Use of model based meta-analysis combining patient-level with summary-level data using multilevel random effects to provide a quantitative assessment of the clinical efficacy (IPSS) profile and competitive positioning of a PDE5 inhibitor (UK369,003) for the treatment of benign prostatic hyperplasia (BPH).
N.H. Prins (1), M. Green (1), S. Haughie (2), P. Johnson (2), S. W. Martin (2)
(1) Pharsight, a Certara company, St. Louis, MO, USA., (2) Pfizer Ltd, Statistics & Pharmacometrics, Sandwich, UK.
Objectives: International Prostate Symptom Score (IPSS) is a questionnaire-based scale for patients suffering from benign prostatic hyperplasia (BPH) to rate their symptom burden ranging from 0 (no symptoms) to 35 (worst symptoms). In two trials, the phosphodiesterase type 5 (PDE5) inhibitor UK-369,003 was tested versus the alpha blocker tamsulosin (in one trial) and placebo (in both trials) in patients who met the criteria for BPH. Substantial variability between trial results for all treatments impeded attempts to build a robust model using only patient-level data. However, with the addition of supporting data from published trial results a robust model was developed that enabled a valid comparison of UK-369,003 to that of placebo and tamsulosin 0.4 mg.
Methods: Patient IPSS time course data from internal studies on UK-369,003, placebo and tamsulosin efficacy in BPH patients were concatenated to a comprehensive literature database containing mean trial outcome on a range of BPH treatments. The literature data included placebo, tolterodine, alfusozin, tamsulosin and PDE5 inhibitors (sildenafil, vardefanil and tadalafil). Using nlme package in SPLUS v8.0, a mixed effects model described IPSS as a function of drug, dose, time, and IPSS at baseline. While the structural components of the model were universal for both trial mean and patient-level data, the multi-level random effects model explicitly separated inter-trial and inter-subject variability [1], while the residual variance was weighted by sample size.
Results: The model fitted the data well and quantified a significant placebo and treatment response. For some treatments the response was best characterized using an Emax model and/or a separate time course. The simultaneous estimation of inter-trial and interindividual random effects provided a significant improvement of model fit. The UK-369,003 dose response predicted that doses greater than 50 mg were expected to be superior to tamsulosin 0.4 mg. At UK-369,003 100 mg the probability of superiority and a 0.5 IPSS point improvement was 83 and 42 % respectively.
Conclusion: Using a multilevel random effects maximum likelihood model based on summary-level and patient-level data enabled a more valid comparison of UK-369,003 to that of placebo and current and potential competitors in the treatment of BPH. To our knowledge, this is the first example of a maximum likelihood multicovariate nonlinear dose/PD model where multilevel random effects have successfully been applied.
References:
[1] Pinheiro & Bates. Mixed-Effects Model in S and S-PLUS. Springer Verlag, New York. 2000.