Supplemental Table S7 summarizes area under the receiver operating characteristic curve (AUC) for various risk modeling approaches, stratified by menopausal status and screening interval.

PURPOSE:Adjustment for a sufficient set of confounders removes bias. When a confounder is unmeasured, approaches exist to estimate bias. Confounders are often correlated, so analyses that ignore correlations overstate bias. METHODS:Using NHANES III, we examined the association between Healthy Eating Index (HEI) and all-cause mortality (n = 2417). A fully adjusted model included tobacco use, sex, age, hypertension, BMI, education, and physical activity. Hazard ratios (HR) that would have been observed-had one of hypertension, BMI, education, or physical activity been "unmeasured"-were estimated by leaving them out. We then performed bias analysis for the unmeasured confounders using uncorrelated bias parameter estimates. RESULTS:The fully adjusted HR comparing HEI Quintile 1 vs. 5 was 1.72 (95% CI 1.24-2.40). After treating variables as "unmeasured" confounders, HRs changed little (range: 1.73-1.98). Bias-adjusted hazard ratios ranged from 1.72 to 1.98 suggesting that substantial unmeasured confounding would be required to explain the associations. CONCLUSIONS:Due to correlations between covariates, the additional bias attributable to unmeasured variables was minimal. QBA produced estimates that often overestimated the impact of the unmeasured confounders. Although QBA is useful for evaluating unmeasured confounding, it may not precisely quantify the strength of bias.

BACKGROUND:Data from validation substudies help quantify bias due to misclassification of an analytic variable by estimating classification parameters. Studies with long enrollment or follow-up may be susceptible to changes in classification parameters over time, which would have important implications for validation substudy design and quantitative bias analysis. Herein, we provide guidance for sampling validation data to account for time trends in classification parameters. METHODS:Using a simulated cohort of 10,000 observations, we induced exposure misclassification under three scenarios: absence of a time trend, but expected change in exposure prevalence; linear time trends in exposure misclassification; and logarithmic time trends in exposure misclassification. Validation sampling was conducted at the beginning, middle, and end of follow-up. These validation data were then used to impute positive and negative predictive values for the entire cohort over the study period as a function of time. We compared these imputed values with the true values. RESULTS:We demonstrated that, in the presence of a time trend, purposeful sampling allows for estimation of the changing positive and negative predictive values over the course of the study. Estimation of predictive values was accurate under both the linear and logarithmic scenarios, and was closest to the truth when a large proportion of exposure/outcome strata was sampled for the validation substudy. CONCLUSIONS:When a time trend in classification parameters exists, designs that allow estimation of time-varying predictive values should be used instead of conventional validation study designs that estimate a single summary classification parameter over the study period.

Supplemental Table S8 presents sensitivity, specificity, positive predictive value, and negative predictive value for various risk modeling approaches, stratified by screening interval.

Abstract Background: Machine learning enables complex risk prediction models, but comparative performance with statistical approaches remains context-dependent. We compared statistical and machine learning models for predicting advanced breast cancer risk. Methods: Using data from 968,178 women (40–74 years) undergoing 2,796,459 annual or 812,126 biennial screening mammograms (2005–2019) in the Breast Cancer Surveillance Consortium, we cross-validated models predicting advanced breast cancer within 12 months (annual) or 24 months (biennial) following screening. Models included conventional logistic regression, regularized regressions [least absolute shrinkage and selection operator (LASSO), elastic net], and machine learning methods (random forests, gradient boosting), considering a modest number of clinical and demographic predictors. Performance was assessed using calibration and area under the receiver operating characteristic curve (AUC). Results: Discrimination was similar across models (AUC 0.677–0.690). Calibration differences were more pronounced. Regularized regressions achieved the most favorable calibration overall and across racial and ethnic groups, with an AUC of 0.689 [95% confidence interval (CI), 0.676–0.701]. Gradient boosting showed a comparable AUC but suboptimal calibration (calibration slope 1.12; 95% CI, 1.04–1.20). Conventional logistic regression had a slightly lower AUC (0.683; 95% CI, 0.671–0.696) and a calibration slope of 0.90 (95% CI, 0.83–0.96). Regression-based approaches were generally well calibrated across racial and ethnic groups (expected-to-observed event ratio 0.96–1.03; calibration intercept −0.03 to 0.04), with some subgroup deviations in calibration slopes (<1). Conclusions: For predicting advanced breast cancers, regularized regression demonstrated similar discrimination and generally more favorable calibration than other approaches. Impact: In settings with rare outcomes and low-dimensional features, regularized regression may offer a practical balance between performance and interpretability.

Supplemental Table S9 shows the final estimated coefficients (on log-odds scale) for unstratified LASSO and Elastic net approaches, averaged across 5 imputed datasets.

Supplemental Table S4 summarizes results from sensitivity analysis for impact of different missing data handling ways on performance metrics.

Supplemental Figure S3 presents screening interval-stratified model calibration for advanced cancer by three metrics: the ratio of expected and observed events (E/O ratio), the calibration intercept, and the calibration slope.

Supplemental Figure S2 presents menopausal status-stratified model calibration for advanced cancer by three metrics: the ratio of expected and observed events (E/O ratio), the calibration intercept, and the calibration slope.

Supplemental Figure S4 summarizes frequency of predictor inclusion in unstratified LASSO and Elastic net approaches across imputed datasets.

Breast cancer is the most frequently diagnosed malignancy in women. The Charlson Comorbidity Index (CCI) is a widely used tool for quantifying multimorbidity. As the population ages, breast cancer incidence and comorbidity prevalence are expected to grow. We therefore investigated the absolute and relative impact of the CCI conditions on 1-year mortality in Danish patients with breast cancer. In this population-based cohort study, we identified women with breast cancer from 1987 to 2022 using the Danish Cancer Registry. We estimated 1-year mortality using the Kaplan–Meier method and adjusted hazard ratios (aHRs) using Cox proportional hazards regression. Analyses were stratified by calendar year, cancer stage, and age at diagnosis. Among 146,397 women with breast cancer (median age: 63.4 years; localized stage: 51

Supplementary Table S1: Meta-analysis results stratified by outcome and DNA source.

Supplementary Table S2: Search Terms used in systematic review

10620 Background: Lynch Syndrome (LS) is a hereditary condition that increases risk for colorectal and other primary cancers. LS arises from pathogenic variants (PVs) in MLH1 , MSH2 , MSH6 , and PMS2 . Gene-specific prevention and surveillance strategies exist, and screening colonoscopy reduces overall mortality. Several barriers to screening colonoscopy are known, however the impact of rurality is not well characterized. Methods: We enumerated a cohort of LS patients residing in Vermont or upstate New York who were seen by the Cancer Genetics Program at the University of Vermont, and for whom regular screening colonoscopies were recommended based on current PV- and age-based NCCN guidelines. We reviewed electronic medical records and abstracted patient characteristics and colonoscopy procedures performed between 2021-2024, capturing most recent practices while avoiding the impact of COVID-19 restrictions. We defined screening compliance as having >1 colonoscopy in the 3-year period, concordant with the minimum expected number of procedures over this time period for all PV groups. We assigned rurality status (metropolitan/micropolitan vs. small town/rural) based on residential ZIP code using Rural-Urban Commuting Area codes. We fit log-binomial and proportional odds regression models to estimate the impact of rurality and recency of a genetics focused clinic visits on colonoscopy adherence and on the number of colonoscopies received, adjusting for age, sex, and PVs. Results: We enrolled 201 LS patients for whom annual, bi- or triennial colonoscopies were recommended. Median age at baseline was 60 years (range: 28-98), 131 (65%) were female, and 58 (29%) resided in a small town/rural setting. Compared with metropolitan/micropolitan, small town/rural residence was associated with a lower probability of having ≥1 screening colonoscopy in the 3-year follow-up period (43% vs. 64%; RR=0.67, 95% CI: 0.49, 0.92). This association did not change substantially upon adjustment for age, sex, and pathogenic variants. Small town/rural residence was also associated with undergoing fewer colonoscopies in the 3-year period (cumulative OR=0.46, 95% CI: 0.25, 0.84). Furthermore, recency of Cancer Genetics Program clinic visit was associated with a higher probability of receiving at least one colonoscopy, independent of rurality ( e.g. , RR for last visit ≤3 years ago, compared with last visit >10 years ago = 1.8, 95% CI: 1.1, 2.8). Conclusions: In a cohort of patients with LS, residing in a rural area was associated with a reduced probability of compliance with screening colonoscopy. Resources should be invested in studies aimed at understanding and ameliorating the mechanisms underlying this association. Shorter time since last clinic visit in the genetics program was associated with a higher likelihood of having a screening colonoscopy, suggesting the importance of genetics longitudinal follow-up for hereditary cancer patients.

Importance:Emerging evidence suggests that cholesterol plays a role in breast cancer (BC) metabolism, raising the possibility that cholesterol-lowering medications, such as statins, may improve BC prognosis. Objective:To evaluate the association between postdiagnosis statin initiation and BC mortality using an emulated target trial approach. Design, Setting, and Participants:This observational cohort study using a target trial framework included 96 924 women diagnosed with stage I to III BC from 2000 to 2021 from Nationwide Danish registries, including the Danish Breast Cancer Group's clinical database and the Danish National Prescription Registry. Women with prior invasive BC or prediagnosis use of cholesterol-lowering medication were excluded. Eligible patients were duplicated into a cloned cohort and assigned to 1 of 2 treatment strategies: statin initiation within 36 months postdiagnosis or no statin initiation. Patients were followed up until deviation from the assigned strategy, emigration, death, 10 years of follow-up, or October 5, 2022. Exposures:Initiation of any statin within 36 months after BC diagnosis. Main Outcomes and Measures:The primary outcome was BC mortality. Secondary outcomes included all-cause mortality. Hazard ratios (HRs) and 95% CIs were estimated using inverse probability of censoring weighted (IPCW) Cox regression models with a robust variance estimator. Results:A total of 66 952 patients with BC were enrolled in the emulated target trial (17 152 [27.6%] were aged 50-59 years); 4851 (7.2%) initiated statins within 36 months after diagnosis. Over 606 266 person-years of follow-up, 9130 patients died from BC and 19 679 from any cause. The 10-year risk of breast cancer mortality was 11.8% among statin initiators and 13.5% among noninitiators, corresponding to a risk difference of 1.7% (95% CI, 0.5% to 3.0%). A similar difference was observed for all-cause mortality (23.3% vs 24.5%; risk difference, 1.2%; 95% CI, -0.1% to 2.5%). IPCW analysis yielded an HR of 0.90 (95% CI, 0.85 to 0.95) for BC mortality and 0.92 (95% CI, 0.85 to 1.00) for all-cause mortality among statin initiators vs noninitiators. Conclusions and Relevance:In this cohort study, among 66 952 women with early BC, postdiagnosis statin initiation was associated with a modest reduction in BC and all-cause mortality. These findings support further investigation into the potential role of statins as an adjunct to standard adjuvant BC treatment.

The anti-cancer potential of low-dose aspirin in long-term breast cancer (BC) survivors remain unknown. We evaluated the association between low-dose aspirin use and BC recurrence and mortality. Women ≥40 years diagnosed with stage I-III BC (1996–2004) were identified from the Danish Breast Cancer Group (DBCG) database and information on aspirin use from the Danish Prescription Registry. We ascertained recurrences from DBCG and via a validated algorithm. We plotted cumulative incidences of recurrence and mortality, accounting for competing risks. Using Cox regression, we estimated hazard ratios (HRs) and 95% confidence intervals (CI), employing landmark analyses at 5-, 10-, and 15-year post-diagnosis. Among 20,509 BC survivors, 4527 developed recurrence over 232,441 person-years of follow-up. The 20-year cumulative incidence of recurrence was lower in users (17.8%) than nonusers (22.4%), with similar trends among 10-year disease-free survivors (9.9% vs. 12.7%). We observed reduced HRs of recurrence (adjusted HR5-year = 0.80, (95% CI = 0.66-0.98); HR10-year = 0.87 (0.73–1.05); HR15-year = 0.82 (0.57–1.17) in aspirin users, but increased HRs of all-cause mortality (HR5-year = 1.08 (0.96–1.21); HR10-year = 1.09 (0.96–1.24); HR15-year = 1.09 (0.80–1.31). The reduced recurrence risk in aspirin users may indicate potential anti-cancer effects of aspirin, though the increased risk of death suggests influence by confounding by indication and competing risks.

Background: Breast cancer is the most commonly diagnosed malignancy among women worldwide. Despite high survival rates, 20%-40% of women will experience a recurrence, with risks extending beyond 20 years after diagnosis. Premenopausal women diagnosed with estrogen receptor (ER) positive disease are prescribed 5-10 years of tamoxifen therapy to prevent a recurrence. 17β-hydroxysteroid dehydrogenase 1 and 2 expression (HSD17B1 and HSD17B2, respectively) regulate the relative concentrations of estrogen metabolites and may modify tamoxifen effectiveness. We evaluated the prognostic and predictive value of these biomarkers. Methods: Premenopausal women diagnosed during 2002-2011 with a first primary stage I-III breast cancer were identified in the Danish Breast Cancer Group database, and categorized based on ER status and receipt of tamoxifen (4600 ER+/TAM+ and 1359 ER−/TAM−). HSD17B1 and HSD17B2 were assayed by immunohistochemistry and scored using automated image analysis [Visiopharm (Hoersholm, Denmark)]. The biomarkers were assessed within ER/TAM strata, to differentiate the enzymes predictive of treatment response from enzymes prognostic for breast cancer recurrence. We used Cox proportional hazards regression and probabilistic bias analysis to account for mismeasurement of biomarker expression and baseline selection bias from tumor sample availability to calculate the hazard ratios (HRs) and 95% simulation intervals (SIs) associating each biomarker with recurrence. Results: 24% of ER+/TAM+ and 13% of ER−/TAM− breast cancers had any HSD17B1 expression. In the bias-adjusted analyses, women diagnosed with tumors positive for HSD17B1 expression had an increased rate of recurrence: HR=1.40 (95% SI: 1.02, 2.92) in the ER+/TAM+ stratum, and HR=1.32 (95% SI: 0.82, 2.74) in the ER−/TAM− stratum. A 10-unit increase in HSD17B2 expression corresponded with a decreased recurrence rate among women in the ER+/TAM+ (HR=0.85, 95% SI: 0.69, 1.05) stratum, but not among women in the ER−/TAM− stratum (HR=1.07, 95% SI: 0.82, 1.42). Conclusions: We observed that HSD17B1 expression was associated with a higher rate of recurrence among premenopausal women, and that HSD17B2 expression was associated with a lower rate of recurrence among premenopausal tamoxifen-treated women diagnosed with ER+ disease. HSD17B1 may be an important prognostic marker of breast cancer recurrence among premenopausal breast cancer patients and HSD17B2 may be predictive of response among tamoxifen-treated premenopausal women diagnosed with ER+ breast cancer. Citation Format: Lindsay Jane Collin, Anders Kjaersgaard, Thomas P. Ahern, Michael Goodman, Lauren E. McCullough, Lance A. Waller, Kristina B. Christensen, Per Damkier, Stephen J. Hamilton-Dutoit, Kristina Lauridsen, Peer M. Chistiansen, Bent Ejlertsen, Henrik Toft Sorensen, Deirdre Cronin-Fenton, Timothy L. Lash. 17β-hydroxysteroid dehydrogenases 1 and 2: potential markers for breast cancer recurrence and tamoxifen resistance among premenopausal women diagnosed with breast cancer in Denmark [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 757.

Supplementary Figure S6: Inverse rank percentile plot.

Abstract Background: We evaluated the impact of systematic bias due to loss of heterozygosity (LOH) and incomplete phenotyping in studies examining the relationship between CYP2D6 variants and breast cancer recurrence among women treated with tamoxifen. Methods: We performed a systematic review of the literature on tamoxifen, CYP2D6 variants, and breast cancer recurrence. A quantitative bias analysis was performed to adjust for LOH and incomplete phenotyping. Bias-adjusted results were then combined in a meta-analysis. Results: Thirty-three studies informed the bias analysis and meta-analysis on CYP2D6 variants and breast cancer recurrence and/or mortality. An unadjusted meta-analysis suggested increased risk of recurrence and/or mortality for poor relative to normal metabolizers [RR = 1.28; 95% simulation interval (SI), 1.04–1.58] with substantial heterogeneity (I2 = 27%; P for heterogeneity = 0.07). Adjusting for LOH and incomplete genotyping resulted in a slight change in the effect estimate and a decrease in heterogeneity (RR = 1.34; 95% SI, 1.10–1.63; I2 = 0%; P for heterogeneity = 0.17). Intermediate metabolizers had a slightly increased risk of recurrence and/or mortality relative to normal metabolizers (RR = 1.15; 95% SI, 1.00–1.34; I2 = 0%; P for heterogeneity = 0.89). Conclusions: Adjusting for biases such as LOH and incomplete genotyping reduced observed heterogeneity between studies. Individuals with poor CYP2D6 phenotypes were at increased risk for breast cancer outcomes compared with those with normal phenotypes. Impact: Reduction in CYP2D6 activity was associated with an increased risk of breast cancer recurrence and/or mortality, and results underscore the importance of quantitatively adjusting for biases when aggregating study results. See related In the Spotlight, p. 221