|Home | About | Journals | Submit | Contact Us | Français|
Disparities in treatment exist for non-white and Hispanic patients with non-small-cell lung cancer, but little is known about disparities in the use of staging tests or their underlying causes.
Prospective, observational cohort study of 3638 patients with newly diagnosed non-small-cell lung cancer from 4 large, geographically-defined regions, 5 integrated health care systems and 13 VA health care facilities.
Median age was 69 years, 62% were men, 26% were Hispanic or non-white, 68% graduated high school, 50% had private insurance, and 41% received care in the VA or another integrated health care system. After adjustment, PET use was 13% lower among non-whites and Hispanics than non-Hispanic whites (RR 0.87, 95% CI 0.77 to 0.97), 13% lower among those with Medicare than those with private insurance (RR 0.87, 95% CI 0.76 to 0.99), and 24% lower among those with an elementary school education than those with a graduate degree (RR 0.76, 95% CI 0.57 to 0.98). Disparate use of PET was not observed among patients who received care in an integrated health care setting, but the association between race/ethnicity and PET use was similar in magnitude across all other subgroups. Further analysis showed that income, education, insurance and health care setting do not explain the association between race/ethnicity and PET use.
Hispanics and non-whites with non-small-cell lung cancer are less likely to receive PET imaging. This finding is consistent across subgroups and not explained by differences in income, education, or insurance coverage.
Accurate staging is crucial among patients with non-small-cell lung cancer (NSCLC), because both prognosis and treatment selection are largely determined by disease stage (1, 2). It is essential to accurately identify those individuals with potentially resectable disease (stages I, II and sometimes IIIA), in whom surgery remains the best option for cure. Most patients with mediastinal lymph node involvement (stages IIIA and IIIB) have limited options for cure, but concurrent chemoradiation can prolong life and palliate symptoms (3). To identify mediastinal metastasis, staging procedures include imaging tests, such as computed tomography (CT) and positron emission tomography (PET), and a growing armamentarium of invasive biopsy procedures.
PET is more accurate than CT for identifying malignant mediastinal lymph nodes (4). Guidelines developed by the American College of Chest Physicians (ACCP) and the National Comprehensive Cancer Network recommend that PET be used to help stage lung cancer in patients who are candidates for curative treatment (5, 6). Three randomized controlled trials in patients with potentially resectable NSCLC found that compared with conventional staging, PET-based staging reduced the frequency of thoracotomy without cure (7–9). However, in the majority of patients with NSCLC who have unresectable disease or are medically inoperable, PET-based staging strategies have not been evaluated in randomized trials. The use of imaging tests for cancer staging is one of the Institute of Medicine’s top 25 priorities for comparative effectiveness research (10).
Studies of disparities in lung cancer care have demonstrated that African Americans and Hispanics are less likely to receive potentially curative surgery (11–13), but the underlying causes of these treatment disparities are incompletely understood, and relatively little is known about disparities in practices for lung cancer staging. One previous study found that the frequency of multi-modality staging (including PET) was especially low among African Americans and persons with low income or educational status (14). We sought to describe variation in PET use among patients with NSCLC, identify socio-demographic and tumor characteristics associated with the use of PET, and examine whether income, education, insurance or health care setting modify or explain disparate use of PET in members of racial and ethnic minority groups.
To examine variation in PET use among patients with NSCLC, we analyzed data collected by the Cancer Care Outcomes Research and Surveillance (CanCORS) Consortium, as part of a prospective observational study of cancer care practices and outcomes for patients with lung and colorectal cancer. The methods of the CanCORS “Share Thoughts On Care” study have been reported previously in detail (15). The CanCORS study, funded by the National Cancer Institute and the VA Health Services Research and Development Service, used data from medical records, patient interviews and physician surveys to explore why some groups of cancer patients are more likely than others to receive recommended treatments and other interventions. For this analysis, we used information found in version 1.8 of the CanCORS core data files. All patients or an appropriate surrogate provided informed consent. Human subjects committees at Stanford University and all participating sites approved the study. See Supplemental Digital Content for additional details about methods and results.
CanCORS used rapid case ascertainment to prospectively enroll incident lung cancer cases in 4 large geographically defined regions, 5 integrated health care delivery systems, and 13 health care facilities of the Veterans Health Administration. Together, these settings capture approximately 10% of the total U.S. population and were responsible for almost 10% of all U.S. lung cancer cases in 2000. Eligible patients were diagnosed with lung cancer between September 1, 2003 and October 14, 2005. For this analysis, we included all CanCORS participants with NSCLC who underwent medical record abstraction.
Professional chart abstractors collected information about variables of interest from the medical records of CanCORS participants, including patient characteristics (age, gender, race, ethnicity, comorbidities, insurance, health care setting), tumor characteristics (size, histology, stage), and use of staging tests (CT, PET, bronchoscopy, mediastinoscopy, endoscopic ultrasound). Collaborative stage was obtained from chart abstraction or tumor registry. For a small proportion of patients, the available stage was registry-classified as local, regional, or distant. Therefore, in this analysis we have used those classifications, considering patients with stage I or II as “local”, stage IIIa or IIIb as “regional”, and stage IV as “distant.” In the base-case analysis, we considered PET to be used for staging when the date of PET was known to be before the date of first treatment. To summarize information about comorbidities, the CanCORS study used the Adult Comorbidity Evaluation-27 (ACE-27), a validated, medical record-based instrument that ranks patients as having no, mild, moderate, or severe comorbidity, according to the most severe comorbid conditions identified (16, 17). We obtained additional information about annual household income and level of education from the CanCORS baseline patient survey.
To allow for non-linear effects and to simplify analysis and interpretation, we dichotomized continuous variables (age, tumor size) at the median value. Similarly, we transformed CanCORS variables for race (17 categories) and ethnicity (7 categories) into a single variable with 3 categories (non-Hispanic white, Hispanic or non-white, and unknown). We also transformed CanCORS variables for primary and secondary insurance, (12 and 14 categories, respectively) by creating a single variable in which patients were categorized as having private insurance (if primary or secondary insurance was private), no health insurance (if both primary and secondary were no insurance) or non-private insurance (all others). We further subdivided patients with non-private insurance as having Medicare (if primary or secondary insurance was Medicare), Medicaid (if primary insurance was Medicaid, and secondary insurance was not Medicare) or other non-private (all others with non-private insurance). In addition, we created a variable to describe whether the site of enrollment was a non-integrated setting, an integrated system (including Kaiser-Permanente members in Northern California and Los Angeles), or the VA Healthcare System. Finally, we simplified the CanCORS histology variable (88 categories) by placing each participant in one of 4 categories: adenocarcinoma, squamous cell carcinoma, other/mixed histology, or NSCLC, not further specified. We recoded all variables prior to data analysis.
We report median, 1st quartile, and 3rd quartile values for continuous variables, and counts and frequencies for categorical variables. To perform between-groups comparisons of categorical variables, we used chi-square tests. To identify patient and tumor characteristics that were independently associated with the use of PET, we performed multiple logistic regression analyses that included all candidate predictor variables. In these analyses, we did not adjust for final stage, because stage could be influenced by whether or not PET was performed. Because use of PET was relatively common, we report both odds ratios and approximate risk ratios, which we estimated by using the method of Zhang and Yu (18).
We tested for clinically relevant subgroup effects by entering all plausible interaction terms one at a time into the full logistic model. We also repeated analyses after stratifying by income, education, insurance coverage and health care setting. We developed additional logistic regression models to examine whether one or more factors mediated (or explained) associations between race/ethnicity and use of PET, including income, education, insurance and health care setting (Appendix Figure). When designing and interpreting these analyses, we used the definition of mediation proposed by Baron and Kenny (19). Briefly, this definition of mediation requires confirming the statistical significance of all of the following: (a) an association between the exposure (e.g. race/ethnicity) and the potential mediator of interest (e.g. income), (b) an association between the exposure and the outcome (e.g. use of PET), and (c) a difference in the beta coefficients for the association between the exposure and the outcome in models that did and did not include the mediator of interest. To formally test the statistical significance of mediator effects, we adapted the method of Preacher and Hayes to generate mean Bootstrap estimates and their 95% confidence limits for the difference between beta coefficients for predictor variables in models that did and did not include the mediator of interest (20).
To test the robustness of our findings, we developed additional logistic regression models that (a) excluded 782 patients who died within 3 months of diagnosis, (b) included adjustment for disease stage, and (c) included patients in whom the timing of PET relative to treatment was unknown among those who were considered to have PET for staging.
Of 4,550 potentially eligible CanCORS participants with lung cancer, we excluded 153 patients who did not have their medical records abstracted, 579 patients with small cell carcinoma, and 180 patients with unknown tumor histology (Figure 1). The resulting sample included 3,638 CanCORS participants with NSCLC, of whom 638 were enrolled at VA sites.
The median age of the sample was 69 years (Table 1). Almost two thirds of participants were men, and 26% were of Hispanic ethnicity or non-white race. Approximately half of participants had moderate or severe comorbidities, 50% had private health insurance, 34% had non-private insurance and 1% had no insurance. Almost 70% graduated high school, and 15% graduated from college. Annual household income was less than $40,000 in 49%. The median tumor size was 40 mm. Adenocarcinoma (35%) was more common than squamous cell histology (24%). Stage distribution was nearly uniform, although slightly fewer participants had stage III/regional disease. Twenty-three percent of patients received care in an integrated health care system and another 18% received care in the Health Care System of the Department of Veterans Affairs.
Over 50% of CanCORS participants received PET (Table 1), including 34% in whom the scan was performed prior to treatment. The timing of PET relative to treatment was unknown in another 13%. Only 5% underwent invasive mediastinal staging (mediastinoscopy, endobronchial ultrasound or endoscopic ultrasound). Just over 50% underwent bronchoscopy, but information about transbronchial needle aspiration (TBNA) biopsy of mediastinal lymph nodes was not recorded. Initial treatment included surgery in 33%, chemotherapy in 26%, radiation in 27% (including 5% who received concurrent chemoradiation) and supportive care only in 19% of participants. The median duration of follow-up from the date of diagnosis was 270 days (IQR 83 to 254 days). Only 26% of the sample was alive at the time of last contact.
Compared to participants who did not receive PET for staging, those who did were more likely to be younger, non-Hispanic and white, and more highly educated (Table 2). They were more likely to have private health insurance and reported greater annual household income. PET recipients were more likely to have local stage disease, to undergo invasive mediastinal staging, to receive surgery, chemotherapy or radiation, and to survive until the end of follow-up.
In a multivariate analysis (Table 3), PET use was approximately 9% lower in patients older than 69 years (estimated risk ratio [RR] 0.91, 95% CI 0.81 to 1.00), 13% lower in non-whites and Hispanics (RR 0.87, 95% CI 0.77 to 0.97), and 17% greater in patients with squamous cell carcinoma compared to adenocarcinoma (RR 1.17, 95% CI 1.04 to 1.30). Compared to patients with private health insurance, PET use was similar among patients with Medicaid (RR 1.02, 95% CI 0.74 to 1.32), significantly lower among patients with Medicare (RR 0.87, 95% CI 0.76 to 0.99), and 30% lower among those without health insurance (RR 0.70, 95% CI 0.39 to 1.12). Compared to those with a graduate or professional degree, PET use was less common among those who completed less than 9 years of school (RR 0.76, 95% CI 0.57 to 0.98) and those who attended but did not graduate from high school (RR 0.80, 95% CI 0.61 to 1.01). Use of PET also increased with increasing household income, but this association was not statistically significant after adjusting for other variables. Finally, compared to patients who received care in non-integrated health care settings, PET was used less frequently among patients who received care in the VA (RR 0.76, 95% CI 0.62 to 0.92) or an integrated health care system (RR 0.88, 95% CI 0.77 to 0.99).
In models that tested for subgroup effects, there were no significant interactions between race/ethnicity and age (p=0.99), gender (p=0.36), comorbidity (p=0.87), insurance (p=0.32), income (p=0.42), education (p=0.18) or health care setting (0.14). In stratified analyses, non-whites and Hispanics were less likely to receive PET across all subgroups in analyses that were stratified by insurance, income or education. However, in an analysis stratified by health care setting, PET was used less frequently among non-whites and Hispanics who received care in a non-integrated setting (RR 0.82, 95% CI 0.71 to 0.96), or VA facility (RR 0.82, 95% CI 0.58 to 1.10), but not among those who received care in an integrated system (RR 1.01, 95% CI 0.77 to 1.30). Further analysis revealed that while the frequency of PET use differed widely across VA facilities (p<0.0001), it was similar for whites and non-whites after adjusting for facility (p=0.84).
Compared to non-Hispanic whites, non-whites and Hispanics were approximately 23% less likely to have private health insurance (RR 0.77, 95% CI 0.67 to 0.89), 19% less likely to have graduated from high school (RR 0.81, 95% CI 0.70 to 0.92), 11% less likely to have an annual household income greater than $60,000 (RR 0.89, 95% CI 0.72 to 1.07), and 22% more likely to receive care in a non-integrated setting (RR1.22, 95% CI 1.04 to 1.41). However, the less frequent use of PET among non-whites and Hispanics remained statistically significant and was unchanged in magnitude in analyses that did and did not adjust for insurance status, income, education and health care setting (Supplemental Digital Content), indicating that these factors did not explain the association between race/ethnicity and PET use.
Results were similar in an analysis that excluded 782 patients who did not survive for at least 3 months following diagnosis, although education was no longer associated with use of PET, while smaller tumor size (RR 0.84, 95% CI 0.75 to 0.93) and VA health care setting (RR 0.70, 95% CI 0.56 to 0.86) were more strongly associated with less frequent PET use. Hispanics and non-whites were even less likely to receive PET when we modified our definition of PET use to include patients in whom the timing of PET was uncertain (RR 0.83, 95% CI 0.75 to 0.91), and when we controlled for primary data collection and recruitment site, rather than health care setting (RR 0.77, 95% CI 0.68 to 0.87) (Supplemental Digital Content).
In a multivariate analysis that also included final stage, patients with local (RR 1.16, 95% CI 1.01 to 1.31) and regional (RR 1.42, 95% CI 1.26 to 1.58) disease were both more likely to receive PET, compared to those with distant disease (Supplemental Digital Content). Including stage did not meaningfully change the magnitude or statistical significance of the effect of any other variable.
In a model looking for differences across racial and ethnic groups, PET use was less frequent among African Americans (RR 0.83, 95% CI 0.70 to 0.97), Asians (RR 0.89, 0.67 to 1.14) and Hispanics (RR 0.87, 95% CI 0.66 to 1.12), compared with non-Hispanic whites. Differences were greater in magnitude and statistically significant for Asians and Hispanics when our definition of PET use included patients in whom the timing of PET was uncertain.
In this analysis of data from a large, prospective, observational study of lung cancer practices and outcomes, PET was used for diagnosis and/or staging in at least one third of all patients with NSCLC, and probably was used for these indications in over 50%. Compared to non-Hispanic whites, PET was used approximately 13% less frequently among non-whites and Hispanics. Use of PET was also less common among older patients, patients with Medicare or no health insurance (compared to patients with private health insurance), patients whose reported annual household income was less than $20,000 (compared to higher income patients), and patients who received care in the VA or another integrated system. Not surprisingly, PET was used more frequently in patients with local or regional disease and those treated by surgery.
One novel aspect of our study is that we attempted to identify subgroups in which the association between race/ethnicity and use of PET was especially strong or weak. In these analyses, we found that the association between race/ethnicity and PET use was not significantly different in subgroups defined by income, education or insurance coverage. While we did not find a statistically significant interaction between race/ethnicity and health care setting, stratified analyses revealed that disparate use of PET was not observed among patients who received care in an integrated health care system, suggesting that access to PET may be more equitable in these settings.
Another innovative aspect of our study is that we adapted well-established methods from social science to identify factors that might explain why PET was used less frequently in non-whites and Hispanics. Somewhat surprisingly, results of these analyses did not support the hypothesis that differences in income, education, insurance coverage or health care setting explained the association between non-white race/ethnicity and less frequent use of PET. Our results suggest not only that race/ethnicity is an independent predictor of PET use, but also that differences in income, education, and insurance coverage are not responsible for the association between race/ethnicity and PET use.
Other factors that might explain the lower frequency of PET use among non-whites and Hispanics include presentation with more advanced disease (21, 22), limited access to imaging centers and hospitals that have PET scanners, patient preferences for less aggressive staging and treatment (23, 24), as well as conscious or unconscious biases of clinicians. Consistent with all of these possibilities, the association between race/ethnicity and PET use was statistically significant and greatest in magnitude amongst African Americans. However, our finding that disparities in PET use were similar for African Americans, Asians and Hispanics makes it unlikely that a single, specific language barrier and/or cultural factor is responsible, as does our finding that disparities were not seen among patients who received care in an integrated health care system. We distinguished between integrated and non-integrated settings to test the hypothesis that the type of delivery system is associated with the likelihood of PET use and found that PET was used more frequently in non-integrated settings, suggesting that financial incentives may be an important determinant of PET use “at the margin”, possibly including both “overuse” in non-integrated settings and “underuse” in integrated systems.
Prior studies have shown that lung cancer surgery is less common in African Americans (11, 25), Hispanics (12), non-elderly Medicare beneficiaries with disabilities (26), octogenarians (27), and those with low incomes (25). Lathan et al. reported that blacks with NSCLC were approximately 25% less likely to undergo invasive staging, defined as receipt of bronchoscopy, mediastinoscopy or thoracoscopy (28). Similarly, Farjah et al. found that only 1% of African Americans underwent tri-modality staging (CT, PET and invasive biopsy), compared with 3% of whites (14). They also found that overall rates of bi-modality staging (CT and PET, or CT and invasive biopsy) were 14% in African Americans and 21% in whites, a difference that is somewhat greater (in relative terms) than the 13% difference in PET use that we observed. Similar to what we reported for PET use, they found a lower likelihood of bi-modality or tri-modality staging among older patients and those with income in the lowest quartile. More recently, Suga et al. used data from the California Cancer Registry to identify 12,395 patients who were diagnosed with NSCLC between 1994 and 2004 (29). In contrast to our results, they found that non-whites were no less likely than whites to receive PET. However, our study included a greater percentage of African Americans (12.7% versus 6.0%), and the overall frequency of PET use was much greater (51% versus 14.6%), reflecting the more recent data collection period. Of note, within-site differences between whites and non-whites in our study were greater in magnitude for the two CanCORS regions in California than any other data collection and recruitment site (data not shown), suggesting the possibility that race and ethnicity-based disparities in PET use in California have become more severe over time as the technology has disseminated more widely.
Strengths of our study include the large, prospectively enrolled and geographically diverse sample, the collection of detailed socio-demographic information with limited amounts of missing data, and use of rigorous analytical methods. Important limitations include the potential for residual confounding, although we adjusted for all available variables that might be associated with both the exposure (race/ethnicity) and the outcome (use of PET). In addition, because the CanCORS study did not collect information about CT and PET results, we were not able to report intermediate outcomes such as upstaging and downstaging, nor were we able to determine rates of futile thoracotomy or changes in treatment planning. Because assignment to PET was non-random and differences in baseline characteristics persisted even after adjustment by propensity score methods (data not shown), we did not attempt to compare survival among patients who did and did not receive PET.
Finally, our study was not designed to determine whether PET was underutilized in non-whites and Hispanics or overused in non-Hispanic whites. ACCP guidelines recommend that PET be used to characterize solitary pulmonary nodules (CT stage IA) when the clinical probability of malignancy is low to moderate (30), and for non-invasive staging in patients with clinical (CT) stage IB-IIIB NSCLC who are candidates for treatment with curative intent (6). In this study, we were not able to determine the appropriateness of PET use because information about CT stage was not available. While PET-based staging in NSCLC has been shown to reduce the frequency of futile thoracotomy, it has not been shown to improve survival, and the potential benefits of PET-based staging have not been demonstrated in patients who are not candidates for surgical cure. Future studies should attempt to identify subgroups of patients with NSCLC who are most likely to benefit from staging that includes PET.
In summary, we found that PET imaging was commonly used for diagnosis and/or staging in patients with NSCLC, but there was substantial variation in use. With the exception of those who received care in an integrated health care system, Hispanics and non-whites with NSCLC were less likely to receive PET. This finding was consistent across subgroups and not explained by differences in income, education, or insurance coverage. PET was also used less frequently among older patients, and among those with Medicare or no health insurance and those with very low income, suggesting that this technology is being differentially applied for reasons unrelated to clinical status.
Supplemental Digital Content 1: Methods: CanCORS Primary Data Collection and Research Sites. .doc
Supplemental Digital Content 2: Sensitivity Analysis. Independent predictors of PET use (PET performed before treatment or timing uncertain). .doc
Supplemental Digital Content 3: Sensitivity Analysis. Independent predictors of PET use (results of model that included final stage). .doc
Supplemental Digital Content 4: Sensitivity Analysis. Results of model that controlled for PDCR site rather than health care setting. .doc
Supplemental Digital Content 5: Analysis of mediator (explanatory) effects
Funding: This work was supported by the Department of Veterans Affairs, Veterans Health Administration, Office of Research and Development, Clinical Sciences Research and Health Services Research and Development Services (CRS 02-164 and HSR 05-101), and the National Cancer Institute (P30 CA014089, U01 CA93324, U01 CA93326, U01 CA93329, U01 CA93332, U01 CA93339, U01 CA93344 and U01 CA93348).
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Disclosure: The authors have no relevant financial conflicts of interest.
Abstract presented at the American Thoracic Society (ATS) International Conference; New Oreleans, LA; May, 2010