The role of modelling in prioritising and planning clinical trials.

OBJECTIVES: To identify the role of modelling in planning and prioritising trials. The review focuses on modelling methods used in the construction of disease models and on methods for their analysis and interpretation. DATA SOURCES: Searches were initially developed in MEDLINE and then translated into other databases. REVIEW
METHODS: Systematic reviews of the methodological and case study literature were undertaken. Search strategies focused on the intersection between three domains: modelling, health technology assessment and prioritisation.
RESULTS: The review found that modelling can extend the validity of trials by: generalising from trial populations to specific target groups; generalising to other settings and countries; extrapolating trial outcomes to the longer term; linking intermediate outcome measures to final outcomes; extending analysis to the relevant comparators; adjusting for prognostic factors in trials; and synthesising research results. The review suggested that modelling may offer greatest benefits where the impact of a technology occurs over a long duration, where disease/technology characteristics are not observable, where there are long lead times in research, or for rapidly changing technologies. It was also found that modelling can inform the key parameters for research: sample size, trial duration and population characteristics. One-way, multi-way and threshold sensitivity analysis have been used in informing these aspects but are flawed. The payback approach has been piloted and while there have been weaknesses in its implementation, the approach does have potential. Expected value of information analysis is the only existing methodology that has been applied in practice and can address all these issues. The potential benefit of this methodology is that the value of research is directly related to its impact on technology commissioning decisions, and is demonstrated in real and absolute rather than relative terms; it assesses the technical efficiency of different types of research. Modelling is not a substitute for data collection. However, modelling can identify trial designs of low priority in informing health technology commissioning decisions.
CONCLUSIONS: Good practice in undertaking and reporting economic modelling studies requires further dissemination and support, specifically in sensitivity analyses, model validation and the reporting of assumptions. Case studies of the payback approach using stochastic sensitivity analyses should be developed. Use of overall expected value of perfect information should be encouraged in modelling studies seeking to inform prioritisation and planning of health technology assessments. Research is required to assess if the potential benefits of value of information analysis can be realised in practice; on the definition of an adequate objective function; on methods for analysing computationally expensive models; and on methods for updating prior probability distributions.