Long-Term Glucocorticoid Use and Cancer Risk: A Population-Based Cohort Study in South Korea

Cancer is one of the major causes of death worldwide (1). From 2006 to 2016, there were 17.2 million patients with cancer worldwide, 8.9 million deaths, and the global incidence of cancer increased by 28% (2). In South Korea, the cancer prevalence has increased very fast, and 229,180 Koreans were newly diagnosed and 78,194 died from cancer in 2016 (3). Furthermore, the incidence of cancer is expected to increase; therefore, cancer prevention is emphasized to achieve a significant reduction in the global burden of disease (4).

Glucocorticoids (GC) are a class of pleiotropic steroid hormones that are commonly prescribed for 1.2% and 1.0% of patients with chronic conditions in the United States and the United Kingdom, respectively (5, 6). GCs have both immunosuppressive and potent anti-inflammatory effects (7), and they inhibit the immune response and production of prostaglandins and leukotrienes, which are the two main products of inflammation (8). Based on its immunosuppressive and anti-inflammatory effects, GCs have been prescribed to patients with chronic obstructive pulmonary disease (COPD) or asthma (9), autoimmune disease (10), rheumatic disease (11), and allergic diseases (12). GCs have also been used in the management of patients with cancer to relieve symptoms such as pain, dyspnea, neurological manifestations, and discomfort from inflammatory reactions and lymphedema (13). GCs have also been widely used in clinical oncology during chemotherapy and radiotherapy to alleviate side effects in patients with nonhematological cancer (14). However, long-term GC therapy is known to cause serious adverse effects such as immunosuppression, and it could impair immunotherapy outcomes (15, 16). Therefore, it is possible that long-term GC use might increase the incidence of cancer; however, there is not enough evidence regarding this issue.

This study aimed to investigate whether long-term GC use is associated with a higher risk of cancer in the population without cancer. In addition, we assessed whether the cancer risk associated with oral GC use is dependent on the daily dose.

The sample cohort database of the National Health Insurance Service (NHIS) was developed to provide data for research, including health surveys and medical surveys of the Korean population. The validated database comprised a stratified random sample of approximately one million people who were registered with the NHIS in 2002, and it was designed to be representative of the national population in terms of demographic and socioeconomic variables. The cohort was dynamic and followed-up until the end of 2015. It was supplemented with additional cohort data, including data regarding infants, to allow for attrition due to death and loss to follow-up. Each year, using stratified extraction methods to ensure that the cohort was representative of the national population, people were added to the cohort to replace those who had died or emigrated during the previous year. The sample cohort database included information regarding individuals' demographics, socioeconomic status, healthcare use, medical history, and cause of death until the end of 2015 (17). The study protocol was approved by the Institutional Review Board of Seoul National University Bundang Hospital (X-1903-531-901) and the Health Insurance Review and Assessment Service (NHIS-2019-2-140).

We assessed a sample cohort of 1,000,000 individuals included in the 2010 cohort database of the NHIS. Next, we excluded individuals who died in 2010 and those who had emigrated between 2011 and 2015 because they might have been diagnosed with cancer after emigrating. In addition, individuals with a history of cancer in 2010 were excluded, as our study focused on new cancer diagnoses between 2011 and 2015.

Individuals prescribed oral GC (prednisolone, methylprednisolone, or dexamethasone) regularly and continuously over ≥30 days were defined as long-term GC users, based on previous report (18). The individuals who were not prescribed any GC or prescribed GC for <30 days were classified as unexposed individuals. Because the longest prescription period of oral GC was from 3 to 7 days in a year for short-term GC users in South Korea, we considered them as unexposed individuals, too. Three types of oral GCs were considered, prednisolone, methylprednisolone, and dexamethasone, and the daily dosages of GCs were divided into two groups, ≥5 mg/day (high-dosage) and <5 mg/day (low-dosage) of prednisolone. Using this dosage calculation, 4 mg/day of methylprednisolone and 0.75 mg/day of dexamethasone were considered equivalent to 5 mg/day of prednisolone, given the potency of GCs (19). Our rationale for dividing the groups using a cut-off value of 5 mg/day was based on previous studies that have reported no significant association with dosages <5 mg/day of prednisolone and all-cause mortality for patients with rheumatoid arthritis (20, 21). As a result, long-term GC users receiving <5 and ≥5 mg/day of prednisolone were classified as the low-dosage group and high-dosage group, respectively.

According to the International Classification of Diseases 10th edition (ICD-10) codes, newly registered diagnoses of any malignancy (C00-C96) from January 1, 2011 to December 31, 2015, were defined as the development of cancer in this study. In details, the cancers were categorized as follows: gastric cancer (C16), esophageal cancer (C15), colorectal cancer (C18–C20), gall bladder and biliary tract cancer (C23–C24), head and neck cancer (C00–C14), brain cancer (C71), liver cancer (C22), pancreatic cancer (C25), lung cancer (C34), bone and articular cartilage cancer (C40–C41), neoplasms of the breast and genital organs (C50–C63), urinary tract cancer (C64–C68), thyroid cancer (C73), and lymphoma or leukemia (C81–C96). The time to cancer diagnosis was calculated starting from January 1, 2011, to the date of diagnosis of cancer, as registered officially in the ICD-10 system. In South Korea, all patients diagnosed with any C-code cancer should be registered in the NHIS database to receive financial coverage of 95% of the total charges for cancer treatment from the NHIS. Therefore, all patients diagnosed with cancer were registered in NHIS database.

The following data were considered confounders in this study: demographic information (age and sex); socioeconomic information [income level and place of residence in 2010 (Seoul, metropolitan cities, and other)]; the 2010 Charlson comorbidity index, which was calculated using registered ICD-10 diagnostic codes in the NHIS database between 2009 and 2010 (Supplementary Table S1); other comorbidities [hypertension (I10), COPD (J44), asthma (J45), allergic disease (T78.4), and autoimmune disease (Supplementary Table S2)]; and total number of hospital and outpatient visit days in 2010. The entire cohort was divided into four groups according to age (18–30, 31–50, 51–70, and ≥71 years old). The cohort in 2010 was also divided into five groups according to the income level (0%–20%, 20%–40%, 40%–60%, 60%–80%, and 80%–100%). Regarding the classification of place of residence, Seoul, the capital city, was assigned a separate category, and the cities of Incheon, Kwangju, Busan, Ulsan, Daegu, and Daejeon were classified as metropolitan cities. The number of hospital days included the number of hospital outpatient clinic visits and days spent admitted to the hospital. For example, if an individual visited a hospital outpatient clinic five times and was admitted to the hospital for 3 days, then the individual was considered to have 8 hospital visit days. In the analysis, the number of hospital and outpatient clinic visit days was categorized into five groups (0 days, 1–7 days, 8–29 days, 30–90 days, and >90 days) and four groups (0–7 days, 8–29 days, 30–90 days, and >90 days), respectively.

The primary endpoint of this study was the new development of cancer from January 1, 2011, to December 31, 2015, among patients registered in the 2010 NHIS sample cohort of South Korea. All individuals without any cancer diagnosis were followed up until December 31, 2015, if they did not die. Individuals who died during follow-up were censored as noncases at the date of death.

We performed multivariable Cox regression analyses to determine whether long-term GC exposure in 2010 was associated with the development of cancer from 2011 to 2015, and compared the results with those of unexposed individuals. All covariates except the Charlson comorbidity index were included in the multivariable model for adjustment to avoid multicollinearity with other comorbidities that were used to calculate the Charlson comorbidity index (Supplementary Table S1). In addition, subgroup analysis according to the daily dosage of GC was performed to investigate whether the daily dosage of GC might affect the main results of this study. In this analysis, the individuals were classified according to the daily dosage of GCs, as follows: the low daily dosage group (<5 mg/day of prednisolone) and high daily dosage group (≥5 mg/day of prednisolone). Next, in the competing risk analyses, we constructed in detail 14 multivariable Cox regression models for the development of cancer during 2011–2015. The 14 types of cancer were each set as endpoints in the multivariable Cox models, and the duration to cancer diagnosis from January 1, 2011, was used for the time-to-event competing risk analyses. These competing risk analyses were performed to investigate what type of cancer risk was most closely related to long-term GC exposure. In addition, subgroup analyses according to sex were performed to analyze the hazard of overall cancer and all 14 cancer types. We also fitted a multivariable Cox regression model for the development of cancer during 2012–2015 (not for 2011–2015) in the entire cohort to assess whether reverse causation bias could have affected the result because there was a short latency time between long-term GC exposure and development of cancer in 2011. Finally, subgroup analyses according to Charlson comorbidity index (0–2 and >2), COPD, asthma, autoimmune disease, rheumatic disease other than autoimmune disease, allergic disease, and liver disease in 2010 were performed to assess whether the comorbid status of individuals or indication bias of long-term GC exposure affected the results of this study. The results of the Cox regression models are presented as hazard ratios (HR) with 95% confidence intervals (CI), and it was confirmed that there was no multicollinearity in all multivariable models of the entire cohort (variance inflation factor of <2.0). It was also confirmed that the central assumption of the Cox proportional assumption was met for all multivariable models using log-minus-log plots.

To enhance the robustness of this study, we performed propensity score (PS) matching, which is an efficient method of reducing the effects of confounders in observational studies (22), using the nearest neighbor method with a 1:5 ratio, without replacement, and a caliper width of 0.2. All covariates were included in the PS model, and a logistic regression analysis was performed to calculate the PS as a logistic model. The absolute value of the standardized mean difference (ASD) was used to evaluate the balance between long-term GC users and the control group before and after PS matching. Through PS matching, we aimed to minimize the ASD of all confounders between the two groups to less than 0.1. Since some important information such as body mass index (BMI), alcohol consumption and smoking status in 2010 were available in only 20% of all individuals (17), they were not included in the PS modeling. They were compared after PS matching among individuals whose data was were available, using Chi-square test. After confirming a good degree of balance between the two groups in the PS-matched cohort, we performed conditional Cox regression via a stratified Cox regression analysis of the development of cancer from 2011 to 2015. All statistical analyses were performed using R version 3.6.3 (R Foundation for Statistical Computing, Vienna, Austria); P < 0.05 was considered statistically significant.

The 2010 sample cohort comprised 1,000,000 individuals. We excluded 173,091 individuals younger than 18 years, 4,487 individuals who died in 2010, 208 individuals who emigrated from 2011 to 2015, and 51,334 individuals who had a history of cancer in 2010. Therefore, 770,880 individuals were included in the analysis (Fig. 1). Among them, 1,602 (0.2%) were long-term GC users and 769,278 (99.8%) were in the unexposed individuals. In the 2010 cohort, 36,157 (4.7%) individuals were newly diagnosed with cancer from January 2011 to December 2015. The median duration from January 1, 2011, to the date of the cancer diagnosis was 2.4 years (interquartile range: 1.1–3.7 years).

Table 1 shows the results of the multivariable Cox regression analyses for cancer development during 2011 to 2015 for the entire sample cohort in 2010 (n = 770,880). In the multivariable model, the risk for cancer development among long-term GC users was 1.23-fold higher than that of the unexposed individuals (HR = 1.23; 95 % CI, 1.06–1.43; P = 0.004; model 1). In addition, the risk of cancer development among high daily dosage long-term GC users was 1.29-fold higher (HR = 1.29; 95% CI, 1.03–1.62; P = 0.030; model 2) than that of the unexposed individuals, whereas that of the low daily dosage long-term GC users was not significantly different (P = 0.096). Supplementary Table S3 shows the results of multivariable Cox regression analyses of the development of cancer during 2012 to 2015 (not for 2011–2015). In the multivariable model, the risk for cancer development among long-term GC users was 1.20-fold higher than that of the unexposed individuals (HR = 1.20; 95% CI, 1.01–1.43; P = 0.042; model 1). In addition, the risk of cancer development among high daily dosage long-term GC users was 1.28-fold higher (HR = 1.28; 95% CI, 1.02–1.61; P = 0.031; model 2) than that of the unexposed individuals, whereas that of the low daily dosage long-term GC users was not significantly different (P = 0.075).

Table 2 shows the results of the competing risk analyses according to the primary cancer sites, which were obtained from the multivariable Cox regression analyses for the development of cancer during 2011–2015. First, the risk of overall cancer in male long-term GC users was 1.28-fold higher (HR: 1.28, 95% CI: 1.06–1.25; P = 0.001) than that in the unexposed individuals, while that was not significantly different (P = 0.144) in female GC users. Additionally, the risks of liver cancer and lung cancer were 1.46-fold (HR:1.46; 95% CI: 1.03–2.07; P = 0.034) and 1.52-fold (HR:1.52; 95% CI: 1.04–2.21; P = 0.029) higher in the long-term GC users than that in the unexposed individuals, respectively.

Table 3 shows the results of multivariable Cox regression analyses of the development of cancer in the subgroups. In individuals whose Charlson comorbidity index was 0 to 2 in 2010, the risk of cancer was 1.34-fold (HR = 1.34; 95% CI, 1.05–1.71; P = 0.020) higher among long-term GC users than among unexposed individuals. In individuals whose Charlson comorbidity index was >2 in 2010, the risk of cancer was 1.23-fold (HR = 1.23; 95% CI, 1.02–1.48; P = 0.031) higher in long-term GC users than in unexposed individuals. In the other subgroup analyses according to comorbidities related to long-term GC exposure, the risk of cancer among long-term GC users who were patients with asthma was 1.31-fold higher than that of unexposed individuals who were asthma patients (HR = 1.31; 95% CI, 1.03–1.66; P = 0.027). All other subgroups of long-term GC users (COPD, autoimmune disease, rheumatic disease other than autoimmune disease, allergic disease, and liver disease) did not show any significant associated between the comorbidity and the risk of overall cancer, compared with unexposed individuals (all P > 0.05). Supplementary Table S4 shows the risk of liver cancer among the long-term GC users with liver disease, and the risk of lung cancer among the long-term GC users with COPD, and asthma. Among the patients with asthma, the risk of lung cancer in the long-term GC users was 2.13-fold higher (HR = 2.13; 95% CI, 1.32–3.46; P = 0.002) than that in the unexposed individuals. However, the risks of lung cancer and liver cancer in the long-term GC users among the patients with COPD (P = 0.091) and those with liver disease (P = 0.177) were not significantly higher than that in unexposed individuals, respectively.

After PS matching, a total of 9,592 individuals (1,599 long-term GC users and 7,991 controls) were included in the final analysis. The demographic and clinical characteristics of the two groups are shown in Table 4, and all ASDs between the two groups after PS matching were <0.1, demonstrating a good degree of balance. However, as shown in Supplementary Table S5, the distribution BMI and alcohol consumption in long-term GC users and unexposed individuals were statistically different in the PS-matched cohort (P < 0.05), while smoking status in the two groups was not statistically different (P = 0.626). Table 5 shows the results of cancer newly diagnosed from 2011 to 2015 in the PS-matched cohort of 2010. In the stratified Cox regression analysis, the risk for cancer development among long-term GC users was 1.25-fold higher than that of the unexposed individuals (HR = 1.25; 95% CI, 1.06–1.48; P = 0.008).

This population-based cohort study showed that long-term GC use was associated with a higher incidence of cancer compared with that of the unexposed individuals. This association was more evident for lung and liver cancer risk. These findings suggested that long-term GC therapy might be a risk factor for cancer, and that long-term GC therapy should be performed carefully because of the cancer risk.

Regarding humoral immunity, long-term GC therapy was reported to be related to a reversible decrease in B-cell counts and antibody responses, even in low dosages (23). Furthermore, long-term GC therapy is known to affect innate immunity by inhibiting macrophage differentiation and suppressing the production of IL1, IL6, TNF, and the pro-inflammatory prostaglandins and leukotrienes by macrophages (24). Because both suppressed innate immunity with enhanced humoral immunity were known as related factors with increased overall cancer risk (25), the influence of immunosuppression may be present in long-term GC users, which might have affected the results of this study. The long-term GC use also inhibits key inflammatory markers such as reactive protein/TNFα, which is known to play a role in carcinogenesis and surpass the antitumor immune responses (26). Moreover, long-term GC therapy is known to suppress the tumoricidal activities of activated macrophages (7); therefore, the risk of cancer might be increased in long-term GC users.

The results of the subgroup analyses of the daily dosage of GC showed that the HR for cancer was higher and significant among the high daily dosage group of long-term GC users, which suggested that the risk of cancer with long-term GC use might be dosage-dependent. In general, the adverse effects of long-term GC therapy are known to be dosage-dependent (27), and our previous retrospective cohort study reported that high daily dosage GC use was significantly associated with a higher risk for 30-day mortality for critically ill patients (28). However, our results should be interpreted carefully, because the 95% CIs in HRs of low daily dosage (0.99–1.50) and high daily dosage GC users (1.14–1.84) were overlapped, suggesting there was no significant difference in HRs for overall cancer development between low daily dosage and high daily dosage GC users. Therefore, the effect of daily dosage in long-term GC users on cancer risk should be evaluated in the future study.

In competing risk analyses, the associations between long-term GC exposure and elevated cancer risk were significant for lung and liver cancer in this study. Recent population-based cohort studies reported that long-term exposure to inhaled corticosteroids in patients with COPD was associated with a lower risk of lung cancer (29, 30). There were differences between these previous studies and ours because we focused on long-term oral GC therapy, whereas previous studies focused on inhaled GC therapy (29, 30). COPD is a well-known risk factor for lung cancer, and inhaled GCs comprise a useful treatment option for preventing inflammatory reactions, which is an important pathophysiologic mechanism underlying the development of lung cancer (31). Furthermore, inhaled GCs are known to have fewer severe adverse effects than orally administered GC (32). In a nested case–control study in Taiwan, the higher risk of squamous lung cancer was significantly associated with recent dose increase of GC among patients with asthma or COPD (33). The study included both the inhaled GC use and oral GC use, and it suggested that long-term oral GC exposure might have a potential risk factor for development of lung cancer. However, the relationship between long-term oral GC exposure and lung cancer risk in population without cancer or patients with COPD should be evaluated in the future with more studies

For liver cancer, there was a very recent study that reported that long-term exposure of oral GC was associated with higher risk of hepatocellular carcinoma in patients who had any one of comorbidities including alcohol-related disease, chronic liver disease, and diabetes mellitus (34). Although the previous study included on the patients with higher risk of liver cancer, our study focused on the individuals without cancer, suggesting that the association of long-term GC exposure might be a potential risk factor for development of liver cancer in healthy individuals without cancer. More study is needed to confirm these associations of long-term GC exposure with liver and lung cancer risk in future.

This population-based cohort study had several limitations. First, some important data, such as BMI, alcoholic consumption, and smoking status, were not included in the analysis for adjustment because they were not present in the entire cohort dataset. In fact, there were statistical differences in distribution of BMI and alcohol consumption in PS-matched cohort as shown in Supplementary Table S4, and it might have effect on the result in this study. Second, we defined the underlying diseases using the ICD-10 codes registered in the NHIS database. The diseases specified by the ICD-10 codes may have differed from the actual underlying diseases. Third, because we assessed the population without cancer in sample cohort, which consisted of many healthy adults, the proportion of long-term GC users was relatively small (0.2%). Fourth, there might have been residual confounders in this study that were not included in the PS model or multivariable adjustment, which could have affected the results of this study. Finally, our findings may have been influenced by the indication bias of GC. For example, although we included some comorbidities such as rheumatic disease and COPD for both PS modeling and multivariable adjustment, the severity of the diseases may differ, leading to differences in GC use. This may have affected the incidence of cancer in long-term GC users.

In conclusion, this population-based cohort study conducted in South Korea showed that long-term GC exposure might be associated with a higher risk of overall cancer. This association was more evident for lung and liver cancer risk. However, there might be unmeasured and potential confounders in this study, therefore, the results should be interpreted carefully, and future studies should be performed to confirm these findings.

No potential conflicts of interest were disclosed.

T.K. Oh: Conceptualization, data curation, formal analysis and writing-original draft. I.-A. Song: Conceptualization, 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.