Insurance Status, Age, and Cervical Cancer Stage at Diagnosis
Insurance Status, Age, and Cervical Cancer Stage at Diagnosis
Our data came from the National Cancer Database (NCDB), a hospital-based cancer registry jointly sponsored by the American Cancer Society and the American College of Surgeons. The NCDB documents approximately 70% of all malignant cancers in the United States, from more than 1400 facilities accredited by the American College of Surgeons' Commission on Cancer. The NCDB contains standardized data elements on patient demographics, insurance status, tumor characteristics, and first course of treatment, as well as census-level socioeconomic factors and facility-level factors.
We selected patients with a first primary invasive tumor of the cervix diagnosed between 2000 and 2007 (n = 76 862). We excluded patients younger than 21 years or older than 85 years from analyses (n = 1427). Because of small sample sizes, we excluded patients who listed "other forms" of government insurance (Bureau of Indian Affairs, Public Health Service; n = 284). We also excluded patients with unknown clinical or pathological stage (n = 5097) or geographic region (n = 315). The total analytic cohort comprised 69 739 patients.
Our primary outcome was American Joint Commission on Cancer stage at diagnosis, defined as clinical stage. If clinical stage was missing, we used pathological stage as a proxy. Independent variables of special focus were age, categorized into 9 groups (21–34, 35–39, 40–44, 45–49, 50–54, 55–59, 60–64, 65–69 and 70–85 years), and insurance type, categorized as Medicaid, Medicare (Medicare alone or with supplemental insurance), uninsured, private insurance (health maintenance organizations, preferred provider organizations), and other or unknown. Because Medicare is available to essentially all US residents aged 65 years and older, but only for permanently disabled individuals younger than 65 years, we dichotomized Medicare patients into age groups: 21 to 64 years and 65 to 85 years.
We also included race/ethnicity and geographic residence (derived from US census classifications). We obtained area-level indicators of education (percentage of adults without a high school diploma) from zip code information in 2000 US Census data and reported them as quartiles of the observed distribution in the general US population.
We conducted analyses with SAS version 9.2 (SAS Institute, Cary, NC). We used the χ test to analyze the relationship between insurance and all other covariates (P < .01). Because of the high prevalence (> 10%) of our outcomes, we used multivariable log binomial models to estimate risk ratios (RRs) and 95% confidence intervals (CIs); we used these to measure statistical significance. We made 2 separate comparisons: stage II versus stage I and stage III or IV versus stage I. These separate comparisons allowed us to examine the risk of moderate early stage (stage II), which may be curable, and advanced stage (III and IV), which may not be curable. We also stratified results by insurance to more closely examine the relationship of stage with insurance and age at diagnosis. We plotted stage at diagnosis against continuous age at diagnosis, and we calculated unadjusted slopes to measure the change in late stage at diagnosis for each additional year of age.
The NCDB only collects data from hospitals approved by the Commission on Cancer, which are located primarily in urban areas and offer more cancer services than other hospitals. Therefore, we compared the NCDB cervical case counts with those of the North American Association of Central Cancer Registries, which covers 98% of incident US cancer cases, and the population-based registries in Surveillance Epidemiology and End Results (SEER 17) data, which cover 28% of incident US cancer cases. We also assessed factors related to missing stage at diagnosis in the NCDB to estimate potential selection bias, with multivariable log binomial analyses.
Screening data are not available in the NCDB, so we used data from the National Health Interview Survey (NHIS) to estimate the prevalence of cervical cancer screening (defined as reporting a Papanicolau test in the past 3 years) among US women by age group and health insurance status. NHIS, an annual nationwide cross-sectional household survey of the civilian, noninstitutionalized population, is designed to provide national prevalence estimates on personal, socioeconomic, demographic, and health characteristics of the US population. It is the official federal monitoring instrument for US cancer screening objectives. Hispanics and African Americans are oversampled in NHIS to improve estimate precision for these populations. Consistent with the NCDB analysis, we conducted assessment of cervical cancer screening with pooled NHIS data for 2000 and 2005 and used comparable age and insurance categories. We also weighted NHIS prevalence estimates to account for the survey's complex population sampling.
Methods
Our data came from the National Cancer Database (NCDB), a hospital-based cancer registry jointly sponsored by the American Cancer Society and the American College of Surgeons. The NCDB documents approximately 70% of all malignant cancers in the United States, from more than 1400 facilities accredited by the American College of Surgeons' Commission on Cancer. The NCDB contains standardized data elements on patient demographics, insurance status, tumor characteristics, and first course of treatment, as well as census-level socioeconomic factors and facility-level factors.
Patients and Measures
We selected patients with a first primary invasive tumor of the cervix diagnosed between 2000 and 2007 (n = 76 862). We excluded patients younger than 21 years or older than 85 years from analyses (n = 1427). Because of small sample sizes, we excluded patients who listed "other forms" of government insurance (Bureau of Indian Affairs, Public Health Service; n = 284). We also excluded patients with unknown clinical or pathological stage (n = 5097) or geographic region (n = 315). The total analytic cohort comprised 69 739 patients.
Our primary outcome was American Joint Commission on Cancer stage at diagnosis, defined as clinical stage. If clinical stage was missing, we used pathological stage as a proxy. Independent variables of special focus were age, categorized into 9 groups (21–34, 35–39, 40–44, 45–49, 50–54, 55–59, 60–64, 65–69 and 70–85 years), and insurance type, categorized as Medicaid, Medicare (Medicare alone or with supplemental insurance), uninsured, private insurance (health maintenance organizations, preferred provider organizations), and other or unknown. Because Medicare is available to essentially all US residents aged 65 years and older, but only for permanently disabled individuals younger than 65 years, we dichotomized Medicare patients into age groups: 21 to 64 years and 65 to 85 years.
We also included race/ethnicity and geographic residence (derived from US census classifications). We obtained area-level indicators of education (percentage of adults without a high school diploma) from zip code information in 2000 US Census data and reported them as quartiles of the observed distribution in the general US population.
Statistical Analyses
We conducted analyses with SAS version 9.2 (SAS Institute, Cary, NC). We used the χ test to analyze the relationship between insurance and all other covariates (P < .01). Because of the high prevalence (> 10%) of our outcomes, we used multivariable log binomial models to estimate risk ratios (RRs) and 95% confidence intervals (CIs); we used these to measure statistical significance. We made 2 separate comparisons: stage II versus stage I and stage III or IV versus stage I. These separate comparisons allowed us to examine the risk of moderate early stage (stage II), which may be curable, and advanced stage (III and IV), which may not be curable. We also stratified results by insurance to more closely examine the relationship of stage with insurance and age at diagnosis. We plotted stage at diagnosis against continuous age at diagnosis, and we calculated unadjusted slopes to measure the change in late stage at diagnosis for each additional year of age.
The NCDB only collects data from hospitals approved by the Commission on Cancer, which are located primarily in urban areas and offer more cancer services than other hospitals. Therefore, we compared the NCDB cervical case counts with those of the North American Association of Central Cancer Registries, which covers 98% of incident US cancer cases, and the population-based registries in Surveillance Epidemiology and End Results (SEER 17) data, which cover 28% of incident US cancer cases. We also assessed factors related to missing stage at diagnosis in the NCDB to estimate potential selection bias, with multivariable log binomial analyses.
Screening data are not available in the NCDB, so we used data from the National Health Interview Survey (NHIS) to estimate the prevalence of cervical cancer screening (defined as reporting a Papanicolau test in the past 3 years) among US women by age group and health insurance status. NHIS, an annual nationwide cross-sectional household survey of the civilian, noninstitutionalized population, is designed to provide national prevalence estimates on personal, socioeconomic, demographic, and health characteristics of the US population. It is the official federal monitoring instrument for US cancer screening objectives. Hispanics and African Americans are oversampled in NHIS to improve estimate precision for these populations. Consistent with the NCDB analysis, we conducted assessment of cervical cancer screening with pooled NHIS data for 2000 and 2005 and used comparable age and insurance categories. We also weighted NHIS prevalence estimates to account for the survey's complex population sampling.
