BreAst cAncer mortAlity reduction After initiAtion of screening progrAm: consistency of effect estimAtes oBtAined using different ApproAches

Effect on breast cancer mortality of a mammography screening program initiated in 1998 was assessed. Two effect estimates were obtained for each of three groups: participants, eligible women and all women of the target age group including those with prior breast cancer diagnosis. Methods: Four approaches were used: 1) observed and projected breast cancer mortality trends for 1998-2004 were compared, 2) breast cancer mortality in the first five years of the program (1998-2003) was compared to mortality for the five years prior to program initiation (1992-1997) restricting numerators of rates to breast cancer deaths occurring among incident cases, 3) observed number of breast cancer deaths among 523,830 program participants was compared to expected number based on breast cancer incidence and survival of nonparticipants, 4) nested case-control study (873 cases, 8730 controls) was done within the 1,054,620 women eligible for screening. Results: Among participants, estimates of breast cancer mortality reduction associated with screening were 35% (95% confidence interval (CI): 23% to 48%) and 41% (95% CI: 28% to 52%). Among eligible women, estimates were 11% (95% CI: 1% to 21%) and 7% (95% CI: 1% to 13%). Among all women of the target age group, estimates were 3% (95% CI: -1% to 15%) and 3% (95% CI: -1% to 6%). Conclusion: The four approaches used provided effect estimates that were consistent within and between groups studied. Effect estimates and their consistency support the view that initiation of this screening program led to a significant reduction in breast cancer mortality within five years of its initiation. A population-based mammography screening program was initiated in Québec, Canada, in 1998. This program invites by letter all women aged 50-69 for biennial screening mammography in designated screening centers. Such community screening programs have been associated with a reduction of breast cancer mortality using a variety of epidemiologic methods that have different strengths and limitations. Possible variation in extent of residual bias of these methods could lead to differences and non-comparability of estimates of screening effects. Thus, in order to estimate the possible effect on breast cancer mortality of the Québec program in the first five years after its initiation, we carried-out four comparisons that provided two estimates of effect for each of three groups of women. Two estimates were obtained for women who participated in the program (participants). These are the only women whose breast cancer mortality can be affected directly by program screening. Among participants, the expected breast cancer mortality reduction associated with screening could reach 35%. Two estimates were also obtained for women who were eligible to the program (eligible women). This group which includes participants as well as non-participants, is of interest because it is the one targeted by the program. The aim of this screening program was to reduce breast cancer mortality in eligible women by 25% after 10 years of operation. Finally, two estimates were obtained for the entire population of women in the target age group (all women of the target age group). This group is of interest because screening is expected to be reflected in overall population age-specific breast cancer mortality statistics. This group includes eligible women as well as women in the target age group who already had a breast cancer diagnosis before program initiation or before becoming eligible to the program once the program had started. Women with prior diagnosis of breast cancer will be responsible for a substantial proportion of breast cancer deaths in the first years after program initiation. MATERIAL AND METHODS OBSERVED AND EXPECTED MORTALITY TRENDS This approach, based on the examination of population breast cancer mortality trends, provided an estimate of the effect of screening among all women of the age targeted by the program. Trends in breast cancer mortality rates observed after the introduction of the program (1998/2004) in all women aged 50 to 74 years were compared with trends expected during this period under the hypothesis that changes in breast cancer mortality seen prior to the introduction of the program had persisted. This analysis focused on the age group 50 to 74 because screening women 50 to 69 could affect breast cancer mortality up to the age of 74 in the five years studied. Observed trends were estimated by weighted loess smoothing applied to the logarithm of age standardized mortality rates observed between 1975 and 2004. Weights were defined as the square of standardized rates divided by their variances. Expected breast cancer mortality trends were obtained by weighted Joinpoint analysis in which weights were the same as those used for Institut national de santé publique du Québec 1 Breast Cancer Mortality Reduction after Initiation of a Screening Program: Consistency of Effect Estimates Obtained Using Different Approaches loess smoothing. Joinpoint analysis was first applied to logarithm of age standardized mortality rates from 1975 to 1998 to identify any major changes in breast cancer mortality that might have occurred prior to 1998 and to estimate the trend in breast cancer mortality in the period immediately preceding 1998. Expected mortality in 2004 was obtained by projecting this Joinpoint trend seen in the period prior to 1998 to the subsequent five years. The ratio of breast cancer mortality in 2004 obtained by loess smoothing to the projected mortality obtained by Joinpoint analysis was estimated. The variance of the logarithm of this ratio was calculated as the sum of the variances of the logarithm of the predicted rates minus twice their covariance which itself was set at its largest possible value (minus the square root of the product of the variances of the logarithm of the two predicted rates). The 95% confidence limits of the logarithm of this ratio were then obtained based on its variance. These limits were exponentiated to obtain corresponding confidence limits for the mortality rate ratio. PRE AND POST MORTALITY RATES This approach, used by Tabar et al., provided an estimate of the effect of screening among eligible women. The breast cancer mortality rate in the first five years of the program (1998-2003) was compared to the rate observed in the preceding five years (19921997). According to this approach, the numerators of the rates include only breast cancer deaths that occur among cases diagnosed in the period under consideration. This restriction implies elimination from the calculation of deaths that occur among women diagnosed before the beginning of the period under consideration. Ideally, the denominators of the rates should also be corrected to exclude women who had a diagnosis of breast cancer before the beginning of the period under consideration because these women are not eligible for screening. However, such women represent only a relatively small proportion of the entire population and this correction would not affect rates materially. All breast cancer deaths that occurred in Québec women in 1992-2003 were identified using the Québec mortality database. Linkage with the Québec Tumour Registry allowed identification of the year of their breast cancer diagnosis. Québec demographic data were used to estimate the denominators of the rates. The ratio of age adjusted breast cancer mortality rate in 19982003 to mortality rate in 1992-1997 was estimated for women aged 50-69. The above mentioned breast cancer mortality rate ratio in women 50-69 years needed to be corrected for secular improvement in treatment and survival to better isolate the effect of screening. A “corrected” rate ratio was obtained by using the breast cancer mortality rate ratios observed among women aged 2049 and 70 years or more. Such correction is based on the idea that the mortality reductions observed in these women reflects the effects of treatment improvement that would also be expected at age 5069. The “corrected” rate ratio intended to better reflect the effect of screening on breast cancer mortality among women aged 50-69 was estimated by dividing the rate ratio observed in this age group by the pooled rate ratio seen in the other age groups. The pooled rate ratio was obtained by calculating the weighted average of the logarithm of the rate ratios for women aged 20-39, 40-49 and ≥ 70 using as weights the inverse of the variances of the logarithm of the rate ratios. The variances of the logarithm of rate ratios in each age group were those obtained by Poisson regression. The variance and 95% confidence limits of the logarithm of the “corrected” ratio were then obtained as well as, by exponentiation, the corresponding 95% confidence limits of the “corrected” ratio itself. OBSERVED AND EXPECTED BREAST CANCER