Herbal Medicine Containing Aristolochic Acid and the Risk of Primary Liver Cancer in Patients with Hepatitis C Virus Infection

Aristolochic acid (AA) is found in some Chinese herbs and occurs naturally in plants in the genera Aristolochia and Asarum, which grow worldwide (1). Prior to stricter regulation in 2003, more than 100 million people used Chinese herbs containing AA in China (2). AA was first identified as a factor in urothelial carcinoma (3, 4). Plants containing AA and AA itself are listed as human carcinogens by the International Agency for Research on Cancer (5, 6). AA increases DNA adduct formation and mutant frequency in experimental animals in both liver and kidney (7). A new study also demonstrated that AA induced liver cancer in mice (8). In Taiwan, hepatocellular carcinomas (HCC) are nearly always associated with AA-mediated mutations, and this AA signature was also identified throughout the Asian world and again among Asian patients treated elsewhere (9). Furthermore, there was a significant dose–response relationship between the consumption of AA and HCC in patients with HBV infection in Taiwan (10).

Persons infected with hepatitis C virus (HCV) had more than 15-fold higher HCC risk than uninfected persons (11). HCV infection is still a major public health burden in Taiwan, the United States, and worldwide, especially in relation to HCC and cancer-related mortality (12). Plants containing AA were prohibited in some countries and in Taiwan since November 2003 (13); however, more than 30% of the population in Taiwan was reportedly prescribed Chinese herbs containing AA between 1997 and 2003 (14). Given the lack of a large population-based study assessing the effect of AA exposure on risk of primary liver cancer (PLC) in HCV-infected individuals, the objective of this research was to investigate whether the use of herbal medicine (HM) containing AA was associated with increased risk of PLC among HCV-infected patients.

National Health Insurance (NHI) Research Database was used for the current analysis, and the study design has been described in detail previously (15). In brief, the NHI program has provided compulsory universal health insurance since 1995 including both traditional Chinese and western medicine for most of the country's population (99%). Data were collected from both hospital admissions and outpatient visits and included following patient characteristics: sex, birth dates, dates of visits, date of admission, date of discharge, and up to three outpatient visit diagnoses or five discharge diagnoses defined by International Classification of Diseases, Ninth Revision classification (ICD-9). Patient prescription information, duration, and drug dosage were also collected. According to the study criteria mentioned above, a population-based cohort of patients ages 18 years or older who had a diagnosis of HCV infection (ICD-9 codes 070.41, 070.44, 070.51, 070.54, 070.7, and V02.62) without HBV infection (ICD-9 codes 070.2, 070.3, and V02.61) between January 1, 1997, and December 31, 2010, was identified and included in the final analysis (Fig. 1; refs. 16, 17). For ensuring comparability, only patients who had visited traditional Chinese medicine clinics between January 1, 1997, and 1 year prior to the diagnosis of PLC or the censor dates were included. In accordance with personal electronic data protection regulations, strict confidentiality guideline was closely followed. The National Health Research Institutes (NHRI) of Taiwan maintains and anonymizes the NHI reimbursement data as files suitable for study. This research was also approved by the Institutional Review Board of the National Taiwan University Hospital.

Following standard prescriptions recommended by the Committee on Chinese Medicine and Pharmacy in Taiwan, HM produced prior to promulgation of new regulations in November 2003 could include the following herbs containing AA: Guang Fang Ji (Aristolochia fangji), Guan Mu Tong (Aristolochia manshuriensis), Ma Dou Ling (fruits of Aristolochia contorta or Aristolochia debilis), Qing Mu Xiang (roots of Aristolochia debilis), Tian Xian Teng (stems and leaves of Aristolochia contorta or Aristolochia debilis), and Xi Xin (Asarum sieboldii or Asarum heterotropoides; ref. 18). Upon the recommendation of ancient Chinese medicine books, these herbs were taken as either single products or components of mixed herbal formulas. The original amount of herbs was determined in the unit of grams for each mixture of HM and the overall dose for each AA-containing herb. The cumulative dose of AA was then calculated by summing the estimated average of AA from each herb for individuals who met the inclusion criteria: (i) 1 year prior to the diagnosis of PLC or the censor dates and (ii) at least 1-year induction time. The AA dose associated with Guan Mu Tong, Guang Fang Ji, Ma Dou Ling, Xi Xin, Tian Xian Teng, and Qing Mu Xiang were respectively estimated as 2.59, 2.04, 0.63, 0.042, 0.026, and 0.009 mg (19–23). Body mass was not consistently available; thus for the analysis, AA exposure was presented as total intake (mg).

Patients with PLC were identified by diagnostic code, ICD-9 code 155.0, and the initial diagnosis used as the index date. The diagnosis of PLC was made following the recommendation of the American Association for the Study of Liver Diseases Practice Guidelines, NHI (24). To improve the validity of the diagnosis, only PLC patients who met the diagnosis criteria aforementioned were included in the analysis. Additional analyses of newly diagnosed patients from January 1, 1999, to December 31, 2013, were performed to ensure that all subjects had a minimum 2-year exposure to HM prior to diagnosis with PLC.

Comorbid diagnoses were identified as potential confounders recorded beginning January 1, 1997, to 1 year before the diagnosis of PLC or the censor dates. These were included by diagnostic code, diabetes (250), hypertension (401), hyperlipidemia (272), nonalcoholic steatohepatitis (571.8 and 571.9), alcohol-related disease (291, 303.0, 303.9, 305.0, 571.0, 571.1, 571.2, and 571.3), liver cirrhosis (ICD-9 codes 571.2, 571.5, 571.6, 572.2, 572.3, 572.4, 572.8, and 573.0), cholelithiasis (574), and chronic obstructive pulmonary disease (491 and 492). Potential confounding prescription medications were evaluated as well including anti-HCV treatments (IFN or ribavirin), aspirin, nonaspirin NSAIDs, statins (i.e., atorvastatin, fluvastatin, lovastatin, pravastatin, rosuvastatin, and simvastatin), metformin, and angiotensin-converting enzyme inhibitors (i.e., benazepril, captopril, cilazapril, enalapril, fosinopril, lisinopril, perindopril, quinapril, and ramipril). In the model, sociodemographic characteristics (sex, age, income, and urbanization level) were also considered. Levels of urbanization in Taiwan are divided into four strata by the NRHI publications, with level I referring to the most urbanized areas and level IV referring to the least urbanized areas.

The Kaplan–Meier method was used to estimate PLC cumulative incidences. The log-rank test was further performed to examine differences in the risk of PLC in the cohort. The direct method to adjust the incidence rates was employing the sex and age distributions for the general population of Taiwan between 1999 and 2013. Finally, the analyses used Cox proportional hazards models that computed the HRs and accompanied 95% confidence intervals (CI) before and after adjusting for variables including sex, age, urbanization level, monthly income, diabetes, and liver cirrhosis (25, 26). The analyses were censored for patients lost to follow-up, and those deaths were recorded in the database. Tests for trend were also applied when appropriate. The categories for each HM containing AA or estimated AA were given ordinal scores (0, 1, 2, …) and then modeled as a continuous variable in the trend analysis.

Sensitivity analyses were performed to evaluate the difference and consistency between exposure to HM containing AA and the risk of PLC as well as to the examination of potential interaction in the subgroup effects. Cox proportional hazards regression models were applied coupled with propensity score matching for the analysis. Induction time, competing mortality, and time-dependent covariates were also accordingly evaluated for potential confounding or time-related biases when examining the associations of interest. A two-tailed P value of 0.05 was set as being significant. SAS statistical software (version 9.4; SAS Institute) was used for all statistical analyses.

A total of 223,467 HCV-infected patients were included in the study cohort, of whom 59.5% had recorded exposure to HM containing AA. Table 1 shows the demographic characteristics, medical conditions, and medication usage of patients.

There were 25,502 PLCs during the follow-up period of 3,052,132 person-years, and the overall incidence rate was 835.5 PLCs per 100,000 person-years. There was a higher risk (adjusted HR, 1.23; 95% CI, 1.20–1.26) of PLC in patients with HCV infection who took HM containing AA. There were also significant dose-dependent trends (P < 0.0001) for those using Guan Mu Tong, Qing Mu Xiang, and Xi Xin (Table 2).

Regarding the cumulative dose of AA, the adjusted HRs were 1.21 (95% CI, 1.18–1.24), 1.48 (95% CI, 1.37–1.59), 1.50 (95% CI, 1.34–1.68), and 1.88 (95% CI, 1.61–2.19) for estimated AA of 1 to 250, 251 to 500, 501 to 1,000, and more than 1,000 mg, respectively, relative to no AA exposure. There was also a significant dose-dependent trend (P < 0.0001; Table 2). The log-rank tests identified significant observed differences (P < 0.001) over the entire Kaplan–Meier curves (Fig 2).

A significant risk of PLC in patients with HCV was also identified using Cox proportional hazards regression with propensity score matching, 2-year induction time, and time-varying changes. Additional covariates in the sensitivity analysis had little effect on the observed associations between AA exposure and the risk of PLC in different models. When the data were further stratified by age, sex, diabetes, nonalcoholic steatohepatitis, alcohol-related disease, liver cirrhosis, anti-HCV treatment, statin, aspirin, or metformin use, the association of interests remained significant in the subgroup analysis. In addition, significant interaction effects were observed for alcohol-related disease (P < 0.0018) and sex (P < 0.0017). This interaction may reflect the greater incidence of alcohol-related disease in men (Table 3).

Higher risk of PLC was observed among HCV-infected patients who took HM containing AA. Furthermore, significant dose-dependent relationships were found between the specific HMs, Guan Mu Tong and Qing Mu Xiang, or overall AA exposure and the risk of PLC in patients with HCV infection.

Despite its profound toxicity and carcinogenicity, the health effects of AA have not been well recognized. This changed in the early 1990s when a cohort of Belgian women developed kidney disease (27) that was later accompanied by the upper urinary tract cancer (2, 3). The unique course of the disease, now known as AA nephropathy (AAN; ref. 13), and its pathophysiology provided insights to solve the long-standing mystery of Balkan endemic nephropathy (BEN). Those diagnosed with BEN were exposed to AA from flour contaminated with Aristolochia clematitis seeds. BEN is now recognized as an example of AAN (28).

Aristolactam-DNA adducts (29) and their consequently AA-specific mutation signature (30, 31) are now accepted biomarkers of AA exposure and effects. These biomarkers have aided in establishing links between AA exposure and upper urinary tract cancer (32, 33), clear-cell renal cell carcinoma (34), bladder cancer (35), and HBV-positive HCC (36) in Taiwan. Thus, AA-associated carcinogenesis is not only limited to the upper urinary tract. Ng and colleagues (7) just presented evidence that the majority of liver tumors sampled in Taiwan were positive for this AA mutational signature. Although they did not present evidence on AA exposure and viral hepatitis, their findings provide additional support for the association between AA exposure and HBV-positive HCC published by Chen and colleagues (8).

In the current analyses, AA exposure as well as Guan Mu Tong and Qing Mu Xiang may increase the risks for PLC in HCV-infected patients in a dose-dependent manner (Table 2). HCV induces mitogenic, angiogenic, and metastatic pathways, blocks cell death, triggers persistent inflammation and ROS production, and dysregulates host lipid metabolism (37). AA may interact with HCV to increase the risk of PLC. Among three significant herbs containing AA, over 84% of Mu Tong herbal formulas contained Guan Mu Tong (Aristolochia manshuriensis), and the estimated average dose of AA was 2.59 mg/g of Guan Mu Tong [18]. Of the top ten herbal formulas prescribed for chronic hepatitis in Taiwan, 2002 (38), two types of herbal formulas containing Mu Tong—Long Dan Xie Gan Tang and Gan Lou Xiao Du Dan—comprised 23.5% and 6.3% of all prescriptions, respectively. Furthermore, Qing Mu Xiang, which is not among the popular chronic hepatitis formulas but contains significant AA, also increases the associated risk of PLC. Thus, HMs intended to improve hepatic distress may actually contribute to an unforeseen hepatic pathogenesis.

The current study has several strengths. First, a large size highly representative population-based computerized database was analyzed as the study population in Taiwan. The possibility of selection bias and losses to follow-up was minimized due to the patients being recruited from an insured general population with approximately 99% population coverage. Second, because the NHI reimbursement database identifies all prescription information prospectively, the possibility of patient recall bias for the intake doses of various HM and AA can be ruled out, though compliance is another issue (see below). Third, the included clinical endpoint, PLC, was well-defined to examine the clinical importance of the study aim through the longitudinal analyses. Fourth, a 1-year minimum induction time was addressed by calculating cumulative AA exposures up to 1 year prior to diagnosis or the censor dates. Finally, sensitivity analyses including analytical designs, additional covariates, and subgroup effects were conducted, and the results remained robust.

The increased risk observed in the current analysis might be associated with a subgroup of those with more severe liver hepatitis, who are more likely to develop liver cancer. If this subgroup were more likely to employ these HMs, a potential behavioral confound would result. Despite this possibility, the results presented indicate the AA dose dependency remained evident in both noncirrhotic or cirrhotic subgroups. Further, AA consumption was assessed in 1997–2003 during which AA was withdrawn from the market, whereas cases of PLC were identified between 1999 and 2013 (ten years after market withdrawal). Therefore, the confounding effect by disease severity, if present, would be diluted.

Several potential limitations should be noted. Because the data on herbal prescriptions are incomplete prior to 1997, information prior to that date was not included, and the data may significantly underestimate AA intake. In addition, the AA estimates presumed that patients were compliant and took all prescribed medications. If there was some degree of noncompliance, the AA consumption would thus be an overestimate. This may bias the current analysis to the null hypothesis. Furthermore, several unmeasured confounders, including body mass index, exposure to aflatoxin B1, alcohol intake, and smoking, which are likewise associated with PLC (39–41), were not included in the database. Nevertheless, the sensitivity analyses on hypertension, hyperlipidemia, alcohol-related disease, nonalcoholic steatohepatitis, and chronic obstructive pulmonary disease as additional covariates were negative, suggesting that these common concerns were not confounders (Table 3).

In summary, the current analyses suggest that AA may increase the efficacy of HCV in the pathogenesis of PLC. The mechanism for this proposed interaction remains unclear, and understanding the process could provide important insight into the oncogenic process and potential points of intervention. Although HM containing AA has largely been banned in Taiwan and other countries, HCV patients with current or past exposure to these HMs should be followed closely.

No potential conflicts of interest were disclosed.

Conception and design: C.-J. Chen, Y.-H. Yang, M.-N. Lai, W.-C. Ho, P.-C. Chen

Development of methodology: C.-J. Chen, Y.-H. Yang, M.-H. Lin, C.-P. Lee, M.-N. Lai, H.-Y. Yang, W.-C. Ho, P.-C. Chen

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C.-J. Chen, M.-H. Lin, C.-P. Lee, Y.-T. Tsan, M.-N. Lai, W.-C. Ho

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C.-J. Chen, Y.-H. Yang, M.-H. Lin, C.-P. Lee, Y.-T. Tsan, M.-N. Lai, P. Doyle, W.-C. Ho, P.-C. Chen

Writing, review, and/or revision of the manuscript: C.-J. Chen, M.-H. Lin, C.-P. Lee, H.-Y. Yang, P. Doyle, W.-C. Ho, P.-C. Chen

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.-H. Lin, C.-P. Lee, W.-C. Ho

Study supervision: Y.-T. Tsan, M.-N. Lai, W.-C. Ho, P.-C. Chen

This work was financially supported by the Innovation and Policy Centre for Population Health and Sustainable Environment (Population Health Research Centre, PHRC), National Taiwan University College of Public Health from the Featured Areas Research Centre Program within the framework of the Higher Education Sprout Project by the Ministry of Education (MOE) in Taiwan. The study is based in part on data from the National Health Insurance Research Database provided by the National Health Insurance Administration, Ministry of Health and Welfare, and managed by NHRI. The interpretation and conclusions contained herein do not represent those of Bureau of National Health Insurance, Department of Health, or NHRI. The authors would like to thank Health Information and Epidemiology Laboratory of Chang Gung Memorial Hospital, Chiayi Branch, for comments and assistance in data analysis.

This study was supported by the Ministry of Science and Technology (MOST-106-2918-I-002-016, MOST-107-3017-F-002-003, and MOST-108-3017-F-002-001), the Featured Areas Research Center Program within the framework of the Higher Education Sprout Project by the Ministry of Education (MOE) in Taiwan (NTU-107L9003 and NTU-108L9003), and Chang Gung Memorial Hospital, Chiayi Branch (CLRPG6G0041). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the article.

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.