Phenotype Discovery and Geographic Disparities of Late-Stage Breast Cancer Diagnosis across U.S. Counties: A Machine Learning Approach

Biological Basis of Breast Cancer-Related Disparities in Precision Oncology Era

Precision oncology is based on deep knowledge of the molecular profile of tumors, allowing for more accurate and personalized therapy for specific groups of patients who are different in disease susceptibility as well as treatment response. Thus, onco-breastomics is able to discover novel biomarkers that have been found to have racial and ethnic differences, among other types of disparities such as chronological or biological age-, sex/gender- or environmental-related ones. Usually, evidence suggests that breast cancer (BC) disparities are due to ethnicity, aging rate, socioeconomic position, environmental or chemical exposures, psycho-social stressors, comorbidities, Western lifestyle, poverty and rurality, or organizational and health care system factors or access. The aim of this review was to deepen the understanding of BC-related disparities, mainly from a biomedical perspective, which includes genomic-based differences, disparities in breast tumor biology and developmental biology, differences in breast tumors' immune and metabolic landscapes, ecological factors involved in these disparities as well as microbiomics- and metagenomics-based disparities in BC. We can conclude that onco-breastomics, in principle, based on genomics, proteomics, epigenomics, hormonomics, metabolomics and exposomics data, is able to characterize the multiple biological processes and molecular pathways involved in BC disparities, clarifying the differences in incidence, mortality and treatment response for different groups of BC patients.

County-Level Socio-Environmental Factors Associated With Stroke Mortality in the United States: A Cross-Sectional Study

We used machine learning methods to explore sociodemographic and environmental determinants of health (SEDH) associated with county-level stroke mortality in the USA. We conducted a cross-sectional analysis of individuals aged ≥15 years who died from all stroke subtypes between 2016 and 2020. We analyzed 54 county-level SEDH possibly associated with age-adjusted stroke mortality rates/100,000 people. Classification and Regression Tree (CART) was used to identify specific county-level clusters associated with stroke mortality. Variable importance was assessed using Random Forest analysis. A total of 501,391 decedents from 2397 counties were included. CART identified 10 clusters, with 77.5% relative increase in stroke mortality rates across the spectrum (28.5 vs 50.7 per 100,000 persons). CART identified 8 SEDH to guide the classification of the county clusters. Including, annual Median Household Income ($), live births with Low Birthweight (%), current adult Smokers (%), adults reporting Severe Housing Problems (%), adequate Access to Exercise (%), adults reporting Physical Inactivity (%), adults with diagnosed Diabetes (%), and adults reporting Excessive Drinking (%). In conclusion, SEDH exposures have a complex relationship with stroke. Machine learning approaches can help deconstruct this relationship and demonstrate associations that allow improved understanding of the socio-environmental drivers of stroke and development of targeted interventions.

Alzheimer`s disease mortality in the United States: Cross-sectional analysis of county-level socio-environmental factors

BACKGROUND

Geographical disparities in mortality among Alzheimer`s disease (AD) patients have been reported and complex sociodemographic and environmental determinants of health (SEDH) may be contributing to this variation. Therefore, we aimed to explore high-risk SEDH factors possibly associated with all-cause mortality in AD across US counties using machine learning (ML) methods.

METHODS

We performed a cross-sectional analysis of individuals ≥65 years with any underlying cause of death but with AD in the multiple causes of death certificate (ICD-10,G30) between 2016 and 2020. Outcomes were defined as age-adjusted all-cause mortality rates (per 100,000 people). We analyzed 50 county-level SEDH and Classification and Regression Trees (CART) was used to identify specific county-level clusters. Random Forest, another ML technique, evaluated variable importance. CART`s performance was validated using a "hold-out" set of counties.

RESULTS

Overall, 714,568 individuals with AD died due to any cause across 2,409 counties during 2016-2020. CART identified 9 county clusters associated with an 80.1% relative increase of mortality across the spectrum. Furthermore, 7 SEDH variables were identified by CART to drive the categorization of clusters, including High School Completion (%), annual Particulate Matter 2.5 Level in Air, live births with Low Birthweight (%), Population under 18 years (%), annual Median Household Income in US dollars ($), population with Food Insecurity (%), and houses with Severe Housing Cost Burden (%).

CONCLUSION

ML can aid in the assimilation of intricate SEDH exposures associated with mortality among older population with AD, providing opportunities for optimized interventions and resource allocation to reduce mortality among this population.

Understanding Disparities in Receipt of Complex Gastrointestinal Cancer Surgery at a Small Geographic Scale

OBJECTIVE

To define neighborhood-level disparities in the receipt of complex cancer surgery.

BACKGROUND

Little is known about the geographic variation of receipt of surgery among patients with complex gastrointestinal (GI) cancers, especially at a small geographic scale.

METHODS

This study included individuals diagnosed with 5 invasive, nonmetastatic, complex GI cancers (esophagus, stomach, pancreas, bile ducts, liver) from the Ohio Cancer Incidence Surveillance System during 2009 and 2018. To preserve patient privacy, we combined US census tracts into the smallest geographic areas that included a minimum number of surgery cases (n=11) using the Max-p-regions method and called these new areas "MaxTracts." Age-adjusted surgery rates were calculated for MaxTracts, and the Hot Spot analysis identified clusters of high and low surgery rates. US Census and CDC PLACES were used to compare neighborhood characteristics between the high- and low-surgery clusters.

RESULTS

This study included 33,091 individuals with complex GI cancers located in 1006 MaxTracts throughout Ohio. The proportion in each MaxTract receiving surgery ranged from 20.7% to 92.3% with a median (interquartile range) of 48.9% (42.4-56.3). Low-surgery clusters were mostly in urban cores and the Appalachian region, whereas high-surgery clusters were mostly in suburbs. Low-surgery clusters differed from high-surgery clusters in several ways, including higher rates of poverty (23% vs. 12%), fewer married households (40% vs. 50%), and more tobacco use (25% vs. 19%; all P <0.01).

CONCLUSIONS

This improved understanding of neighborhood-level variation in receipt of potentially curative surgery will guide future outreach and community-based interventions to reduce treatment disparities. Similar methods can be used to target other treatment phases and other cancers.

Geographical Patterns and Risk Factor Association of Cardio-Oncology Mortality in the United States

Cardio-oncology mortality (COM) is a complex issue that is compounded by multiple factors that transcend a depth of socioeconomic, demographic, and environmental exposures. Although metrics and indexes of vulnerability have been associated with COM, advanced methods are required to account for the intricate intertwining of associations. This cross-sectional study utilized a novel approach that combined machine learning and epidemiology to identify high-risk sociodemographic and environmental factors linked to COM in United States counties. The study consisted of 987,009 decedents from 2,717 counties, and the Classification and Regression Trees model identified 9 county socio-environmental clusters that were closely associated with COM, with a 64.1% relative increase across the spectrum. The most important variables that emerged from this study were teen birth, pre-1960 housing (lead paint indicator), area deprivation index, median household income, number of hospitals, and exposure to particulate matter air pollution. In conclusion, this study provides novel insights into the socio-environmental drivers of COM and highlights the importance of utilizing machine learning approaches to identify high-risk populations and inform targeted interventions for reducing disparities in COM.

Understanding Factors Leading to Surgical Attrition for "Resectable" Gastric Cancer

OBJECTIVES

We used a novel combined analysis to evaluate various factors associated with failure to surgical resection in non-metastatic gastric cancer.

METHODS

We identified factors associated with the receipt of surgery in publicly available clinical trial data for gastric cancer and in the National Cancer Database (NCDB) for patients with stages I-III gastric adenocarcinoma. Next, we evaluated variable importance in predicting the receipt of surgery in the NCDB.

RESULTS

In published clinical trial data, 10% of patients in surgery-first arms did not undergo surgery, mostly due to disease progression and 15% of patients in neoadjuvant therapy arms failed to reach surgery. Effects related to neoadjuvant administration explained the increased attrition (5%). In the NCDB, 61.7% of patients underwent definitive surgery. In a subset of NCDB patients resembling those enrolled in clinical trials (younger, healthier, and privately insured patients treated at high-volume and academic centers) the rate of surgery was 79.2%. Decreased likelihood of surgery was associated with advanced age (OR 0.97, p < 0.01), Charlson-Deyo score of 2+ (OR 0.90, p < 0.01), T4 tumors (OR 0.39, p < 0.01), N+ disease (OR 0.84, p < 0.01), low socioeconomic status (OR 0.86, p = 0.01), uninsured or on Medicaid (OR 0.58 and 0.69, respectively, p < 0.01), low facility volume (OR 0.64, p < 0.01), and non-academic cancer programs (OR 0.79, p < 0.01).

CONCLUSION

Review of clinical trials shows attrition due to unavoidable tumor and treatment factors (~ 15%). The NCDB indicates non-medical patient and provider characteristics (i.e., age, insurance status, facility volume) associated with attrition. This combined analysis highlights specific opportunities for improving potentially curative surgery rates.

Prioritizing Screening Mammograms for Immediate Interpretation and Diagnostic Evaluation on the Basis of Risk for Recall

PURPOSE

The aim of this study was to develop a prioritization strategy for scheduling immediate screening mammographic interpretation and possible diagnostic evaluation.

METHODS

A population-based cohort with screening mammograms performed from 2012 to 2020 at 126 radiology facilities from 7 Breast Cancer Surveillance Consortium registries was identified. Classification trees identified combinations of clinical history (age, BI-RADS® density, time since prior mammogram, history of false-positive recall or biopsy result), screening modality (digital mammography, digital breast tomosynthesis), and facility characteristics (profit status, location, screening volume, practice type, academic affiliation) that grouped screening mammograms by recall rate, with ≥12/100 considered high and ≥16/100 very high. An efficiency ratio was estimated as the percentage of recalls divided by the percentage of mammograms.

RESULTS

The study cohort included 2,674,051 screening mammograms in 925,777 women, with 235,569 recalls. The most important predictor of recall was time since prior mammogram, followed by age, history of false-positive recall, breast density, history of benign biopsy, and screening modality. Recall rates were very high for baseline mammograms (21.3/100; 95% confidence interval, 19.7-23.0) and high for women with ≥5 years since prior mammogram (15.1/100; 95% confidence interval, 14.3-16.1). The 9.2% of mammograms in subgroups with very high and high recall rates accounted for 19.2% of recalls, an efficiency ratio of 2.1 compared with a random approach. Adding women <50 years of age with dense breasts accounted for 20.3% of mammograms and 33.9% of recalls (efficiency ratio = 1.7). Results including facility-level characteristics were similar.

CONCLUSIONS

Prioritizing women with baseline mammograms or ≥5 years since prior mammogram for immediate interpretation and possible diagnostic evaluation could considerably reduce the number of women needing to return for diagnostic imaging at another visit.

Risk factors and geographic disparities in premature cardiovascular mortality in US counties: a machine learning approach

Disparities in premature cardiovascular mortality (PCVM) have been associated with socioeconomic, behavioral, and environmental risk factors. Understanding the "phenotypes", or combinations of characteristics associated with the highest risk of PCVM, and the geographic distributions of these phenotypes is critical to targeting PCVM interventions. This study applied the classification and regression tree (CART) to identify county phenotypes of PCVM and geographic information systems to examine the distributions of identified phenotypes. Random forest analysis was applied to evaluate the relative importance of risk factors associated with PCVM. The CART analysis identified seven county phenotypes of PCVM, where high-risk phenotypes were characterized by having greater percentages of people with lower income, higher physical inactivity, and higher food insecurity. These high-risk phenotypes were mostly concentrated in the Black Belt of the American South and the Appalachian region. The random forest analysis identified additional important risk factors associated with PCVM, including broadband access, smoking, receipt of Supplemental Nutrition Assistance Program benefits, and educational attainment. Our study demonstrates the use of machine learning approaches in characterizing community-level phenotypes of PCVM. Interventions to reduce PCVM should be tailored according to these phenotypes in corresponding geographic areas.

Geographic Variation and Risk Factor Association of Early Versus Late Onset Colorectal Cancer

The proportion of patients diagnosed with colorectal cancer (CRC) at age < 50 (early-onset CRC, or EOCRC) has steadily increased over the past three decades relative to the proportion of patients diagnosed at age ≥ 50 (late-onset CRC, or LOCRC), despite the reduction in CRC incidence overall. An important gap in the literature is whether EOCRC shares the same community-level risk factors as LOCRC. Thus, we sought to (1) identify disparities in the incidence rates of EOCRC and LOCRC using geospatial analysis and (2) compare the importance of community-level risk factors (racial/ethnic, health status, behavioral, clinical care, physical environmental, and socioeconomic status risk factors) in the prediction of EOCRC and LOCRC incidence rates using a random forest machine learning approach. The incidence data came from the Surveillance, Epidemiology, and End Results program (years 2000-2019). The geospatial analysis revealed large geographic variations in EOCRC and LOCRC incidence rates. For example, some regions had relatively low LOCRC and high EOCRC rates (e.g., Georgia and eastern Texas) while others had relatively high LOCRC and low EOCRC rates (e.g., Iowa and New Jersey). The random forest analysis revealed that the importance of community-level risk factors most predictive of EOCRC versus LOCRC incidence rates differed meaningfully. For example, diabetes prevalence was the most important risk factor in predicting EOCRC incidence rate, but it was a less important risk factor of LOCRC incidence rate; physical inactivity was the most important risk factor in predicting LOCRC incidence rate, but it was the fourth most important predictor for EOCRC incidence rate. Thus, our community-level analysis demonstrates the geographic variation in EOCRC burden and the distinctive set of risk factors most predictive of EOCRC.

Variation in and Factors Associated With US County-Level Cancer Mortality, 2008-2019

IMPORTANCE

The association between cancer mortality and risk factors may vary by geography. However, conventional methodological approaches rarely account for this variation.

OBJECTIVE

To identify geographic variations in the association between risk factors and cancer mortality.

DESIGN, SETTING, AND PARTICIPANTS

This geospatial cross-sectional study used county-level data from the National Center for Health Statistics for individuals who died of cancer from 2008 to 2019. Risk factor data were obtained from County Health Rankings & Roadmaps, Health Resources and Services Administration, and Centers for Disease Control and Prevention. Analyses were conducted from October 2021 to July 2022.

MAIN OUTCOMES AND MEASURES

Conventional random forest models were applied nationwide and by US region, and the geographical random forest model (accounting for local variation of association) was applied to assess associations between a wide range of risk factors and cancer mortality.

RESULTS

The study included 7 179 201 individuals (median age, 70-74 years; 3 409 508 women [47.5%]) who died from cancer in 3108 contiguous US counties during 2008 to 2019. The mean (SD) county-level cancer mortality rate was 177.0 (26.4) deaths per 100 000 people. On the basis of the variable importance measure, the random forest models identified multiple risk factors associated with cancer mortality, including smoking, receipt of Supplemental Nutrition Assistance Program (SNAP) benefits, and obesity. The geographical random forest model further identified risk factors that varied at the county level. For example, receipt of SNAP benefits was a high-importance factor in the Appalachian region, North and South Dakota, and Northern California; smoking was of high importance in Kentucky and Tennessee; and female-headed households were high-importance factors in North and South Dakota. Geographic areas with certain high-importance risk factors did not consistently have a corresponding high prevalence of the same risk factors.

CONCLUSIONS AND RELEVANCE

In this cross-sectional study, the associations between cancer mortality and risk factors varied by geography in a way that did not correspond strictly to risk factor prevalence. The degree to which other place-specific characteristics, observed and unobserved, modify risk factor effects should be further explored, and this work suggests that risk factor importance may be a preferable paradigm for selecting cancer control interventions compared with risk factor prevalence.

PSOWNNs-CNN: A Computational Radiology for Breast Cancer Diagnosis Improvement Based on Image Processing Using Machine Learning Methods

Early diagnosis of breast cancer is an important component of breast cancer therapy. A variety of diagnostic platforms can provide valuable information regarding breast cancer patients, including image-based diagnostic techniques. However, breast abnormalities are not always easy to identify. Mammography, ultrasound, and thermography are some of the technologies developed to detect breast cancer. Using image processing and artificial intelligence techniques, the computer enables radiologists to identify chest problems more accurately. The purpose of this article was to review various approaches to detecting breast cancer using artificial intelligence and image processing. The authors present an innovative approach for identifying breast cancer using machine learning methods. Compared to current approaches, such as CNN, our particle swarm optimized wavelet neural network (PSOWNN) method appears to be relatively superior. The use of machine learning methods is clearly beneficial in terms of improved performance, efficiency, and quality of images, which are crucial to the most innovative medical applications. According to a comparison of the process's 905 images to those of other illnesses, 98.6% of the disorders are correctly identified. In summary, PSOWNNs, therefore, have a specificity of 98.8%. Furthermore, PSOWNNs have a precision of 98.6%, which means that, despite the high number of women diagnosed with breast cancer, only 830 (95.2%) are diagnosed. In other words, 95.2% of images are correctly classified. PSOWNNs are more accurate than other machine learning algorithms, SVM, KNN, and CNN.