The incremental cost-effectiveness ratio (ICER) of FIT versus gFOBT is based on the studies identified through the systematic literature search described above. In most of these studies the incremental cost-effectiveness ratio (ICER) was already calculated and explicitly given. For the remaining studies the incremental costs and effects of FIT versus gFOBT was calculated and divided to obtain the ICER:

ICER= DCosts/D Effects

For the estimation of the overall costs of the different screening technologies most of the cost-effectiveness models adopted a payer’s perspective and hence only included the direct costs of the screening programme, the follow-up tests and the treatment of a detected adenoma or cancer. Heitman et al. (2009 and 2010) {14, 15} also added indirect costs of the patient time and of the transport. 

Adjustment of price levels:

The results of the cost-effectiveness models were not adjusted to one price level or one point in time. Since most studies (12 of 16) were conducted in or after 2008 this might cause non-significant changes to the ranking of the ICERs in the tables but is unlikely to change the overall order dramatically.

Sensitivity analyses:

An analysis of the robustness of the results was not given by all studies, especially not on the gFOBT-vs.-FIT ICER since the primary aim of some studies was to compare not only FIT and gFOBT but also other screening strategies, or no screening at all.

1. Studies life years gained as the outcome measure:

The result of the study of van Ballegooijen et al. (2003) revealed that FIT with a 98% specificity dominates gFBOT (Hemeoccult II) and that this holds even if the compliance rate decreases from 100% to 60% for gFOBT and 90% for FIT {10}.

In a Monte Carlo simulation Parekh et al. (2008) shows that FIT dominates a no screening strategy in 100% of the simulations whereas gFOBT dominates a no-screening strategy in only around 95% of the simulations {18}. A direct comparison of FIT and gFOBT using a Monte Carlo simulation was not conducted.

Van Rossum et al. (2011) present the results of a deterministic sensitivity analysis, with the result that only very low CRC incidence and very high estimated costs of colonoscopy have a negative impact on the cost-effectiveness of FIT {19}.

Another deterministic sensitivity analysis was conducted by Berchi et al. (2004) using different assumptions about the participation rate, the costs of gFOBT, costs of colonoscopy, costs of CRC treatment, quality of FIT and natural history of disease {12}. The results suggest a robustness of the finding that FIT is more effective but also more costly than gFOBT.

Lejeune et al.’s (2010) sensitivity analysis also shows that FIT remained more effective and more costly for all parameter variations, but that the ICER increases if the price for gFOBT is decreased by 50% or if the sensitivity of FIT decreases. The ICER decreases if the price for FIT decreases by 50%, if the participation rate of FIT is increased by 10%, if the lead time associated with FIT increases or if FIT`s sensitivity increases {17}.

A probabilistic sensitivity analysis of the model designed by Heresbach et al. (2010) shows that FIT is more effective than gFOBT in 97.8% and dominant in 8.9% of the simulations {16}.

Hassan et al. (2011) also conducted a probabilistic sensitivity analysis with the result that annual FIT had the highest net-benefit in 20% of the simulations, compared to other strategies such as sigmoidoscopy every five years, of colonoscopy every 10 years which had the highest net-benefit in 40% and 26% of the simulations respectively {13}.

In the publication of Zauber (2010) no sensitivity analysis is reported {25}.

1. Studies using QALYs gained as the outcome measure:

The univariate sensitivity analysis of Heitman et al. (2009) showed that changing the costs or the screening interval (biennial instead of annual) did not change the dominance of FIT-mid (representing studies with “mid-range” test performance) over gFOBT {14}.

In the study published by Heitman et al. one year later (2010) no sensitivity analysis of the direct comparison of FIT versus gFOBT was published, but the probabilistic analysis shows that in nearly 100% of the simulations FITmid was cost saving and more effective compared to no screening, but no direct comparison of {15}.

The results of the model presented by Whyte et al (2012) are highly sensitive to several model parameters such as uptake, endoscopy costs and FIT thresholds. In the probabilistic sensitivity analysis FIT remains cost-effective compared to gFOBT at an assumed willingness-to-pay threshold of 20,000 £ {23}.

Wilschut et al. (2011) conducted a deterministic sensitivity analysis on different parameters. FIT (50ng/ml) for the age group 45-80 years remained most cost-effective when attendance rate increased to 100%, quality of life loss was taken into account, fatal complications decreased, FOBT costs and colonoscopy costs were halved, and when complication and treatment costs were halved or doubled. FIT (50ng/ml) for the age group 50-80 years became more cost-effective for all other parameter changes {24}.

In the one-way sensitivity analysis of Sharp et al. (2012) the most influential parameters were the discount rate, the costs of the tests and the costs of managing CRC. The ICERs for FIT and gFOBT compared to no screening always remained below the chosen notional cost-effectiveness threshold (45,000 € per QALY). The probabilistic sensitivity analysis showed that uncertainty was greatest for FIT, but all cost-effectiveness ratios remained below the notional threshold and the incremental QALY gains for FIT exceeded those for gFOBT in most simulations {20}.

The model of Telford et al. (2010) was sensitive to variations in the sensitivity of the tests to detect advanced adenoma, the costs of the test, the compliance with screening and the costs of cancer care. The probabilistic sensitivity showed that the ranking of strategies did not change and that no strategy was dominant {22}.

In the model of Sobhani et al. (2011) the participation rate seems to have a very high impact on the results, whereas changing the compliance rate for colonoscopy did not change the results much. Throughout the one-way sensitivity analyses the 3-sample FIT with a cut-off level of 50ng/ml was consistently the most cost-effective choice {21}.

It is difficult to provide an overall statement on the robustness of the ICER estimations because neither the methods of sensitivity analysis nor the reporting of the results are standardised. The matter is further complicated by that fact that in some of the studies FIT is not directly compared to gFOBT but both are compared to no screening. As a consequence, the sensitivity analyses focus on the robustness of the cost-effectiveness compared to no screening.

In conclusion, the deterministic sensitivity analyses of the studies using life years gained as the outcome measure show that the model results are quite robust, especially to changes in compliance and participation rate, but might be sensitive to other assumptions, such as that the CRC incidence is very low, the costs of a colonoscopy or a gFOBT is very high or that the sensitivity of FIT decreases. The probabilistic sensitivity analyses suggest that the results are rather robust.

The sensitivity analyses of the studies using QALYs gained as the outcome measure show no common pattern and therefore no qualified statement about influential parameters can be given.

Six studies reported the result of a deterministic sensitivity analysis {10, 12, 17, 19, 21, 24}, four used a probabilistic method {13, 15, 16, 18}and four studies used both, a deterministic and a probabilistic analysis {14, 20, 22, 23}. Only one study did not report a sensitivity analysis {25}.