Clinical performance

Description of terms used to describe test effectiveness 

gFOBT screening has been proven to be effective in reducing colorectal cancer mortality (Hewitson et al. 2007). In randomised trials the reduction in cause-specific mortality ranged from 15% (Hardcastle et al. 1996) to 33% (Mandel et al. 1993). Such a large variance in mortality can be explained by test differences, different numbers of faecal samples, different intervals between invitation cycles (one-year or two-year) and different responses to invitation associated with the characteristics and composition of the population screened. The sensitivity and specificity quoted for a test will therefore be influenced both by the test’s analytical characteristics and the context in which the test is used and evaluated.

gFOBTs come in two forms, the conventional form with normal sensitivity and the more sensitive variety, Hemoccult SENSA, in which the sample is hydrated before analysis. Hemoccult SENSA performs quite differently from the gFOBTs used in European trials (Hardcastle et al. 1996; Kronborg et al. 1996) and is both more sensitive and less specific. Comparison of the clinical performance of gFOBT and iFOBT is complex because iFOBTs can have different levels of specificity and sensitivity indeed they may have variable positive cut-off concentrations. Changes in cut-off concentrations result in different clinical performance characteristics.

Although only one population-based RCT has been described with iFOBT (van Rossum et al. 2008), the many published diagnostic accuracy studies provide information on the comparative ability of current tests to distinguish subjects with or without colorectal cancer and adenoma and can be considered an acceptable indication of the satisfactory performance of iFOBT in population screening (Burch et al. 2007). 

CRC mortality rates

It will be several years before the impact of population screening on CRC mortality becomes observable, and many more years before the full effect is achieved. The timing of a reduction depends on the natural history of the disease, and the ‘lead time’ due to screening (i.e. the time by which screening advances the date of diagnosis) as well as on the time taken to cover the target population. It will also depend on the quality of screening. 

Methods to evaluate the impact of screening on CRC mortality include analyses of population trends, cohort studies (aggregated or individual-based) and case-control studies.

Population trends 

Mortality from CRC has been decreasing in many European countries since the mid 1990’s, (Karim-Kos et al. 2008). Analyses of the routinely produced age-gender specific population rates over time will be subject to limitations due to the dilution of the effect of screening from deaths occurring in cases diagnosed prior to the introduction of screening, and/or at an age below which invitations begin. This can be overcome by use of refined CRC mortality rates in which such deaths are excluded. However, the rates will also be confounded by other factors such as cohort effects on underlying incidence, and by the effects of improvements in treatment and/or the stage of diagnosis of symptomatic disease on survival and mortality. Thus whilst a lack of any reduction in population mortality rates several years after the introduction of a screening programme should be a cause for concern, it is difficult to use such trends to quantify the effect, and attempts to do so should take account of the factors discussed above. 

Cohort studies 

In some settings, the introduction of population screening will have been phased in such a way as to facilitate comparisons of populations invited at different time points. Such a model has been used in Finland (see Ch. 2, Sect. 2.6.4). In the absence of such a system, comparisons can be made between geographical areas (regions invited/not invited to screening) or between the same population in different time periods before and after the introduction of screening. Both types of comparison are liable to possible bias due to underlying differences in the risk in the populations/time-periods. This may – under certain circumstances – be compensated for by including also a comparison group from geographic areas where no invitational program existed from before the introduction of screening. Cohort studies using aggregated data need estimates of incidence in order to avoid dilution effect discussed above. 

These biases can be avoided by individual-based cohort studies in which deaths and cancer registrations are linked to screening histories.