Anemia May Increase the Overall Risk of Cancer: Findings from a Cohort Study with a 12-Year Follow-up Period in South Korea

Cancer is the leading cause of death worldwide (1). During 2006 to 2016, globally, cancer accounted for 17.2 million patients and 8.9 million deaths, with an increase of 28% in overall cancer incidence (2). The incidence rate of new cancer cases per population is expected to increase in the future, making prevention key to reducing the associated global burden of the disease (3).

Anemia is a chronic condition in which a person's hemoglobin levels are lower than required by their physiological needs. It affects approximately one third of the world's population (4). Iron deficiency anemia (IDA) is the most common type of anemia (5). Some evidence suggests that anemia may increase the risk of cancer, as iron deficiency can lead to carcinogenesis. In fact, iron deficiency has been associated with an increased risk of esophageal (6, 7), gastric (8, 9), and colorectal (10, 11) cancers. A separate study has reported that the overall cancer risk was significantly increased among patients with IDA in the Taiwanese population (12); however, this study included patients with IDA defined based on the International Classification of Diseases (ICD) codes, making it likely that not all patients were captured (12). Moreover, other types of anemia such as hemolytic and aplastic anemia may also increase the overall risk of cancer (13, 14), suggesting that screening for anemia based on serum hemoglobin levels may be a suitable approach to cancer risk assessment. Nevertheless, no previous study has assessed cancer risk in individuals with anemia diagnosed based on hemoglobin levels that are evaluated as part of a national health screening program for the adult population. In South Korea, all adults may undergo standardized medical examination provided by the National Health Insurance Service (NHIS); serum hemoglobin level assessments are included in this standardized medical examination.

Using data from the South Korea NHIS, we aimed to investigate whether anemia was associated with an increased overall risk of cancer. In addition, we examined whether this association differed according to cancer type or anemia severity.

This study involved human participants and was conducted in accordance with the guidelines provided by the relevant ethics boards. The Institutional Review Board (X-1911-579-902) and the Health Insurance Review and Assessment Service in South Korea approved the study protocol. The informed consent requirement was waived due to the retrospective nature of the present study, which involved anonymized data derived from the NHIS database.

The National Health Screening Cohort (NHIS-HEALS) was the data source for this study (15). As the sole public insurance provider in South Korea, the NHIS collects information on the residents' demographic and socioeconomic characteristics, disease diagnoses (classified using the ICD-10 codes), and any prescribed treatments. NHIS subscribers aged ≥40 years are recommended to undergo standardized health examinations every 2 years (16); the associated findings are used to create the NHIS-HEALS database used for medical research. This cohort comprises 514,795 individuals that underwent standardized medical examinations between 2002 and 2003 and followed up until 2015. The database contains information regarding body mass index (BMI), laboratory test results including serum hemoglobin levels, and lifestyle parameters (exercise habits, alcohol consumption, and smoking status). In the present study, we included adult individuals who underwent a standardized medical examination during 2002 and 2003. However, individuals who died during 2002 and 2003 or had missing data on hemoglobin levels or had a history of any cancer type diagnosed before or during 2002 and 2003 were excluded from the analysis.

Included individuals were divided into “anemia” and “control” groups. Anemia was defined according to the World Health Organization (WHO) criteria as hemoglobin levels of <12 and <13 g/dL for women and men, respectively. The severity of anemia was categorized as mild (hemoglobin ≥ 11 g/dL), moderate (hemoglobin 8–10.9 g/dL), or severe (hemoglobin < 8 g/dL), according to the same criteria (17). Serum hemoglobin concentration was measured using the cyanmethemoglobin method. For individuals with two hemoglobin measurements obtained during 2002 and 2003, the level obtained in 2003 was used to diagnose and classify anemia.

Newly registered diagnoses of any malignancy (ICD-10 codes of C00-C96) recorded from January 1, 2004, to December 31, 2015 (12 years), were classified as cancer cases in this study, which were further categorized as follows: gastric (C16), esophageal (C15), colorectal (C18–C20), gall bladder and biliary tract (C23–C24), head and neck (C00–C14), brain (C71), liver (C22), pancreatic (C25), lung (C34), bone and articular cartilage cancers (C40–C41), neoplasms of the breast and genital organs (C50–C63), urinary tract (C64–C68), and thyroid (C73) cancers, and lymphoma or leukemia (C81–C96). The NHIS provides the diagnosis of neoplasms of the breast and genital organs (C50–C63) as one group of ICD-10 codes (C_), because it is considered sensitive information in the NHIS database. Therefore, prostate cancer, cervical cancer, and breast cancers were all included in this category. The time to cancer diagnosis was calculated from January 1, 2004, to the date of cancer diagnosis, as registered in the ICD-10 system. In South Korea, all patients diagnosed with any C-code cancer are registered in the NHIS database to receive financial coverage of 95% of cancer-related treatment costs. Therefore, it is expected that all patients diagnosed with cancer are registered in the NHIS database. In cases of double cancer diagnoses, we included only the first diagnosis in the analysis. For example, an individual diagnosed with breast cancer in 2006 and lung cancer in 2012 was included as a breast cancer case in the analysis of the overall cancer development risk and time to cancer diagnosis. However, for site-specific cancer analyses, each diagnosis was considered an event, and time to diagnosis was calculated for each event even in cases that involved multiple diagnoses. Accordingly, an individual diagnosed with breast cancer in 2006 and lung cancer in 2012 was included in time-to-event analyses for both breast and lung cancers.

The following information was collected as covariates for this study: demographic characteristics (age, sex, and BMI), socioeconomic status (residence and annual income level), presence of comorbidities (underlying disability; Charlson comorbidity index; presence of mild to moderate and severe liver disease), and lifestyle information (smoking status, alcohol consumption, and exercise frequency).

The place of residence was classified into three groups (Seoul, other metropolitan cities, and other areas). BMI was categorized into four groups (<18.5, 18.5–24.9, 25.0–29.9, and >30 kg/m²). The annual income level was categorized into five groups [0%–20% (lowest), 20%–40%, 40%–60%, 60%–80%, and 80%–100% (highest)], and the underlying disability was divided into two groups (mild to moderate, and severe disability). In South Korea, all physical disabilities must be registered in the NHIS for the affected individuals to receive various benefits. Disabilities were categorized into six levels based on their severity. Patients with disabilities of the first (most severe) through the third levels were classified as a “severe” disability group, and those with disabilities of the fourth through the sixth (mildest) levels were classified as a “moderate” disability group. Smoking status was divided into four groups: nonsmoker, previous smoker, current smoker, and unknown (no-response group). Alcohol consumption was divided into four groups [nondrinker, mild drinker, heavy drinker, and unknown (no-response group)]. The “mild” drinker group was defined as alcohol consumption of ≤70 g and ≤20 g per week in men and women, respectively; meanwhile, the “heavy” drinker group was defined as alcohol consumption of >70 g and >20 g per week in men and women, respectively. Exercise frequency was divided into six groups [no exercise, 1–2 times per week, 3–4 times per week, 5–6 times per week, exercise almost every day, and unknown (no-response group)]. The Charlson comorbidity index was calculated using the registered ICD-10 codes from 2002 to 2003 (Supplementary Table S1).

The participants' clinicopathologic characteristics were presented as the mean values with SD for continuous variables and counts with percentages for categorical variables. The incidence of cancer was presented per 1,000 person-years. First, we performed 1:1 propensity score (PS) matching between the anemia and control groups to reduce the impact of confounding (18). For PS matching, the nearest neighbor method was used without replacement with a caliper of 0.25. All covariates were adjusted for in the PS model including demographic characteristics, socioeconomic status–related information, presence of comorbidities, and lifestyle information, and logistic regression analysis was performed to calculate the PSs. To account for comorbidities, both the Charlson comorbidity index value and the underlying diseases considered by it were included in the PS model for adjustment. The absolute value of the standardized mean difference (ASD) was used to evaluate the balance between the groups before and after PS matching. The ASD was set at <0.1, to confirm adequate balance between the groups. Subsequently, we performed the Cox proportional hazards regression analysis to assess cancer incidence in the PS-matched cohort. In this time-to-event analysis, the “event” was cancer diagnosis, and the period of observation was set between January 1, 2004, and the date of cancer diagnosis. Next, we performed Cox regression analyses to assess the impact of anemia on the risk of each cancer type in the PS-matched cohort. In this analysis, the neoplasms of the breast and genital organs were dichotomized based on sex, as some of these cancers are sex-specific (e.g., ovarian and prostate cancer). Because Cox regression modeling is a time-to-event analysis, patients were censored due to death or emigration. We performed sensitivity analyses based on the duration of follow-up, as observation time may affect findings. In sensitivity analyses, the development of cancer was evaluated from January 1, 2004, to December 31, 2008 (5 years), and from January 1, 2004, to December 31, 2012 (9 years), separately.

In a separate sensitivity analysis, we investigated the association between anemia and cancer development during 2005 to 2015 (not 2004–2015) in the PS-matched cohort to avoid reverse causation bias, as for individuals diagnosed in 2004, the latency period was short (19). Finally, we constructed a multivariable Cox regression model for the development of cancer in the entire cohort to (i) determine whether the results obtained from the PS-matched cohort were generalizable to the entire cohort and to (ii) determine the risk of cancer in the anemia group after adjustments for other covariates important in this context. All covariates were included in the multivariate Cox model.

Using multivariable Cox regression modeling, we performed subgroup analyses to investigate whether mild, moderate, and severe anemia types were associated with the development of cancer. Another multivariable Cox regression model (subgroup analyses; model 2) was constructed to avoid multicollinearity with multivariable model 1, which was constructed to investigate whether anemia was associated with the risk of cancer in the entire cohort. We confirmed the absence of multicollinearity in all multivariable models that involved the entire cohort (variance inflation factor of <2.0). The results of the Cox regression were presented as HRs with 95% confidence intervals (CI). The C-statistic was used to identify the C-index of the multivariable Cox regression model. All statistical analyses were performed using R software (version 4.0.3, R packages, R Project for Statistical Computing). P values of <0.05 were considered indicative of a statistically significant finding.

A total of 514,795 individuals underwent a standardized medical examination between 2002 and 2003. Among them, 1,320 and 570 individuals died during 2002 and 2003 and lacked data on their serum hemoglobin levels, respectively, and were excluded. In addition, 2,887 and 55,714 individuals with cancer diagnosis made before or during 2002 and 2003 were excluded. Finally, a total of 454,304 adults were included in the analysis, and the mean follow-up time for cancer diagnosis was 11.6 years (SD, 1.7 years). In the final sample, 41,947 (9.2%) and 412,357 (90.8%) individuals were in the anemia and control groups, respectively. After PS matching, a total of 83,886 individuals (41,943 individuals per group) were included in the analysis (Fig. 1). The clinicopathologic characteristics of the participants before and after PS matching are presented in Table 1. After PS matching, all ASDs of covariates were of <0.1, suggesting that adequate balance was achieved through PS matching.

Table 2 presents cancer incidence data before and after PS matching. In the anemia group of the PS-matched cohort, cancer incidence during 2004 to 2015 was 397 per 1,000 person-years; the corresponding control group value was 389 per 1,000 person-years. Cox regression analysis revealed that the risk of cancer during 2004 to 2015 in the anemia group was 3% higher than that in the control group (HR, 1.03; 95% CI, 1.01–1.05; P = 0.023). Similarly, in the sensitivity analysis of cancer development during 2005 to 2015 (Supplementary Table S2), the risk of cancer in the anemia group was 3% higher than that in the control group (HR, 1.03; 95% CI, 1.00–1.05; P = 0.044). In sensitivity analyses according to the length of follow-up, the risk of cancer during 2004 to 2012 in the anemia group was 3% higher than that in the control group (HR, 1.03; 95% CI, 1.01–1.06; P = 0.027). However, the risk of cancer during 2004 to 2008 in the anemia group was not significantly different from that of the control group (P = 0.442).

Table 3 presents the association between anemia and the risk of each cancer type after PS matching. The risk of gastric (HR, 1.29; 95% CI, 1.17–1.43; P < 0.001), esophageal (HR, 1.37; 95% CI, 1.01–1.85; P = 0.044), lung (HR, 1.14; 95% CI, 1.02–1.26; P = 0.016), breast and genital (HR, 1.25; 95% CI, 1.15–1.36; P < 0.001), and thyroid cancers (HR, 1.14; 95% CI, 1.02–1.28; P = 0.026), lymphoma or leukemia (HR, 1.63; 95% CI, 1.29–2.06; P < 0.001), and other cancer types (HR, 1.17; 95% CI, 1.09–1.26; P < 0.001) was higher in the anemia than in the control group. In addition, the incidence of neoplasms of breast and genital organs was higher among men (HR, 1.25; 95% CI, 1.15–1.36; P < 0.001) than among women (HR, 1.12; 95% CI, 1.03–1.21; P = 0.006).

Table 4 presents the results of the multivariable Cox regression model for cancer development during 2004 to 2015 for the entire cohort. The risk of cancer in the anemia group was 3% higher than that in the control group (HR, 1.03; 95% CI, 1.01–1.05; P < 0.001; model 1). In subgroup analyses, moderate anemia was associated with a cancer risk that was 4% higher (HR, 1.04; 95% CI, 1.01–1.07; P = 0.021; model 2) than that associated with the control group, whereas mild (P = 0.055) and severe anemia (P = 0.339) were not associated with an increased risk of cancer.

This population-based cohort study showed that anemia was associated with a slightly increased overall risk of cancer in South Korea. This association was significant for gastric, esophageal, lung, breast and genital, and thyroid cancers, and lymphoma and leukemia. Moreover, in subgroup analysis, moderate anemia was significantly associated with the risk of cancer. These findings are based on the diagnosis of anemia made using serum hemoglobin levels obtained during a standard health examination, with estimates adjusted for potential confounders such as lifestyle and clinical parameters. Our study's definition of anemia has both strengths and weaknesses. The strength is that it encompassed all individuals with anemia; all individuals (including asymptomatic individuals) were screened by measuring serum hemoglobin levels during a standard health examination. The weakness is that this definition did not allow to distinguish between anemia types such as IDA, aplastic anemia, and hemolytic anemia.

In our study, the prevalence of anemia defined by the WHO criteria among the South Korean adult population during 2002 to 2003 was 9.2%. Using the same criteria, a similar prevalence of anemia was found in China (9.7%; ref. 20); the corresponding value for the United States was 5.6% (21). Although the present study did not evaluate the proportion of IDA among all anemia cases, a previous study reported IDA as the main cause of anemia, in particular, among women of reproductive age and in older adults of both sexes in South Korea (22).

The impact of anemia on cancer risk was most evident for lymphoma and leukemia in this study. Patients with lymphoma often develop anemia even before chemotherapy and in the absence of bone marrow involvement (23). Anemia is a presenting feature in approximately 40% of patients with Hodgkin's lymphoma and is considered an important prognostic factor for outcomes of therapy (24, 25). Moreover, anemia is common among patients with leukemia (26). There are multiple factors responsible for anemia in patients with lymphoma or leukemia (27). Several inflammatory mediators can be activated by lymphoma or leukemia, including IL1, IFNγ, and TNF, which inhibit erythropoiesis (28, 29). Moreover, dysregulation of iron metabolism and abnormal utilization of iron due to leukemia or lymphoma can result in the bone marrow response to erythropoietin (30–32). However, previous studies focused on the relationship between anemia and diagnosed leukemia or lymphoma; thus, the role of anemia in the development of leukemia or lymphoma remains to be elucidated. The present findings suggest that low-serum hemoglobin levels might be a potential risk factor for the development of leukemia or lymphoma.

In the present study, anemia was associated with an increased risk of gastric, esophageal, and lung cancers. A previous study reported that iron deficiency was a common condition in patients with gastric or esophageal cancer (33). Moreover, previous epidemiologic studies reported that iron deficiency was associated with an increased risk of gastric and esophageal cancers (6–8). In addition, in the present study, anemia diagnosed based on serum hemoglobin levels was associated with an increased risk of gastric and esophageal cancers, suggesting that adults diagnosed with anemia should be monitored for cancer markers in the early period of cancer development. Moreover, previous studies have reported that anemia is common in patients with lung cancer, in whom it is also associated with poor survival outcomes (34, 35). The present study findings are consistent with those of previous studies regarding lung cancer (34, 35); future studies are required to elucidate the mechanism of anemia involvement in the development of solid tumors such as gastric, esophageal, and lung cancers.

A previous study reported that the overall risk of cancer may be increased in patients with IDA (12); the present study findings are consistent with those of this previous study. In the analysis, the authors adjusted for age, sex, and duration of follow-up (12); by contrast, in the present study, we adjusted for factors including age, sex, BMI, socioeconomic and lifestyle characteristics, family and individual history of cancer, and comorbidities. As a result, the present estimates may be more robust than those previously reported (12). Nevertheless, it should be noted that our definition of anemia was different from that previously used, which focused on IDA. The present findings should therefore be interpreted with caution and verified in future studies.

In this study, there was no evidence of an association between anemia severity and cancer risk; the highest risk of cancer was observed in the moderate anemia group. Individuals with severe anemia might be at a higher risk of overall mortality rather than cancer. Guinn and colleagues reported that severe anemia was associated with increased myocardial ischemia and mortality rates in patients who declined transfusion (36). Furthermore, anemia with a lower baseline serum hemoglobin level was a risk factor for increased mortality among patients with chronic heart failure (37). These findings suggest that patients with severe anemia might be at a high risk of mortality due to causes such as cardiovascular disease, which could eliminate the potential for increases in cancer risk. Further studies are required to understand the relationship between anemia and cancer risk.

This study has several limitations. First, we defined the underlying diseases using ICD-10 codes registered in the NHIS database. The diseases specified by the ICD-10 codes may differ from the observed underlying diseases. Second, residual confounding might have affected the presented estimates. Third, the type of anemia was not considered in this study. IDA, aplastic anemia, and hemolytic anemia may affect cancer risk differently; the impact of these types of anemia on cancer risk should be elucidated in future studies. Fourth, because the NHIS provides the diagnosis of neoplasms of the breast and genital organs (C50–C63) as one group of ICD-10 codes (C_), we could not distinguish between prostate, cervical, and breast cancer risk. Finally, because we used data on serum hemoglobin levels obtained during 2002 to 2003, we could not confirm whether individuals diagnosed with anemia at this time point remained anemic during the whole study period (2004–2015).

In conclusion, in the present study, anemia was independently associated with an increased overall risk of cancer in the South Korean population. This association was significant for gastric, esophageal, lung, and thyroid cancers, neoplasms of breast and genital organs, and lymphoma and leukemia. Further research is required to elucidate the role of anemia in the development of cancer.

No disclosures were reported.

T.K. Oh: Conceptualization, data curation, formal analysis, methodology, writing–original draft. I.-A. Song: Conceptualization, data curation, supervision, methodology, writing–review and editing.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.