Successful drug development depends on making wise decisions about portfolios, clinical trials, marketing, etc. We’re continuously faced with the challenge of deciding whether to continue development or stop it. To support those decisions, we gather data, typically through clinical trials. We analyze the data from those clinical trials, and then we use these analyses to build models that we then use to predict what may happen in the next trial. The data collected from these in-house trials are “internal data” or “proprietary data.”
In addition to your internal data, external data are accessible through many sources including published articles in PubMed, the FDA website, and ClinicalTrials.gov. Using external data to aid decision making is cost-effective because your competitors have already paid for the research to generate the data. But, perhaps more importantly, using external data is necessary to inform key decisions.
Combining aggregate level data from multiple studies
Companies rarely share individual level data. But, they all publish most of their aggregate level trial results. Meta-analysis is the statistical method for combining data from multiple studies. Preferably, you perform a meta-analysis on data from systematically searched and selected sources, collected in an actively maintained database. This full process is a systematic review or evidence synthesis.
An introduction to meta-analysis
Meta-analysis requires data aggregation. What does that mean? Each clinical trial has a number of patients in them. Some trials have more subjects than others. And each subject in each trial will contribute one or more data points regarding the effects of a drug. Each patient will also contribute information regarding covariates such as age, body weight, etc.
The process of aggregation summarizes these data. So for each of these trials, we summarize the drug effects and covariates. In addition, we give more importance to the bigger trials than to smaller trials in the analysis. Thus, we assign each study a weight, typically based on the inverse of the variance. That means that we integrate the size of the trial, the number of subjects, as well as the variability in the trial.
When we run our meta-analysis, we combine statistics from different trials to identify a parameter that describes the effects in these trials. A regression analysis could also be performed to describe how covariates― a drug, dose, or demographic factor― impact that drug effect.
The historical context for meta-analysis
Gene Glass, a social scientist, introduced the term meta-analysis in the mid-1970s. And meta-analysis is still heavily used in social sciences, not just in medicine. In 1904, Karl Pearson was one of the first to statistically combine medical data from previous analyses of the inoculation of soldiers to prevent typhoid fever.
Since the 1990s, meta-analysis has become a cornerstone of evidence-based medicine. Most meta-analyses in the medical literature evaluate the effects of approved drugs. Over the last ten years, using meta-analysis to support drug development decisions has increased in popularity.
Types of meta-analysis in drug development
Several different types of meta-analysis are used to inform drug development. The most utilized type of meta-analysis is pairwise meta-analysis, which examines interventions or trial arms in pairs. This approach has the advantage of being relatively fast and easy. The major drawback to pairwise meta-analysis is that it only considers paired intervention-versus-control evidence. Thus, it cannot make indirect comparisons of drugs that haven’t been compared in a clinical trial.
Network meta-analysis combines studies in a network and builds a statistical framework to support indirectly comparing drugs that may not have been evaluated head-to-head in clinical trials.
Model-based meta-analysis (MBMA) extends upon network meta-analysis. MBMA combines information on a drug given at multiple doses and time points as well as multiple drugs with the same mechanism of action in a statistical framework that integrates models inside models. This framework enables “borrowing information” across different trials or different drugs. MBMA incorporates dose and duration and uses standard pharmacology models and assumptions. It can include trial-level covariate relationships on the dose-response models to account for between-trial differences in patient populations. It also allows us to simultaneously model multiple endpoints and potentially link biomarkers to clinical outcomes.
Like network meta-analysis, MBMA can provide indirect comparisons. However, because MBMA uses longitudinal dose-response models for individual drugs or drug classes and incorporates covariate effects in these models, we can use MBMA to evaluate new scenarios and simulate the probability of clinical trial success.
What questions can be answered with model-based meta-analysis?
MBMA can answer questions related to your program’s competitive landscape, disease/trial characteristics, and drug characteristics. You may want to know the key comparators or key competitor compounds in a certain therapeutic area. Or you might wonder how many trials are available for a specific endpoint.
You can also use MBMA to gain insights into disease pathology and clinical trials that have been conducted in that indication. This approach can reveal the major covariates for trials in terms of populations, baseline values, and demographics. In addition, you can examine typical placebo effects and their variability. And, you can evaluate the heterogeneity within outcomes and how the trials were conducted.
Lastly, MBMA can provide competitive intelligence on the dose-response and time-course of drugs in the same class or other classes. Ultimately, these models can help you differentiate and position your drug between existing and developing competitors.
While some of these questions can be answered with internal (proprietary) data, most require external data.
Take home messages
MBMA combines aggregate level data from published sources in a formal, quantitative framework. These aggregate level data are typically attained from published articles in PubMed, ClinicalTrials.gov, the FDA website, and scientific conferences.
This innovative approach can help answer key questions in drug development. We can leverage MBMA to predict Phase 3 trial results including comparisons of drugs that were not compared in trials before. Multiple endpoints can be modeled, even simultaneously, and multiple scenarios with those endpoints can be assessed in simulations. We can use those simulations to predict the probability of a drug being superior or non-inferior to a competitor in a head-to-head clinical trial. The insights gained from MBMA can be used to optimize clinical trial designs and marketing/commercial strategies.
If you’re interested in learning more, please read a white paper that we’ve written on MBMA.