Liver‐related long‐term outcomes of thiazolidinedione use in persons with type 2 diabetes

Abstract Background & Aims Studies have described prominent histologic improvement in patients with nonalcoholic steatohepatitis (NASH) using thiazolidinedione (TZD); however, these were all short term with moderate sample size, no liver‐related long‐term outcomes could be noted. Methods This retrospective cohort study enrolled patients with newly diagnosed type 2 diabetes mellitus (T2DM) from Taiwan's National Health Insurance Research Database between 1 January 2000 and 31 December 2013. We matched TZD users and nonusers at a 1:1 ratio through propensity score matching. This study included 5095 paired TZD users and nonusers. Cox proportional hazard models were used to compare the risks of cirrhosis, hepatic decompensation, hepatic failure and all‐cause mortality between TZD users and nonusers. The Kaplan‐Meier method was used to compare the cumulative incidence of these main outcomes. Results The incidence rates of cirrhosis, hepatic decompensation, hepatic failure and all‐cause mortality during follow‐up were 0.77 vs 1.95, 1.43 vs 1.75, 0.36 vs 0.70, and 4.89 vs 3.78 per 1000 person‐years between TZD users and nonusers. The adjusted hazard ratios of cirrhosis, hepatic decompensation, hepatic failure and all‐cause mortality were 0.39 (95% confidence interval [CI]: 0.21‐0.72), 0.86 (95% CI: 0.52‐1.44), 0.46 (95% CI: 0.18‐1.17) and 1.18 (95% CI: 0.87‐1.61) respectively. Conclusions Our study demonstrated that TZD use could significantly lower the risk of cirrhosis. In clinical settings, TZD use might be able to improve liver‐related long‐term outcomes.


| INTRODUC TI ON
Owing to a sedentary lifestyle and westernized diet, the prevalence of type 2 diabetes mellitus (T2DM) and nonalcoholic fatty liver disease (NAFLD) has dramatically increased worldwide. According the IDF diabetes atlas, globally, diabetes mellitus (DM) cases increased from 151 million in 2000 to 425 million in 2017, representing an approximately 2.8-fold increase in 17 years. 1 In Taiwan, DM cases also increased from 707 000 in 2000 2 to 1 958 000 in 2017, representing an approximately 2.77-fold increase in 17 years. NAFLD is a new epidemic and is the most common cause of chronic liver disease 3 ; its estimated worldwide prevalence is approximately 15%-30%. 4 People with T2DM frequently have dyslipidaemia and NAFLD. Approximately 40%-70% of patients with T2DM have NAFLD 5,6 ; in Taiwan, approximately 43.3% of T2DM have NAFLD. 7 NAFLD can progress to nonalcoholic steatohepatitis (NASH), hepatic fibrosis, cirrhosis and even hepatocellular carcinoma (HCC) 8 ; it can also aggravate cardiovascular events in patients with T2DM. 9 Furthermore, in patients with NAFLD, diabetes can increase the risks of hepatic complications and death. 10 Thiazolidinediones (TZDs) are one of the most promising medications for treating NAFLD; studies have revealed histological improvement in patients with NASH, and fibrosis was even attenuated in some patients. [11][12][13][14][15] TZDs bind and activate the nuclear receptor of peroxisome proliferator-activated receptor gamma (PPARγ) with strong insulin-sensitizing activity. They ameliorate insulin resistance by acting on adipose tissue, muscle and liver to increase glucose utilization and decrease glucose production. They can also increase adiponectin levels, reduce free fatty acid influx, increase fatty acid oxidation, and then decrease liver fat and attenuate hepatic inflammation. 16 But until now, most of the TZD studies in patients with NAFLD or NASH are short term with few number of participants, no long-term liver outcomes were noted. Therefore, we performed this nationwide cohort study to evaluate the liver-related outcomes of TZD use in persons with type 2 diabetes.

| Study design and patients
In Taiwan, the National Health Insurance (NHI) programme has been implemented since 1995. At least 99% of the 23.5 million population of Taiwan are registered in this insurance programme. 17 The National ants of the NHI programme in 2000. This cohort study was conducted using the LHID2000. All information that could be used to identify individuals or care providers was encrypted. This study was approved by the Research Ethics Committee of China Medical University and Hospital (CMUH104-REC2-115) and was granted a waiver of informed consent.
We selected patients with type 2 diabetes (T2DM) diagnosis ascertained through the presence of the ICD-9-CM code 250.xx in at least two outpatient records over 1 year or in one inpatient record in the LHID2000 between 1 January 2000 and 31 December 2013. We excluded patients diagnosed with DM before 1 January 2000, to ensure that only incident T2DM cases were included.
This study excluded individuals younger than 30 years or older than 80 years, having follow-up less than 180 days and those diagnosed with type 1 diabetes (Table S1)

| Procedures
We defined the first date of TZD use by our patients as the index date. Those who had not used any TZD in the observation period were considered TZD nonusers. Each TZD nonuser was randomly assigned an index date according to the corresponding index date of a TZD user. The TZDs examined in this study included pioglitazone and rosiglitazone (Troglitazone and ciglitazone were not used in Taiwan). The covariates analysed in multivariable models included baseline demographics (we grouped the diagnoses of overweight, abnormal weight gain, and BMI 25-29 as overweight; obesity, BMI 30-39, obesity complicated pregnancy as obesity; severe obesity, BMI ≥ 40, and bariatric surgery status for obesity as severe obesity), comorbidities diagnosed 1 year before the index date, and medications including antidiabetic agents, antihypertensive drugs, statin and aspirin. We used the Charlson comorbidity index (CCI) to quantify patients' comorbidity profiles 20 and the Diabetes Complications Severity Index (DCSI) score 21 to define the severity of diabetes. CCI and DCSI scores were calculated according to patients' records in the NHIRD 1 year before the index date.

| Liver-related long-term outcomes
Using ICD-9-CM codes in medical records, we assessed the incidence rates of cirrhosis, hepatic decompensation (the composite of oesophageal varices, ascites, hepatic encephalopathy and jaundice), 22 oesophageal varices, abdominal ascites, hepatic encephalopathy, jaundice, hepatic failure and HCC to determine liver-related long-term outcomes. We did a sensitivity analysis by excluding patients diagnosed with cirrhosis, oesophageal varices, hepatic ascites, hepatic encephalopathy, jaundice, hepatic failure, HCC or death within 365 days after index date.

| Statistical analyses
Propensity score matching was used to optimize comparability between TZD users and nonusers. 23 The propensity score was estimated for every patient using a nonparsimonious multivariable logistic regression, with TZD use as the dependent variable. In all, 26 clinically relevant covariates were used as independent variables (Table 1) in the regression. The nearest-neighbour algorithm was applied to construct matched pairs, assuming that a proportion of 0.995-1.0 was perfect.
Cox proportional hazard models were used to compare the outcomes between TZD users and nonusers. All analyses were conducted using an intention-to-treat approach in accordance with the initial TZD assignment, irrespective of subsequent changes to other antidiabetic medications. The results are expressed as hazard ratios (HRs) with 95% confidence intervals (CIs). To calculate the risk of mortality, we censored patients at the time of death or the end of study, whichever occurred first. To calculate the risks of other investigated outcomes, we censored patients on the respective events or at the end of follow-up on 31 December 2013, whichever occurred first.
Using the Kaplan-Meier method, we compared the cumulative incidence of cirrhosis over time between TZD users and nonusers.
We performed subgroup analysis according to pre-specified strata of clinical interest to assess effect modification. The subgroup strata included overall rosiglitazone and pioglitazone use; sex; CCI score; oral antidiabetic drugs; insulin; antihypertensive drugs and statin. We calculate the P for interaction to see the different effects of variables in the same subgroup.
A two-tailed P value less than .05 was considered significant. SAS version 9.2 was used for analyses.

| Main outcomes
In this study, 96 (1.88%) TZD users and 76 (1.49%) nonusers died during the follow-up period (incidence rate: 4.89 vs 3.78 per 1000 person-years). Multivariable models showed that TZD users had no significant difference of mortality (aHR: 1.18, P = .27; Table 2). For liver-related outcomes, TZD users appeared to have lower risks of cirrhosis (incidence rate: 0.77 vs 1.95 per 1000 person-years; aHR: 0.39, P = .002; Table 2); TZD users appeared to have no significant difference in the risks of hepatic decompensation (aHR: 0.86, P = .58), hepatic failure (aHR: 0.46, P = .10) and HCC (aHR:1.22, P = .35) during the follow-up period. Figure 2 delineates the cumulative incidence rates of cirrhosis between TZD users and nonusers, which were determined using the Kaplan-Meier method. Table 3 presents the results of subgroup analysis of cirrhosis between TZD users and nonusers. Compared with TZD nonusers, rosiglitazone and pioglitazone users, men, patients undergoing insulin treatment, using ≧3 oral antidiabetic drugs, and statin nonusers had a significantly lower risk of cirrhosis. We used the p for interaction to compare rosiglitazone and pioglitazone on the effect of cirrhosis, which show no significant difference (P = .9209).

| Sensitivity test
After excluding patients with liver-related events or death within 365 days after index date, the adjusted HR of cirrhosis in TZD users was 0.37 (95% CI: 0.20-0.65, P = .0006) compared with nonusers.
TZD users had a significantly lower risk of cirrhosis (Tables S2 and S3).

| D ISCUSS I ON
Our study demonstrated that TZD use in T2DM could significantly decrease the risk of cirrhosis. Subgroup analysis revealed that both rosiglitazone and pioglitazone could lower the risks of cirrhosis with no significantly different effects between these two drugs. and fibrosis was noted. 15 The systemic review and meta-analysis of TZD use in patents with NASH has revealed that TZD could reduce liver fat, normalize aminotransferase levels and improve histological steatosis, ballooning and inflammation. 24 These researches indicated that TZD use in patients with NASH could attenuate hepatic injury, inflammation and even fibrosis. However, these studies were all short-term clinical trials with a moderate sample size; no large series study of long-term liver-related outcomes has been conducted.
To the best of our knowledge, our study is the first large series cohort study investigating liver-related long-term outcomes of TZD use in patients with T2DM. Though our patients were not image or histology confirmed NAFLD or NASH cases; however, based on epidemiological studies, at least 50% of patients with T2DM exhibit NAFLD. 5 In addition, we excluded patients with previous viral hepatitis and alcoholism to make our population more similar to patients with NAFLD. This study indicated that TZD use in patients with T2DM might be able to prevent the development of cirrhosis; T2DM patients with the risk of hepatic injury could use TZD to prevent bad liver-related long-term outcomes.
Through the activation of PPARγ, TZD can reduce insulin resistance, sequester fatty acid in adipose tissue, and alleviate fat storage, steatosis and ballooning in the liver. TZDs also can activate AMP-activated protein kinase (AMPK) and reduce hepatic fat content. 25 TZD use can increase adiponectin and reduce high-sensitivity C-reactive protein (hs-CRP), TNFα, IL-1β and IL-6 levels, which can reduce hepatic inflammation. 26,27 In preclinical studies, TZDs bind to PPARγ and thus inhibit the activation of hepatic stellate cells, reduce extracellular matrix production, decrease transforming growth factor β1 expression, attenuate matrix remodelling, and protect for tissue repair, fibrosis and even cirrhosis. 28 TZD can induce cell cycle arrest and apoptosis and inhibit cancer cell proliferation and invasion through the activation of PPARγ. 29 Some studies have revealed that TZD use may decrease the risk of HCC in T2DM. 30,31 However, another study provided contrasting results. 32 A meta-analysis disclosed that TZD does not decrease the risk of HCC, 33 which is consistent with our result.
Our study has several strengths. We recruited patients from the NHIRD, which covers approximately 99% of the population of Taiwan.
This might be able to decrease the risk of selection bias in the study.
We used medical records instead of self-reports, which might decrease recall bias and more correctly censor the incident rates of the main outcomes. The events noted within 6 months after the index date were F I G U R E 2 Cumulative incidence of cirrhosis between Thiazolidinediones users and nonusers in T2DM through Kaplan-Meier excluded to decrease the possibility of latent morbidity and mortality.
We did a sensitivity test to exclude patients with liver-related events or death within 365 days after the index date, which also revealed that TZD could significantly lower the risk of cirrhosis.
Our study has some limitations. First, the NHIRD does not contain detailed information on patients' lifestyle, height and body weight, and family history; all of which might influence the measured outcomes. Although we took many codings to include overweight, obesity and severe obesity as covariates in analysis, many patients' obesity might not be recorded in the database, which could lead to the underestimation of the prevalence of obesity in our study. Second, the use of ICD-9 codings for the censoring of cases in administrative databases has been criticized about its accuracy. In this study, the algorithm of using ICD-9 to define type 2 diabetes and cirrhosis has been validated in the Taiwan National with inevitable biases certainly existed. A larger randomized control study should be conducted to observe the liver-specific endpoints after TZD use in patients with T2DM.

| CON CLUS IONS
Our nationwide cohort study revealed that compared with TZD nonuse, TZD use in type 2 diabetes could significantly lower the risk of cirrhosis. In clinical settings, TZD use in T2DM patients might be able to improve their liver-related long-term outcomes.

ACK N OWLED G EM ENTS
This study was funded by grants from the Ministry of Health and funders; no writing support was from the funders. The corresponding authors had complete access to all data in the study and had final responsibility for the decision to submit for publication. This manuscript was edited by Wallace Academic Editing.

CO N FLI C T O F I NTE R E S T
The authors do not have any disclosures to report.

Specific author contributions: Dr Ming-Chih Hou and Chih-Cheng
Hsu had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of data analysis.