Chondroitin erence in shape and scale parameters between studies and assumed constant heterogeneity across treatment comparisons. We used vague initial values and analyzed both models using two chains and 90,000 iterations after a burn in of 30,000 iterations. We evaluated convergence by monitoring trace plots, autocorrelations, and Monte Carlo error which describes the variability of each estimate due to the simulations.We evaluated goodness of fit of the Weibull and log logistic models by visually evaluating linearity of diagnostic data plots and comparing the deviance information criteria between models, where smaller DIC values indicate a better chemical screening fitting model.12 We also used DIC to evaluate the goodness of fit of fixed effect versus random effect models. We conducted sensitivity analysis by varying the initial parameter values. We used the parameter estimates obtained from the posterior summary to derive hazard rates, hazard ratios, and PFS curves for each therapy. We projected PFS for timepoints beyond those reported in the trials by assuming a rhein Weibull distribution and constant hazard. We estimated the mean duration of PFS for each treatment by calculating the area under each projected PFS curve.
Results Randomized controlled trials and quality assessment In the systematic review, we identified seven RCTs that met our inclusion criteria. We based our analysis on the endpoint of PFS, described as the duration of time from randomization until disease progression or death from any cause, and is a better measure of clinical benefit in CLL than response rates. We were unable to conduct analysis using an endpoint of overall survival due to lack of reported data in the RCT of alemtuzumab and chlorambucil13 Fig. 3 presents the PFS data observed in the individual trials. The same therapies are ZD6474 presented with the same color. The last therapy listed in each individual trial was considered the baseline comparator in that trial. During data abstraction, we excluded two studies due to trial heterogeneity.
We excluded a trial that compared bendamustine with chlorambucil14 because the chlorambucil treatment protocol allowed for a maximum of six chlorambucil treatment cycles, whereas other studies in the comparison network allowed for a maximum of 12 treatment cycles. The survival proportions in the chlorambucil arm of the Knauf, et al. study wassignificantly lower than those in the other trials possibly because patients randomized to chlorambucil in the Knauf, et al. study received fewer treatment cycles and a lower cumulative drug dosage compared to the chlorambucil dosed patients in the other trials. In the simulation, this difference may artificially enhance the magnitude of treatment effect for the comparator drug, bendamustine. We excluded a trial that compared fludarabine with chlorambucil15 because the patient population was older, and older patients have higher competing mortality risks compared with their younger counterparts, and this difference in mortality risks could bias the relative treatment estimates. Table 1 summarizes the five studies included in the comparison network and describes sample sizes, dosing regimens, and base line patient characteristics. Overall, the patients were younger, the majority had good performance status based on the East.