Search

Background: Machine learning (ML) approaches facilitate risk prediction model development using high-dimensional predictors and higher-order interactions at the cost of model interpretability and transparency. We compared the relative predictive performance of statistical and ML models to guide modeling strategy selection for surveillance mammography outcomes in women with a personal history of breast cancer (PHBC). Methods: We cross-validated seven risk prediction models for two surveillance outcomes, failure (breast cancer within 12 months of a negative surveillance mammogram) and benefit (surveillance-detected breast cancer). We included 9,447 mammograms (495 failures, 1,414 benefits, and 7,538 nonevents) from years 1996 to 2017 using a 1:4 matched case-control samples of women with PHBC in the Breast Cancer Surveillance Consortium. We assessed model performance of conventional regression, regularized regressions (LASSO and elastic-net), and ML methods (random forests and gradient boosting machines) by evaluating their calibration and, among well-calibrated models, comparing the area under the receiver operating characteristic curve (AUC) and 95% confidence intervals (CI). Results: LASSO and elastic-net consistently provided well-calibrated predicted risks for surveillance failure and benefit. The AUCs of LASSO and elastic-net were both 0.63 (95% CI, 0.60-0.66) for surveillance failure and 0.66 (95% CI, 0.64-0.68) for surveillance benefit, the highest among well-calibrated models. Conclusions: For predicting breast cancer surveillance mammography outcomes, regularized regression outperformed other modeling approaches and balanced the trade-off between model flexibility and interpretability. Impact: Regularized regression may be preferred for developing risk prediction models in other contexts with rare outcomes, similar training sample sizes, and low-dimensional features. [ABSTRACT FROM AUTHOR]

Background: Previous reports suggested risk of death and breast cancer varied by comorbidity and age in older women undergoing mammography. However, impacts of functional limitations remain unclear. Methods: We used data from 238,849 women in the Breast Cancer Surveillance Consortium-Medicare linked database (1999-2015) who had screening mammogram at ages 66-94 years. We estimated risk of breast cancer, breast cancer death, and non-breast cancer death by function-related indicator (FRI) which incorporated 16 claims-based items and was categorized as an ordinal variable (0, 1, and 2+). Fine and Gray proportional sub-distribution hazards models were applied with breast cancer and death treated as competing events. Risk estimates by FRI scores were adjusted by age and NCI comorbidity index separately and stratified by these factors. Results: Overall, 9,252 women were diagnosed with breast cancer, 406 died of breast cancer, and 41,640 died from non-breast cancer causes. The 10-year age-adjusted invasive breast cancer risk slightly decreased with FRI score [FRI = 0: 4.0%, 95% confidence interval (CI) = 3.8-4.1; FRI = 1: 3.9%, 95% CI = 3.7-4.2; FRI = 2: 3.5%, 95% CI = 3.1-3.9). Risk of non-breast cancer death increased with FRI score (FRI = 0: 18.8%, 95% CI = 18.5-19.1; FRI = 1: 24.4%, 95% CI = 23.9-25.0; FRI = 2: 39.8%, 95% CI = 38.8-40.9]. Risk of breast cancer death was low with minimal differences across FRI scores. NCI comorbidity index-adjusted models and stratified analyses yielded similar patterns. Conclusions: Risk of non-breast cancer death substantially increases with FRI score, whereas risk of breast cancer death is low regardless of functional status. [ABSTRACT FROM AUTHOR]

Background: To reduce colorectal cancer incidence and mortality, experts recommend surveillance colonoscopy 3 years after advanced adenoma removal. Little is known about adherence to that interval. Methods: We describe patterns of and factors associated with subsequent colonoscopy among persons with =3 adenomas and/or =1 adenoma with villous/tubulovillous histology in four U.S. integrated healthcare delivery systems. We report Kaplan-Meier estimators of the cumulative percentage of patients undergoing colonoscopy 6 months to 3.5 years after an index colonoscopy with high-risk findings. Combining data from three healthcare systems, we used multivariable logistic regression with inverse probability of censoring weights to estimate ORs and 95% confidence intervals (CI) for associations between patient characteristics and receipt of subsequent colonoscopy. Results: Among 6,909 persons with advanced adenomas, the percent receiving a subsequent colonoscopy 6 months to 3.5 years later ranged from 18.3% (95% CI: 11.7%-27.8%) to 59.5% (95% CI: 53.8%-65.2%) across healthcare systems. Differences remained significant in the multivariable model. Patients with =3 adenomas were more likely than those with 1 to 2 villous/tubulovillous adenomas to undergo subsequent colonoscopy. Subsequent colonoscopy was also more common for patients ages 60-74 and less common for patients ages 80 to 89 compared with those ages 50 to 54 years at their index colonoscopy. Sex, race/ethnicity, and comorbidity index score were generally not associated with subsequent colonoscopy receipt. Conclusions: Colonoscopy within the recommended interval following advanced adenoma was underutilized and varied by healthcare system, age, and number of adenomas. Impact: Strategies to improve adherence to surveillance colonoscopy following advanced adenomas are needed. [ABSTRACT FROM AUTHOR]

The article discusses a study describing patterns of colorectal cancer screening uptake in a U.S. insured population. Topics discussed include assessment of temporal trends and patient characteristics with screening uptake, increase in proportion of individuals initiating colorectal cancer screening via colonoscopy in 2008 to 2010, and identification of risk factors for colorectal cancer.

The author reflects on development of the National Cancer Institute's (NCI) Population-Based Research Optimizing Screening through Personalized Regimens (PROSPR) consortium for evaluation of cancer screening process. Topics discussed include creation of central data repository to facilitate research evaluating the colorectal cancer screening, description of research centers focused on colorectal cancer screening, and Institute of Medicine's goals with respect to evidence-based cancer screening.

The article presents abstracts of three case studies focused on international research collaborations including the study of Dr. Carl M. Kitahara on early growth and thyroid cancer incidence in Denmark, the study of Dr. Sara Tamers on smoking tobacco and its implications on stressful life events (SLE) in France and the study of Dr. Mohammed H. Al-Temimi on the survival of head and neck cancer patients in Eastern and Central Europe.

The article presents a study on the impact of obesity on follow-up after an abnormal screening mammogram. It examined the association between body mass index (BMI) and timeliness of follow-up after 241,222 abnormal screening mammograms carried out on 201,470 women in the Breast Cancer Surveillance Consortium. It finds that actual differences in days to follow-up across BMI groups were small and unlikely to be clinically significant.

The article presents a report on opportunities for cancer epidemiology and control in low and middle-income countries. According to the authors, the key factors in cancer being a leading cause of death globally are the increasing numbers of older people in developing countries and the increasing prevalence of risk factors such as smoking, physical inactivity, and obesity. In response to the cancer epidemic, the International Cancer Prevention Interest Group was established in 2007.

Loss of estrogen receptor β (ERβ) expression in the gut is associated with colorectal cancer (CRC) initiation and progression. Germline single-nucleotide polymorphisms (SNP) in genes for the sex-steroid hormone receptors are not strongly associated with CRC risk; however, these SNPs have not previously been evaluated in relation to survival after diagnosis. We enrolled 729 women, ages 50 to 74, diagnosed with invasive CRC between 1997 and 2002 in 13 counties covered by the Seattle-Puget Sound Surveillance Epidemiology and End Results cancer registry. Participants provided germline DNA. We selected 99 tag-SNPs for the androgen receptor (AR), ERα (ESR1), ERβ (ESR2), and progesterone receptor (PGR) genes. Mortality outcomes were ascertained from the National Death Index. During a median of 6.6 years of follow-up, 244 deaths occurred (161 from CRC). We identified 20 SNPs (12 of ESR2 and 8 of PGR) for replication in 1,729 women diagnosed with incident invasive CRC (555 deaths; 405 from CRC) from three prospective cohort studies that participate in the Genetics and Epidemiology of Colorectal Cancer Consortium. Three correlated SNPs in the promoter of ESR2 (rs2987983, rs3020443, and rs2978381) were statistically significant predictors of CRC-specific and overall survival. Minor alleles of each were associated with improved survival [for rs2987983, CRC-specific HR, 0.77; 95% confidence interval (CI), 0.60-0.99 in the initial study, and HR, 0.79; CI, 0.64-0.98 in replication]. No associations were noted for SNPs of AR, ESR1, or PGR. SNPs in the promoter of may be important to pathways related to the association between ERβ and tumor progression and ESR2 metastasis. [ABSTRACT FROM AUTHOR]

Background: Machine learning (ML) approaches facilitate risk prediction model development using high-dimensional predictors and higher-order interactions at the cost of model interpretability and transparency. We compared the relative predictive performance of statistical and ML models to guide modeling strategy selection for surveillance mammography outcomes in women with a personal history of breast cancer (PHBC)., Methods: We cross-validated seven risk prediction models for two surveillance outcomes, failure (breast cancer within 12 months of a negative surveillance mammogram) and benefit (surveillance-detected breast cancer). We included 9,447 mammograms (495 failures, 1,414 benefits, and 7,538 nonevents) from years 1996 to 2017 using a 1:4 matched case-control samples of women with PHBC in the Breast Cancer Surveillance Consortium. We assessed model performance of conventional regression, regularized regressions (LASSO and elastic-net), and ML methods (random forests and gradient boosting machines) by evaluating their calibration and, among well-calibrated models, comparing the area under the receiver operating characteristic curve (AUC) and 95% confidence intervals (CI)., Results: LASSO and elastic-net consistently provided well-calibrated predicted risks for surveillance failure and benefit. The AUCs of LASSO and elastic-net were both 0.63 (95% CI, 0.60-0.66) for surveillance failure and 0.66 (95% CI, 0.64-0.68) for surveillance benefit, the highest among well-calibrated models., Conclusions: For predicting breast cancer surveillance mammography outcomes, regularized regression outperformed other modeling approaches and balanced the trade-off between model flexibility and interpretability., Impact: Regularized regression may be preferred for developing risk prediction models in other contexts with rare outcomes, similar training sample sizes, and low-dimensional features., (©2023 American Association for Cancer Research.)

Background: Previous reports suggested risk of death and breast cancer varied by comorbidity and age in older women undergoing mammography. However, impacts of functional limitations remain unclear., Methods: We used data from 238,849 women in the Breast Cancer Surveillance Consortium-Medicare linked database (1999-2015) who had screening mammogram at ages 66-94 years. We estimated risk of breast cancer, breast cancer death, and non-breast cancer death by function-related indicator (FRI) which incorporated 16 claims-based items and was categorized as an ordinal variable (0, 1, and 2+). Fine and Gray proportional sub-distribution hazards models were applied with breast cancer and death treated as competing events. Risk estimates by FRI scores were adjusted by age and NCI comorbidity index separately and stratified by these factors., Results: Overall, 9,252 women were diagnosed with breast cancer, 406 died of breast cancer, and 41,640 died from non-breast cancer causes. The 10-year age-adjusted invasive breast cancer risk slightly decreased with FRI score [FRI = 0: 4.0%, 95% confidence interval (CI) = 3.8-4.1; FRI = 1: 3.9%, 95% CI = 3.7-4.2; FRI ≥ 2: 3.5%, 95% CI = 3.1-3.9). Risk of non-breast cancer death increased with FRI score (FRI = 0: 18.8%, 95% CI = 18.5-19.1; FRI = 1: 24.4%, 95% CI = 23.9-25.0; FRI ≥ 2: 39.8%, 95% CI = 38.8-40.9]. Risk of breast cancer death was low with minimal differences across FRI scores. NCI comorbidity index-adjusted models and stratified analyses yielded similar patterns., Conclusions: Risk of non-breast cancer death substantially increases with FRI score, whereas risk of breast cancer death is low regardless of functional status., Impact: Older women with functional limitations should be informed that they may not benefit from screening mammography., (©2021 American Association for Cancer Research.)