A population-based cohort study in Taiwan--use of insulin sensitizers can decrease cancer risk in diabetic patients?

A population-based cohort study in Taiwan––use of insulin sensitizers

can decrease cancer risk in diabetic patients?

C.-H. Kao

*,

, L.-M. Sun

, P.-C. Chen

, M.-C. Lin

, J.-A. Liang

, C.-H. Muo

,

S.-N. Chang

& F.-C. Sung

introduction

Thiazolidinediones (TZDs) are insulin sensitizers that bind and activate peroxisome proliferator-activated receptors gamma (PPARγ), a nuclear hormone receptor [1, 2] Conficting results have been reported regarding the

relationship between TZDs’ use and subsequent cancer risk. A number of studies have found that TZDs modulate several cancer cell lines and probably inhibit tumor growth,

progression, differentiation, and metastasis [3–5]. In contrast, an earlier study suggested a possible association between development of cancer and the use of TZDs, particularly rosiglitazone [6]. Experimental studies have revealed a link between pioglitazone and rat bladder cancer [7]. A

subsequent PROactive (prospective pioglitazone clinical trial in macro-vascular events) study similarly detected more human bladder cancer cases than expected in the pioglitazone group [8]. The US Food and Drug

Administration was concerned with these results, and in September 2010 announced an ongoing investigation into the possible risk of TZDs in humans. Recent research into reports of adverse events related to drug use also found a defnite signal for bladder cancer associated with

pioglitazone [9].

TZDs are currently widely used as oral agents for the

treatment of type 2 diabetes. Thus, even a small magnitude of hazard could have important clinical implications, and such fndings may hold interest for the general public as well as the medical profession. A population-based, large-scale study may help clarify this controversy. Therefore, we explored the issue using the database from the National Health Insurance (NHI) system of Taiwan.

data source

We obtained data on reimbursement claims from the Taiwanese NHI system, a mandatory health care plan that has provided affordable healthcare to all residents since March 1995. Since 2007, this system has covered >99% of the population. The NHI contracts with >90% of hospitals and clinics in Taiwan, and provides comprehensive medical services including outpatient and inpatient care, dental care, physical therapy, preventive care, and prescriptions. The National Health Research Institute is responsible for administering the NHI Research Database (NHIRD), which includes numerous randomly selected claims that are representative of the entire population, and which is used for administrative purposes and research. The data we used were a sub-dataset composed of one million randomly selected subjects (∼5% of the entire population), drawn from the larger pool of all enrollees registered with the NHI from 1996 to 2009. The details of this population-based cohort database have been published previously [10]. Diagnoses were coded using the International

Classifcation of Diseases, 9th Revision, Clinical Modifcation (ICD-9-CM).

study sample

We identifed patients who were ≥20 years of age with newly diagnosed type 2 diabetes (ICD-9-CM code 250.xx) and who were treated with antidiabetic agents between 2001 and 2009, excluding the subjects with

potential type 1 diabetes (code numbers 250.x1 and 250.x3). Subjects with cancer (code numbers 140.xx-208.xx) diagnosed before their date of diabetes diagnosis were further excluded, leaving 22 910 diabetic patients for data analysis. For patients with diabetes, the potential comparison subjects were persons with the same sex and birth year who had never been diagnosed with diabetes and cancer before the year of diagnosis of their matched diabetic counterparts. To identify the comparison group, we stratifed the one million NHI benefciaries using the distribution of birth year, sex, and calendar year of diabetes diagnosis among the diabetes patients. Within the strata, the comparison subjects were randomly selected in a 4:1 ratio to diabetic patients.

Among the diabetic subjects, those who have been prescribed for

rosiglitazone and pioglitazone before the study’s end date formed the TZDs group; all others were included in the non-TZDs group.

statistical analysis

We compared the event-free probabilities for cancer among the TZDs, non-TZDs and the non-diabetic comparison groups using Kaplan–Meier

survival curves constructed using the current age as the time scale. The differences between these groups were tested using the log-rank test. The incidence density of cancer was calculated by using the number of cancer cases as the nominator and follow-up person-years as the denominator. Multivariable Cox proportional hazards models with the current age as the time scale and adjusted for sex were carried out to assess the association between anti-diabetic medications and the risk of developing cancer, quantifed as hazard ratios (HRs) with 95% confdence intervals (CI). For the diabetic group, the entry time of this study was age at the frst prescription of anti-diabetic drugs, and the follow-up ended at the age of diagnosis of malignant cancer (ICD-9-CM codes 140.xx-208.xx, the observed end-point of this study), withdrawal from NHI program, or the termination of this study on 31 December 2009, whichever came frst. For subjects in the TZDs group previously treated with other anti-diabetic drugs, the follow-up duration before initiation of TZDs treatments was included in the non-TZDs group. For the non-diabetic comparison group, the entry time was age at the same year of diagnosis of their matched diabetic counterparts. The defnition of the end of follow-up was the same as that applied to the diabetic group, but for the comparison subjects later diagnosed with diabetes, the follow-up ended at the age of diabetes diagnosis. Two sets of Cox regression analyses were carried out. One was to estimate the cancer risk for TZDs and non-TZDs groups, respectively, using the non-diabetic comparison group as the reference group. The other estimated the risk for the TZDs group relative to the non-TZDs group. We also evaluated the dose–response relationship between TZDs and the risk of cancer. The dosage of TZDs was classifed by the tertile of average prescription dose per year, which were at levels of <0.69 g, 0.69–5.29 g, and >5.29 g. Further Cox regression analysis was carried out by stratifying time since the initiation of treatments in order to evaluate whether the risk of cancer associated with TZDs and other anti-diabetic drugs differed over time. All analyses were carried out using SAS statistical software (version 9.1 for Windows; SAS Institute, Inc., Cary, North Carolina), and the

signifcance level was set to 0.05.

results

Among the 22 910 subjects with diabetes, 4158 patients had been treated with TZDs (Table 1). More than half of the

diabetic patients were men (55.1%) and aged between 45 and 64 years (54.2%). The frequency distribution of age and sex

was the same between the diabetic group and the comparison group after matching. However, patients in the TZDs group were younger than those in the non-TZDs group with the mean age of 54.2 years versus 57.0 years.

During the follow-up period, 2952 patients in the

comparison group, 781 in the non-TZDs group and 119 in the TZDs group had been diagnosed with cancer (Table 2). The Kaplan–Meier curves of cumulative event-free probability by age of cancer were similar among the non-TZDs, TZDs, and the comparison groups before 60 years of age (Figure 1). After that, the event-free probability was higher in the non-diabetic comparison group and the TZDs group than in the non-TZDs group after 80 years of age (log-rank test among the three groups, P < 0.001).

Compared with the non-diabetes group, the diabetes group had an increased risk of developing cancer (adjusted HR = 1.20; 95% CI = 1.11–1.29) (Table 2). The adjusted HR was 1.20 (95% CI = 1.11–1.30) for the non-TZDs group and 1.18 (95% CI = 0.98–1.42) for the TZDs group. The elevated risks of cancer in the non-TZDs group, relative to the comparison group, were observed in all the subgroups of age, except those ≥75 years of age at their frst prescriptions. The risk of cancer associated with the TZDs group was greater among patients starting their prescription at 45–54 years of age (adjusted HR = 1.62, 95% CI = 1.16–2.27). The adjusted HR was at the same strength for patients aged <45 years at their frst prescription but not statistically signifcant (95% CI = 0.88–2.99). A

statistically non-signifcant decrease in the risk in the TZDs group was observed among subjects whose frst prescription was at ≥75 years of age. (HR = 0.75, 95% CI = 0.37–1.51). No signifcant association was observed in the analysis comparing the risk of cancer in the TZDs with the non-TZDs group.

Site-specifc cancer risks are presented in Table 3. When

using the non-diabetes comparison group as a reference group, the risks were substantially increased in the non-TZDs diabetes group for colorectal cancer (HR = 1.27), liver cancer (HR = 1.79), and pancreatic cancer (HR = 2.78). In the TZDs group, the risks increased for colorectal cancer (HR = 1.82), pancreatic

cancer (HR = 5.45), and lymphoma (HR = 2.50). Among the diabetic patients, no signifcant fndings were observed for the risk of cancer in the TZDs group relative to the non-TZDs group, but a marginally lower risk of liver cancer was found in the TZDs group.

Analysis of the TZD dose showed that compared with the non-diabetes comparison group, the risk of cancer increased in the TZDs group prescribed for the lowest dosage (<0.69 g/ year) (HR = 1.40; 95% CI = 1.09–1.80), but decreased to null for other TZD dosage, suggesting no dose–response trend on the risk of cancer (Table 4). Similar results were observed when we used non-TZDs diabetic patients as a reference group. Stratifed analysis by time since the frst prescription showed that the risks of cancer in both the non-TZDs and TZDs groups were the greatest within 1 year after initiation of prescription, when compared with the comparison group (supplementary Table S5, available at Annals of Oncology online).

discussion

The results from this population-based cohort study indicated that diabetic patients had a signifcant increase in the overall cancer risk, especially for patients not using TZDs. The risk depends on the cancer site. TZD users had signifcantly higher risks of colorectal, pancreatic cancers, and lymphoma.

Non-TZD users had signifcantly higher risks of colorectal, liver, and pancreatic cancers. For patients with diabetes, the use of

TZDs decreased the risk of liver cancer with a marginal statistical signifcance.

Cancer has been the leading cause of death in Taiwan since 1982. The age-adjusted incidence rate has increased steadily since then, and in 2007, 270 new cases per 100 000 individuals were reported in the general population [11]. This trend differs from that of the United States, where data from Surveillance Epidemiology and End Results showed that the overall cancer incidence rate decreased by 0.7% per year between 1999 and 2006 [12]. Because cancer continues to be a challenge for public health in Taiwan, it has come to the attention of the government, resulting in population-based investigations

regarding cancer-preventive epidemiology. The NHI program provides comprehensive health care coverage, and the NHIRD contains data on ambulatory service records, hospital service records, and prescription claims. This database enabled us to select patients for study who were representative of the

underlying population. Previously, we used the data to evaluate the risk of malignancy for patients with end-stage renal

disease, and uncovered a number of positive fndings, which have been published [13]. The current study used a similar design in an attempt to determine whether the use of TZDs is

associated with the risk of cancer. The factors affecting cancer incidence in the diabetic

population are diverse and complex. Epidemiologic evidence suggests that the risk of cancer is increased in diabetic patients [14, 15]. Type 2 diabetes and cancer are well-known to share many risk factors. However, most observational studies have not examined the potential association between anti-diabetic medication use and cancer risk. Evidence from some

observational studies suggests that some medications used to treat hyperglycemia are associated with either an increased or reduced risk of cancer [14]. To our knowledge, this was the frst population-based study conducted in Taiwan to compare cancer rates among diabetic patients who were either using TZDs or those who were not. The data of large samples of TZD users (4158 patients) and non-TZD users (18 752

patients) among the type 2 diabetic patients were compared. To create a comparison group, we randomly matched each patient with four persons from the general population without diabetes, based on the demographic variables (year birthday, sex, and index year).

Using the non-diabetes comparison group as a reference, our data revealed that the risk of cancer for diabetic patients

(TZDs + non-TZDs) was signifcantly higher than for the nondiabetes cohort by a 20% difference. The risks in the TZDs

group and the non-TZDs group were of similar strength. However, the HRs were statistically signifcant in the non-TZDs group but not in the non-TZDs group possibly because of a relatively small sample size of the TZDs group. We thought

that diabetes itself rather than anti-diabetic medications should be accounting for the risk of cancer. In contrast, studies have indicated that TZDs inhibit tumor growth, progression, and metastasis in numerous types of cancer by activating individual PPARγ, and may potentially play a chemopreventive role [3, 4,

16]. Although PPARγ activators show anticancer effects on cell

lines, testing in human clinical trials with advanced cancer patients has met with limited success [17]. If we focused on the

diabetes group patients, we found that TZDs’ use reduced overall cancer risk only by 2% compared with non-TZDs’ use. With regard to the age difference, the fgure revealed some specifc patterns and implied that a higher cancer risk can be expected for diabetic patients >60 years of age compared with non-diabetic subjects, and a lower cancer risk can be expected in the TZDs group than in the non-TZDs group after 80 years of age.

For site-specifc cancer risk, our study showed that non-TZDs diabetic patients were at a signifcantly higher risk of colorectal, liver, and pancreatic cancers. For diabetic patients using TZDs, the risk of colorectal and pancreatic cancers and lymphoma was signifcantly higher. It is compatible with earlier studies, which found that diabetes is a risk factor for colorectal, liver, and pancreatic cancers, as well as lymphoma [14, 15, 18, 19]. PPARγ indeed has many faces and is well known for its role in the cellular proliferation [20]. Activation of individual PPARγ has been implicated in breast, cervical, colon, prostate, and lung cancers [3, 16, 17]. Once activated, PPARγ preferentially binds with retinoid X receptor α and signal antiproliferative, antiangiogenic, and prodifferentiation pathways in these cell types, thus making PPARγ a highly useful target for down-regulation of carcinogenesis [21, 22]. Our results did not reveal any chemopreventive effect by TZDs in these cancers, and the shared risk factors for cancer and diabetes may counteract its effect. On the other hand, earlier investigators found that TZDs promote growth in colon

tumors with mutations in the APC gene [23, 24]. Furthermore, the published trials on TZDs demonstrate that they produce substantial body weight gain [25, 26], which consequently

increases the risk of cancer of the breast, colon, prostate, endometrium, and kidney [27]. This factor may also offset the effect of chemopreventive role of TZDs on colon and prostate cancers, as shown by our results.

Recent studies on TZDs, particularly pioglitazone, have

shown that the medication may increase bladder cancer risk, as evidenced by a higher-than-expected frequency of bladder cancer among pioglitazone users [9, 28, 29]. However, the current fndings did not confrm this pattern, and are consistent with the report from Tseng [30].

When we focused on the diabetic patients, for those taking anti-diabetic drugs other than TZDs, our study revealed no signifcant difference in the overall cancer risk when compared with the TZDs group, but a marginally signifcantly higher risk of liver cancer in the non-TZDs group. The fndings may support the potential chemopreventive effect of TZDs on liver development of cancer in diabetic patients, which has been suggested by Chang et al. [31]. Evidence from in vivo studies has also shown that TZDs inhibit tumor formation in the liver [32], providing a plausible biological basis for these fndings. For the dose–response relationship, Table 4 shows that the TZDs group signifcantly increased overall cancer risk at the lowest dose level when compared with non-diabetic subjects. It suggests that longer than 1-year medication of TZDs is

preferred when development of cancer is concerned in diabetic patients. For the treatment duration, our data revealed that the signifcantly higher risk was only observed among patients taking TZDs or non-TZDs <1 year, or taking non-TZDs between 5 and 6 years. One study from Taiwan found that diabetes <2-year duration is associated with pancreatic cancer and long-standing diabetes was not a risk factor for pancreatic cancer [33], and our fndings are partially compatible with this. The study was subject to some limitations, which must be mentioned. First, the NHIRD does not provide detailed

information on patients such as their smoking habits, alcohol consumption, body mass index (BMI), physical activity,

socioeconomic status, and family history of cancer. All of these are major risk factors for numerous cancers, and could

plausibly be associated with diabetes and anti-diabetic medications (either TZDs or non-TZDs). An earlier

epidemiologic study of type 2 diabetes in Taiwan also showed that BMI, physical activity, and cigarette smoking were possible risk factors for newly diagnosed diabetes [34]. These shared risk factors for cancer and diabetes can induce the higher cancer incidence for TZDs and non-TZDs groups at the

baseline, which may hinder the possible chemopreventive effect of cancer for TZDs. However, we still got some signifcantly different results between TZDs and non-TZDs groups in diabetic patients. Second, the evidence derived from a cohort study is generally of a lower methodological quality than that

from randomized trials because a cohort study design is subject to many biases related to adjustment for confounds. Despite

our meticulous study design with adequate control of

confounding factors, a key limitation was that bias could still remain because of possible unmeasured or unknown

confounders (e.g. the difference in the stage of the disease or in risk factors between TZD users and non-TZD users). Third, the diagnoses in NHI claims primarily serve the purpose of administrative billing, and do not undergo verifcation for scientifc purposes. We were unable to contact the patients directly to obtain more information on their use of TZDs because of the anonymity assured by the identifcation

numbers. Furthermore, prescriptions for the study drugs issued before 1996 were excluded from our analysis. This omission could have resulted in the underestimation of the cumulative dosage and may have weakened the observed association. However, the data that we obtained on TZDs prescriptions and cancer diagnoses were highly reliable. Last, the small number of cancers in the TZD group, particularly in the analysis stratifed by age and on the incidence of cancer at different sites, suggests cautious interpretation of the results of the study. Failure to fnd an association for types of cancer may be due to the small number of cancer cases.

In conclusion, this population-based retrospective cohort study found that diabetes is signifcantly associated with an increase in the risk for overall cancer incidence, and TZDs

or non-TZDs correlated with the increased risks for some sites of cancer. However, the use of TZDs is probably associated with a decreased incidence of liver cancer risk in diabetic patients. These fndings may be partially explained by a biologically plausible link between type 2 diabetes and cancer outcomes, as well as some plausibly pharmacologic mechanisms. However, underlying mechanisms must still be explored and identifed. Additional large population-based unbiased studies are required, and it would be essential to confrm our current fndings before drawing any frm conclusions.

funding

This work was supported by study projects (DMR-101-061 and DMR-101-080) at China Medical University Hospital, by

Taiwan’s Department of Health Clinical Trial and Research Center for Excellence (DOH101-TD-B-111-004), and by Taiwan’s Department of Health Cancer Research Center for Excellence (DOH101-TD-C-111-005).

disclosures