MICA: Model Based Network Meta-Analysis for Pharmacometrics and Drug-Development

In the development of new drugs, studies are conducted to compare the relative benefits of the drug at different doses with placebo and/or other active drugs (which may also be at different doses). Furthermore, the health outcomes may be measured repeatedly over time. In order to decide whether to take the new drug forward into larger clinical trials, the results from all studies that have been conducted on a new drug are combined in meta-analysis to obtain a pooled estimate of the effect of the drug against placebo or active comparator drugs. Recently methods have been developed to allow for relative benefits to depend on dose and time of measurement in meta-analysis that compares the new drug with placebo (or another drug). However, there may be more than one comparator drug, and they have been measured at various different doses and times. Network meta-analysis is a technique that allows one to compare the relative benefits of multiple drugs that have been compared in randomised clinical trials, where not all drugs have been included in every study. This study aims to combine models of the relationships for the relative health benefits with dose and time, with network meta-analysis. This will allow us to combine information from studies comparing different drugs at different doses and different times, even though those studies may not have included the same dose and times.

Decisions as to which drugs to take forward into clinical trials, has substantial impact on all patients. Drug companies have limited resources, and so the decision to invest in one promising drug may come at the expense of another. It is therefore important to make drug-development decisions based on as much available evidence as possible. The methods developed in this project will allow as much existing evidence from comparative studies as possible to contribute to drug development decisions. Furthermore, we will explore the possibility of also incorporating evidence from studies that only sudy a single drug, or studies that compare drugs that we are not directly interested in, but that could help us understand the form of the relationships over dose and time.

The methods we will develop may require some strong assumptions. It is therefore very important to check whether those assumptions hold, and a key part of this work will be to look at methods to check assumptions and to check how well the models developed fit to the observed data. Decisions should be based on the most robust model predictions, and sensitivity to any assumptions explored.

This project will be a collaboration with project partner Pfizer, who will provide the datasets and expertise in dose and time course modelling. The University of Bristol team brings expertise in network meta-analysis, assessing model fit and consistency, and statistical computing. The collaboration is designed to ensure that the methods developed will be relevant to the needs of drug-development organisations, and the interaction with Pfizer will allow the methods to be used by that organisation, and publications and disemination plans will introduce the methods more widely. This approach will help the methods be used by industry to better invest their resources into drugs to improve patient health based on a better summary of the available evidence.

Early phase trials compare drugs at different doses with placebo and/or other active drugs, often with repeated outcome measures over time. Drug development decisions are often informed by a meta-analysis of such studies. Recently methods have been developed to incorporate relationships with dose and time, termed Model-Based Meta-Analysis (MBMA). However, there may be more than one comparator drug, and the comparators may also have been measured at various different doses and times. Network meta-analysis is a technique that allows relative effects to be estimated across multiple treatments that have been compared in trials that form a connected network of treatment comparisons. Dose can be incorporated as a separate treatment, but this may lead to sparsely populated networks. This study aims to incorporate models for dose and time within a network meta-analysis framework. This will allow us to combine information from studies comparing different drugs at different doses and different times, even though those studies may not have included the same dose and times. Furthermore, we will explore the possibility of also incorporating evidence from single arm studies, and/or studies that compare drugs that we are not directly interested in, but that could help us understand the form of the relationships over dose and time.

The models we develop may require some strong assumptions. It is therefore important to check whether those assumptions hold, and a key part of this work will be to look at methods to check model fit, validity of model assumptions, in particular checks for inconsistency. The methods developed will allow the synthesis of all the relevant available evidence, and assessment of the robustness of the model predictions, to better inform drug-development decisions.
The industry project partner Pfizer will ensure that the methods developed will be relevant to the needs of drug-development organisations, facilitating the impact of the research.

The methodology to be developed is planned to have commercial value. The project partner (Pfizer) has identified specific areas of pharmacometrics and drug-development where there is a need for network meta-analysis methodology to be developed. The industry partner (Pfizer) will directly benefit from this research in the development of methods for unmet industry needs to inform the drug-development process. The researcher will spend 12 one-week visits at Pfizer where they will interact with researchers at Pfizer, and present the results from their research. This will directly disseminate the methods to those who are involved in the drug-development process, and can be used by Pfizer staff to inform future development decisions. The decision to invest in a new drug that later is approved by reimbursement agencies, brings a new treatment option to patients, with associated benefits in health related quality of life. Conversely, the decision not to invest in a particular drug, frees resources to invest instead in more promising drugs, which again if later approved by reimbursement agencies, are expected to lead to increases in health-related quality of life. Improvements in the methods that inform such decisions may lead to new beneficial drugs being made available to patients more quickly. The methods may also be used to identify whether further evidence is needed to be collected in future studies, before a decision on a particular drug can be made. The research methods will be made available to those working in drug-development during the project, however, impact on patients would be expected several years after the project due to the long duration of the drug-development process.

It is expected that the methodology developed will be of value to other pharmaceutical companies, as well as Pfizer. The research will be published in peer reviewed journals, and presented at conferences, targeting statisticians in the pharmaceutical industry. The impact is therefore expected to be realised across the pharmaceutical industry, and not just the project partner.

The methodology is not restricted to early development. It is frequently the case in meta-analysis and network meta-analysis, that treatments/interventions differ in dose/intensity, and that lumping over dose/intensity can be the cause of heterogeneity. The methods developed here, may therefore have implications more generally amongst statisticians and health economists working in meta-analysis and network meta-analysis.