Statistical assessment of learning curve effects in health technology assessment
New treatments are being constantly developed and introduced into practice. The randomised controlled trial is widely seen as the ‘gold standard’ method of evaluating treatments. However, its use in evaluating surgical techniques has been contentious. Since surgeons may not be as experienced in the new technique, the comparison may be biased, failing to represent the true worth of the new technique. This ‘learning curve effect’ is one of the major reasons for the low usage of clinical trails in surgery. A literature review of fibreoptic intubation, which was acknowledged to have a learning curve effect, was performed with a view to measuring the learning curve. However, it was not possible to ascertain the approximate magnitude of the learning curve features. A number of statistical methods had been suggested as suitable for modelling of learning curves in health technologies. A comparison of Bayesian and maximum likelihood estimation hierarchical models and generalised estimating equations was made using data from trainees performing fibreoptic intubation. It was concluded that statistical methods that account for the hierarchical structure of learning data should be used. The use of non-linear hierarchical models for modelling learning curve effects was evaluated and it was recommended that Bayesian non-linear hierarchical modelling should be the preferred statistical method to model learning data, where feasible. Finally, a randomised controlled trial was assessed for the existence of learning effect. A pragmatic three step approach was used to identify the existence of a learning curve effect in each outcome. Twenty-one outcomes were evaluated with four outcomes concluded to have a learning curve effect. The trial analysis was repeated using Bayesian hierarchical models, to account for the existence of learning. In the last chapter the results were discussed and areas for further research were highlighted.