Negative binomial meta-regression analysis of combined glycosylated hemoglobin and hypoglycemia outcomes across eleven Phase III and IV studies of insulin glargine compared with neutral protamine Hagedorn insulin in type 1 and type 2 diabetes mellitus.

OBJECTIVES: This analysis first modeled the interaction between hypoglycemia and glycosylated hemoglobin (HbA1c) in clinical trials that compared insulin glargine (glargine) with human neutral protamine Hagedorn insulin (NPH) in patients with type 1 or type 2 diabetes mellitus. The model was then used to compare rates of hypoglycemia associated with use of these insulins.
METHODS: Patient-level data from all randomized Phase III/IV clinical trials sponsored by the manufacturer of glargine that compared glargine and NPH and were available in May 2004 were included in the model. In addition, MEDLINE, EMBASE, and BIOSIS were searched for comparative randomized controlled trials of glargine and NPH using the terms insulin glargine, HOE 901, neutral protamine Hagedorn insulin, and NPH insulin. Studies were excluded from the analysis if patient-level data were not available. Unadjusted rates of symptomatic, confirmed, and severe hypoglycemia were compared with those derived from negative binomial regression analysis, which stratified the results by HbA1c at end point (with last observation carried forward), treatment, and duration of diabetes. In addition, the analysis was stratified by Phase III studies (which focused on determining tolerability and efficacy before regulatory approval) and Phase IV studies (which compared the clinical efficacy of the 2 insulins). The first month of the study was not included in the analysis because of continual adjustment of the insulin dose and maintenance of previous NPH in some studies.
RESULTS: Eleven sponsored randomized trials were included in the model (total of 5074 patients). Four other sponsored trials were not included because the databases were not finalized, and 3 investigator-initiated trials were not included because patient-level data were unavailable. Rates of hypoglycemia had a curvilinear relationship with HbAlc, increasing at lower end-point HbAlc values. In combined analyses of the studies of type 1 and type 2 diabetes, unadjusted rates of hypoglycemia were lower for glargine than NPH: 6.1% lower for all symptomatic hypoglycemia, 21.6% lower for confirmed hypoglycemia, and 23.9% lower for severe hypoglycemia (all, P < 0.05). When modeled using the negative binomial distribution with end-point HbA1c as a covariate, the corresponding results were 9.1% (P < 0.05), 26.6% (P < 0.001), and 30.0% (P = 0.08), respectively. When only Phase IV trials were analyzed, the relative reductions with glargine were 16.2% (P < 0.01), 40.8% (P < 0.01), and 46.8% (P < 0.05). The results of the separate analyses of studies of type 1 and type 2 diabetes were comparable.
CONCLUSIONS: Based on the results of this analysis, calculated unadjusted hypoglycemia event rates appear to underestimate the differences between glargine and NPH. In most of the present analyses, unadjusted rates were significantly lower with glargine than NPH. Adjustment for end-point HbA1c resulted in greater relative reductions in the risk of hypoglycemia for glargine compared with NPH. The adjusted risk reduction with glargine was highest in the Phase IV studies. Copyright 2007 Excerpta Medica, Inc.