Research Design and Methods
Patients between the ages of 21 and 70 years, with the diagnosis of type 2 diabetes within the previous 2 months and who were treatment naive were recruited from Parkland Memorial Hospital inpatient and outpatient services or by self-referral to the Clinical Diabetes Research Clinic at University of Texas Southwestern. Patients with type 1 diabetes–related antibodies, a baseline A1C level < 7%, an elevated serum creatinine level, a clinical history of heart failure, a history of lactic acidosis, untreated proliferative diabetic retinopathy, any life-threatening conditions, or use of more than two alcoholic drinks/day or illicit drug use within the 6 months before enrollment were excluded. Women who were pregnant or desired to become pregnant were not enrolled. The study was approved by the institutional review board of the University of Texas Southwestern, and written informed consent was obtained from all subjects preceding the start of the study.
This was an open-label randomized trial comparing triple oral therapy with an insulin plus metformin regimen. After enrollment, insulin and metformin were initiated in all patients for a 3-month lead-in treatment period. This lead-in period had a dual purpose: 1) to hom*ogenize the glycemic control of the study population at the time of randomization; and 2) to expose all subjects to an insulin-based treatment regimen that would serve as a real-life comparison for treatment satisfaction and lifestyle impact assessment after randomization.
Diabetes education and nutritional counseling were provided to all patients at enrollment in the study and reinforced at the time of randomization. Upon enrollment, treatment was started with 0.2 unit · kg−1 · 24 h−1 of Insulin NovoLog Mix 70/30 with Flex Pen delivery, divided into two equal doses to be injected immediately before breakfast and supper. Metformin was started at a dose of 500 mg/day and increased weekly by increments of 500 mg/day to a goal dose of 1,000 mg twice daily. Results of this study period were published previously.[17]
After 3 months of treatment, patients were randomly assigned to either continue insulin and metformin or begin triple oral therapy. Treatment assignment was determined with a stratified, block randomization scheme programmed by the biostatistician (B.A.H.) using SAS Proc Plan software. The randomization was stratified by race (African American or non–African American) and BMI (< 35 kg/m2 or ≥35 kg/m2), generating four blocked, randomized lists of treatment assignments, one for each stratum. The principal investigator assigned treatment sequentially from these randomized lists as the participant reached the randomization visit.
Patients randomly assigned to triple oral therapy continued metformin and started 1.25 mg glyburide twice daily and 15 mg daily pioglitazone. Pioglitazone was titrated monthly to a final dose of 45 mg daily. Titration of insulin and glyburide (up to the highest clinically effective dose of 10 mg daily) was performed by the study physician throughout the study, based on home blood glucose monitoring logs targeting a fasting blood glucose level of 70–110 mg/dl and postprandial blood glucose level of < 140 mg/dl. All patients were asked to monitor blood glucose at least twice daily, regardless of the group assignment. Initiation and dose adjustment of antihypertensive and lipid-lowering agents were allowed if medically necessary. Patients were followed at the Clinical Diabetes Research Clinic at University of Texas Southwestern monthly for the first 4 months, at 6 months after randomization, and every 3 months thereafter for a total of 36 months. "Treatment failure," a predefined study end point, was defined as A1C > 8% and confirmed by a second reading, occurring after maximization of the glyburide dose or adequate insulin dose adjustments. Volunteers randomly assigned to the triple oral group who reached this end point were transitioned to insulin and metformin treatment, whereas those randomly assigned to insulin continued with the same treatment. Follow-up after treatment failure continued as scheduled.
Measurements
A1C was performed at each visit, using high-performance liquid chromatography in the Clinical Diabetes Laboratory at University of Texas Southwestern. Routine chemistry studies, hematology, and a lipid panel were performed by a commercial laboratory (Quest Diagnostics, Irving, TX).
Weight, blood pressure, hypoglycemic events, and compliance were assessed at every visit. Mild hypoglycemic episodes were defined as symptoms indicative of low blood glucose accompanied by a documented capillary blood glucose value of < 70 mg/dl. Severe hypoglycemia was defined as symptoms of hypoglycemia that required assistance from another individual for treatment, regardless of capillary blood glucose level. Patients were instructed to return their unused medications at every visit for inventory and estimation of patient compliance. We reported the average compliance of all study medications in each group.
QoL was measured at randomization and also 6 and 18 months later using the modified Diabetes Quality of Life Clinical Trial Questionnaire (supplementary material available in an online appendix at https://care.diabetesjournals.org/cgi/content/full/dc09-0653/DC1). This questionnaire addresses several areas with respect to diabetes QoL: satisfaction with treatment, impact of treatment, worry about future effects of diabetes, and worry about social issues,[18] in addition to a hypoglycemia worry scale, a lifestyle flexibility scale, and five separate questions concerning the patient's treatment satisfaction with insulin and perception of their own health.[19] Answers are in the form of a Likert scale score of 1–5, with a lower score demonstrating greater impact, worry, or satisfaction. For patients randomly assigned to triple oral therapy, questions regarding treatment satisfaction with insulin were omitted. For each subscale, the mean of individual item scores was reported. This questionnaire was chosen because it addresses illness-specific issues, as well as insulin treatment issues[20] to best identify excess disease burden due to insulin treatment.
Statistical Analysis
For continuous variables, we computed means ± SD and 95% CI. For categorical variables, we computed percentages. To compare weight gain and A1C control in the presence of missing data due to loss of follow-up or treatment failure, we adopted two strategies. The first strategy was to estimate the slope of the treatment effects using a linear mixed model, with random effects accounting for the correlation among multiple observations from each subject. Then we compared treatment effects based on slope estimation. For this strategy we used all available observations. The second strategy was to use a t test based on complete data from subjects who finished the study ("completers" analysis). Mild and severe hypoglycemic event rates were compared among groups with Poisson regression models using a general estimating equations approach to incorporate the repeated measurements. Responses to the QoL questionnaire at 0, 6, and 18 months were compared between and within groups as repeated measures using mixed models. Statistical significance was declared at 5%.
As specified a priori in the protocol, data collected after treatment failure were not included in the per-protocol analysis described above. To confirm our results, we also performed an intention-to-treat analysis, in which all data were analyzed as randomized. All results presented below are consistent with those obtained under the intention-to-treat analysis.
