The role of socially patterned risk factors in childhood cancer incidence and survival

Abstract

A primary goal of childhood cancer research is to understand the role of non-genetic, socially patterned, risk factors in incidence and survival. Many unanswered questions persist in the literature that, if answered, may help address this aim. Such questions include: do pregnancy-related exposures contribute to the rise in childhood cancer incidence over time; is socioeconomic status (SES) associated with childhood cancer incidence; and what is the underlying role of social versus biological factors in explaining racial disparities in childhood cancer survival? This dissertation addressed each of these questions by leveraging population-based data from multiple existing data sources, and by employing advanced statistical methods, in three separate investigations. In manuscript 1, we conducted a time series ecologic analysis at the county-level to test the hypothesis that the temporal rise in childhood cancer incidence is due to secular trends in established pregnancy-related risk factors including older maternal age, higher birthweight, and smaller family size. We linked population-based cancer registry data from the Surveillance, Epidemiology, and End Results (SEER) 9 database (1975-2013) to natality data from the National Center for Health Statistics (1970-2013). We compared the crude average annual percent change (AAPC) in incidence of combined (all diagnoses) and individual cancers among children, ages 0-4 years, to AAPCs adjusted for pregnancy-related and sociodemographic (race/ethnicity and poverty) factors. AAPCs were estimated from Poisson mixed models. In crude models, we found a statistically significant temporal rise in incidence of combined childhood cancers (AAPC, 0.71%; 95% CI, 0.55, 0.86), ALL (0.78%; 0.49, 1.07), AML (1.86%; 1.13, 2.59), CNS tumors (1.31%; 0.94, 1.67), and hepatoblastoma (2.70%; 1.68, 3.72). Contrary to our hypothesis, AAPCs remained statistically significantly above 0% in models fully adjusted for county-level characteristics, though AAPCs were attenuated towards the null for AML (1.62%; 0.38, 2.87) and hepatoblastoma (2.36%; 0.71, 4.04). Therefore, we did not find conclusive evidence that secular trends in established pregnancy-related risk factors account for the temporal rise in cancer incidence rates among children, ages 0-4 years. In manuscript 2, we tested whether SES, measured at multiple levels of exposure, is associated with incidence of childhood cancers after accounting for established demographic and pregnancy-related risk factors. We conducted a population-based case-cohort study using the Minnesota birth registry, 1989-2014, as the source cohort. Cases, ages 0-14 years, were identified in the Minnesota Cancer Surveillance System and linked to birth records through probabilistic record linkage. Controls were 4:1 frequency matched on birth year (2,947 cases, 11,907 controls). We measured individual-level SES using maternal education, and we measured neighborhood-level SES using a census tract composite index. Associations between SES and childhood cancer incidence were tested with logistic mixed models. In crude models, we found that higher maternal education was adversely associated with incidence of combined childhood cancers (OR, 1.08; 95% CI, 1.04, 1.13), ALL (OR, 1.10; 95% CI, 1.02, 1.19), CNS tumors (OR, 1.12; 95% CI, 1.04, 1.21), and neuroblastoma (OR, 1.15; 95% CI, 1.02, 1.30). These associations were attenuated towards the null, and no longer statistically significant, after adjusting for established demographic and pregnancy-related risk factors. Similar patterns were observed for neighborhood-level SES. A protective association, robust to covariate control, was detected between higher maternal education and hepatoblastoma risk (adjusted OR, 0.70; 95% CI, 0.51, 0.98). Overall, results suggest that, unlike for some adult cancers that show strong socioeconomic gradients, associations between SES and many childhood cancers appear to be explained by established demographic and pregnancy-related risk factors (i.e. non-Hispanic white race/ethnicity, older maternal age, and higher birthweight). In manuscript 3, we tested whether SES contributes to (i.e. mediates) black-white racial disparities in childhood cancer survival. We used population-based survival data from the SEER 18 database for black and white children, ages 0-19 years, diagnosed 2000-2011 (N=27,741). Race was recorded in SEER through medical record abstraction. We measured SES using a validated census tract composite index, and we tested treatment (first-course cancer-directed surgery and radiation) and distal stage at diagnosis as secondary potential mediators. We used the inverse odds weighting (IOW) method to test for mediation among combined and individual childhood cancers. Results showed that whites have a significant survival advantage over blacks for combined childhood cancers, leukemias, lymphomas, CNS tumors, neuroblastoma, and non-rhabdomyosarcoma soft tissue sarcomas (NRSTS). Significant black-white mortality hazard ratios ranged from 1.44 (95% CI, 1.11, 1.87) for NRSTS to 1.91 (95% CI, 1.52, 2.41) for astrocytomas. SES significantly mediated the race-survival association for combined childhood cancers and leukemias, accounting for 20% of the disparity for combined childhood cancers (indirect effect HR (iHR), 1.12; 95% CI, 1.08, 1.15), 44% for ALL (iHR, 1.18; 95% CI, 1.08, 1.29), and 28% for AML (iHR, 1.16; 95% CI, 1.04, 1.29). Treatment and stage significantly mediated race-survival associations for astrocytomas (27%; iHR, 1.19; 95% CI, 1.01, 1.40) and neuroblastoma (32%, iHR, 1.16; 95% CI, 1.04, 1.30). Overall, we found evidence that SES contributes to black-white racial disparities in childhood cancer survival, particularly for childhood leukemias. However, we could not rule out the possibility that non-social factors also contribute to survival differences by race, particularly for astrocytomas and neuroblastoma. Taken together, findings from these three investigations suggest that while continued study of socioeconomic exposures may generate limited new insight into childhood cancer etiology, this line of research may be fruitful for understanding and, ultimately, addressing differences in childhood cancer outcomes.