The National Health and Nutrition Examination Survey (NHANES) studies are cross-sectional, complex samples of the U.S. civilian non-institutionalized population conducted by the National Center for Health Statistics (NCHS) of the Centers for Disease Control and Prevention (CDC). Additional details on the NHANES survey, sampling methodologies, and design have previously been published (12–15). All participants provided written informed consent and the NCHS institutional review board approved each NHANES cycle. The data from the survey interview and physical examination within continuous NHANES (1999–2018, n = 102,956) were analyzed. Our analysis was limited to participants who aged 20 years or older with available malignancy information and family income to poverty ratio (PIR) data (n = 49,720; Figure 2).

The primary exposure variable was SES, which was assessed based on the PIR. The PIR is a ratio of self-reported household income that accounts for household income according to household or family size, household age composition, and year. Hence, PIR ≤ 1 entails that the household is below the poverty line, and conversely a PIR > 1 entails that the household is above the poverty line. We separated participants into 3 groups: low-income (i.e., at or below the poverty ratio), middle-income (i.e., above the PIR to <4), and high-income (i.e., PIR, ≥ 4) adults. A PIR of ≤1 means that a person is ≤100% below the federal poverty level/threshold. The middle- and high-income groups were assessed by eligibility for subsidies according to the Patient Protection Affordable Care Act (16).

The following variables were self-reported: age, sex, race/ethnicity, marital status, citizenship status, educational level, insurance status, alcohol status, smoking status, physical activity, and family income. Height and weight were measured using standard protocols. Race/ethnicity categories were defined as follows: non-Hispanic white, non-Hispanic black, Mexican American, and other. Body mass index (BMI) (weight in kilograms divided by height in meters squared) was calculated as a BMI of at least 25.0 and obesity as a BMI of at least 30.0. Educational attainment was categorized as less than high school graduate, high school graduate or a general educational development certificate, and greater than high school. Alcohol consumption (none/ <2 drinks per week/≥2 drinks per week), smoking status (non-smoker/former smoker/current smoker), physical activity (none/1–3 times per week/4 or more times per week) were measured through self-reported questionnaires.

All analyses used the NHANES sampling weights and accounted for the other aspects of the complex survey design. A value of p < 0.05 was used as a cutoff to indicate statistical significance. Statistical analysis was performed using both SPSS statistical software (Version 24, IBM corp.) and STATA Statistics/Data Analysis (Version 16.0, Stata Corp.).

Descriptive data are presented as mean ± standard deviation (SD) for continuous variables or numbers and their proportions for categorical variables. Chi-square test statistics were used to examine differences in categorical variables, and ANOVA was used to examine differences in normal continuous variables. The exact chi-square tests with Bonferroni's correction for multiple comparisons were applied to compare the prevalence rates of the cancers between the three groups. To visually illustrate changes in cancer prevalence during consecutive surveys, we calculated the prevalence of each outcome by descriptive statistics.

Multivariate logistic regression was used to determine associations between cancer, demographics, and risk factors. Model 1 included adjustment for age (40–59 vs. 20–39, 60+ vs. 20–39), race/ethnicity (white vs. black, Hispanic/Mexican vs. black, and other vs. black), marital status (married vs. not married), health insurance (covered vs. not covered), an education level (high school or equivalent vs. less than high school, greater than high school vs. less than high school), citizenship status (US citizenship vs. non-US citizenship), and PIR (PIR ≤1.0 vs. PIR 1.0–4.0, PIR ≥4.0 vs. PIR 1.0–4.0); Model 2 adjusted for model 1 variables plus an additional adjustment for BMI (25.0–29.9 vs. <25.0, ≥30.0 vs. <25.0), drinking status (<2 drinks/d vs. non-drinker, ≥2 drinks/d vs. non-drinker), smoking status (former smoker vs. non-smoker, current smoker vs. non-smoker), and physical activity (moderate vs. never, vigorous vs. never).

The prevalence of prostate cancer was higher in middle-income (3.61% [n = 464] vs. 1.83% [n = 84], p < 0.001) and high-income groups (3.36% [n = 227] vs. 1.83% [n = 84], p < 0.001) than in low-income group (Figure 3A). No statistically significant difference was found between lung cancer and income levels (Figure 3B). Compared with low-income group, a higher prevalence of colon and rectum cancer occurs in middle-income (0.87% [n = 112] vs. 0.48% [n = 22], p = 0.012) and high-income groups (0.89% [n = 58] vs. 0.48% [n = 22], p = 0.018; Figure 3C). However, only middle-income group remained statistically significant (p < 0.0167) after Bonferroni correction.

The prevalence of breast cancer was higher in middle-income (2.86% [n = 390] vs. 2.00% [n = 117], p < 0.001) and high-income participants (3.48% [n = 218] vs. 2.00% [n = 117], p < 0.001) than in low-income participants (Figure 4A). The prevalence of lung cancer was lower in high-income participants than middle-income participants (0.10% [n = 6] vs. 0.29% [n = 39], p = 0.014; Figure 4B). We found no statistically significant relationship between income levels and the prevalence of colon and rectum cancer (Figure 4C).

In the high-income group, the prevalence of prostate cancer was increased from 2.952% (n = 96) in 1999–2008 to 3.967% (n = 131) in 2009–2018 (p = 0.025). Lung cancer prevalence was decreased from 0.246% (n = 8) in 1999–2008 to 0.151% (n = 5) in 2009–2018 (p = 0.389). The prevalence of colon and rectum cancer was slightly decreased from 1.046% (n = 34) in 1999–2008 to 0.727% (n = 24) in 2009–2018 (p = 0.168) (Figure 5 and Supplementary Figure S1).

In the middle-income group, the prevalence of prostate cancer in 1999–2008 (3.748%, n = 234) is basically the same trend as observed in 2009–2018 (3.477%, n = 230; p = 0.411). For lung cancer, the prevalence was the same in 1999–2008 (0.384%, n = 24) and 2009–2018 (0.438%, n = 29; p = 0.633). Colon and rectum cancer prevalence was non-significantly decreased from 0.929% (n = 58) in 1999–2008 to 0.816% (n = 54) in 2009–2018 (p = 0.492; Figure 5).

Below the federal poverty level, the prevalence of prostate cancer was increased from 1.562% (n = 31) in 1999–2008 to 2.041% (n = 53) in 2009–2018 (p = 0.232). The prevalence of lung cancer was decreased from 0.302% (n = 6) in 1999–2008 to 0.231% (n = 6) in 2009–2018 (p = 0.640). The prevalence of colon and rectum cancer in 1999–2008 (0.453%, n = 9) was comparable to that in 2009–2018 (0.500%, n = 13; p = 0.820; Figure 5).

For the high-income stratum, the prevalence of breast cancer was increased from 3.289% (n = 103) in 1999–2008 to 3.679% (n = 115) in 2009–2018 (p = 0.400); the prevalence of colon and rectum cancer was increased from 0.415% (n = 13) in 1999–2008 to 0.672% (n = 21) in 2009–2018 (p = 0.167). There were no significant trends for the prevalence of lung cancer between the two groups (p = 0.998; Figure 6).

For the middle-income stratum, breast cancer prevalence was increased from 2.743% (n = 184) in 1999–2008 to 2.977% (n = 206) in 2009–2018 (p = 0.414) and colon and rectum cancer from 0.760% (n = 51) in 1999–2008 to 0.910% (n = 63) in 2018 (p = 0.336). In contrast, lung cancer prevalence was decreased from 0.373% (n = 25) in 1999–2008 to 0.202% (n = 14) in 2009–2018 (p = 0.063; Figure 6).

For the low-income stratum, the prevalence of breast cancer was increased from 1.743% (n = 45) in 1999–2008 to 2.207 (n = 72) in 2009–2018 (p = 0.208), lung cancer from 0.194% (n = 5) in 1999–2008 to 0.276 % (n = 9) in 2009–2018 (p = 0.523), and colon and rectum cancer from 0.620% (n = 16) in 1999–2008 to 0.828% (n = 27) in 2009–2018 (p = 0.355; Figure 6).

Adjusting the models for demographic variables in men, the odds of reporting lung cancer were reduced in the highest resource population over time (odds ratio [OR], 0.533; 95% CI, 0.281–1.013; p = 0.055). While, no significant change was observed in prostate cancer (OR, 0.915; 95% CI, 0.765–1.095; p = 0.332) or colon and rectum cancer (OR, 1.104; 95% CI, 0.780–1.561; p = 0.577; Supplementary Tables S1–S3).

In models adjusted for demographics of women, the high-income group had higher odds of reporting breast cancer (OR, 1.203; 95% CI, 1.001–1.446; p = 0.049). Conversely, the odds of reporting lung cancer (OR, 0.308; 95% CI, 0.126–0.753; p = 0.010) and colon and rectum cancer (OR, 0.683; 95% CI, 0.453–1.030; p = 0.069) were reduced (Supplementary Tables S4–S6). When cancer risk factors were included in the model, the odds of high-income group reporting colon and rectum cancer (OR, 0.225; 95% CI, 0.087–0.582; p = 0.002) remained low over time, but no statistically significant variation in the odds of reporting breast cancer (OR, 1.105; 95% CI, 0.839–1.455; p = 0.478) or lung cancer (OR, 0.472; 95% CI, 0.085–2.641; p = 0.393) (Supplementary Tables S10–S12).

Both logistic regression analysis models suggest that, in general, older ages tended to be associated with a greater likelihood of reporting cancers. The ORs of cancer ranged from 6.145 (95% CI, 2.558–14.763) to 13.072 (95% CI, 1.634–104.581) for40–59 years age group and from 15.722 (3.749–65.930) to 92.179 (12.182–697.521) for the oldest age group (60+ years) as compared to the youngest age group (20–39 years) (Supplementary Tables S1–S12).

In the fully adjusted model of women (model 2), married vs. non-married individuals had a lower probability of reporting a colon and rectum cancer (OR, 0.567; 95% CI, 0.330–0.977; p = 0.041). In prostate cancer and breast cancer, health insurance covered vs. not covered participants had higher odds of reporting cancers (OR ranged from 1.912 [95% CI, 1.140–3.205] to 3.255 [95% CI, 1.926–5.503]) in both model 1 and model 2. Those with US citizenship had a higher probability of reporting breast cancer when compared with those without US citizenship (model 1: OR, 1.591 [95% CI, 1.041–2.432]; model 2: OR, 1.859 [95% CI, 1.017–3.398]) (Supplementary Tables 1–12).

In the first model of men, as compared to black participants, Hispanic and Mexican participants had a lower probability of reporting prostate cancer (OR, 0.270; 95% CI, 0.188–0.388), lung cancer (OR, 0.073; 95% CI, 0.010–0.564), and colon and rectum cancer (OR, 0.394; 95% CI, 0.181–0.858; Supplementary Tables S1–S3). When cancer risk factors were included in the second model, the odds of Hispanic and Mexican participants reporting prostate cancer (OR, 0.270; 95% CI, 0.135–0.541; p < 0.001) remained low over time, but no statistically significant variation was reported in the odds of reporting lung cancer (p = 0.983) or colon and rectum cancer (p = 0.078) (Supplementary Tables S7–S9).

In the first model of women, as compared to black participants, white participants had a higher probability of reporting breast cancer (OR, 1.550; 95% CI, 1.252–1.919) and colon and rectum cancer (OR, 1.629; 95% CI, 1.095–2.422; Supplementary Tables S4, S6). In a multivariable model adjusted for cancer risk factors (model 2), the association was still statistically significant (OR ranged from 1.849 [95% CI, 1.310–2.609] to 2.167 [95% CI, 1.088–4.314]) (Supplementary Tables S10, S12).

In the first model, people with a college degree or above are more likely to have prostate cancer (OR, 1.312; 95% CI, 1.079–1.596) and breast cancer (OR, 1.293; 95% CI, 1.049–1.593) than those without a high school diploma or the General Educational Development (GED) certificate (Supplementary Tables S1, S4). This association no longer reached statistical significance after adjustment for cancer risk factors (Model 2) (Supplementary Tables S7, S10).

The prevalence of prostate cancer was higher in middle (3.61% [n = 464]) and high-income group (3.36% [n = 227]) than in low-income group (1.83% [n = 84], all ps < 0.001). In the high-income group, the prevalence of prostate cancer was increased from 2.952% in 1999–2008 to 3.967% in 2009–2018 (p = 0.025). Health insurance covered vs. not covered participants had higher odds of reporting prostate cancer in both model 1 and model 2. People with a college degree or above are more likely to have prostate cancer than those without a high school diploma or GED certificate. Prostate cancer incidence has been robustly correlated with markers of access to care in multiple studies: regions with higher income and educational attainment have higher prostate cancer incidence, which is attributable to increased use of prostate-specific antigen testing (17–20). Prostate carcinoma was positively associated with income, being married, coffee consumption and physical activities from a previously conducted case-control study in Taiwan (21). The results from the National Prostate Cancer Register of Sweden suggested that men with prostate cancer were more often married, educated, and wealthier when compared with men in the control group (22). Accordingly, in low—and middle-income countries, the prognosis of cancer patients is usually poor, because when compared with patients in high-income countries, they have relatively low awareness of cancer, late diagnosis, and lack or unfair access to affordable treatment services (23). These all contribute to a considerable reduction in survival and thus affect prevalence rates.

The prevalence of breast cancer was higher in middle-income (2.86% [n = 390]) and high-income participants (3.48% [n = 218]) than in low-income participants (2.00% [n = 117], all ps < 0.001). People with a college degree or above are more likely to have breast cancer than those without a high school diploma or GED certificate. Similarly, a study from the University of Texas MD Anderson Cancer Center showed that participants were relatively highly educated and wealthier than the national average, which may reflect the epidemiology of breast cancer (24). It has also been reported that factors related to low education and low income are a greater obstacle to participating in clinical trials (24, 25). People with a high school degree or lower were less likely to get screened than those with at least a bachelor's degree.

Our study also shows that white participants had a higher probability of reporting breast cancer as compared to black participants in demographic variables adjusted and risk factors adjusted models. Likewise, Rafeek et al. also found that non-white participants were less willing to receive molecular tests or tumor biopsies, even if they were financially secure and used to guide them to use approved drugs for treatment (24). If widespread, these attitudes are likely to lead to a lower detection rate in cancer outcomes.

In our study, after adjusting the models for demographic variables in men, the odds of reporting lung cancer were reduced in the highest resource population over time. We also found that the prevalence of lung cancer in women was lower in high-income participants than middle-income participants in unadjusted and adjusted models. Similarly, a recent paper published by Patel et al. analyzed the incidence trend of lung cancer in California over the past 28 years and found that the increase of female lung adenocarcinoma was more obvious in areas with low SES in the community (26). Consistent with our study, some studies have shown that the risk of lung cancer is negatively correlated with SES factors, such as education, income, and occupation (27–29). SES is related to health status in many ways, such as social resources, physical and psychosocial stressors, and health-related behaviors.

Similar findings were reported in our study, when compared with a low-income group of men (0.48%), a higher prevalence of colon and rectum cancer occurs in middle-income (0.87%, p = 0.012) and high-income groups (0.89%, p = 0.018). Health insurance covered vs. not covered participants had higher odds of reporting colon and rectum cancer in model 1. Indeed, multiple studies have shown that lack of insurance and other socioeconomic factors have been associated with lower colorectal cancer screening rates (30–37). Therefore, when compared with low-income people, high-income people may have more rational health behavior, better understand their symptoms, and better communicate with medical staff. In consequence, the former may have a higher chance of early cancer detection. Based on these considerations, the prevalence of colon and rectum cancer is not the actual prevalence, which could be reasonably considered to be the detection rate. Data from the Behavioral Risk Factor Surveillance System show that the screening rates are persistently low among low-income individuals, people with medical subsidies or no medical insurance, adults with short years of education, ethnic minorities, and residents in rural areas (38, 39). Therefore, we now recognize that the relationship between SES and health may reflect two-way causality (i.e., from better health to higher SES and from higher SES to better health).

Our study has several limitations. First, because this was primarily a cross-sectional study, causality could not be ascertained. Second, the data rely exclusively on self-reported information, which can lead to underreporting or overreporting. In the context of low- and middle income, the reported incidence rate is more susceptible to the bias in the case determination. In the context of high income, the incidence rate of the report is more reliable. However, studies have illustrated that the self-report results of NHANES are reliable and valid (40). Third, the results may not be mature enough due to the rare patients with cancer. These results remained highly interpretable and were still displayed. Further studies investigating this subject and confirming our findings are therefore needed.