Breast cancer mortality using mammography with or without clinical breast examination* (15)
| Age, yr | Range of follow-up, yr | No. and design of studies | Women who die from breast cancer, %† | Women who are not screened: risk of dying of breast cancer per 1000 (15) | Relative risk (95% CI)‡ | Women who are screened: risk of dying from breast cancer per 1000 (95% CI) | Absolute effect per 1000 (95% CI) | Number needed to screen (95% CI) | GRADE rating of certainty of evidence | |
|---|---|---|---|---|---|---|---|---|---|---|
| Mammogram with or without clinical breast examination | Usual care, % | |||||||||
| 40–49 | 17.7 to 25.7 | 8 RCTs (23), (33)– (37)¶ | Unavailable†† | 0.4 | 3.85 | 0.85 (0.78 to 0.93) | 3.27 (3.00 to 3.58) | 0.58 fewer (0.27 fewer to 0.85 fewer) | 1724 (1176 to 3704) | ⊕⊕○○ LOW‡‡ |
| 50–59§ | 18 to 30 | 6 RCTs (33), (35)– (37)¶ | Unavailable†† | 0.5 | 5.00 | 0.85 (0.78 to 0.93) | 4.25 (3.90 to 4.65) | 0.75 fewer (from 0.35 fewer to 1.10 fewer) | 1333 (909 to 2857) | ⊕○○○ VERY LOW§§ |
| 60–69 | 13.1 to 30 | 4 RCTs (23), (33), (35)– (36)** | Unavailable†† | 0.6 | 6.15 | 0.85 (0.78 to 0.93) | 5.23 (4.80 to 5.72) | 0.92 fewer (from 0.43 fewer to 1.35 fewer) | 1087 (741 to 2326) | ⊕⊕○○ LOW‡‡ |
| 70–74 | 13.2 to 13.6 | 2 RCTs (23), (35) | Unavailable†† | 1.0 | 10.31 | 0.85 (0.78 to 0.93) | 8.76 (8.04 to 9.59) | 1.55 fewer (from 0.72 fewer to 2.27 fewer) | 645 (441 to 1389) | ⊕○○○ VERY LOW ¶¶ |
Note: CI = confidence interval, GRADE = Grading of Recommendations Assessment, Development and Evaluation system, RR = relative risk.
↵* This table presents screening outcomes based on short-case accrual methods. The systematic review conducted for this guideline presents screening outcomes using both short- and long-case accrual methods. (15) In short-case accrual (initial + 7 subsequent screens; 23 years of follow-up), only those cases diagnosed during the screening period are included (median 7 yr; range 3 to 12 yr), whereas long-case accrual (initial + 4 subsequent screens; 14 years of follow-up) includes all cases diagnosed to the end of the follow-up period. Short-case accrual reduces bias from contamination because women in the control group would not have been screened until the trial was over, while long-case accrual may underestimate the benefits of screening as women in the control group are more likely being screened after the trial. (15)
↵† The baseline risk in the control groups may not have been representative of all included studies because the numerators or denominators were either unclear or not reported.
↵‡ A subgroup analysis of relative risk by age was assessed based on published methodology. (31), (32) No difference in RR among subgroups was detected and true differences resulting from age were deemed unlikely. The use of the all-ages RR data rather than focusing on each decade of age aligns with this assessment. It is appropriate to use control event rates relevant to each age group to determine corresponding risks, with attendant implications on absolute benefit (calculations provided in Appendix 4).
↵§ Presents results with control group at moderate baseline risk.
↵¶ The Stockholm and Gothenburg trials were considered quasi-randomized. (33)
↵** The Stockholm trial was considered quasi-randomized. (23)
↵†† Complete data were not available. Numerators or denominators were either unclear or not reported for all included studies.
↵‡‡ Very serious concerns about risk of bias because randomization and allocation concealment were either not reported or had serious deficiencies.
↵§§ Very serious concerns about risk of bias because randomization and allocation concealment were either not reported or had serious deficiencies and there were serious concerns about inconsistency, as heterogeneity may be moderate (I2 = 26%; p value = 0.24) and there were serious concerns about imprecision because, although the number of events and total population are large (> 300 threshold for events), the 95% CIs include the null and do cross appreciable benefit (RR 0.75).
↵¶¶ Very serious concerns about risk of bias because randomization and allocation concealment were either not reported or had serious deficiencies and there were serious concerns about imprecision because, although the total population is large (> 2000), the 95% CIs include the null and do cross appreciable benefit (RR 0.75).