Abstract
BACKGROUND: Central nervous system (CNS) tumors rank among the top 5 cancers diagnosed in young adults aged 20 to 39 years at diagnosis and show a clear male excess in incidence. It is unknown whether sex differences in survival persist across histologic types and depend on the treatment received.
METHODS: From the National Cancer Database (2004-2016), young adults (aged 20-39 years) who had been diagnosed with CNS tumors were identified. Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated as measures of association between sex and death via Cox regression. An inverse odds weighting mediation analysis was performed with treatment received as a mediator.
RESULTS: There were 47,560 cases (47% male). Males had worse overall survival than females for 9 of 16 histologic types, including diffuse astrocytoma, glioblastoma, and meningioma (all P < .05). Males had an increased risk of death after a brain tumor diagnosis overall (HR, 1.47; 95% CI, 1.41-1.53) and for 8 histologies. There was a significant association between male sex and death overall that was mediated by treatment received (indirect-effect HR, 1.17; 95% CI, 1.15-1.18), but no single histology had a significant indirect effect. All histologies examined in mediation analyses had significant direct effects for sex. The excess mortality due to sex was 20% for all CNS tumors combined and 34% among males with CNS tumors.
CONCLUSIONS: Overall, treatment received may mediate a portion of the association between sex and death after a CNS tumor, but sex itself appears to be a stronger risk factor for death in this study.
| Original language | English (US) |
|---|---|
| Pages (from-to) | 1616-1625 |
| Number of pages | 10 |
| Journal | Cancer |
| Volume | 128 |
| Issue number | 8 |
| DOIs | |
| State | Published - Apr 15 2022 |
Bibliographical note
Funding Information:Research reported in this publication was supported by the National Cancer Institute of the National Institutes of Health under award T32CA163184 (principal investigators Michele Allen and Kristin J. Moore) and was administered by the University of Minnesota Medical School Program in Health Disparities Research and the University of Minnesota School of Public Health. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. This work is also supported by the Children's Cancer Research Fund (Lindsay A. Williams).
Funding Information:
Research reported in this publication was supported by the National Cancer Institute of the National Institutes of Health under award T32CA163184 (principal investigators Michele Allen and Kristin J. Moore) and was administered by the University of Minnesota Medical School Program in Health Disparities Research and the University of Minnesota School of Public Health. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. This work is also supported by the Children's Cancer Research Fund (Lindsay A. Williams). Individuals aged 20 to 39 years who had been diagnosed with brain and central nervous system (CNS) tumors (heretofore referred to as brain tumors) were identified in the NCDB (2004-2016), a clinical oncology database derived from hospital registry data from more than 1500 facilities across the United States. The NCDB is not a population-based resource but captures approximately 70% of cancer cases in the United States.9 To be included in our analysis, cases had to have a brain/CNS tumor histologic grouping as described below and had to have information on sex. We have defined young adults as those aged 20 to 39 years of age and have excluded 15- to 19-year-olds, who are generally considered adolescents and are grouped with children by the Children's Oncology Group and the National Cancer Institute. Individuals aged 20 to 39 years who had been diagnosed with brain and central nervous system (CNS) tumors (heretofore referred to as brain tumors) were identified in the NCDB (2004-2016), a clinical oncology database derived from hospital registry data from more than 1500 facilities across the United States. The NCDB is not a population-based resource but captures approximately 70% of cancer cases in the United States.9 To be included in our analysis, cases had to have a brain/CNS tumor histologic grouping as described below and had to have information on sex. We have defined young adults as those aged 20 to 39 years of age and have excluded 15- to 19-year-olds, who are generally considered adolescents and are grouped with children by the Children's Oncology Group and the National Cancer Institute. Histologic groupings (Supporting Figure 1) were based on the Central Brain Tumor Registry of the United States Brain and Central Nervous System Tumor Histology Groupings, which use International Classification of Diseases for Oncology, Third Edition codes.10 Ambiguous histologies and histologies with fewer than 25 cases in each strata for analyses were excluded. The histologic types included in the analysis were as follows: pilocytic astrocytoma, diffuse astrocytoma, anaplastic astrocytoma, unique astrocytoma variants, glioblastoma, oligodendroglioma, anaplastic oligodendroglioma, oligoastrocytic tumors, ependymal tumors, other malignant glioma, choroid plexus tumors, neuronal and mixed neuronal-glial tumors, medulloblastoma, primitive neuroectodermal tumors, meningioma, and germ cell tumors (Supporting Table 1). The site codes for tumors included in the analysis were C70.0, C70.1, C70.9, C71.0 to C72.5, C72.8, and C72.9. Invasive and noninvasive tumors were included in the analyses. This analysis used publicly available data without personal identifiers; therefore, the study was exempt from University of Minnesota institutional review board consideration. Descriptive statistics were calculated for clinical and demographic characteristics stratified by patient sex with variables available in the NCDB, including the following: age at diagnosis (20-24, 25-29, 30-34, or 35-39 years); race/ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, Asian/Pacific Islander, or American Indian/Alaska Native); year of diagnosis (2004-2008, 2009-2013, or 2014-2016); insurance status (not insured, private insurance/managed care, Medicaid, Medicare, other government, or unknown); education, which was defined as the number of adults in the patient's zip code who did not graduate from high school (quartiles 1-4 [Q1-Q4]; based on the year of diagnosis); income, which was defined as the median household income for each patient's area of residence by zip code (Q1-Q4; based on the year of diagnosis); urban or rural county of case (metro, rural, or urban; based on the year of diagnosis); great circle distance, which was defined as the distance in miles between the patient's residence and the hospital that reported the case (0-8.9, 9-22.3, 22.4-60.9, or >60.9); Charlson/Deyo comorbidity score (0 or ≥1); treatment received, which was defined as a combination treatment based on surgery (yes or no), chemotherapy (yes or no), and radiation (yes or no) (treatment: none; surgery only; surgery and radiation; surgery and chemotherapy; radiation only; chemotherapy only; radiation and chemotherapy; and radiation, chemotherapy, and surgery); and vital status (dead or alive). All variables were selected on the basis of their relationship with the exposure and outcome. Quartiles of median income and educational attainment (Q1-Q4) were assigned on the basis of the year of diagnosis to prespecified quartiles calculated by NCDB and presented in the data dictionary. For those diagnosed between 2004 and 2007, the quartiles of median income were as follows: Q1, <$30,000; Q2, $30,000 to $34,999; Q3, $35,000 to $45,999; and Q4, ≥$45,000. The quartiles of educational attainment were as follows: Q1, ≥29.0%; Q2, 20.0% to 28.9%; Q3, 14.0% to 19.9%; and Q4, <14.0%. For cases diagnosed in 2008-2011, the median income was categorized as follows: Q1, <$38,000; Q2, $38,000 to $47,999; Q3, $48,000 to $62,999; and Q4, ≥$63,000. The educational attainment categories included the following: Q1, ≥21.0%; Q2, 13.0% to 20.9%; Q3, 7.0% to 12.9%; and Q4, <7.0%. Finally, for those cases diagnosed between 2012 and 2016, the median income categories included the following: Q1, <$47,227; Q2, $40,227 to $50,353; Q3, $50,354 to $63,332; and Q4, ≥$63,333. The quartiles of educational attainment were as follows: Q1, ≥17.6%; Q2, 10.9% to 17.5%; Q3, 6.3% to 10.8%; and Q4, <6.3%. χ2 tests were used to assess sex differences in treatment received (Supporting Table 2). Kaplan-Meier survival curves and log-rank P values were used to compare survival differences by sex overall and stratified by tumor histology. Cox proportional hazards models were used to calculate hazard ratios (HR) and 95% confidence intervals (CIs) for the association between sex and death, with females used as the referent group. No violation of the proportional hazards assumption was identified when a time × sex interaction term was used in the models (all P >.05). We conducted a mediation analysis using an inverse odds weighting (IOW) method for histologic types that had a statistically significant association between sex and death, with treatment received used as the mediator. The method has been described elsewhere, and code is available from Nguyen et al.11-13 This method uses a weighted Cox proportional hazards model allowing for an estimation of the association between sex and death (direct effect) independent of treatment received. The weights were estimated from a logistic regression model for treatment received in association with sex, with females serving as the referent group (weight of 1). Males were assigned the value of the inverse odds from the logistic model. IOW first leg results demonstrated an association between sex and treatment in all tumors combined when the exposure, sex, was regressed onto the mediator, treatment, with adjustments made for age at diagnosis, race/ethnicity, year of diagnosis, insurance status, urban/rural status, distance between hospital and home, and income of the case's zip code (Supporting Table 3). The total-effect β was calculated from a Cox proportional hazards model without the IOW specification. The indirect effect of sex associated with death, operating through treatment received, was calculated by subtraction of the β from the direct effect from the β of the total effect (βindirect = βtotal – βdirect). Bootstrapping (1000 replications) was used to calculate standard errors for 95% CIs. A significant indirect effect was interpreted as evidence of mediation by treatment in the association between sex and death. Cox regression analyses were conducted in SAS version 9.4 (SAS Institute, Cary, North Carolina), mediation analyses were conducted in Stata version 16.0 (StataCorp, College Station, Texas), and figures were generated in R and GraphPad Prism 9.0.0 (GraphPad Software, La Jolla, California). Statistical significance was determined with 2-sided hypotheses with P <.05. Because this was observational research, we did not adjust for multiple comparisons.14
Publisher Copyright:
© 2022 American Cancer Society
Keywords
- mediation analysis
- public health
- sex differences
- survival disparities
- young adult brain tumors
