BACKGROUND: Baseline hypoglycemia rates are generally not collected or included as a covariate in statistical models used for analyzing hypoglycemia data. The objective of the present study was to examine the effect of adjusting for baseline hypoglycemia on estimation efficiency and statistical power. SUBJECTS AND METHODS: A post hoc analysis of data from 15 insulin trials, including patients with type 1 diabetes mellitus (T1DM) (n=210), previously insulin-treated type 2 diabetes mellitus (T2DM) (n=1,511), or T2DM and previously insulin-naive (n=1,075). Hypoglycemic episodes were analyzed with a negative binomial regression model. RESULTS: Baseline nocturnal hypoglycemia rate was significantly correlated with post-baseline nocturnal hypoglycemia rate in previously insulin-treated patients with T1DM and T2DM (correlation range, 0.37-0.63; P<0.001). Adjusting for baseline hypoglycemia resulted in a reduction in the SE for negative binomial regression for previously insulin-treated patients with T1DM and T2DM (range, 2.2-11.8%) and increased statistical power. Modeling of lengthening the lead-in period increases the correlation between baseline and post-baseline hypoglycemia event rate and statistical power. CONCLUSIONS: Baseline hypoglycemia rate is significantly correlated with post-baseline hypoglycemia rate for patients with diabetes treated with insulin prior to randomization. The length of the lead-in period can impact correlations between baseline and post-baseline data, and adjustment for baseline hypoglycemia may improve the estimation efficiency for hypoglycemia data analyses in clinical trials.
BACKGROUND: Baseline hypoglycemia rates are generally not collected or included as a covariate in statistical models used for analyzing hypoglycemia data. The objective of the present study was to examine the effect of adjusting for baseline hypoglycemia on estimation efficiency and statistical power. SUBJECTS AND METHODS: A post hoc analysis of data from 15 insulin trials, including patients with type 1 diabetes mellitus (T1DM) (n=210), previously insulin-treated type 2 diabetes mellitus (T2DM) (n=1,511), or T2DM and previously insulin-naive (n=1,075). Hypoglycemic episodes were analyzed with a negative binomial regression model. RESULTS: Baseline nocturnal hypoglycemia rate was significantly correlated with post-baseline nocturnal hypoglycemia rate in previously insulin-treated patients with T1DM and T2DM (correlation range, 0.37-0.63; P<0.001). Adjusting for baseline hypoglycemia resulted in a reduction in the SE for negative binomial regression for previously insulin-treated patients with T1DM and T2DM (range, 2.2-11.8%) and increased statistical power. Modeling of lengthening the lead-in period increases the correlation between baseline and post-baseline hypoglycemia event rate and statistical power. CONCLUSIONS: Baseline hypoglycemia rate is significantly correlated with post-baseline hypoglycemia rate for patients with diabetes treated with insulin prior to randomization. The length of the lead-in period can impact correlations between baseline and post-baseline data, and adjustment for baseline hypoglycemia may improve the estimation efficiency for hypoglycemia data analyses in clinical trials.
Authors: M J Davies; D Russell-Jones; J-L Selam; T S Bailey; Z Kerényi; J Luo; J Bue-Valleskey; T Iványi; M L Hartman; J G Jacobson; S J Jacober Journal: Diabetes Obes Metab Date: 2016-08-12 Impact factor: 6.577
Authors: R M Bergenstal; H Lunt; E Franek; F Travert; J Mou; Y Qu; C J Antalis; M L Hartman; M Rosilio; S J Jacober; E J Bastyr Journal: Diabetes Obes Metab Date: 2016-08-03 Impact factor: 6.577