[2]

Mortality from breast cancer began falling in England and Wales from around 1990, before the programme could have been expected to have a major impact.[7] By 1994, in the age group 55-69 years, where mortality would be affected by screening, there was a 12% reduction compared with prescreening rates in the late 1980s. An explanation for some of this reduction may be the use of adjuvant tamoxifen, which by 1990 was in widespread use for women aged over 50 years. It has been pointed out, however, that there is no direct evidence for such an effect and that by 1993 some reduction in mortality (albeit rather small) would have been expected from screening.[8]

Breast cancer mortality in this period may also have been affected by changes in stage at presentation, possibly because of increased publicity about breast cancer during the introduction of the screening programme, and by birth cohort effects.[9] Cohort effects may cause mortality to increase or decrease in different age groups and result in the true reduction attributable to screening and improvements in treatment being over or underestimated. The term "improvements in treatment" is used here to include effects of both changes in treatment and other factors including earlier presentation outside the age range of women invited for screening and structural changes in the NHS after the Calman-Hine report.[10]

We used an age cohort model based on mortality data for 1971-89 (which would not have been affected by screening) to predict the mortality from breast cancer for 1990-8. By comparing the observed mortality in different age groups with that predicted by the model we estimated the separate effects of screening and of improvements in treatment and other factors.

Method

We obtained data on mortality from breast cancer for England and Wales from the database of deaths at the Office for National Statistics for 1971-99 by single year of age and calendar year of death. We reconstructed these data into five year age groups (from 40-44 to 75-79 years) and cohort groups from 1896 to 1946. We corrected the annual numbers of deaths for coding changes that occurred in 1984 and 1993 as a result of revised interpretation of WHO rule 3 and for other procedural changes in 1993; the overall adjustment was only a few percentage points for women aged 50-74 but greater for younger and older age groups.[7]

We assumed that use of tamoxifen, other treatment changes, and screening would not have had any substantial impact on mortality until 1990 and that age and birth cohort effects were the major influences on mortality from 1971 to 1989.

We fitted an age cohort model using five year age groups from 40-44 to 75-79 for 1971-89 using log linear Poisson modelling with the generalised linear models command in the statistical package STATA. Although we considered that an age cohort model was likely to give the best fit we also used age (alone) and age period models for comparison, according to the methods of Clayton and Shifflers.[11] The age cohort model was highly significant (P [much less than] 0.001) when tested against the age model alone. Other models, such as an age period model did not fit the data well. Although the age cohort model provided a good fit to the data, the 1946 cohort estimate was based on only few data and had wide confidence limits. To provide a better estimate for this cohort we extrapolated the rate ratios for the 1896 to 1941 cohorts to 1946 by fitting a quadratic model, after noting that the rate ratios increased up to the 1926 cohort and decreased thereafter. The final model was used to predict the mortality from breast cancer in the absence of screening or treatment effects for 1990-8 for each five year age group.

From the mid-1980s the use of tamoxifen in women with breast cancer increased rapidly. By 1990 in the Thames regions over 90% of women aged over 50 years and over 50% below the age of 50 received tamoxifen.[12] Results of randomised controlled trials suggest a fairly uniform effect of tamoxifen across the age range 50-79 years.[13] The effect of screening, in contrast, will be more age dependent because, of the age range of women invited. Screening will reduce mortality in older age groups as time progresses, influencing mortality in women aged 55-74 by 2000 and those aged 55-79 by 2005.