Modeling Adverse Event rates of Opioids for the Treatment of Osteoarthritis Pain using Literature Data
F. Ezzet (1), K. Prins (2), M. Boucher (3)
(1) Pharsight a Certara Company, St Louis, MO, USA; (2) Pharsight a Certara Company, Netherlands; (3) Pfizer, Sandwich, UK
Objectives: To characterize adverse event (AE) and dropout profiles of Opioids for the treatment of Osteoarthritis (OA) Pain using literature data.
Methods: A database was constructed using data from scientific literature of randomized controlled clinical trials investigating safety and efficacy of opioids. Attention was focused on dropout rates due to AE's and proportions reporting events of constipation and nausea. Using proportions together with sample size, the response is a binomial variable, and was thus modeled using a mixed effect general linear model with glme, Splus. The function operates on log(p/(1-p)), a linear function of model covariates, where p is the probability of an event. Inter-study random effect enters the model as an additive term. The main covariates of interest were opioid strength (none, moderate or strong) and treatment dose. Other effects investigated, included treatment formulation and study duration. Model diagnostics were explored to evaluate goodness of fit.
Results: The database included about 40 studies on 12 treatments involving over 12000 OA patients. There were a sufficient number of studies using moderate opioids (e.g. Tramadol and Codeine) and strong opioids (e.g. Oxycodone). With the outcome variable as a proportion, when converted to binomial, is equivalent to having access to patient level response from individual trials. The resulting large sample size would thus significantly increase power and precision of model estimates. Three models were established, which determined that strong opioids increase the chances of constipation, nausea and dropout rates. Using placebo as a reference group, strong opioids have odds ratios of 7.7, 5.6 and 5.3, respectively. For moderate opioids, the odds ratios were 3.3, 2.7 and 3.3 and for Non-opioids they were 1.5, 1.1 and 1.2. Inference on influence of dose is usually limited due to dose ranges investigated. However, the dropout model indicated a dose effect for moderate and strong opioids. Diagnostic plots indicated adequacy of the model fit.
Conclusion: The models established that rates of AE's and dropouts increase significantly with the strength of opioids. While benefits of Meta analysis using public literature are well established [1,2] , models for proportions have the added advantage of increased statistical power, a consequence of using subject level information.
References:
[1] Ezzet, F., Ravva, P., Tensfeldt, T. Model-Based Literature Meta-Analysis: Virtues and Limitations, ACoP, Tucson, Arizona, March, 2008 (http://tucson2008.go-acop.org/pdfs/17-Ezzet_FINAL.pdf )
[2] Ezzet F. The Role of Literature-Based Disease Progression Models to
Support Knowledge Management and Decision-Making in Clinical Drug Development., AAPS National Biotechnology Conference, May 2008, Toronto, Canada
