Single-study-based economic evaluations have the potential advantage of the internal validity of the trial design and the advantage of joint collection of data on both resource use and effectiveness. However, the aims of the underlying trials and economic evaluations, may differ in significant respects, which can lead to problems concerning the suitability or comparability of trial-based economic analyses. Despite the aims of single-study-based economic evaluations generally being somewhat different than model-based economic evaluation, trial-based economic analyses may provide individual-level analysis of the impact of the screening technology and its comparator(s). On the other hand, it should be kept in mind that in many instances modelling is needed, e.g., to estimate final outcomes from the intermediate outcomes measured during a trial, or to extrapolate to the envisaged population. As an alternative to single-study-based economic evaluations, ‘stage-shift models’ or ‘natural-history models’ can be used to synthesise information from numerous studies and sources.

Stage-shift models (sometimes referred to as ‘shallow’ models) generally compare the distribution of disease stages associated with screening (e.g., observed in screening trials) with the distribution of disease stages at diagnosis (at a later point in time) seen in the absence of a screening programme (e.g., observations from cancer registers or screening-trial comparator arms) {39}. In the case of CRC, relevant stages could include absence of disease, pre-cancerous lesions, different stages of cancer. In stage-shift models the effect of screening is modelled by shifts to less severe stages, e.g., using information that, given systematic screening, proportionally more people will be diagnosed with pre-cancerous lesions or at an early stage of cancer that may be more operable or still curable. One common feature is that modelling starts at the time of the first screening and models estimate costs and effects as a function of disease stage at screening or diagnosis.

An alternative to stage-shift models is to use ‘natural-history models’ to estimate the development of cancer and its precursor stages in a population that is followed from an early age onwards (e.g., starting in young adulthood or at birth). For each member of the modelled cohort, the development of disease is tracked, i.e., the model determines at each point in time whether the person remains disease free, has a precursor lesion, or has developed a certain stage of cancer. Such epidemiological models of the natural history of the disease, sometimes referred to as ‘deep models’, have the advantage of serving as a basis for ‘applying’ or ‘plugging-in’ a diverse range of screening strategies, diagnostic procedures, treatments, and associated costs and effects. One limitation is that extensive epidemiological data of high quality are required. Natural-history models also usually cover the early stages of the disease process and there are often gaps in knowledge that hinder modelling. On the other hand, lead-time and length-time issues (e.g., the question whether screening participants may encounter a net survival benefit or beneficial treatment, or whether he/she is only diagnosed earlier), may be efficiently addressed using natural-history models, if suitable input data are available.

The differences and similarities between the studies with respect to the type of model used can be found from the Appendix (see table “Study designs”).