(247 / 250) BackgroundPharmaceuticals are usually granted a marketing authorisation on the basis of randomised controlled trials (RCTs). Occasionally the efficacy of a treatment is assessed without a randomised comparator group (either active or placebo).
ObjectiveTo identify and develop a taxonomic account of economic modelling approaches for pharmaceuticals licensed without randomised control trial data.
MethodsWe searched PubMed, the websites of UK health technology assessment (HTA) bodies and the ISPOR Scientific Presentations Database for assessments of the 74 indications granted a marketing authorisation by the FDA or EMA from Jan-1999 to May-2014 without RCT data. The outcome of interest was the approach to modelling efficacy data.
ResultsFifty-one unique models were identified based on 29 peer-reviewed articles, 30 health technology appraisals (HTAs), and 15 ISPOR abstracts concerning 30 indications (44 indications had not been modelled). Also notable was the low rate of submission to HTA agencies (28/98).The majority of models (43/51) were based on 'historical controls' -comparisons to previous meta-analysis or pooling of trials (5), individual trials (19), registries/case series (14), or expert opinion (7). Other approaches used the patient as their own control, performed threshold analysis, assumed time on treatment was added to overall survival, or performed cost-minimization.
ConclusionsThere is considerable variation in the quality and approach of models constructed for drugs granted a marketing authorisation without a RCT. The most common approach is of a naïve comparisons to historical data (using other trials / registry data as a control group), with considerable scope for bias.
Key points for decision makers When pharmaceuticals are licensed without comparative data, economic models are generally constructed using a historical control Even within evaluations using the same method, the quality and appropriateness of an approach varies The appropriateness of historical comparison for modelling using uncontrolled study data requires further methodological evaluation Article: Word count: 2261