Sociodemographic Survival Disparities for Lung Cancer in the United States, 2000-2016

Abstract

Background

Understanding the impact of patient and tumor characteristics on lung cancer survival can help build personalized prognostic models and identify health disparities.

Methods

We identified 557 555 patients aged 25 years and older diagnosed with lung or bronchus carcinoma from the Surveillance, Epidemiology, and End Results database, 2000-2016. We estimated hazard ratios (HR) for demographic (sex, age, race and ethnicity), tumor (stage, histology, year of diagnosis), and geographic characteristics (census tract–level urbanicity, socioeconomic status [SES]), as well as selected interactions, on the rate of lung cancer–specific death using multivariable proportional hazards models.

Results

Women had a higher survival (lower hazard) of lung cancer–specific death than men (HR = 0.83, 95% confidence interval [CI] = 0.82 to 0.83). Hazards differed by race and ethnicity. Regional (HR = 2.41, 95% CI = 2.37 to 2.44) and distant (HR = 6.61, 95% CI = 6.53 to 6.69) tumors were associated with a lower survival (higher hazard) than localized tumors. Small cell tumors were associated with a lower survival (HR = 1.19, 95% CI = 1.18 to 1.20) than non–small cell tumors. Patients diagnosed after 2009 had lower hazards (HR = 0.86, 95% CI = 085 to 0.86) than those diagnosed 2000-2009. Lung cancer–specific survival did not depend on urbanicity after adjusting for census tract–level SES, but survival decreased with decreasing census tract–level SES. Differences in survival between non-Hispanic Black and White patients were greater for younger patients and localized tumors and increased with census tract–level SES. Differences by sex were greatest for young patients and localized tumors.

Conclusions

Disparities in survival after lung cancer diagnosis remain, with intersectional patterns suggesting differential access to and quality of care. Efforts are needed to ensure that high-risk groups receive guideline-concordant treatment.

Lung cancer remains the leading cause of cancer-related death in the United States (1). Though lung cancer incidence and mortality rates have been declining in the United States since their peaks (approximately 1990 for men and 2000 for women), largely because of reductions in smoking (2), lung cancer still has a comparatively low 5-year survival rate of 19% (1,3). Previous research has identified many factors to be prognostic for survival in lung cancer, including tumor stage, histology (especially small cell lung cancer vs non–small cell lung cancer), various biomarkers, comorbidities, and behavioral risk factors like smoking (4-8).

Researchers have also identified many factors driving health disparities in lung cancer survival, particularly by race and ethnicity (9). Studies looking for differences in mutation frequency by ancestry markers have not consistently identified biological explanations for survival differences (10-13). Instead, differences in survival by race and ethnicity, which are social constructs, are likely driven in large part by inequal access to and quality of care, which may then result in differences in stage at diagnosis and thus survival (14). Indeed, racial disparities disappear in some populations with universal medical coverage (15). The picture is far from simple, though, with patterns of comorbidities also playing a role, particularly in overall survival (16,17). Beyond racial and ethnic disparities, it has long been known that men have been shown to have worse lung cancer survival than women, even after controlling for smoking (18-20), and prognosis, treatment, and survival can vary widely geographically (21,22).

Estimates of survival disparities are usually measured independently and do not account for interactions between different patient characteristics. However, cancer survival may depend on the interaction between multiple patient and tumor characteristics. For example, is survival for Black women well approximated by combining independent estimates for Black people and women overall, or is it distinct? Are racial and ethnic disparities consistent across stage of diagnosis? To better understand disparities in lung cancer survival, we estimated how survival disparities by race and ethnicity and by sex differed across patient demographic (sex, age, race and ethnicity), tumor (stage, histology, year of diagnosis), and geographic (census tract–level urbanicity, socioeconomic status [SES]) characteristics. Here, we estimate disparities in lung cancer–specific survival to better understand the contributions of these characteristics to death because of lung cancer.

Methods

Data

The Surveillance, Epidemiology, and End Results (SEER) program at the National Cancer Institute collects information of cancer patients from 18 cancer registries, covering approximately one-third of the US population (23). We obtained lung cancer data from the SEER program from 2000 through 2016 (SEER 18) linked to census tract–level SES and urbanicity variables using the SEER*Stat program (24). We limited our sample to primary carcinomas in the lung and bronchus (International Classification of Diseases–10 Clinical Modification C34.0-C34.9, with International Classification of Diseases Oncology–3 histological codes 8000-8579). Additionally, we restricted the dataset to include patients aged 25 years or older and who survived at least 1 month beyond diagnosis, resulting in a dataset of 557 555 lung cancer patients. The median follow-up among those who did not die of their cancer or other causes was 37 months. A comparison of follow-up by covariate is given in Supplementary Table 1 (available online) (with a sensitivity analysis of main results given in Supplementary Table 2, available online, censoring all patients at 18 months). In this analysis, we censored the sample at 5 years (60 months) to concentrate on 5-year survival and to minimize the impact of potential time-varying effects of covariates on survival (25).

Patient sex was characterized by SEER from medical records. Race and ethnicity of patients was characterized by SEER as Hispanic, non-Hispanic American Indian and Alaskan Native, non-Hispanic Asian and Pacific Islander, non-Hispanic Black, non-Hispanic Other or Unknown (not one of the above or not specified), and non-Hispanic White. SEER abstracts race from medical records and uses the North American Association of Central Cancer Registries Hispanic Identification Algorithm to determine Hispanic ethnicity. We grouped small cell vs non–small cell lung cancers by International Classification of Diseases Oncology–3 histological codes. Small cell lung cancer included histologic codes 8002 and 8041-8045; non–small cell lung cancer included histologic codes 8000-8001, 8003-8040, and 8046-8579 (26). We used SEER summary stage (2000) to categorize patent tumor stage as localized, regional, distant, or unknown. We distinguished between tumors diagnosed in 2000-2009 vs those diagnosed in 2010-2016. Census tract urbanicity variables were fixed by SEER at 2000 levels for tumors diagnosed 2000-2005 and at 2010 levels for tumors diagnosed 2006-2016; census tracts were categorized as all urban (100% urban, ie, 100% of population in the census tract is living in an urban area), mostly urban (50%-99% urban), mostly rural (1%-49% urban), all rural (0% urban), and unknown. Census tract SES indices were estimated by SEER through data provided by the Census American Community Survey, and quintiles of the SES indices were created by SEER to have equal populations in each quintile (in the full registry). An analysis of missingness by covariate is given in Supplementary Table 3 (available online).

Statistical Analysis

Our outcome measure was time from lung cancer diagnosis to lung cancer–specific death [death attributed to the primary tumor (27)]. Lung cancer is caused primarily by smoking, and lung cancer patients often suffer from smoking-related comorbidities that put them at a higher risk of death because of causes other than lung cancer. Accordingly, estimating disparities for overall survival may be misleading because of the competing risks of cancer-specific mortality and other-cause mortality (28). Thus, we considered death from other causes to be a separate, competing risk (a censoring event from the perspective of cancer-specific death). We grouped our potential predictors of cancer survival into patient demographic (sex, age, race and ethnicity), tumor (stage, small cell vs non–small cell histology, year of diagnosis), and geographic (census tract–level urbanicity and SES quintile) characteristics of lung cancer patients.

We used Cox proportional hazards models to the estimate instantaneous hazard of lung cancer–specific death and the impact of the above covariates on lung cancer–specific survival. We assessed the proportional hazards assumption by plotting the Schoenfeld residuals against survival time for each covariate in the model. We estimated hazard ratios (HRs) for the covariates in univariable, bivariable (accounting for tumor stage or for SES), and multivariable (fully adjusted) models. We also considered interactions between race and ethnicity and each of the other variables (investigating 1 interaction at a time but accounting for all the other variables) and between sex and the other variables. Patients with missing data (ie, those with an unknown level of a covariate) are not excluded; hazard ratios for those with unknown values are computed but not reported because they are not interpretable. The analyses were done in R (R Foundation for Statistical Computing, Vienna, Austria) version 4.0 using the survival package (29,30). All P values are for the statistical significance of hazard ratios, ie, the effect of the covariate on cancer-specific survival in the given Cox proportional hazards model, or for a ratio of hazard ratios, and and all P values represent 2-sided comparisons. P values less than .05 were considered statistically significant.

Results

Patient demographic, tumor, and geographic characteristics of our analysis sample are reported in Table 1. Cumulative incidence plots for the population overall and by covariate are given in Figure 1. The underlying risk table is given in Supplementary Table 4 (available online). Analogous figures and tables by each level of sex and race and ethnicity are given in Supplementary Figures 1-7 (available online) and Supplementary Tables 5-11 (available online).

Figure 1.

Cumulative incidence of lung cancer–specific death overall and by covariate. A) Cumulative incidence distinguishes between the fraction of individuals who are alive, who have died from their cancer, or who have died from other causes. The subsequent plots present fraction of the population that has died from their cancer by (B) sex, (C) age at diagnosis, (D) race and ethnicity, (E) tumor stage, (F) histology, (G) era of diagnosis, (H) census tract–level urbanicity, and (I) census tract–level socioeconomic status quintile. A/PI =Asian and Pacific Islander; AI/AN = American Indian and Alaskan Native; NH = non-Hispanic.

Table 1.

Number of patients diagnosed with lung or bronchus carcinoma in the SEER cancer registry (2000-2016) by patient demographic, tumor, and geographic characteristic^a

Characteristics	No. (%)
All	557 555 (100)
Demographic
Sex
Men	294 998 (52.9)
Women	262 557 (47.1)
Age, y
Younger than 50	31 240 (5.6)
50-59	97 181 (17.4)
60-69	168 882 (30.2)
70-79	170 720 (30.6)
80 and older	89 532 (16.1)
Race and ethnicity
Hispanic	30 152 (5.4)
NH American Indian and Alaskan Native	1924 (0.3)
NH Asian and Pacific Islander	33 123 (5.9)
NH Black	63 362 (11.4)
NH other or unknown	731 (0.1)
NH White	428 263 (76.8)
Tumor
Tumor stage
Localized	98 123 (17.6)
Regional	133 212 (23.9)
Distant	295 854 (53.1)
Unknown	30 366 (5.5)
Histology
Non–small cell	486 601 (87.3)
Small cell	70 954 (12.7)
Era of diagnosis, y
2000-2009	333 322 (59.7)
2010-2016	224 233 (40.2)
Geographic, census tract
Urbanicity
All urban	339 688 (60.9)
Mostly urban	106 017 (19.0)
Mostly rural	39 706 (7.1)
All rural	38 648 (6.9)
Unknown	33 496 (6.0)
Socioeconomic status quintile
1 = lowest	109 402 (19.6)
2	112 841 (20.2)
3	108 564 (19.5)
4	101 580 (18.2)
5 = highest	85 928 (15.4)
Unknown	39 240 (7.0)

a

NH = non-Hispanic; SEER = Surveillance, Epidemiology, and End Results.

Hazard of Lung Cancer–Specific Death

All patient demographic, tumor, and geographic characteristics were statistically significantly associated with lung cancer–specific survival in our univariable models (Table 2). Except for 2 racial and ethnic categories (non-Hispanic American Indian and Alaskan Native and non-Hispanic Black), all variables were significantly associated with lung cancer–specific survival in the bivariable models adjusting for stage. In the bivariable models adjusting for census tract–level SES, all variables were significantly associated with lung cancer–specific survival except for 2 racial and ethnic categories (Hispanic and non-Hispanic American Indian and Alaskan Native) and many of the levels of urbanicity. Except for non-Hispanic American Indian and Alaskan Native and non-Hispanic Black race and ethnicity and all rural urbanicity, all other variables were significantly associated with survival in the multivariable models, which accounted for all characteristics simultaneously.

Table 2.

Hazard ratios (HRs) and confidence intervals (CIs) for patient demographic, tumor, and geographic characteristics in proportional hazards models for lung cancer–specific survival^a

	Univariable		Bivariable (with tumor stage)		Bivariable (with SES)		Multivariable
Characteristic	HR (95% CI)	Pb	HR (95% CI)	Pb	HR (95% CI)	Pb	HR 95% CI	Pb
Demographic
Sex
Male	1.00 (Referent)	(Referent)	1.00 (Referent)	(Referent)	1.00 (Referent)	(Referent)	1.00 (Referent)	(Referent)
Female	0.82 (0.81 to 0.82)	<.001	0.84 (0.83 to 0.84)	<.001	0.82 (0.82 to 0.83)	<.001	0.83 (0.82 to 0.83)	<.001
Age, y
Younger than 50	0.84 (0.83 to 0.85)	<.001	0.70 (0.69 to 0.71)	<.001	0.83 (0.82 to 0.84)	<.001	0.69 (0.68 to 0.70)	<.001
50-59	0.91 (0.90 to 0.92)	<.001	0.79 (0.78 to 0.79)	<.001	0.90 (0.89 to 0.91)	<.001	0.77 (0.76 to 0.77)	<.001
60-69	0.90 (0.90 to 0.91)	<.001	0.85 (0.84 to 0.85)	<.001	0.90 (0.89 to 0.90)	<.001	0.83 (0.83 to 0.84)	<.001
70-79	1.00 (Referent)		1.00 (Referent)		1.00 (Referent)		1.00 (Referent)
80 and older	1.29 (1.28 to 1.20)	<.001	1.29 (1.28 to 1.30)	<.001	1.30 (1.29 to 1.31)	<.01	1.34 (1.33 to 1.35)	<.001
Race and ethnicity
Hispanic	1.02 (1.01 to 1.03)	.01	0.93 (0.92 to 0.95)	<.001	0.99 (0.98 to 1.01)	.70	0.94 (0.93 to 0.95)	<.001
NH American Indian and Alaskan Native	1.07 (1.01 to 1.12)	.02	1.03 (0.98 to 1.09)	.22	1.04 (0.99 to 1.10)	.14	1.05 (0.99 to 1.10)	.08
NH Asian and Pacific Islander	0.90 (0.89 to 0.92)	<.001	0.77 (0.76 to 0.78)	<.001	0.93 (0.91 to 0.94)	<.001	0.82 (0.81 to 0.83)	<.001
NH Black	1.09 (1.08 to 1.20)	<.001	1.00 (0.99 to 1.01)	.50	1.03 (1.02 to 1.04)	<.001	1.01 (1.00 to 1.02)	.13
NH White	1.00 (Referent)	(Referent)	1.00 (Referent)	(Referent)	1.00 (Referent)	(Referent)	1.00 (Referent)	(Referent)
Tumor
Tumor stage
Localized	1.00 (Referent)	(Referent)	—	—	1.00 (Referent)	(Referent)	1.00 (Referent)	(Referent)
Regional	2.39 (2.36 to 2.43)	<.001	—	—	2.38 (2.35 to 2.41)	<.001	2.41 (2.37 to 2.44)	<.001
Distant	4.68 (4.60 to 4.76)	<.001	—	—	6.43 (6.43 to 6.51)	<.001	6.61 (6.53 to 6.69)	<.001
Histology
Non–small cell	1.00 (Referent)	(Referent)	1.00 (Referent)	(Referent)	1.00 (Referent)	(Referent)	1.00 (Referent)	(Referent)
Small cell	1.52 (1.50 to 1.53)	<.001	1.16 (1.15 to 1.16)	<.001	1.50 (1.49 to 1.51)	<.001	1.19 (1.18 to 1.20)	<.001
Era of diagnosis, y
2000-2009	1.00 (Referent)	(Referent)	1.00 (Referent)	(Referent)	1.00 (Referent)	(Referent)	1.00 (Referent)	(Referent)
2010-2016	0.86 (0.85 to 0.87)	<.001	0.85 (0.85 to 0.86)	<.001	0.86 (0.85 to 0.86)	<.001	0.86 (0.85 to 0.86)	<.001
Geographic (census tract)
Urbanicity
All urban	1.00 (Referent)	(Referent)	1.00 (Referent)	(Referent)	1.00 (Referent)	(Referent)	1.00 (Referent)	(Referent)
Mostly urban	1.01 (1.00 to 1.02)	<.001	1.03 (1.02 to 1.04)	<.001	1.01 (1.00 to 1.02)	.11	1.02 (1.01 to 1.02)	<.001
Mostly rural	1.03 (1.02 to 1.04)	<.001	1.05 (1.04 to 1.06)	<.001	1.00 (0.98 to 1.01)	.61	1.02 (1.01 to 1.03)	.003
All rural	1.08 (1.07 to 1.10)	<.001	1.09 (1.08 to 1.10)	<.001	1.02 (1.00 to 1.03)	.01	1.01 (1.00 to 1.03)	.05
SES quintile
1 = lowest	1.30 (1.29 to 1.31)	<.001	1.28 (1.27 to 1.29)	<.001	—	—	1.29 (1.28 to 1.30)	<.001
2	1.24 (1.22 to 1.25)	<.001	1.23 (1.21 to 1.24)	<.001	—	—	1.22 (1.21 to 1.24)	<.001
3	1.17 (1.16 to 1.18)	<.001	1.17 (1.16 to 1.18)	<.001	—	—	1.16 (1.15 to 1.18)	<.001
4	1.10 (1.09 to 1.12)	<.001	1.10 (1.09 to 1.12)	<.001	—	—	1.10 (1.09 to 1.11)	<.001
5 = highest	1.00 (Referent)	(Referent)	1.00 (Referent)	(Referent)	—	—	1.00 (Referent)	(Referent)

a

The univariable models are not adjusted for other covariates, the bivariable models are adjusted for tumor stage or census tract–level SES, and the multivariable model is adjusted for all covariates. HR > 1 implies a lower lung cancer–specific survival than the referent group. A dash indicates that it is not applicable. NH = non-Hispanic; SES = socioeconomic status.

b

P values indicate whether hazard ratio are significantly different from 1 (2-sided) in the specified Cox proportional hazards model at a level of significance of .05.

In the multivariable model, women had longer lung cancer–specific survival than men (ie, lower hazard, HR = 0.83, 95% confidence interval [CI] = 0.82 to 0.83). The age gradient in survival was in the expected direction, with patients diagnosed younger than 50 years having greater survival (lower hazard, HR = 0.69, 95% CI = 0.68 to 0.70) compared with those diagnosed between ages 70 and 79 years and with patients aged 80 years and older having a much lower lung cancer–specific survival (higher hazard, HR = 1.34, 95% CI = 1.33 to 1.35). There were no significant disparities for non-Hispanic American Indiana and Alaskan Native and non-Hispanic Black patients after adjusting for the other variables. Non-Hispanic Asians and Pacific Islander patients had a greater survival than non-Hispanic White patients (HR = 0.82, 95% CI = 0.81 to 0.83), as did Hispanic patients (HR = 0.94, 95% CI = 0.93 to 0.95).

Tumor stage, as expected, had the largest impact on cancer survival, with regional (HR = 2.41, 95% CI = 2.37 to 2.44) and distant (HR = 6.61, 95% CI = 6.53 to 6.69) tumors having substantially lower lung cancer–specific survival than localized tumors. Tumors with small cell histology were associated with a lower survival (HR = 1.19, 95% CI = 1.18 to 1.20) compared with tumors with non–small cell histology. Patients diagnosed after 2009 had longer survival times (HR = 0.86, 95% CI = 0.85 to 0.86) compared with those diagnosed 2000-2009.

After adjusting for the other variables (specifically SES, as the bivariable model demonstrated), there was little impact of urbanicity on cancer survival. Lung cancer–specific survival decreased as SES decreased (with HRs up to 1.29, 95% CI = 1.28 to 1.30, for the lowest vs highest SES quintiles).

Interactions Between Sex and Other Covariates

Hazard ratios for models estimating interactions between sex and one other variable while adjusting for the rest are given in Table 3. In the previous section, we found that men had a higher hazard of lung cancer–specific death and thus lower survival than women. Here, we see that there was generally an age gradient in this sex disparity, with a greater difference in survival (ie, the HR ratio is further from 1.0) between younger women and men compared with older patients, with a greater disparity (HR ratio farther from 1.0) for those aged younger than 50 years (HR ratio 0.81) compared with those aged 80 years and older (HR ratio 0.87). Although the difference in the HR ratio is not large, there is an impact at the population level: the excess lung cancer deaths in men compared with women among those aged 80 years and older would be 234 per 10 000 but 792 per 10 000 among those aged younger than 50 years. With the exception of non-Hispanic Asian and Pacific Islander patients, who have the largest survival disparity by sex (HR ratio = 0.75), the interactions between sex and race and ethnicity, though largely statistically significant, may not be clinically meaningful. Survival disparities by sex were greater for localized tumors (HR ratio = 0.74) compared with regional (0.83) and distant (0.84) tumors. There were no significant differences in lung cancer–specific survival disparities by sex by urbanicity or SES.

Table 3.

Hazard ratios (HRs) and confidence intervals (CIs) for patient demographic, tumor, and geographic characteristics in proportional hazards models for lung cancer–specific survival with an interaction between sex and the corresponding characteristic^a

Characteristic	Male	Female
	HR (95% CI)	HR (95% CI)	Ratiob	Pc
Demographic
Age, y
Younger than 50	0.70 (0.69 to 0.72)	0.57 (0.55 to 0.58)	0.81	<.001
50-59	0.79 (0.78 to 0.80)	0.62 (0.61 to 0.63)	0.79	<.001
60-69	0.85 (0.84 to 0.86)	0.69 (0.68 to 0.70)	0.82	<.001
70-79	1.00 (Referent)	0.85 (0.84 to 0.86)	0.85	(Referent)
80 and older	1.32 (1.30 to 1.34)	1.15 (1.14 to 1.17)	0.87	.003
Race and ethnicity
Hispanic	0.96 (0.95 to 0.98)	0.77 (0.75 to 0.78)	0.80	<.001
NH American Indian and Alaskan Native	1.03 (0.96 to 1.11)	0.90 (0.83 to 0.97)	0.87	.53
NH Asian and Pacific Islander	0.86 (0.84 to 0.87)	0.64 (0.63 to 0.66)	0.75	<.001
NH Black	1.02 (1.01 to 1.04)	0.83 (0.82 to 0.84)	0.81	<.001
NH White	1.00 (Referent)	0.84 (0.83 to 0.85)	0.84	(Referent)
Tumor
Tumor stage
Localized	1.00 (Referent)	0.74 (0.72 to 0.76)	0.74	(Referent)
Regional	2.28 (2.24 to 2.32)	1.89 (1.85 to 1.92)	0.83	<.001
Distant	6.23 (6.13 to 6.33)	5.21 (5.13 to 5.30)	0.84	<.001
Histology
Non–small cell	1.00 (Referent)	0.82 (0.81 to 0.83)	0.82	(Referent)
Small cell	1.15 (1.14 to 1.17)	1.01 (1.02 to 1.03)	0.88	<.001
Era of diagnosis, y
2000-2009	1.00 (Referent)	0.84 (0.83 to 0.85)	0.84	(Referent)
2010-2016	0.87 (0.86 to 0.88)	0.70 (0.70 to 0.71)	0.81	<.001
Geographic, census tract
Urbanicity
All urban	1.00 (Referent)	0.83 (0.82 to 0.84)	0.83	(Referent)
Mostly urban	1.02 (1.00 to 1.03)	0.84 (0.83 to 0.85)	0.83	.87
Mostly rural	1.02 (1.00 to 1.03)	0.85 (0.83 to 0.87)	0.83	.67
All rural	1.01 (0.99 to 1.03)	0.85 (0.83 to 0.86)	0.84	.52
Socioeconomic status quintile
1 = lowest	1.30 (1.28 to 1.32)	1.08 (1.06 to 1.09)	0.83	.33
2	1.24 (1.22 to 1.26)	1.02 (1.00 to 1.04)	0.82	.06
3	1.17 (1.15 to 1.18)	0.98 (0.96 to 0.99)	0.84	.73
4	1.11 (1.09 to 1.13)	0.92 (0.90 to 0.93)	0.83	.17
5 = highest	1.00 (Referent)	0.84 (0.83 to 0.85)	0.84	(Referent)

a

Each model is adjusted for the remaining variables, but only 1 interaction is investigated in each model. NH = non-Hispanic.

b

The ratio column indicates the ratio between the hazard ratio for women vs men for each level of the covariate (eg, among patients with distant tumors, the hazard ratio for women vs men is 5.21 to 6.23 = 0.84).

c

The P value indicates whether the female-to-male ratio for each level of the covariate is significantly different from 1 in the specified proportional hazard model (2-sided, at a level of significance of .05) from that of the referent level (eg, comparing the female-to-male ratio for patients with distant tumors (ratio = 0.84) to that of patients with localized tumors (ratio = 0.74) has a P value less than .01, indicating that the sex disparity is statistically significantly greater (HR further from 1) among patients with localized tumors compared with those with distant tumors.

Interactions Between Race and Ethnicity and Other Covariates

Hazard ratios for models estimating an interaction between race and ethnicity and one other variable while adjusting for the other variables are given in Table 4. We do not include non-Hispanic American Indian and Alaskan Native patients in this analysis because the smaller sample size (Supplementary Table 8, available online) led to no significant interactions.

Table 4.

Hazard ratios (HR) and confidence intervals (CI) for patient, tumor, and geographic characteristics in proportional hazards models for lung cancer–specific survival with an interaction between race and ethnicity and the corresponding characteristic^a

Characteristic	NH White	Hispanic			NH Asian and Pacific Islander			NH Black
	HR (95% CI)	HR (95% CI)	Ratiob	Pc	HR (95% CI)	Ratiob	Pc	HR (95% CI)	Ratiob	Pc
Demographic
Sex
Male	1.00 (Referent)	0.96 (0.95 to 0.98)	0.96	(Referent)	0.86 (0.84 to 0.87)	0.86	(Referent)	1.02 (1.01 to 1.04)	1.02	(Referent)
Female	0.84 (0.83 to 0.85)	0.77 (0.75 to 0.78)	0.91	<.001	0.64 (0.63 to 0.66)	0.76	<.001	0.83 (0.82 to 0.84)	0.99	<.001
Age, y
Younger than 50	0.69 (0.68 to 0.70)	0.56 (0.53 to 0.59)	0.81	<.001	0.55 (0.53 to 0.58)	0.80	.44	0.73 (0.70 to 0.75)	1.05	.01
50-59	0.77 (0.76 to 0.78)	0.70 (0.67 to 0.72)	0.91	.004	0.60 (0.58 to 0.62)	0.78	.02	0.79 (0.78 to 0.81)	1.03	.01
60-69	0.83 (0.83 to 0.84)	0.80 (0.78 to 0.82)	0.96	.67	0.66 (0.64 to 0.67)	0.79	.02	0.84 (0.82 to 0.85)	1.01	.49
70-79	1.00 (Referent)	0.97 (0.94 to 0.99)	0.97	(Referent)	0.82 (0.80 to 0.84)	0.82	(Referent)	1.00 (0.98 to 1.02)	1.00	(Referent)
80 and older	1.34 (1.32 to 1.35)	1.28 (1.24 to 1.33)	0.96	.67	1.18 (1.15 to 1.22)	0.89	<.001	1.27 (1.23 to 1.30)	0.95	.006
Tumor
Tumor stage
Localized	1.00 (Referent)	0.94 (0.89 to 1.00)	0.94	(Referent)	0.85 (0.80 to 0.90)	0.85	(Referent)	1.16 (1.12 to 1.21)	1.16	(Referent)
Regional	2.43 (2.40 to 2.47)	2.25 (2.17 to 2.33)	0.93	.60	2.06 (1.98 to 2.13)	0.85	.95	2.57 (2.51 to 2.64)	1.06	<.001
Distant	6.73 (6.64 to 6.82)	6.33 (6.20 to 6.47)	0.94	.95	5.45 (5.34 to 5.56)	0.81	.13	6.67 (6.56 to 6.79)	0.99	<.001
Histology
Non–small cell	1.00 (Referent)	0.94 (0.92 to 0.95)	0.94	(Referent)	0.81 (0.80 to 0.82)	0.81	(Referent)	1.02 (1.01 to 1.03)	1.02	(Referent)
Small cell	1.19 (1.18 to 1.21)	1.14 (1.10 to 1.19)	0.96	.37	1.08 (1.03 to 1.13)	0.90	<.001	1.12 (1.08 to 1.15)	0.93	<.001
Era of diagnosis (year)
2000-2009	1.00 (Referent)	0.94 (0.92 to 0.95)	0.94	(Referent)	0.83 (0.82 to 0.85)	0.83	(Referent)	1.02 (1.01 to 1.03)	1.02	(Referent)
2010-2016	0.86 (0.86 to 0.87)	0.82 (0.80 to 0.84)	0.95	.32	0.69 (0.67 to 0.70)	0.80	.003	0.85 (0.83 to 0.86)	0.99	<.001
Geographic, census tract
Urbanicity
All urban	1.00 (Referent)	0.94 (0.92 to 0.95)	0.94	(Referent)	0.81 (0.80. 0.82)	0.81	(Referent)	1.00 (0.99 to 1.01)	1.00	(Referent)
Mostly urban	1.01 (1.00 to 1.02)	0.99 (0.95 to 1.03)	0.98	.03	0.88 (0.84 to 0.91)	0.87	.002	1.03 (1.00 to 1.06)	1.02	.22
Mostly rural	1.02 (1.00 to 1.03)	0.96 (0.87 to 1.05)	0.94	.91	0.92 (0.80 to 1.05)	0.90	.12	1.03 (0.98 to 1.08)	1.01	.71
All rural	1.01 (1.00 to 1.02)	0.90 (0.81 to 1.01)	0.89	.45	0.89 (0.75 to 1.05)	0.88	.33	1.03 (0.98 to 1.08)	1.02	.54
SES quintile
1 = lowest	1.30 (1.28 to 1.31)	1.21 (1.17 to 1.24)	0.93	.43	1.11 (1.06 to 1.15)	0.85	.03	1.30 (1.28 to 1.32)	1.00	.03
2	1.23 (1.21 to 1.24)	1.15 (1.12 to 1.19)	0.94	.61	0.97 (0.94 to 1.01)	0.79	.40	1.23 (1.21 to 1.26)	1.00	.04
3	1.17 (1.15 to 1.18)	1.11 (1.07 to 1.15)	0.95	.99	0.97 (0.94 to 1.00)	0.83	.15	1.15 (1.11 to 1.18)	0.98	.01
4	1.19 (1.09 to 1.12)	1.04 (1.00 to 1.08)	0.94	.73	0.89 (0.86 to 0.92)	0.81	.91	1.14 (1.10 to 1.18)	1.03	.30
5 = highest	1.00 (Referent)	0.95 (0.91 to 0.99)	0.95	(Referent)	0.81 (0.79 to 0.83)	0.81	(Referent)	1.06 (1.01 to 1.11)	1.06	(Referent)

a

Each model adjusted for the remaining variables, but only 1 interaction is investigated in each model. NH = non-Hispanic; SES = socioeconomic status.

b

The ratio columns indicate the ratio between the hazard ratio for that race and ethnicity group and that for NH White patients for each level of the covariate (eg, among patients with distant tumors, the hazard ratio for NH Blacks vs NH Whites is 6.67 to 6.73 = 0.99) .

c

The P value indicates whether the ratio for each level of the covariate is significantly different from 1 in the specified proportional hazard model (2-sided, at a level of significance of .05) from that of the referent level (eg, comparing the NH Black to NH White ratio for patients with distant tumors; ratio = 0.99) to that of patients with localized tumors (ratio = 1.16) has a P value less than .01, indicating that the race and ethnicity lung cancer–specific survival disparity is statistically significantly greater (hazard ratio further from 1) among patients with localized tumors compared with those with distant tumors.

The survival disparity between non-Hispanic Black and non-Hispanic White patients was not meaningful different between female (HR ratio = 0.99) and male patients (HR ratio = 1.02). Non-Hispanic Black patients had modestly worse survival than non-Hispanic White patients (HR ratio = 1.05) among those aged younger than 50 years, and the disparity shrunk with age, flipping direction for patients aged 80 years and older (HR ratio = 0.95). The survival disparity between non-Hispanic Black and non-Hispanic White patients depended greatly on tumor stages, with greatest differences for localized tumors (HR ratio = 1.16) and the least for distant tumors (0.99). There were no significant differences in lung cancer–specific survival between non-Hispanic Black and non-Hispanic White patients across the urbanicity gradient. However, non-Hispanic Black patients also had modestly worse survival compared with non-Hispanic White patients in higher SES census tracts (HR ratio = 1.06 in the highest quintile vs 1.00 in the lowest).

Corresponding analyses for Hispanic patients and non-Hispanic Asian and Pacific Islander vs non-Hispanic White patients may also be found in Table 4. Disparities vs non-Hispanic White patients decreased with age for both groups, but there were few other interactions that could be considered clinically meaningful.

Discussion

In this study, we estimated the association of patient demographic, tumor, and geographic characteristics on lung cancer survival in the United States and determined the extent to which differences in lung cancer–specific survival by sex and by race and ethnicity are intersectional and depend on other characteristics. The strongest predictors of survival were the tumor characteristics (stage, small-cell histology, and era of diagnosis), followed by patient demographic characteristics (age, sex, and race and ethnicity). The impacts of geographic, census tract–level characteristics were driven by SES. This work can help us understand health disparities more specifically as well as improve personalized predictions of lung cancer prognosis based on patient demographic and tumor characteristics and their interactions. Groups that currently have low lung cancer–specific survival can be targeted with risk reduction interventions, screening to enhance early detection, and increased attention to guideline-concordant treatment.

Many of our findings highlight the importance of access to and quality of care. Non-Hispanic Black patients are known to be diagnosed with more advanced tumors and to have differential receipt of any treatment and guideline-concordant treatment (9,14,31). In our bivariable analysis of the impact of race and ethnicity on survival, non-Hispanic Black patients were no different from non-Hispanic White patients after adjusting for tumor stage, emphasizing the importance of time to diagnosis in observed survival disparities. Community SES also plays an important role, as accounting for census tract–level SES in bivariable analyses reduced non-Hispanic Black vs non-Hispanic White patient disparities and increased the effect of tumor stage on cancer survival. Interestingly, the impact of non-Hispanic Black race and ethnicity on survival varied significantly (if generally modestly) by tumor stage, sex, age, and census tract–level SES. Disparities in early stage lung cancers have been highlighted by previous work (9,32,33). Disparities in lung cancer–specific survival between non-Hispanic Black and non-Hispanic White patients were greater for younger ages and localized tumor stage; these results may suggest that non-Hispanic Black patients are receiving poorer quality care when they would be considered lower risk cases. At the same time, the SES findings are suggestive of an issue of access: there are greater differences in non-Hispanic Black vs non-Hispanic White patients lung cancer survival in the most affluent census tracts, where one would expect the highest quality care to be located. Although we did not adjust for insurance status in this analysis, it has also been shown to be a significant source of disparity in cancer preventive care and treatment (34,35).

The patterns of racial and ethnic disparities across demographic, tumor, and geographic characteristics were somewhat different for each racial and ethnic group. Non-Hispanic Asian and Pacific Islander patients had a wider difference in survival by sex than any other racial and ethnic group. Hispanic patients had few differences compared with non-Hispanic White patients, except for by age. Racial and ethnic differences were greater for younger ages compared with older ages for Hispanic, non-Hispanic Asian and Pacific Islander, and non-Hispanic Black patients compared with non-Hispanic White patients. Although non-Hispanic American Indian and Alaskan Natives are a vulnerable group that deserves further study, the sample size in the SEER data did not support the same depth of analysis.

It has long been understood that there are complicated patterns by sex when it comes to smoking, lung cancer susceptibility, and survival (18–20). Women smoke less than men but may be more susceptible to lung cancer for a given level of exposure. Nevertheless, consistent with previous research, we find that women have longer lung cancer survival than men. Although the increased lung cancer–specific survival for women does differ statistically significantly across patient demographic and tumor characteristics (particularly for those at lowest risk of death, including younger ages and those with localized tumors), the ratio of female-to-male hazard ratios across other characteristics stayed within a fairly narrow band (all HR ratios were between 0.74 to 0.88). Unlike for disparities by non-Hispanic Black vs non-Hispanic White patients, we see no impact of SES, underscoring the more likely role of sex differences in smoking and comorbidities rather than access to and quality of care.

Our analysis found that increasing age at lung cancer diagnosis was strongly associated with lower survival, as expected. Previous work has demonstrated that increasing age is associated with differences in treatment, particularly surgery, which may reflect multiple considerations, such as perceived higher risk of complications or futility of treatment at advanced age (36,37). Increasing age is also associated with a higher burden of comorbid conditions (16), which can impact treatment eligibility and postoperative complications and mortality. Recent advances in lung cancer treatment have helped increase the duration of postdiagnostic survival (38–40). Although specific treatments were not included in this analysis, these advances are nonetheless broadly reflected in our results, with substantial decreases in the hazard of lung cancer–specific death in 2010-2016 compared with 2000-2009.

The SEER cancer registry provides a large sample of lung cancer patients as well as the opportunity to investigate lung cancer–specific rather than all-cause survival. However, no individual behavior (eg, smoking) or environmental exposure data are available, precluding the inclusion of modifiable risk factors into the analysis. Behaviors such as smoking status may confound the true associations between the nonmodifiable risk factors (sociodemographic and tumor characteristics) and survival in our models. Our work also did not specifically address treatment and interactions of treatment with patient demographic and tumor characteristics. Future work incorporating behavior, environmental exposure, genetic risk, and treatment is critical to improve our understanding of disparities in lung cancer survival and the differential impact of treatment improvements, access to care, and quality of care on lung cancer outcomes.

Our work provides a closer look at the impact of demographic patient, tumor, and geographic characteristics, and their interactions, on lung cancer–specific survival. A more granular understanding of health disparities can help improve awareness and draw attention to systematic issues in access to and quality of care.

Funding

This work was supported by the National Cancer Institute at National Institutes of Health (grant number U01CA253858 to RM and JJ).

Notes

Role of the funder: The funder had no role in the design of the study; the collection, analysis, and interpretation of the data; the writing of the manuscript; and the decision to submit the manuscript for publication.

Disclosures: The authors have nothing to disclose.

Author contributions: AFB: conceptualization, formal analysis, writing—review & editing. JME: conceptualization, data curation, formal analysis, writing—original draft. JJ: writing—review & editing. RM: writing—review & editing.

Data Availability Statement

Data were derived from a source in the public domain (23).