Biostatistical Design and Analyses of Long-Term Animal Studies Simulating Human Postmenopausal Osteoporosis
- PDF / 10,956,903 Bytes
- 10 Pages / 573.839 x 774.959 pts Page_size
- 80 Downloads / 177 Views
47
Biostatistical Design and Analyses of Long-Term Animal Studies Simulating Human Postmenopausal Osteoporosis
Ludwig A. Hotborn, PhD Unit of Bioinformatics, Hannover Universitv.
Roche Diagnostics GmbH, Pharma Research Penzberg, Penzberg. Germany, and Institute of Pharmacology and Toxicology, Heidelberg h i v e m i @ , Mannheim. Germany
Key Words lbandronate; Osteoporosis; Many- to-one comparison; Minimum-effective dose; Confidence interval for the ratio; Pharmacological study Correspondence Address Ludwig A. Hothorn. PhD. Bioinformatics, Hannover LJniversi@,Herrenhaeuserstrasse 2 , D-30419 Hannover: Germany (e-mail: hothorn@bioinf. uni-hannoverde).
Presented at the Third International DlA Workshop on "Statistical Methodology in Non-Clinical Ri3D." Barcelona, Spain, September 2002.
Using three well-designed experimental studies as illustration, we demonstrate that the biostatistical design and analysis of long-term animal studies simulating human osteoporosis should be analogous to the design and analysis of randomized clinical trials. This principal is in accordance with the recommendations fiom the International Conference on Harmonisation guidelines concerning statistical principles in clinical trials (I). An important element of biostatistical study design is sample size. The three studies that are described herein used an a-prion sample size estimation for the one-way layout that included controls and several treatment and dose groups. In these k-sample designs, with at least one control group, both the multiple comparison
INTRODUCTION To date, the biostatistical design and analysis of published pharmacological studies has been highly varied. This has not been assisted by the fact that only a few relevant publications exist that recommend biostatistical procedures (2,3), and those that do exist recommend considerably different methodology compared to clinical trials. Surprisingly, neither international recommendations nor guidelines are available for the statistical design and analysis of pharmacological studies. Some studies do exist, particularly those relating to long-term animal investigations, for which the biostatistical design and analysis is similar to that of a randomized clinical trial (RCT). However, essential differences remain between clinical and experimental studies. An important example is that quantitative decisionmaking (with an upper boundary of false-positive and false-negative rates) based on measurements from invasive techniques, for example,
procedure and trend tests within procedures for identification of the minimal-effective dose are recommended. Although p-values in pharmacology are quite common, confidence intervals should be used according to their interpretation for both statistical significance and clinical relevance. The use of one-sided confidence intervals for both the difference and the ratio to control for proving either superiority or at least noninferiority is demonstrated by real data examples. Relevant and relatively straightforward sofiare is available for biost
Data Loading...
