Explainable artificial intelligence (XAI) for exploring spatial variability of lung and bronchus cancer (LBC) mortality rates in the contiguous USA

Machine learning (ML) has demonstrated promise in predicting mortality; however, understanding spatial variation in risk factor contributions to mortality rate requires explainability. We applied explainable artificial intelligence (XAI) on a stack-ensemble machine learning model framework to explore and visualize the spatial distribution of the contributions of known risk factors to lung and bronchus cancer (LBC) mortality rates in the conterminous United States. We used five base-learners—generalized linear model (GLM), random forest (RF), Gradient boosting machine (GBM), extreme Gradient boosting machine (XGBoost), and Deep Neural Network (DNN) for developing stack-ensemble models. Then we applied several model-agnostic approaches to interpret and visualize the stack ensemble model's output in global and local scales (at the county level). The stack ensemble generally performs better than all the base learners and three spatial regression models. A permutation-based feature importance technique ranked smoking prevalence as the most important predictor, followed by poverty and elevation. However, the impact of these risk factors on LBC mortality rates varies spatially. This is the first study to use ensemble machine learning with explainable algorithms to explore and visualize the spatial heterogeneity of the relationships between LBC mortality and risk factors in the contiguous USA.

Differential Second Primary Malignancy Occurrence After Breast Cancer According to HER2 Status: a SEER Population-Based Analysis

Background: The survival improvement in breast cancer (BC) renders the long-termsurvivorsan increasedprobability of second primary malignancy (SPM), and thus excess mortality. Although previous evidence has indicated various predictorsofSPM, little is known whether SPM incidencevaries by HER2 status of first BC. Methods: Based on BC patients registered between 2010-2018 in the NCI SEER database, we utilized standardized incidence ratio (SIR) and Poisson regression to quantify SPM occurrence compared with the general population. Then, adjusted for competing death risk, cumulative incidence function and Gray’s test were adopted to estimate the probability of SPM. Subsequent proportional subdistribution hazards regression was executedto identify the HER2 status impact on SPM risk. Finally, survival analysiswas performed.Results: A total of 409,796 first BC patients were includedand 18,283 were identified with at least one SPM. The SIR of SPM after HER2+ BC was significantly lower than HER2- BC (1.03 vs 1.13; RR, 0.92; 95% CI, 0.88-0.96; p<0.001). But the predominantly declining SPM risk was only observed for second BC (RR, 0.89; 95% CI, 0.82-0.96; p=0.003) and lung and bronchus cancer (RR, 0.84; 95% CI, 0.74-0.95; p=0.007). Further competing risk analysis verified the protective effect of HER2 positivity status on SPM occurrence.The 5-year cumulative incidence of SPM following HER2+ and HER2- BC were 5.16% and 4.09%, respectively (p<0.001). In addition, among patients suffering from SPM,HER2 positivity status contributed tobetter overall survival. Conclusion: After considering intrinsic incremental risk with age and adjusting for competing risk of death, our study demonstrated that HER2+ BC patients had lower SPM occurrence, remarkable for second BC and lung and bronchus cancer. The disparity implies the relation between SPM occurrence and therapeutic along with genetic factors underlying BC HER2 status.

Patterns of the risk for subsequent primary cancer among survivors of adult-onset cancers in the United States.

12060 Background: The number of cancer survivors who develop new cancers is projected to grow in the US. Few studies, however, have provided a comprehensive overview of the contemporary pattern in the risk of subsequent primary cancer (SPC) among survivors of adult-onset cancers. Herein, we evaluate overall and type-specific risks of SPCs among adult-onset cancer survivors by first primary cancer types and sex. Methods: We assessed the excess risk of SPCs among 1,442,374 persons aged 20-84 years who were diagnosed with first primary cancers from 1992-2010 and survived ≥5 years in the 12 Surveillance, Epidemiology, and End Results registries. We expressed the risks using excess absolute risk (EAR) per 10,000 person-years and standardized incidence ratio (SIR) by first primary cancer types and sex, compared to those expected in the general population. We also estimated percent contributions of each specific type of SPCs to the total EAR for all first primary cancers combined by sex. Results: The overall risk of SPCs was higher than expected for 24 of the 34 first primary types among male survivors and for 28 of the 35 first primary types among female survivors. The greatest SIR and EAR were estimated after laryngeal cancer in both men (SIR = 1.74, 95%CI = 1.67-1.82; EAR = 159.3, 95%CI = 143.6-175.5) and women (SIR = 2.48, 95% CI = 2.26-2.73; EAR = 202.7; 95%CI = 171.8-236). There were 290 type-specific associations with significantly higher risk of SPC, 36% of which being reciprocal, predominantly among smoking-associated, HPV-associated, and hematologic cancers. The SIRs in men ranged from 1.05 (95%CI = 1.00-1.10; EAR = 1.69) for lung/bronchus cancer after colorectal cancer to 73.9 (95%CI = 58.3-92.3; EAR = 23.3) for anal cancer after Kaposi sarcoma; and in women the SIRs ranged from 1.08 (95%CI = 1.02-1.15; EAR = 0.36) for pancreatic cancer after breast cancer to 19.9 (95%CI = 15.0-26.0; EAR = 39.5) for oral cavity/pharyngeal cancer after laryngeal cancer. For all first primary cancers combined, lung/bronchus cancer comprised the greatest proportion of the total EAR of SPCs, 34.6% in men and 29.1% in women, followed by urinary bladder (11.8%) and oral cavity/pharynx (7.5%) in men and by corpus uterus (12.9%) and colorectum (7.6%) in women. Conclusions: Despite the substantial heterogeneity in the risk of SPCs across the first primary types, only a few cancers comprised a considerable proportion of the total excess risk among survivors. Better understanding of contributing factors to these patterns will inform survivorship care plans and care delivery.

Lung cancer: Improving screening in rural clinic.

e13060 Background: Lung cancer is the leading cause of overall cancer-related deaths in both men and women in the United States. In 2009, cancer surpassed cardiovascular disease as the leading cause of death in North Carolina. Between 2010-2014, the age-adjusted incidence rate for lung and bronchus cancers in North Carolina was 70 per 100000 persons per year. Between 2010-2014, the age-adjusted mortality rate for lung and bronchus cancers in North Carolina was 50.6 per 100000 persons per year. It is documented that about half of lung and bronchus cancer cases were diagnosed at the distant stage and about 60 percent of deaths occurred in adults ages 65-84. In 2011, 32 percent of cancer death was from lung/bronchus cancer in Robeson County. A chart review in a rural primary care clinic identified patients not being appropriately screened for targeted intervention. Methods: Our retrospective chart review at Lumberton Medical Clinic, a rural outpatient Internal Medicine Clinic, reviewed 91 records from adults aged 55 to 80 years old during the timeframe of September 2017 through August 2018. Patients with a known history of lung cancer were excluded from this study. Patient records were assessed for compliance with USPTF lung cancer screening guidelines. USPTF recommends adults aged 55 to 80 who have a 30 pack-year smoking history and currently smoke or have quit within the past 15 years to have an annual low-dose computed tomography of the chest to screen for lung cancer. Results: The review showed that during the timeframe studied, 42% of patients who qualified received appropriate screening, while 58% of those qualified were found to have no documentation of screening. Following data analysis, intervention to increase screening rates has been initiated. This involves provider education, posters, and individualized letters mailed to patients found deficient during the study. The success of this direct patient outreach effort will be measured over six months. Conclusions: It is the responsibility of providers to emphasize the importance of proper lung cancer screening. This retrospective review found that a large percentage (58%) of adults were not being adequately screened in our rural clinic. Direct outreach is underway to increase compliance rates in this high-risk population served by our rural clinic.