Abstract
PURPOSE:
Controversy continues regarding the impact of screening mammography on breast cancer outcomes. We evaluated late-stage cancer rate and overall survival for different screening intervals using a real-world institutional research data mart.
MATERIALS AND METHODS:
Patients having both a cancer registry record of new breast cancer diagnosis and pre-diagnosis screening history between 2004 and 2019 were identified from our institutional research breast data mart. Time interval between the two screening mammograms immediately preceding diagnosis and the time to cancer diagnosis were determined. Screening interval was deemed “annual” if ≤15 months, “biennial” if >15 and ≤ 27 months, “intermittent” if > 27 months, and “baseline” if only one pre-diagnosis screen was known. Primary endpoint was late-stage cancer (TNM stage IIB or worse) and secondary endpoint was overall survival. Association of screening interval and late-stage cancer was analyzed using multivariable logistic regression adjusting for pre-diagnosis characteristics. Proportional hazards regression was used for survival analysis. Potential lead time was analyzed using survival from a uniform fixed timepoint.
RESULTS:
8,145 patients with breast cancer had pre-diagnosis screening mammography in the timeframe. The percentage of late-stage cancers diagnosed increased significantly with screening interval with 9%, 14% and 19% late stages for annual, biennial, and intermittent groups (p<0.001). The trend persisted regardless of age, race, and menopausal status. Biennial and intermittent groups had substantially worse overall survival than the annual screened group, with relative hazards of 1.42 (95%CI: 1.11–1.82) and 2.69 (95%CI: 2.11–3.43) respectively, and 1.39 (95%CI: 1.08–1.78) and 2.01 (95%CI: 1.58–2.55) after adjustment for potential lead time.
CONCLUSIONS:
Annual mammographic screening was associated with lower risk of late-stage cancer and better overall survival across clinical and demographic subgroups. Our study suggests benefit of annual screening for women aged 40 and older.
Introduction
Randomized mammography screening trials have shown a significant reduction in breast cancer mortality in participating patients [1–3]. Since the trials published in the late 20th century, mammography screening has become policy in many middle- and high-income countries. However, in practice, screening frequency and age range vary among countries and among providers [4, 5].
While the trials established the major strategic issues of screening efficacy and effectiveness, the tactical issues of age range and screening frequency can be informed by observational data within working mammography programs in routine healthcare [6]. Several papers published by the Breast Cancer Surveillance Consortium (BCSC) have examined the issue of annual versus biennial screening and have concluded that premenopausal women benefit from annual screening. There was weaker evidence for benefit in post-menopausal women, and the benefit was mainly confined to those using hormone replacement therapy [7–9]. A recent analysis of population screening data in Sweden found that attendance at two successive (biennial) screening mammograms conferred a greater reduction in risk of breast cancer mortality than did attendance at only one of the two [10].
The American College of Radiology and several other organizations recommend annual population screening with mammography from age 40 [11]. This contrasts with recommendations from other bodies, which recommend initiating screening at a later age and with longer screening intervals [1,3]. The US Preventive Services Task Force (USPSTF) recommendations for 2023 are biennial screening starting at age 40 [12]. Given these differences, there is clearly still a need for evidence on start age and screen interval.
We have developed an institutional research breast data mart that can address this and other questions about breast cancer detection and treatment. The mart contains electronic medical record (EHR), radiology information system (RIS), pathology database (PD) and cancer registry (CR) breast-related data on nearly 500,000 women who received services from 2004 to 2020 and was developed within our existing research data warehouse, Neptune [13,14]. In this paper, we used a set of these data to examine the effect of adherence to an approximate annual schedule for mammography screening, in comparison with longer screening intervals, on cancer outcomes, specifically the fraction of late-stage cancers and all-cause mortality. We also examined outcomes by three subcategories: age, race, and menopausal status.
Methods
Institutional review board (IRB) approval was obtained for this retrospective observational HIPPA compliant review. At the time of this project, the mart contained information for 12,145 patients entered into the cancer registry (CR) as having a new diagnosis of breast cancer between January 1,2004 to December 31,2020. Of these, 8,145 patients aged 40 and older had at least one screening examination prior to diagnosis. Two screenings occurring within 260 days of each other were considered to be from the same screening episode. Patients with only 1 screen exam performed within 260 days of the cancer diagnosis date were classified as “baseline”. This definition included true baselines and those whose prior outside facility exam was irretrievable either due to being old enough to have been destroyed by the outside facility or not locatable because the patient could not recall the location of the study. Figure 1 shows how the current cohort was identified.
Figure 1:
Cohort development flowchart.
*Cancer Registry
**Electronic Health Record
In our institution from 2004–2012 full field digital mammography (FFDM) screening was performed for all cases. Digital breast tomosynthesis (DBT) screening using combination FFDM with DBT began in 2012 with full transition to DBT based screening in 2014. In 2016, screening transitioned completely to synthetic mammography (SM) with DBT.
Our primary aim was to determine the effect of screening interval on cancer stage at presentation and secondary aim was subsequent overall survival. Dependent variables were stage at diagnosis (with subcategories by tumor grade) and years of survival post diagnosis (based on CR data element “last known contact” through the end of the study period, or death, if sooner). Time over the fixed window of calendar time was considered to verify the robustness of findings with respect to potential lead-time bias. Independent variables included age at diagnosis, race, first degree family history and menopausal status. We classified regularity of screening adherence into four categories: baseline- only one screening episode prior to diagnosis; “annual”- ≤15 months between the two most recent screenings; “biennial”- >15 months and ≤ 27 months between the two most recent screenings; and “intermittent”- > 27 months between the two most recent screenings. Adherence intervals were based on prior similar evaluation from the BCSC publication of Miglioretti et al (9). Part of the rationale to select a 3-month window past the 12 and 24 month annual and biennial strict definition was to accommodate instances of slight clinical variance that naturally occurs in any screening program but to avoid selecting dates too close to the 6-month midpoint mark (i.e. 18 months) between annual and biennial. To verify that our conclusions were robust, we also considered the association of ungrouped screening interval (in days) and late-stage cancer.
Late-stage cancer was defined as American Joint Commission on Cancer (AJCC) TNM stage IIB or worse, based on survival data and was obtained from the CR [15]. The specific classification applied in the cancer registry for each new cancer was based on the version of AJCC classification in use at the time of diagnosis and according to the North American Association of Central Cancer Registries (NAACCR) guidance.
Statistical analysis was performed using non-parametric and semi-parametric methods implemented in SAS v.9.4 (SAS Institute, Cary, NC). Likelihood ratio tests (proc freq) were used for patient characteristic distribution comparisons and screening adherence groups. Logistic regression (proc logistic) was used to assess trends of late-stage cancer by adherence group (primary covariate) and significant (at 0.05 level) patient characteristics (adjustment covariates). To verify the robustness of the unadjusted association of screening interval with late-stage cancer, the Kruskal-Wallis test (proc npar1way) was used to compare continuous values of screening interval between cancer stages and grades. To control for possible confounding in the association of late-stage cancer and adherence groups, we used multivariable logistic regression with late-stage as an outcome, the adherence groups as a primary predictor, and other variables as adjusting covariates (age, race, menopausal status, and first-degree relatives with breast cancer).
Overall survival was analyzed using the methods for incompletely observed time-to-event data. The Kaplan-Meier approach estimated median follow-up time after diagnosis as well as the overall survival curves for different groups and strata. Log-rank test was used to evaluate the unadjusted trend in overall survival by the adherence group (proc lifetest). Cox proportional hazards model (proc phreg) was used to assess the association of post-diagnosis survival with screening adherence group adjusting for other significant (at 0.05 level) patient characteristics. As a check on the possibility of lead time bias, we used the proportional hazard model for right truncated time (proc phreg) to analyze the time between the last screen and diagnosis by group and estimated survival from a fixed date for all cancers, rather than the date of diagnosis (Appendix A). Unknown data for each characteristic were included in the tables.
Results
Through detailed state department of health mandated reconciliation efforts, our CR has 100% case ascertainment for reportable cases. Appendix B summarizes excluded cases. Table 1 summarizes characteristics of the 8,145 patients in the final analysis dataset including 1,470 (18%) patients aged ≤50 years at diagnosis and 4,312 (53%) aged ≥60; 7,287 (89%) White, 6,224 (76%) postmenopausal and 2,305 (28%) with first-degree relatives with breast cancer. Cancer was most frequently early-stage (1–2A) grade 2 in 2,065 (25%) with late-stage (2B-5) in 1,121 (14%). Screening interval was “annual” in 3,369 (41%), “biennial” in 1,340 (16%), “intermittent” in 1,129 (14%), and “baseline” in 2,307 (28%).
Table 1.
Distribution of patients’ characteristics in the study population and across the screening adherence groups.
| Patient characteristics | Overall | Baseline | Annual (⩽15 months) | Biennial (>15 and ⩽27 months) | Intermittent (>27 months) | ||
|---|---|---|---|---|---|---|---|
|
| |||||||
| Overall | 8145 (100%) | 2307 (100%) | 3369 (100%) | 1340 (100%) | 1129 (100%) | ||
|
| |||||||
| Age (years) * | |||||||
| 40–49 | 1470 (18%) | 644 (28%) | 458 (14%) | 201 (15%) | 167 (15%) | ||
| 50–59 | 2363 (29%) | 635 (28%) | 962 (29%) | 423 (32%) | 343 (30%) | ||
| 60–69 | 2456 (30%) | 641 (28%) | 1057 (31%) | 424 (32%) | 334 (30%) | ||
| 70–79 | 1384 (17%) | 271 (12%) | 686 (20%) | 231 (17%) | 196 (17%) | ||
| 80+ | 472 (6%) | 116 (5%) | 206 (6%) | 61 (5%) | 89 (8%) | ||
|
| |||||||
| Race * | |||||||
| White | 7287 (89%) | 2052 (89%) | 3101 (92%) | 1187 (89%) | 947 (84%) | ||
| Black | 676 (8%) | 196 (8%) | 207 (6%) | 126 (9%) | 147 (13%) | ||
| Other | 182 (2%) | 59 (3%) | 61 (2%) | 27 (2%) | 35 (3%) | ||
|
| |||||||
| Menopausal status | |||||||
| Pre- | 1277 (16%) | 496 (21%) | 426 (13%) | 207 (15%) | 148 (13%) | ||
| Post- | 6224 (76%) | 1501 (65%) | 2733 (81%) | 1064 (79%) | 926 (82%) | ||
| Unknown | 644 (8%) | 310 (13%) | 210 (6%) | 69 (5%) | 55 (5%) | ||
|
| |||||||
| First-degree Relatives with Breast Cancer * | |||||||
| No | 5840 (72%) | 1722 (75%) | 2313 (69%) | 977 (73%) | 828 (73%) | ||
| Yes | 2305 (28%) | 585 (25%) | 1056 (31%) | 363 (27%) | 301 (27%) | ||
|
| |||||||
| Stage * | |||||||
| Unknown | 125 (2%) | 43 (2%) | 48 (1%) | 14 (1%) | 20 (2%) | ||
| 0 | 1866 (23%) | 534 (23%) | 829 (25%) | 303 (23%) | 200 (18%) | ||
| 1 | 3847 (47%) | 939 (41%) | 1756 (52%) | 635 (47%) | 517 (46%) | ||
| 2A | 1186 (15%) | 376 (16%) | 425 (13%) | 207 (15%) | 178 (16%) | ||
| 2B | 538 (7%) | 193 (8%) | 153 (5%) | 89 (7%) | 103 (9%) | ||
| 3 | 483 (6%) | 184 (8%) | 133 (4%) | 73 (5%) | 93 (8%) | ||
| 4 | 100 (1%) | 38 (2%) | 25 (1%) | 19 (1%) | 18 (2%) | ||
|
| |||||||
| Tumor Grade * | |||||||
| Stage 0 and Unknown | G1 | 68 (1%) | 21 (1%) | 25 (1%) | 13 (1%) | 9 (1%) | |
| G2 | 577 (7%) | 110 (5%) | 264 (8%) | 109 (8%) | 94 (8%) | ||
| G3 | 381 (5%) | 88 (4%) | 172 (5%) | 69 (5%) | 52 (5%) | ||
| Unknown | 965 (12%) | 358 (16%) | 416 (12%) | 126 (9%) | 65 (6%) | ||
|
|
|||||||
| Early-stage Invasive (Stage 1–2A) | G1 | 262 (3%) | 52 (2%) | 123 (4%) | 44 (3%) | 43 (4%) | |
| G2 | 2034 (25%) | 427 (19%) | 900 (27%) | 372 (28%) | 335 (30%) | ||
| G3 | 999 (12%) | 194 (8%) | 441 (13%) | 180 (13%) | 184 (16%) | ||
| Unknown | 1738 (21%) | 642 (28%) | 717 (21%) | 246 (18%) | 133 (12%) | ||
|
|
|||||||
|
Late-stage Invasive (Stage 2B-4) |
G1 | 17 (0%) | 2 (0%) | 0 (0%) | 9 (1%) | 6 (1%) | |
| G2 | 331 (4%) | 99 (4%) | 112 (3%) | 45 (3%) | 75 (7%) | ||
| G3 | 345 (4%) | 113 (5%) | 85 (3%) | 63 (5%) | 84 (7%) | ||
| Unknown | 428 (5%) | 201 (9%) | 114 (3%) | 64 (5%) | 49 (4%) | ||
distribution of characteristic differs across the annual, biennial, and intermittent groups (p<0.005 by the likelihood ratio test).
Because of the characteristics of the baseline group, all factors in Table 1 have significantly different distribution across groups. Excluding baselines, increasing screening interval was associated with younger age (p<0.001) and non-White race (p<0.001). The annual screening group had a larger proportion of first-degree relatives having breast cancer. There was no meaningful association of menopausal status with screening adherence (p>0.01). Longer screen intervals (grouped or continuous) were more highly associated with high grade cancers and late-stage cancers overall (p<0.001).
Table 2 summarizes the distribution of stage and tumor grade by patient characteristics and screening adherence group. Late-stage prevalence varied significantly across age, race, and adherence groups (p<0.001, by likelihood ratio test). There was no meaningful association of late stage of any grade with menopausal status or family history (p>0.2). The prevalence of late-stage cancer was 9%, 14%, and 19% for the annual, biennial, and intermittent groups respectively (p<0.001) and remained substantial and significant for late-stage high grade (grade 3) tumors (with the respective prevalence levels being 3%, 5%, and 7%, p<0.001).
Table 2.
Prevalence of late-stage (stage IIB or higher) cancer (overall and for specific tumor grades) by patient characteristics and screening adherence.
| Patient characteristics | Cancer Patients # | Patients with Late-Stage Cancer |
|||
|---|---|---|---|---|---|
| Any grade** # (%) | Grade 1 # (%) | Grade 2 # (%) | Grade 3 # (%) | ||
|
| |||||
| Overall | 8145 | 1121 (14%) | 17 (0%) | 331 (4%) | 345 (4%) |
|
| |||||
| Age (years) * | |||||
| 40–49 | 1470 | 235 (16%) | 3 (0%) | 58 (4%) | 73 (5%) |
| 50–59 | 2363 | 362 (15%) | 5 (0%) | 97 (4%) | 98 (4%) |
| 60–69 | 2456 | 309 (13%) | 3 (0%) | 109 (4%) | 96 (4%) |
| 70–79 | 1384 | 147 (11%) | 3 (0%) | 43 (3%) | 52 (4%) |
| 80+ | 472 | 68 (14%) | 3 (1%) | 24 (5%) | 26 (6%) |
|
| |||||
| Race * ,+ | |||||
| White | 7287 | 964 (13%) | 15 (0%) | 292 (4%) | 284 (4%) |
| Black | 676 | 133 (20%) | 1 (0%) | 31 (5%) | 53 (8%) |
| Other | 182 | 24 (13%) | 1 (1%) | 8 (4%) | 8 (4%) |
|
| |||||
| Menopausal Status | |||||
| Premenopausal | 1277 | 174 (14%) | 2 (0%) | 48 (4%) | 56 (4%) |
| Postmenopausal | 6224 | 845 (14%) | 15 (0%) | 272 (4%) | 272 (4%) |
| Unknown | 644 | 102 (16%) | 0 (0%) | 11 (2%) | 17 (3%) |
|
| |||||
| First-degree Relatives with Breast Cancer | |||||
| No | 5840 | 815 (14%) | 16 (0%) | 241 (4%) | 239 (4%) |
| Yes | 2305 | 306 (13%) | 1 (0%) | 90 (4%) | 106 (5%) |
|
| |||||
| Screening Adherence * ,+ | |||||
| Baseline screen | 2307 | 415 (18%) | 2 (0%) | 99 (4%) | 113 (5%) |
| Annual (⩽15 months) | 3369 | 311 (9%) | 0 (0%) | 112 (3%) | 85 (3%) |
| Biennial (>15 and ⩽ 27 months) | 1340 | 181 (14%) | 9 (1%) | 45 (3%) | 63 (5%) |
| Intermittent (>27 months) | 1129 | 214 (19%) | 6 (1%) | 75 (7%) | 84 (7%) |
statistically significant association with late stage cancer (p<0.001 by the likelihood ratio test)
including unknown grade
statistically significant association with grade 3 late-stage cancer (p<0.001 by likelihood ratio test)
The robustness of the trend by screening adherence was further supported by the differences in screening intervals times between groups of patients with early-stage (median 412 days, 95% CI: 407–415), late-stage low grade (1–2) (median 469 days, 95% CI: 439–503), and late-stage high grade (3) cancer (median 615 days, 95% CI: 518–710), p<0.001 by Kruskal-Wallis test.
Table 3 illustrates the consistency of the trend in the proportion of late-stage cancer with screening adherence for specific patient characteristics. The proportions of late-stage cancer remained increasing for all individual characteristic levels without any characteristic being a significant effect modifier (p>0.5 for the interaction term in the logistic regression with late-stage cancer as outcome). Adjusting for all considered characteristics within the multivariable logistic regression, higher rates of late stage with longer screen interval remained significant (p<0.001), with odds ratios of 1.5 (95% CI: 1.2–1.8) for biennial and 2.2 (95% CI: 1.8–2.6) for intermittent compared to the annual group. The baseline group also had a significantly higher risk of late-stage cancer than the annual group (odds ratio 2.1, 95% CI: 1.8–2.4).
Table 3.
Trend in prevalence of late-stage cancer across the three screening adherence groups (without baseline) by patient characteristics.
| Patient Characteristics* | Annual (⩽15 months) | Biennial (>15 and ⩽27 months) | Intermittent (>27 months) |
|---|---|---|---|
|
| |||
| Overall | 311/3369 (9%) | 181/1340 (14%) | 214/1129 (19%) |
| Age (years) | |||
|
| |||
| 40–49 | 54/458 (12%) | 41/201 (20%) | 35/167 (21%) |
| 50–59 | 99/962 (10%) | 68/423 (16%) | 67/343 (20%) |
| 60–69 | 89/1057 (8%) | 44/424 (10%) | 61/334 (18%) |
| 70–79 | 53/686 (8%) | 19/231 (8%) | 31/196 (16%) |
| 80+ | 16/206 (8%) | 9/61 (15%) | 20/89 (22%) |
| Race | |||
|
| |||
| White | 277/3101 (9%) | 155/1187 (13%) | 171/947 (18%) |
| Black | 29/207 (14%) | 20/126 (16%) | 34/147 (23%) |
| Other | 5/61 (8%) | 6/27 (22%) | 9/35 (26%) |
| Menopausal status | |||
|
| |||
| Premenopausal | 36/426 (8%) | 38/207 (18%) | 24/148 (16%) |
| Postmenopausal | 249/2733 (9%) | 133/1064 (13%) | 180/926 (19%) |
| Unknown | 26/210 (12%) | 10/69 (14%) | 10/55 (18%) |
| First-degree relatives | |||
|
| |||
| No | 211/2313 (9%) | 131/977 (13%) | 159/828 (19%) |
| Yes | 100/1056 (9%) | 50/363 (14%) | 55/301 (18%) |
the increasing trend is observed for all levels of characteristics, without any significant effect modification (p>0.5 for interaction terms in the logistic regression with late-stage as outcome)
Table 4 shows numbers of cases, known deaths, death rate (per 1,000 patients/year), and median follow-up after diagnosis by adherence group. Median time to diagnosis from screening ranged from 24 to 28 days across all adherence groups. Figure 2A shows corresponding survival probability by time. In the univariate model, the biennial and intermittent groups had significantly higher rates of death than the annual group (p<0.001 for Cox model or Log-rank test), with hazard ratios of 1.42 (95%CI: 1.11–1.82) and 2.69 (95% CI: 2.11–3.43) respectively. The baseline group also showed an elevated hazard rate relative to the annual group (hazard ratio 1.32, 95% CI: 1.1–1.61).
Table 4.
Cases, follow-up durations and known all-cause deaths by adherence group.
| Adherence group | Cases | Median follow-up time (years) | Known Deaths (rate per 1,000 patients per year) |
|---|---|---|---|
| Baseline | 2,307 | 6.77 | 232 (15.3) |
| Annual (≤15 months) | 3,369 | 5.13 | 191 (10.8) |
| Biennial (>15 and ≤ 27 months) | 1,340 | 4.64 | 92 (14.8) |
| Intermittent (>27 months) | 1,129 | 3.36 | 103 (25.6) |
| All | 8,145 | 5.12 | 619 (14.3) |
Figure 2.
Survival over time from the cancer diagnosis for the four adherence groups. (A) including all ages (B) cohort 40–49 years of age (1,470 with 45 deaths) (C) cohort aged 50 years and older (6,675 with 574 deaths).
Survival curves for adherence groups within the primary screening-age strata are shown in Figure 2B and 2C. Among 1,470 women aged 40–49 at the time of diagnosis, only 45 (3.6%) deaths were known (over the follow-up time; median 5.7 years) offering a limited ability to differentiate between the adherence groups. There were 574 (8.6%) deaths known among 6,675 women aged ≥50 years (over the follow-up time; median 5 years) with the distribution illustrating a substantial difference among the adherence groups (p<0.001), with the annual group showing the best survival, intermittent the worst, and biennial and baseline groups being very similar.
In the multivariable model adjusting for age, race, menopausal status, and first-degree relatives with breast cancer, longer screen interval remained strongly associated (p<0.001) with poorer post-diagnosis survival, with the adjusted hazard ratios for screening adherence groups being similar to the unadjusted ones reported above (i.e., hazard ratio 1.48 for biennial, and 2.04 for intermittent).
For all groups, > 75% of cancers were diagnosed within two months of a screening mammogram, suggesting that all groups had high proportions of screen-detected cancers, and that all groups would be likely to be similarly affected by lead time. However, to address the potential for unobserved differential lead time, we also compared post-diagnosis survival from the same date of December 31, 2009, for all women rather than individual diagnosis date. Details are provided in Appendix A. Results confirmed the best post-diagnosis survival in the annual adherence group, followed by biennial (hazard ratio 1.39, 95% CI: 1.08–1.78, p=0.010), then baseline 1.7(hazard ratio 1.98, 95% CI:1.63–2.41, p<0.001), and with the worst survival in the intermittent adherence group (hazard ratio 2.01, 95% CI: 1.58–2.55, p<0.001).
Discussion
Data from 16 years of screening mammography (which includes FFDM, DBT plus FFDM, and DBT plus SM) showed that annual screening is associated with lower likelihood of having late-stage cancer, and higher overall survival, than does biennial screening or intermittent screening. This benefit of annual screening persisted across all age, race, and menopause status categories. Our results show unequivocally the advantages of annual asymptomatic breast cancer screening starting at the age of 40.
This result builds upon knowledge from a study from the Breast Cancer Surveillance Consortium [9]. They found that the benefit of annual screening in terms of stage was clear in pre-menopausal women, but in post-menopausal women, the benefit was smaller and mainly confined to women using hormone replacement therapy. However, that study also demonstrated a borderline significant increase in risk of invasive tumors measuring >15 mm associated with biennial screening in post-menopausal women not using hormone replacement therapy. Thus, one would not rule out a benefit in postmenopausal women based on the Breast Cancer Surveillance Consortium results. One would, however, conclude that any such effect is weaker in post-menopausal than in pre-menopausal women. Our results agree with this, although without statistically significant difference between pre- and post-menopausal women. It is worth noting that Moorman et al [16], also found a significant tendency to diagnose breast cancer at an earlier stage with annual screening in postmenopausal women.
Our results show a benefit for annual versus biennial screening, which are consistent with population level results from Duffy et al [24] and Morell et al [25] both of whom demonstrated a significant mortality benefit by adherence to recommended screening intervals compared to cohorts who were screened less frequently or inconsistently. Our results also are consistent with CISNET modelling [17] and recommendations by the National Comprehensive Cancer Network, the American Congress of Obstetricians and Gynecologists, and the American College of Radiology. It is contrary to the USPSTF, which recommends biennial screening for women 40 and older. Recently, Woloshin and colleagues [18] argued against screening in women aged under 50, asserting that there is no significant effect on mortality of screening in this age group, citing the 2013 Nordic Cochrane review to support this [18, 19]. However, the latter is out of date, as it does not include the most recent results from the UK Age Trial [20], and because it relies heavily on the Canadian National Breast Screening Trial, for which there is evidence that the randomization was comprised [21, 22]. The more recent trial results, such as those of the UK Age Trial, and observational findings on mammography screening in women aged 40–49 are consistent with our results [23]. Those classified as baseline represented a significant fraction of screen detected cancers (28%) in this cohort. Because this group included both true baseline patients and those of any age whose prior study was so old as to have been irretrievable, the cohort was subject to both prevalence and incidence detection.
We examined our data for possible sources of bias by confirming that the effect on the late-stage cancer persisted for different thresholds for the adherence groups as well as for the difference in time between screening (women with late-stage cancer having on average, longer gaps between screening exams, p<0.001). We also computed the median time between cancer diagnosis and the preceding screening, and found that all adherence groups had approximately 28 day median values. Thus, the stage differences at diagnosis are unlikely to be due to differences in detection mode or time since last screen. One might expect our survival results to be influenced by lead time, since adherence to annual screening confers a greater probability of cancer being screen detected sooner. However, more than 75% of cancers in all groups were diagnosed within two months of a screening mammogram, suggesting that all groups had similarly high proportions of screen-detected cases and therefore comparable lead times. To avoid lead time bias altogether, we also calculated survival from a fixed date. Results confirmed the poorer post-diagnosis survival in the biennial group (hazard ratio of 1.3, p<0.001) and intermittent group (hazard ratio of 2.02, p<0.001) compared with the annual group (Appendix A). Longer follow-up will afford an opportunity to explore issues of lead time in detail.
Our study is limited in that our data are from a single facility within a single organization. The organization has academic and community medical centers in rural, suburban, and urban settings. While the patient population does not necessarily reflect the distribution of ethnicity and race across the entire country, our subcategory analyses show consistent results across both race and age. Therefore, we believe our results should hold for different patient populations. Evaluating baseline health across the population was outside the scope of this project and could theoretically confer a bias toward those participating in more frequent screening. There is a possibility that a patient had a screening mammogram outside organization and would be misclassified into the wrong adherence group — into a group with a longer interval. Given the nature of the regional health insurance environment, this is unlikely to occur. Subjects having screening which was not recorded in our data mart would mean that some subjects were misclassified as receiving less intensive screening than they actually did receive. Assuming that the risk of diagnosing cancer at late stage is lower for those participating in more frequent screening based on our results of such, this would be expected to lead to underestimation of differences among adherence groups. Finally, we note that despite the large data size, strong and consistent results, and adjustments for multiple factors related to both cancer risk and screening strategy, a large prospective, randomized clinical trial would be needed to unequivocally prove that increasing screen interval increases all-cause mortality. It is unlikely that such a study will be performed, thus our results add to the body of data demonstrating the relationship.
Conclusions
Annual screening is associated with a significantly lower risk of diagnosis of late-stage breast cancer and a significant survival benefit for patients ages 40 and older, with the trend persisting across age, menopausal status and race.
Supplementary Material
Context Summary.
Key Objective:
Does the frequency of mammographic screening affect the stage at diagnosis and patient overall survival?
Knowledge Generated:
Extracting data from our institutional breast cancer data mart, in 8,145 patients with breast cancer, those women who were screened annually had a significantly lower percentage of late-stage cancers than those screened biennially or with longer intervals; this association held regardless of age, race and menopausal status. Annually screened women also had significantly better overall survival.
Relevance (Dr. Miller):
Screening guidelines must balance the benefits of additional detection with the potential harms and costs associated with false positives. This observational study complements results from modelling studies and supports annual screening for most women.
Acknowledgments:
The authors would like to thank the following for their contributions to this work: Jonathan Silverstein, MD, Richard Morgan, Amy Klym, Sharon Winters
Funding:
This project was supported in part by award number P30CA047904 from the National Cancer Institute.
Footnotes
List of where and when the study has been presented in part elsewhere: Presented as a poster at the San Antonio Breast Cancer Symposium, December, 2023.
References:
- 1.US Preventive Services Task Force. Screening for breast cancer: U.S. Preventive Services Task Force recommendation statement. Ann Intern Med. 2009. Nov 17;151(10):716–26, W-236. [DOI] [PubMed] [Google Scholar]
- 2.Lauby-Secretan B, Scoccianti C, Loomis D, et al. ; International Agency for Research on Cancer Handbook Working Group. Breast-cancer screening--viewpoint of the IARC Working Group. N Engl J Med. 2015. Jun 11;372(24):2353–8. [DOI] [PubMed] [Google Scholar]
- 3.Schünemann HJ, Lerda D, Quinn C, et al. ; European Commission Initiative on Breast Cancer (ECIBC) Contributor Group. Breast Cancer Screening and Diagnosis: A Synopsis of the European Breast Guidelines. Ann Intern Med. 2020. Jan 7;172(1):46–56. [DOI] [PubMed] [Google Scholar]
- 4.Klabunde CN, Ballard-Barbash R; for the International Breast Cancer Screening Network. Evaluating Population-Based Screening Mammography Programs Internationally. Semin Breast Dis. 2007. Jun;10(2):102–107. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Dowling EC, Klabunde C, Patnick J, Ballard-Barbash R; International Cancer Screening Network (ICSN). Breast and cervical cancer screening programme implementation in 16 countries. J Med Screen. 2010;17(3):139–46. [DOI] [PubMed] [Google Scholar]
- 6.Hellquist BN, Duffy SW, Abdsaleh S,et al. Effectiveness of population-based service screening with mammography for women ages 40 to 49 years: evaluation of the Swedish Mammography Screening in Young Women (SCRY) cohort. Cancer. 2011. Feb 15;117(4):714–22. [DOI] [PubMed] [Google Scholar]
- 7.White E, Miglioretti DL, Yankaskas BC, et al. Biennial versus annual mammography and the risk of late-stage breast cancer. J Natl Cancer Inst. 2004. Dec 15;96(24):1832–9. [DOI] [PubMed] [Google Scholar]
- 8.Kerlikowske K, Zhu W, Hubbard RA, et al. ; Breast Cancer Surveillance Consortium. Outcomes of screening mammography by frequency, breast density, and postmenopausal hormone therapy. JAMA Intern Med. 2013. May 13;173(9):807–16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Miglioretti DL, Zhu W, Kerlikowske K, et al. ; Breast Cancer Surveillance Consortium. Breast Tumor Prognostic Characteristics and Biennial vs Annual Mammography, Age, and Menopausal Status. JAMA Oncol. 2015. Nov;1(8):1069–77. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Duffy SW, Tabár L, Yen AM, et al. Beneficial Effect of Consecutive Screening Mammography Examinations on Mortality from Breast Cancer: A Prospective Study. Radiology. 2021. Jun;299(3):541–547. [DOI] [PubMed] [Google Scholar]
- 11.Lee CH, Dershaw DD, Kopans D, et al. Breast cancer screening with imaging: recommendations from the Society of Breast Imaging and the ACR on the use of mammography, breast MRI, breast ultrasound, and other technologies for the detection of clinically occult breast cancer. J Am Coll Radiol. 2010. Jan;7(1):18–27. [DOI] [PubMed] [Google Scholar]
- 12.Screening for breast cancer US Preventive Services Task Force recommendation statement. JAMA. 2024; doi: 10.1001/jama.2024.5534 [DOI] [PubMed] [Google Scholar]
- 13.Visweswaran S, McLay B, Cappella N, et al. An atomic approach to the design and implementation of a research data warehouse. J Am Med Inform Assoc. 2022;29(4):601–608. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Zuley ML, Silverstein J, Logue D, et al. Organizational Breast Cancer Data Mart: A Solution for Assessing Outcomes of Imaging and Treatment. JCO Clin Cancer Inform. 2024;8, e2300193. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Amin MB, Edge S, Greene F, Byrd DR, Brookland RK, Washington MK, Gershenwald JE, Compton CC, Hess KR, et al. (Eds.). AJCC Cancer Staging Manual (8th edition). Springer International Publishing: American Joint Commission on Cancer; 2017. [Google Scholar]
- 16.Moorman SEH, Pujara AC, Sakala MD, et al. Annual Screening Mammography Associated With Lower Stage Breast Cancer Compared With Biennial Screening. AJR Am J Roentgenol. 2021. Jul;217(1):40–47. [DOI] [PubMed] [Google Scholar]
- 17.Hendrick RE, Monticciolo DL, Biggs KW, Malak SF. Age distributions of breast cancer diagnosis and mortality by race and ethnicity in US women. Cancer. 2021. Dec 1;127(23):4384–4392. [DOI] [PubMed] [Google Scholar]
- 18.Woloshin S, Jørgensen KJ, Hwang S, Welch HG. The New USPSTF Mammography Recommendations - A Dissenting View. N Engl J Med. 2023. Sep 21;389(12):1061–1064. [DOI] [PubMed] [Google Scholar]
- 19.Gøtzsche PC, Jørgensen KJ. Screening for breast cancer with mammography. Cochrane Database Syst Rev. 2013. Jun 4;2013(6):CD001877. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Duffy SW, Vulkan D, Cuckle H, et al. Effect of mammographic screening from age 40 years on breast cancer mortality (UK Age trial): final results of a randomised, controlled trial. Lancet Oncol. 2020. Sep;21(9):1165–1172. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Yaffe MJ, Seely JM, Gordon PB, Appavoo S, Kopans DB. The randomized trial of mammography screening that was not—A cautionary tale. J Med Screen. 2022; Mar;29(1): 7–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Seely J Failed randomization and explanation for the outlier results of the Canadian Breast Cancer Screening Studies. Borealis, V2. University of Ottawa – Dataverse.https://dataverse.scholarsportal.info/dataset.xhtml?persistentId=doi:10.5683/SP3/2DEY36. Accessed May 5, 2023. [Google Scholar]
- 23.Wilkinson AN, Ellison LF, Billette JM, Seely JM. Impact of Breast Cancer Screening on 10-Year Net Survival in Canadian Women Age 40–49 Years. J Clin Oncol. 2023. Oct 10;41(29):4669–4677. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Duffy SW, Tabár L, Yen AMF, Dean PB, Smith RA, Jonsson H, Törnberg S, Chiu SYH, Chen SLS, GHH, Ku MMS, Hsu CY, Ahlgren J, Maroni R, Holmberg L, Chen TH. Beneficial Effect of Consecutive Screening Mammography Examinations on Mortality from Breast Cancer: A Prospective Study Radiology 2021. 299:3, 541–547. [DOI] [PubMed] [Google Scholar]
- 25.Morrell S, Taylor R, Roder D, Robson B, Gregory M, Craig K. Mammography service screening and breast cancer mortality in New Zealand: a National Cohort Study 1999–2011. Br J Cancer 2017;116(6):828–839. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.