Analysis of Clinical Trial DataSTAT 5400H.Kousse, T. Li, H. Somal, F. YangDecember 6, 2010AbstractA clinical trial was conducted to assess the safety and efficacy of a new med-ication for the management of blood glucose in diabetes patients. Two treatments(drug and placebo) were administered to two different groups of diabetes patientsfor twelve weeks. Based on the design of the trial, we analyze whether i) treatmentis associated with improved control of blood glucose levels over the twelve-weektrial, and ii) whether drug treatment is associated with an improvement over andabove that associated with placebo.1 IntroductionA clinical trial was conducted to assess the safety and efficacy of a new medication forthe management of diabetes using multi-center, randomized, placebo controlled, andparallel design. The primary objective was to discern the efficacy of drug treatmentfrom placebo in improving long-term (several months) patient blood glucose control,as measured by glycated hemoglobin levels (HbA1c). Higher HbA1c is associatedwith increased risk for diabetes-related complications. Data on primary efficacy vari-able (HbA1c), secondary efficacy variable (Fasting Blood Glucose), and several safetyvariables were collected at week zero (pre-dose) and week twelve (post-dose).Based on the design of the trial, we justify an analysis to determine whether i) treat-ment (drug or placebo) is associated with significantly improved blood glucose con-trol as indicated by a change in HbA1c and/or Fasting Blood Glucose levels over thetwelve-week trial, and ii) drug treatment is associated with significant improvementover and above that associated with placebo. The analysis comprises three steps:1. Use baseline (pre-dose) measurements to validate randomization by comparing pa-tients in the two treatment groups across all variables.2. Detect for changes in HbA1c and Fasting Blood Glucose levels from pre-dose topost-dose within each treatment group.3. Discern between the effects of drug treatment and placebo on changes in HbA1cand Fasting Blood Glucose levels over the course of the trial.1Employing the analysis described herein, we conclude that i) patients randomly ad-ministered either treatment experienced improved control of blood glucose levels asreflected by the twelve-week change in primary and secondary efficacy measures, andii) patients administered drug treatment experienced an improvement over and abovethat associated with placebo. In the following sections we detail the methods employedto arrive at this conclusion, provide the theoretical justification and quantitative valida-tion of model assumptions, and summarize the results of our analysis.2 Methodology2.1 Trial DesignOf primary interest is the effect of drug intervention on long-term blood glucose con-trol in diabetes patients. Changes in blood glucose experienced by patients in the trialmust be considered relative to the inherent variation in blood glucose levels observedin the population of all diabetes patients. Therefore, clinicians seek to quantify inher-ent variation for the purpose of qualifying treatment effects. Conducting the trial atthree different centers is a means of blocking or separating inherent variation in bloodglucose levels from sytematic variation introduced by ”nuisance factors,” or sources ofvariability that are not of primary interest. Such ”nuisance” factors may be as subtle asthe time of day treatment is administered and the behaviours of employees at the center.In theory, blocking controls for such systematic sources of variation; thus, isolating the”true” variation.This clinical trial can be classified as a two-factor randomized block experiment(Table 1), where the factors and respective levels are:a. Treatment (factor-of-interest)i. Placeboii. Drugb. Center (blocking factor)i. Center 1ii. Center 2iii. Center 3For each patient, measurements are repeated (one at pre-dose and post-dose, respec-tively) so observations are dependent in time. The measure ∆HbA1cicollapses data,taking advantage of the repeated measures structure to make data amenable to random-ized block analysis (Table 1).2.2 AnalysisWith the specified trial design, we analyze the effect of drug intervention via a three-step approach.2Pre-dose Post-doseCenter 1 Center 2 Center 3 Center 1 Center 2 Center 3DrugHbA1c1HbA1c1HbA1c1HbA1c1HbA1c1HbA1c1..................HbA1c9HbA1c10HbA1c10HbA1c9HbA1c10HbA1c10PlaceboHbA1c1HbA1c1HbA1c1HbA1c1HbA1c1HbA1c1..................HbA1c9HbA1c10HbA1c11HbA1c9HbA1c10HbA1c11Table 1: Two-factor randomized block design of clinical trial. Note the repeated mea-sures (over time) structure.1. Use baseline (pre-dose) measurements to validate randomization by comparing pa-tients in the two treatment groups across all variables.Validating the randomization assumption reasonably assures that patientsin both treatment groups are selected from the same underlying popula-tion (diabetes patients); thus, facilitating comparison between groups.Therefore, we compare the two treatments groups using a t-test for twoindependent samples on all baseline measurements.2. Detect for changes in HbA1c and Fasting Blood Glucose levels from pre-dose topost-dose within each treatment group.For each patient, we take the twelve-week change in efficacy variablesas the primary and secondary response measures (Table 2). Using thetwelve-week change takes advantage of the ”automatic blocking” intro-duced when repeated measurements are taken on the same experimentalunit (patient) over time. For each treatment group, we separately utilizea t-test for pairwise dependent observations to determine if treatment isassociated with a significant improvement (over time) in blood glucosemanagement.3. Discern between the effects of drug treatment and placebo on changes in HbA1cand Fasting Blood Glucose levels.Only if significant changes are detected from pre-dose to post-dose, needwe proceed to the question of whether drug is more effectivethan placeboin improving blood glucose management. Furthermore, if the random-ization assumption is valid, we may address this question using simpletechniques. Based on the results of steps 1 and 2, we use the two-sample3pooled t-test to determine if the change in blood glucose levels observedin patients administered the drug is significantly different from the cor-responding change observed in patients administered the placebo.HbA1cPatient Pre-dose Post-dose ∆HbA1c ∆1 HbA1c1HbA1c1∆1Drug............∆D
View Full Document
Unlocking...