Skip to main content

Table 1 Ten-point checklist of main methodological problems affecting studies of the effect of mammography screening programmes on the incidence of advanced breast cancer

From: The impact of mammography screening programmes on incidence of advanced breast cancer in Europe: a literature review

Point # Issue Problem Consequence Potentially affected studies (reference number)
1 Follow-up time The time window available to observe a decrease (if any) in ABCR is narrow and closes rapidly. In the Two-County trial, ABCR in the study group began to decrease 4 years after randomization and stabilized at a lower level on the 8th year [2]. The ABCR is expected to increase with the prevalence screening, it may fall in the years immediately following the prevalence screen, and will likely be stable at the end of screening in a cohort of women. In trend and dynamic population analysis, in the absence of an individual time zero (time at entry), the effect is confounded and the effect of screening on ABCR is underestimated. This is particularly applicable to estimates of annual percent change. [8, 12, 13, 19, 34, 37, 41]
2 Exposure time The target population is a dynamic one (but the same holds true for cohort studies). Because there is a latency for the effect of screening on ABCR to take place, at any point in time there are women (i.e., new quinquagenarians, new immigrants, and late attendees) with insufficient exposure time. The effect of screening on ABCR is underestimated, due to a disproportionate influence of prevalence screens. All studies
3 Pace of implementation Public health screening programmes are implemented gradually, in a markedly stepwise fashion, since large populations are divided in distinct administrative units each targeted by an independent local plan of action. The effect of screening on ABCR is diluted. Until implementation is completed, there are women who are diagnosed with breast cancer before being invited, and who greatly contribute to ABCR. [8, 14, 15, 19, 29, 30, 32, 33, 36,37,38,39, 44]
4 Prevalence effect The prevalence screen may be associated with a transient increase in ABCR [13]. During a stepwise implementation of the programme, when the time elapsed from the start is theoretically sufficient to see a decrease in ABCR, this is counteracted by an opposite effect due to newly enrolled women – especially if invitations increase over time. [8, 14, 15, 19, 29, 30, 32, 33, 36,37,38,39, 44]
5 Reference incidence (i) The reference (or underlying) incidence rate, with which to compare the rate observed after the introduction of screening, is not known with precision [49]. The rate can be estimated using the rate observed in the last few years before screening, assuming its stability over time, or by linear extrapolation of a pre-existing trend. The second approach is arguably preferable, but both are dependent on underlying assumptions about trends or absence of trends in incidence, and results can vary depending on these assumptions. All studies
6 Reference incidence (ii) Whatever incidence rate is being used as a reference, its validity decreases with increasing number of years of observation due to uncontrollable changes (or in the pace of such changes) in the underlying risk of breast cancer. Assessing the long-term effect of screening on ABCR is subject to considerable uncertainty and there is potential for inaccuracy in either direction (overestimation or underestimation of effect). [8, 12, 13, 19, 34, 37, 41]
7 Definition of advanced cancer There is no agreed definition of advanced breast cancer [50], even though there is general agreement that large or metastatic cancers are ‘late stage’. The definition is chosen based on differing criteria. The pT information alone, which is the most available one, is direct and relatively unaffected by biases due to confounding. Conversely, multiple-stage data are more meaningful, since the effect of screening may differ across different categories of advanced cancers. All studies
8 Stage migration The introduction of sentinel lymph node biopsy between mid-1990s and mid-2000s caused a substantial increase in the registered incidence of node-positive breast cancer (stage migration bias) [18]. The use of pN staging is problematic in studies of trends in ABCR over the last two decades, since changes in the risk of node-positive cancer cannot be adjusted for stage migration. The increase in node-positive disease is likely to be population-specific and will depend on the rate of change of local surgical policy. However, reductions in node-positive disease as a results of screening are likely to be underestimated rather than overestimated due to the stage migration. [12,13,14, 19, 29,30,31,32,33,34,35,36,37,38,39,40,41,42,43]
9 Missing data on tumour stage Whatever staging system is being used, the introduction of a screening programme tends to bring an improved quality of breast cancer registration, with a sharp decrease in the proportion of unknown-stage cancers. Because more cases are increasingly placed in all known-stage categories, an apparent increase in all stage-specific rates occurs – including ABCR. [8, 15, 30, 32, 33, 38, 39]
10 Statistical approach The statistical approach is not standardised, and includes the provision of purely descriptive information and the use of methods which are difficult to interpret, such as joinpoint analysis. Descriptive information does not allow evaluation of the magnitude and significance of observed changes in ABCR. Methods like the joinpoint analysis are useful for assessing the points in time when ABCR begins to decrease and when it stabilizes, but may be misleading when used to assess the significance of the trend. Also, the important issue is arguably what happened to ABCR following the screening rather than at what point a change occurred in the direction of a trend, which is affected by both confounding and analytic assumptions. [8, 12, 13, 19, 29, 35, 40,41,42,43]