A Prospective, Randomized Comparison of the Metabolic Effects of Pioglitazone or Rosiglitazone in Patients With Type 2 Diabetes Who Were Previously Treated With Troglitazone

This was an open-label, randomized comparison of PIO and ROSI in patients previously stabilized on TROG in combination with various other glucose-lowering medications. All patients were derived from a single diabetes center associated with a major teaching hospital serving a large metropolitan, Midwestern city. As outlined by the Declaration of Helsinki and Title 21 of the U.S. Code of Federal Regulations, the center’s Human Subjects Research Committee reviewed and approved the protocol. Participating subjects gave written informed consent. Any patient currently taking TROG and with stable liver function assessed by liver enzymes (alanine and aspartate aminotransferase) was eligible to participate. Information regarding standard demography and concurrent medications at the time of randomization was compiled in all participants. Before stopping TROG therapy, baseline assessment of weight and clinical laboratory values were completed. These measures were repeated after 4 months of randomized therapy with either PIO or ROSI and included liver enzymes (alanine and aspartate aminotransferase), HbA_1c, and fasting plasma lipids (total cholesterol, LDL cholesterol, HDL cholesterol, and triglycerides). Concurrent lipid-lowering therapy was held constant during the 4-month study period. Daily thiazolidinedione dose conversion was accomplished using each manufacturer’s marketed dosage forms and the knowledge that all three compounds demonstrated relatively linear, dose-dependent pharmacokinetics (1,2,12). For patients currently taking TROG 600 mg/day, either PIO 45 mg/day or ROSI 4 mg b.i.d. was used; patients taking TROG 400 mg/day were given either PIO 30 mg/day or ROSI 4 mg/day; and patients taking TROG 200 mg/day were given either PIO 15 mg/day or ROSI 2 mg/day.

The major end points of this study were weight, HbA_1c, and plasma lipids. Total cholesterol and triglycerides were measured on a Vitros 950 (Ortho Clinical Diagnostics, Rochester, NY); HDL cholesterol and LDL cholesterol (when required) were analyzed using a Dimension RXL (Dade, Wilmington, DE). LDL cholesterol concentrations were estimated using the Friedewald equation when triglyceride levels were <300 mg/dl (13). LDL cholesterol was measured directly when triglycerides exceeded 300 mg/dl. HbA_1c was analyzed using the DCA2000 latex immunoglutination inhibition system (Bayer, Tarrytown, NY)

Univariate analysis of baseline and follow-up demography and clinical laboratory end points, both between and within the PIO and ROSI treatments, were accomplished using Pearson χ², unpaired, or paired Student’s t test where appropriate. Changes in HbA_1c, weight, triglyceride, total cholesterol, LDL cholesterol, and HDL cholesterol over time were computed as values at month 4 minus baseline values. General linear models (SAS, Cary, NC) were used to assess the impact and potential interactions of various independent variables on the change in the outcome measures listed above. Independent variables included gender, age, thiazolidinedione (PIO or ROSI), thiazolidinedione dose, days exposed to TROG, and concurrent medication present or absent HMG-CoA reductase inhibitor, insulin, sulfonylurea, and/or metformin). Initial models screened for the significance of independent variables and their interaction in the models. Final models contained only those factors identified as significant from the initial models. One-way ANOVA (SPSS, Chicago, IL) using the Dunnett C procedure was used to assess significance of change in cholesterol from baseline in four subgroups: those taking either thiazolidinedione alone (PIO or ROSI) or in combination with an HMG-CoA reductase inhibitor. Statistical significance was considered at P < 0.05.

After withdrawal of TROG and after a 2-week washout period, a cohort of 186 TROG-treated patients were randomized to either PIO or ROSI therapy. An equal number of patients from each assigned group subsequently refused to start randomized treatment, stopped thiazolidinedione therapy of their own accord, switched care to a different clinic, or were withdrawn before completing the protocol (14 patients in the PIO group, 16 patients in the ROSI group). Many of these patients ascribed their actions to concern about potential liver toxicity with their new therapy. Other participants (17 patients in the PIO group, 12 patients in the ROSI group) were excluded because of incomplete or unusable data (i.e., did not complete baseline assessment, follow-up assessment, or both and/or nonfasting blood draw). A total of 127 randomized patients (67 in the PIO group, 60 in the ROSI group) completed follow-up and had usable data. After review and based on available data, this group did not seem to be significantly different from those who were excluded from analysis.

The baseline characteristics of this cohort are listed in Table 1. There was no significant difference in gender mix, age, weight, fasting plasma lipids (total cholesterol, LDL cholesterol, HDL cholesterol, triglycerides), or HbA_1c between the ROSI and PIO groups. Additionally, no differences in the proportions of concurrent medications used by the patients in these groups were apparent (Table 1). Maximal PIO dose (45 mg/day) was used in 76.1% (51 of 67 patients) of the PIO cohort; similarly, ROSI (4 mg b.i.d.) was used in 76.7% (46 of 60 patients) of the ROSI cohort. The midrange PIO dose (30 mg/day) was used in 22.4% (15 of 67 patients) of the PIO group, and the midrange ROSI dose (4 mg/day) was used in 16.7% (10 of 60 patients) of the ROSI group. The remaining subjects in each group received the low dose of each drug. The mean ± SD exposure to randomized therapy for the PIO and ROSI groups was 138.8 ± 30.4 and 134.0 ± 30.1 days, respectively (P > 0.3).

The mean changes from baseline for weight, HbA_1c, total cholesterol, HDL cholesterol, LDL cholesterol, and triglycerides are shown in Fig. 1. Both groups experienced significant weight gain versus their baseline starting point (P < 0.01). No change in glycemic control was apparent at the 4-month assessment point when compared with initial HbA_1c. Changes in lipid profile were detected; however, these were only significant for the PIO group versus baseline and when compared with changes noted in the ROSI group.

Equivalent weight gain after conversion from TROG was noted in both the PIO and ROSI groups. However, significant effects of concurrent use of sulfonylurea (P = 0.027) and/or metformin (P = 0.047) were noted in the final model, using change in weight adjusted for thiazolidinedione, concurrent sulfa, and concurrent metformin. Results from general linear model analysis are shown in Table 2, showing that thiazolidinedione and concurrent HMG-CoA reductase inhibitor had significant effects with respect to change in cholesterol. No significant statistical interaction between thiazolidinedione and HMG-CoA reductase inhibitor was detected. However, somewhat opposite effects on cholesterol were noted between the two groups. Specifically, a much greater effect in both direction and quantity was seen in those treated with PIO alone or with concurrent HMG-CoA reductase inhibitor versus that seen in the respective ROSI groups (Table 3).

The need to convert patients to either PIO or ROSI because of the withdrawal of TROG from worldwide markets provided a unique opportunity to study potential differential effects of a new class of insulin-sensitizing agents. Other groups have published clinical observations within this same paradigm; however, to date, no studies have randomly assigned patients to either of the alternate therapies (9,10). Our prospective, randomized conversion study demonstrated similar effects of PIO and ROSI with respect to weight gain and glycemic control after conversion from TROG. However, differences in lipid profile were detected. Because a significant number of the study patients were concurrently taking medications that could have had effects on serum lipid levels, further analysis of data were warranted. Using change in cholesterol as the dependent variable, we used statistical models that adjusted for demographic and co-medication factors and demonstrated that two factors studied, use of thiazolidinedione and HMG-CoA reductase inhibitor, had significant effects. Subgroup analysis of HMG-CoA reductase inhibitor use by thiazolidinedione use demonstrated that conversion from TROG to ROSI or PIO resulted in differential lipid profile changes that seem to be influenced by the presence of HMG-CoA reductase inhibitor.

Although our study was not designed to elucidate possible mechanisms underlying these clinical observations, recent studies and clinical case reports implicate one possible mechanism that is consistent with our observations. TROG is known to induce hepatic cytochrome P450 3A (CYP3A), whereas PIO and ROSI seem to lack significant enzyme-induction effects (1,2,14–16). TROG is also known to decrease serum concentrations of many CYP3A substrates, including some of the HMG-CoA reductase inhibitors, atorvastatin and presumably simvastatin (17– 21). A short-term, prospective, crossover study in normal volunteers documented significant decreases in atorvastatin serum concentrations with concurrent TROG exposure; however, no change in lipid lowering from atorvastatin was noted (19,20). Conversely, worsening of lipid profile has been noted in dyslipidemic patients when TROG was added to existing atorvastatin or simvastatin therapy (17,18). In these cases, TROG-related CYP3A enzyme induction may have resulted in decreased statin exposure and worsening plasma lipid levels. When considering our study, the opposite situation may have been detected, considering that the preponderance of concurrent statin use in our study participants were CYP3A substrates (Table 3). Loss of enzyme induction with the discontinuation of TROG, in statin-treated patients, may have enhanced the observed reduction in lipids associated with PIO conversion and blunted the increase in lipids seen with conversion to ROSI (see Table 3).

Our project incorporates a randomization step in an attempt to control for potential bias inherent to retrospective or prospective analysis of clinical data. In this context, we report that significant changes in lipid profile were apparent despite similar glycemic control and weight gain when patients were randomly converted from TROG therapy to either PIO or ROSI. These changes were apparent after 4 months of therapy and persisted after adjusting for various demographic and concurrent medication factors. Although the mechanisms and clinical significance of these differences are yet to be defined, future thiazolidinedione studies should consider the potential confounding effects of concurrent medications commonly used in the diabetes patient population.

We thank Marcia Madden, ANP; Jodie Duntley, CNS; Kathleen Hall, MD; Mohammad Fareed, MD; and the staff of the Hennepin Center for Diabetes and Endocrinology for their various contributions to this project. We also thank Leigh Hamersten and Clifford Hill for their contribution with respect to the data collection.

Portions of these data were presented in abstract form at the 61st Scientific Sessions of the American Diabetes Association, Philadelphia, Pennsylvania, 22–26 June 2001.