Dr Wicher Bergsma
Social Statistics group
Email: w.p.bergsma@lse.ac.uk
Personal webpage
Social Statistics group research website
Summary of impact
Pharmaceutical companies use clinical trials to assess the safety and benefit-risk evaluation of newly developed drugs. It is hard to obtain sufficient numbers of patients to test the drugs on and obtain a valid assessment of their safety. Statistical methodology developed at LSE has led to a methodology for which potentially fewer patients are needed. Furthermore, it allows pharmaceutical companies to obtain a more detailed risk profile of new drugs. GSK has adopted the marginal modelling methodology as one of a number of approaches they will be using for improving drug testing.
Underpinning research
Background:
Very often the data collected by researchers involve dependent observations, without, however, the investigators having any substantive interest in the nature of the dependencies. Although these dependencies are not important for the answers to the research questions concerned, they must still be taken into account in the analysis. Standard statistical estimation and testing procedures assume independent and identically distributed (iid) observations, and need to be modified for observations that are clustered in some way. Marginal models provide the tools to deal with these dependencies without having to make restrictive assumptions about their nature.
Research insights gained:
A comprehensive approach to marginal modelling is developed in the book by Bergsma, Croon and Hagenaars, 2009. The book contains innovative statistical models, as well as methods to estimate parameters and their standard errors in these models using maximum likelihood (ML). Furthermore, a software package was developed (Bergsma and Van der Ark, 2009) as a companion to the book, which can be used to fit and test marginal models. Advantages of ML estimators compared to competing methods include the easy availability of goodness-of-fit statistics, asymptotic efficiency and the possibility to take latent variables into account.
Application of research to impact:
In clinical trials, typically repeated measurements are made on subjects to compare adverse effects of treatment in an active and a control group. The traditional methodology to analyse data from such trials is to do separate tests for a difference for each time point, and use Bonferroni-Holm corrections to account for multiple testing.
The case study had two components. In the first component, Linear Categorical Marginal Models (LCMMs) were applied to a Phase III clinical trial of a candidate meningococcal paediatric vaccine. It was possible to obtain a better understanding of the relative safety of the vaccine than with the traditional method. Firstly, significant effects were detected for more symptoms (the symptoms considered were pain, redness, and irritability) and for each symptom significant effects were found for more days (measurements were taken on the four days following vaccination). Secondly, it was possible to assess how differences in symptoms of active and control groups vary with time.
The second component of the case study consisted of a simulation study. This showed potentially improved power of LCMMs compared to the traditional method. In particular, LCMMs allow power to be directed at salient alternatives, while the traditional method is less flexible and relatively good power only for certain general alternatives.
The advantages of the marginal modelling for assessing safety of drugs in clinical trials is given in Bergsma, Aris and Tibaldi (2013).
References to the research
Details of the impact
The case study involves the application of state-of-the-art marginal modelling techniques developed at the Statistics Department of the LSE to a Phase III clinical trial of a candidate meningococcal paediatric vaccine from GlaxoSmithKline (GSK) Biologicals in Brussels.
The methodology developed at LSE was found to have important advantages compared to classical methods for assessing safety. Firstly, it has more power, so that smaller numbers of patients are needed to detect whether a drug is safe or not. Hence, the new methodology will eventually allow faster development cycles of drugs. Secondly, classical methods allow detection of adverse effects, but do not allow detailed assessment of how adverse effects vary over time. For example, pain may take time to develop after taking a drug, and this may be problematic for classical methods to detect. Thus, the new methodology allows a better risk profile to be obtained. This includes a better understanding of side effects which is beneficial for patients, and drug prescription can be better tailored to patients’ needs. The method was successfully applied to data from the aforementioned clinical trial, and will therefore also be used by GSK in future trials.