Impact indicators

Achievement of the objective of screening for breast cancer, i.e. mortality reduction, is inevitably long-term. Ascertainment of impact on mortality demands (a) that follow up of the screened cohorts continues over extended periods of time, (b) that data on vital status and disease-free interval be vigorously sought and recorded despite the problems of follow up, and (c) that adequate links exist between programme data and other relevant data sources, e.g. medical records, pathology registers, death certificate information. Models for evaluating the impact of screening on mortality have not yet been fully developed. Given that this area of analysis is still evolving, a frequently used alternative is to identify and monitor early surrogate measures that can possibly predict outcome.

Analysis of breast cancer mortality

The objective of a breast cancer screening programme is to detect the tumour as early as possible to facilitate effective treatment and thereby reduce the mortality due to the disease. Continuous evaluation of the programme is necessary to ensure that it is as effective as expected. Difficulties in determining the impact of population screening for breast cancer entail the application of observational research designs, the absence of readily available control groups or control areas and the lack of individual data.

In the past decades, breast cancer incidence rates have steadily risen in many countries while breast cancer mortality rates have remained stable. Recently though, in several countries, mortality rates have been shown to level off or decline16,17 _{especially in countries where}

population-based screening programmes for breast cancer have been introduced in the late 1980s or early 1990s. An important question now is the relative contribution of screening to the reported declines in mortality.18_{Establishing a relationship with screening is not straightforward}

since in some countries with screening programmes, declines in mortality started already before screening was introduced, and declines also occurred in non-screened age groups and in some countries without a national screening programme.14_{This observation gives rise to questions}

about the potential contribution of other determinants of breast cancer mortality, in particular treatment advances.19,20 _{Thus, the challenge for researchers in this area is to tease out the}

relative contributions of screening and nonscreening factors to the reported declines in mortality.21

A first step in the evaluation of screening is to look at trends in breast cancer mortality. However, especially when data come from population statistics, the potential impact of service screening on breast cancer mortality will take many years to emerge, starting from a few years after the introduction of a programme but taking decades to show a full effect.11_{The delay is caused by}

the fact that it usually takes a number of years before a screening programme is fully implemented and most screening programmes are not able to correct national or regional mortality statistics for breast cancers diagnosed in women before the start of the screening programme.22,23_{Further delays are due to the lack of information on the screening history of}

individual women. No corrections at the individual level can be made for the phased implementation of the screening programme and the varying participation behaviour of women invited when individual data are not available.4

To estimate the effect of the screening programme based on a comparison of the trend in the breast cancer mortality in areas with and without a programme additional questions should be considered. The most complicated is how the control area should be selected? What aspects should be prioritised with respect to comparability of the areas – risk factor pattern for breast cancer (often unknown), treatment programmes for breast cancer, accessibility to health care, etc.

So far, the majority of studies on the impact of service screening have compared trends in breast cancer mortality, either between geographical regions or over time.4,5,9,24_{The estimates for the}

observed reductions from these studies vary from 12%25 _{to as much as 50%}26_{, with differing}

periods after the introduction of screening. However, ultimate proof that a service-based screening programme is effective can never be based solely on the analysis of trends, since factors other than screening may also be (partially or wholly) responsible for the changes in

breast cancer mortality. A few studies were able to identify a temporary group of contemporaneous controls that were not (yet) invited for screening but experienced similar exposure to breast cancer risk factors and treatment.7,10,27,28 _{The effects estimated in these}

studies, in the order of a 15 to 25% breast cancer mortality reduction, can therefore be directly attributed to the screening programme. Unfortunately, this advantage is time-limited since the control groups were also invited eventually and these studies will therefore not be able to assess the long-term impact of the service screening programme. Lastly, several studies have used modelling to estimate the expected breast cancer mortality in the absence of screening.23,28,31

By comparing the observed versus the expected numbers of breast cancer deaths, the models yielded estimates of the potential impact of the screening programme in the long-term in the order of 13 to 29%.

The number of studies on service screening efficacy, comparing breast cancer mortality in women actually screened to unscreened women, is limited so far. As expected, the results from these studies generally show higher effect estimates, even though the range is wide and varies from a 19%6_{to 63%}26_{reduction in breast cancer mortality.}

The disadvantage of using breast cancer mortality as the endpoint in evaluation of a screening programme is that it takes many years before an effect can be expected. It takes years until the study population is screened for the first round and many more years until it is possible to see an effect of the intervention. It is recommended to try to estimate the proportion of the study population exposed to the intervention from the start of the screening programme to be able to estimate when it is realistic to expect an effect.

Analysis of surrogate indicators

An attractive alternative to analysing breast cancer mortality is to identify early surrogate indicators and follow their development over time. Several characteristics have been indicated to predict a reduction in the breast cancer mortality e.g.

• Interval cancer rate*

• Breast cancer detection rate*

• Stage at diagnosis of screen-detected cancers

• Proportion of screen-detected invasive cancers ≤ 10 mm • Proportion of screen-detected cancers that are invasive

• Proportion of screen-detected cancers with lymph node metastases

* cf Glossary of terms

After ascertainment, confirmation and classification of interval cancer cases identified, the following additional measures can be calculated, as outlined in paragraph 1.8:

• Number of interval cancers per 10,000 women screened negative by time since last screening examination

• The interval cancer rate in a defined period after screening expressed as a proportion of the background (expected) breast cancer incidence rate in the absence of screening. Please note that proportions for e.g. the first and second year after screening should not be considered cumulative.

• Age-specific interval cancer rates • Round-specific interval cancer rates

• Association of interval cancer rates with other performance indicators of screening such as participation rate, recall/additional imaging rate and positive predictive value of screening mammography and of each investigation undertaken as further assessment of screen- detected lesions

• Sensitivity and impact of the screening programme

Table 33lists those early surrogate indicators for which acceptable and desirable levels could reasonably be specified in a European context. Each screening programme could decide to expand this table to include other surrogate indicators.

The acceptable and desirable levels suggested for initial screening are limited to the implementation phase of the screening programme. As a screening programme progresses over

time, an increasing proportion of the initial screening examinations will come from women entering the programme at the lower age limit (e.g. 50-51 year old). This shift in age distribution will affect the outcome of the surrogate indicators for initial screens.

Most indicators are calculated as a proportion of ‘total cancers screen-detected’ or ‘total invasive cancers screen-detected’. Cancers with unknown size or nodal status should be included in the denominator, even though the proportions calculated will seem too low if numbers unknown are high.

Table 33: Early surrogate indicators by which the impact of a breast screening programme is assessed

Surrogate Acceptable Desirable Screening indicator level level programme

50-69 Interval cancer rate* /

Background incidence rate* (%)

• 0-11 months 30% < 30% • 12-23 months 50% < 50% Breast cancer

detection rate*

• Initial screening 3xIR > 3xIR • Subsequent-regular screening 1.5xIR > 1.5xIR Stage II+/Total cancers

screen-detected (%)

• Initial screening NA < 30% • Subsequent-regular screening 25% < 25% Invasive cancers ≤10 mm/

Total invasive cancers screen-detected (%)

• Initial screening NA ≥ 25% • Subsequent-regular screening ≥ 25% ≥ 30% Invasive cancers/ 90% 80-90% Total cancers screen-detected (%)

Node-negative cancers/ Total invasive cancers screen- detected (%) • Initial screening NA > 70% • Subsequent-regular screening 75% > 75% IR = background incidence NA = not applicable * cf Glossary of terms

Proportions versus rates

Ideally, comparison of prognostic factors (size, stage) should be presented as rates per population screened, as opposed to proportions. Rates allow consideration of changes in the characteristics of cancers detected by screening.

In the early phase of a screening programme, when most examinations are prevalence screens, a high proportion of small or early-stages cancers will be detected (and, in consequence, a decreased percentage of advanced cancers). Similarly, significant ‘overdiagnosis’ of small lesions would lead to a decreased percentage of advanced cancers, although the absolute rates

may be unchanged. Expression of results as the percentage reduction in the incidence of advanced cancers requires an estimate of the incidence of advanced cancers that would have been observed in the absence of a screening programme.11

In principle, the calculation of the overall reduction of advanced cancers is straightforward: the aim of mammography screening is to detect breast cancers in an early localised stage in order to remove them and to prevent progression into an advanced potentially lethal stage.32_Thus,

effective screening should lead to a decline of advanced tumours among the screened women already in the time after the first screening round. Day et al.33_{and IARC}11_{request a reduction of}

at least 30 %.

For the assessment of the reduction, the advanced interval cancers from the interval after the first screening round and the advanced screen-detected cancers of the second screening round have to be summed up and compared with the background incidence rate of advanced tumours; correspondingly, the advanced interval cancers from the interval after the second screening round and the advanced screen-detected cancers of the third screening round have to be considered and so on. This type of combining the data is somewhat unfamiliar, since interval cancers are usually related to the screening round before the respective interval (e.g., for sensitivity assessment).

Formally, the relative reduction may be computed by: (2 x b – (v + s)) / 2 x b = 1 – (v + s) / 2 x b where

v denotes the rate of advanced interval breast cancers per 1000 screened women;

s denotes the rate of advanced screen-detected breast cancers per 1000 screened women in the screening round subsequent to the respective interval;

b denotes the background incidence rate of advanced breast cancers per 1000 women. Twice the background rate b arises from the fact that one accumulates the advanced stages within the screening over a two-year period (if the screening interval is two years).

Even though it is rather simple to translate the suggested levels for surrogate indicators (in proportions) to rates, setting acceptable and desirable levels always involves estimating the background incidence. The problems in obtaining a valid estimate for background incidence have been outlined before (paragraph 1.2).

Determining a target for the reduction of the rate of advanced tumours by screening has the same problem. In addition to the screening-internal quantities s and v, the external quantity b enters into the formula. This quantity may be different from country to country and is dependent upon prevalence and effectiveness of opportunistic screening.