Bayesian approaches to use historical data in the analysis of clinical trials

bayesian approach

The BAYES 2018 Bayesian Biostatistics workshop took place from the 20th to the 22nd of June, and was an extraordinary chance to discuss how Bayesian methodology can be used within the pharma industry.

One relevant topic that was discussed was the use of historical controls in the analysis of randomized clinical trials with presentations by Nicky Best (GlaxoSmithKline), Maxine Bennett (MRC Biostatistics Unit), Leonhard Held and Isaac Gravestock (both University of Zurich). This blog post gives a short overview of some of the key points presented by the above speakers.

When designing clinical trials historical data of previous clinical trials is always used. For example, to determine the variance and the clinically relevant effect size for a sample size calculation or to collect information on recruitment rates and population sizes. However, historical data seems to be rarely used in the analysis of clinical trials.

One possibility to incorporate historical data into the analysis of a clinical trial is to replace or supplement controls by historical controls. When applying this approach there are several assumptions that need to hold for the historical control group(s). Examples are: the historical control group(s) must (1) have received the same precisely defined standard treatment as the randomized controls in the current study, (2) have been part of a recent clinical study with the same subject inclusion and exclusion criteria. Further requirements can be found in [1].

Various Bayesian methods exist to incorporate historical control data from a single study into the analysis of trial data. While the most trivial pooling approach simply assumes that the historical controls are equal to the randomized controls from the current study, there are other methods which down weight the historical information. These include the Power prior [2], Hierarchical Power prior [2], Modified Power prior [3] and Commensurate prior [4].

While the power prior and commensurate prior can be extended to incorporate historical control data from multiple clinical trials into the analysis of a new study [5, 6], there are also two meta analytical approaches. The retrospective meta analytical combined (MAC) analysis performs a meta-analysis of historical and current data combined. This could even be a non-Bayesian analysis. The prospective meta analytical-predictive (MAP) analysis performs a meta-analysis of the historical data to form a MAP prior which is then combined with the current data using the Bayesian rule. The meta-analytic prior (MAP) has been used in several published studies and is currently the gold standard [7].

By using historical control data in the analysis of clinical trials we can increase statistical power or reduce the required sample size. With an increasing number of trials taking place in small populations (e.g. rare disease studies, paediatrics studies, studies in difficult to recruit therapeutic populations) and difficulties to meet evidential standards, an increasing demand of methods to re-use patient data from previous clinical trials is expected in the coming years.

Bayesian Study Design & Interim Analyses


At Quanticate our Statistical Consultants are experienced in clinical study designs and have delivered multiple trial analyses using Bayesian methods. For more information Submit a Request for Information and a member of our team will be in touch with you shortly.





[1] Pocock (1976) The Combination of Randomized and Historical Controls in Clinical Trials. J. Chron. Dis 29:175-188

[2] Ibrahim JG and Chen M-H (2000) Power prior distributions for regression models. Statistical Science 15:46-60

[3] Neuenschwander B et al. (2009) A note on the power prior. Stat. Med. 28:3562-3566

[4] Hobbs et al. (2011) Hierarchical commensurate and power prior models for adaptive incorporation of historical information in clinical trials. Biometrics 67:1047-1056

[5] Duan (2005) A Modified Bayesian Power Prior Approach with Applications in Water Quality Evaluation. PhD thesis

[6] Hobbs B et al. (2012) Commensurate Priors for Incorporating Historical Information in Clinical Trials Using General and Generalized Linear Models. Bayesian Analysis 7(3):639-674

[7] Neuenschwander B, Capkun-Niggli G, Branson M and Spiegelhalter DJ (2010) Summarizing Historical Information on Controls in Clinical Trials. Clinical Trials 7(1):5-18

Subscribe to the Blog