Joint modeling of survival and longitudinal non-survival data: current methods and issues. Report of the DIA Bayesian joint modeling working group.

Explicitly modeling underlying relationships between a survival endpoint and processes that generate longitudinal measured or reported outcomes potentially could improve the efficiency of clinical trials and provide greater insight into the various dimensions of the clinical effect of interventions included in the trials. Various strategies have been proposed for using longitudinal findings to elucidate intervention effects on clinical outcomes such as survival. The application of specifically Bayesian approaches for constructing models that address longitudinal and survival outcomes explicitly has been recently addressed in the literature. We review currently available methods for carrying out joint analyses, including issues of implementation and interpretation, identify software tools that can be used to carry out the necessary calculations, and review applications of the methodology.

Indirect treatment comparison between fixed-dose-combinations of amlodipine/losartan and amlodipine/valsartan in blood pressure control.

AIMS: This study compared blood pressure (BP) changes after 8 weeks of therapy between a fixed-dose combination (FDC) of amlodipine/losartan and amlodipine/valsartan using a network meta-analysis because no trials directly compared amlodipine/losartan with other FDCs. METHODS: A systematic search identified six randomised controlled trials (amlodipine/losartan-3, amlodipine/valsartan-3) of FDCs and their component monotherapies. Conventional fixed-effects methods were used to conduct the comparisons. The primary and secondary effect measures were the changes in sitting diastolic and systolic blood pressure (sitDBP, sitSBP) at 8 weeks post-randomisation. RESULTS: The estimated amlodipine/valsartan - amlodipine/losartan difference (95% confidence interval) in sitDBP reduction was -1.27 mmHg, (-5.7, 2.2) for lower dosages and -0.45 mmHg, (-3.7, 2.7) for higher dosages; for sitSBP, the values were -3.74 mmHg, (-9.0, 2.9) for lower dosages and 0.2 mmHg, (-6.2, 6.0) for higher dosages. The confidence of a greater reduction in BP (fixed difference = 0) on amlodipine/losartan 5/50 than on amlodipine/valsartan 5/80 was 77% for sitDBP and 89% for sitSBP. The corresponding confidence for the higher doses was 61% for sitDBP and 48% for sitSBP. The findings support asserting with (fixed) 95% confidence that the BP reduction on amlodipine/valsartan 5/80 exceeds the amlodipine/losartan 5/50 reduction by at most 1.6 mmHg for sitDBP, and at most 1.26 mmHg for sitSBP. The corresponding upper bounds for the higher dosages were 2.31 mmHg (sitDBP) and 5.38 mmHg (sitSBP). CONCLUSIONS: The BP lowering effect with amlodipine/losartan and amlodipine/valsartan was comparable. Potential superiority of the reductions realised with amlodipine/valsartan relative to amlodipine/losartan, are unlikely to be clinically material.

Bayesian adaptive determination of the sample size required to assure acceptably low adverse event risk.

An emerging concern with new therapeutic agents, especially treatments for type 2 diabetes, a prevalent condition that increases an individual's risk of heart attack or stroke, is the likelihood of adverse events, especially cardiovascular events, that the new agents may cause. These concerns have led to regulatory requirements for demonstrating that a new agent increases the risk of an adverse event relative to a control by no more than, say, 30% or 80% with high (e.g., 97.5%) confidence. We describe a Bayesian adaptive procedure for determining if the sample size for a development program needs to be increased and, if necessary, by how much, to provide the required assurance of limited risk. The decision is based on the predictive likelihood of a sufficiently high posterior probability that the relative risk is no more than a specified bound. Allowance can be made for between-center as well as within-center variability to accommodate large-scale developmental programs, and design alternatives (e.g., many small centers, few large centers) for obtaining additional data if needed can be explored. Binomial or Poisson likelihoods can be used, and center-level covariates can be accommodated. The predictive likelihoods are explored under various conditions to assess the statistical properties of the method.

Timing of futility analyses for 'proof of concept' trials.

'Proof of concept' (PoC) trials often are carried out to determine if a treatment is biologically active or inactive. If the predictive probability of rejecting a null hypothesis of no treatment effect on completion of the trial calculated using information accumulated before trial completion is sufficiently low, then the trial could be terminated and resources redirected more productively. If the predictive probability is high, some time might be gained by proceeding with further steps in drug development without affecting the course of the PoC trial in any way. We describe how the interim evaluation timing affects the ability to reach a decision to stop or continue as a function of various assumptions about the true parameters and about the interim outcomes. An important practical finding is that there is no point to carrying out an evaluation before accumulating about 40 per cent of the planned observations. Consequently, if trial costs are mostly at startup or for recruitment of subjects for extended periods of observation, it is unlikely that much benefit will be realized by considering the possibility of early termination for futility except possibly for administrative efficiencies with respect to resource and staff allocation.