last update of this page: 29 July 2005
click here for abstract of article preceding this additional discussion
This chapter concentrates strongly on the balance of favourable and unfavourable effects as a determinant for the best upper age limit for a breast cancer screening programme. The reason for it is that, at least as far as the Netherlands are concerned, the discussion on screening older women was dominated by the issue of age discrimination and there was little possibility for cost-effectiveness considerations. Following are some additional sensitivity analyses with respect to the balance of favourable and unfavourable effects that were used for the decision on extension of the Dutch breast cancer screening programme to higher ages. After that I present some considerations on cost-effectiveness of screening older women.
further sensitivity analyses concerning the balance of favourable and unfavourable health effects of breast cancer screening at older ages
In addition to the analyses of the previous part of this chapter, the following model assumptions have been varied.
- After observing that there is good reason to assume that the Two Counties Study has shown a substantially better screening performance than so far in the Dutch national screening programme in terms of detection of particularly smaller cancers(Fracheboud et al. 1997), it was considered that there is also reason to assume that the influence of the Dutch programme on mortality is less substantial than in the Swedish randomised trials(Nystrm et al. 1993). Therefore next to the basic scenarios, also a 15% lower improvement of prognosis was assumed than originally estimated from the Swedish trials.
- The basic scenarios assume that the utilities of life years spent in the health states that are possibly affected by screening, are according to the point estimates from our quality of life study.(de Haes et al. 1991) As a variant, a worst case scenario based on uncertainty of the utility estimates was created. Instead of the point estimates, the side of the range of the individual estimates was chosen that gives the model prediction of screening effects that is least favourable for screening.
- It has been observed in the Nijmegen pilot project that mortality from other causes than breast cancer is lower among those participating to screening than among other women of the same age in the same population. This difference is very large just after screening, which is clearly due to death often being preceded by a period of bad health that precludes participation to screening. But also up to several years after screening, there is still a substantial difference in mortality. We have estimated that, apart from the risk of dying from breast cancer, women of 65 years and older who participate in the Nijmegen breast cancer screening pilot project, have around 1 year more life expectancy than the average population of the same age in that project. Therefore we added a model variant that assumed that per breast cancer death prevented there was 1 more life year gained than in the basic scenario that did not assume any association between participation to screening and mortality from other causes than breast cancer.
Figure 8.2 shows the expected number of 5% discounted QALYs gained of different scenarios by the number of two-yearly screenings that is added after the screening programme of women aged 50-69.
Figure 8.2 shows that optimistic scenarios for screening have no clear upper boundary for the age from which the unfavourable health effects of breast cancer screening outweigh the favourable effects. The decreasing slopes by age are not only caused by a decrease of the number of QALYs gained by screening, but also to a major extent by the decrease of the population with increasing age. Under the most pessimistic scenario, there remains a positive balance of favourable and unfavourable health effects up to around the age of 75.
It was considered to be prudent to extend the age range of the Dutch national breast cancer screening programme with three invitations from an age range of 50-69 to 50-75. Evaluation of the findings from screening women aged 70-75 is expected to give a better estimate of the age dependency of the preclinical screen-detectable period, thus answering the question whether reality is closer to the optimistic or to the pessimistic duration scenario. In case reality appears to be closer to the optimistic scenario, and if the same line of reasoning would be continued, then further extension of the screening programme would be appropriate.
efficiency of breast cancer screening at older ages
The line of reasoning by which screening is continued up to the age at which the unfavourable effects start outweighing the favourable effects is questionable. The cost per QALY gained of screenings that are close to the age at which no QALYs are gained, is very high because close to that age there are already very few QALYs being gained. Therefore, from a perspective of maximising utility gained given a certain amount of expenses, decision making would be based on a different line of reasoning. Then the question to be answered is: How are the invitations of the screening programme to be scheduled in order to achieve the most QALYs gained at a certain cost?
The model that has been developed for the Dutch screening programme was used to calculate expected health effects of a range of possible programmes. Two model variants with and without further increasing preclinical duration of the disease after age 65 (resp. pessimistic and optimistic, and line C and B of figure 8.2) were considered. Invitation schedules that include women younger than 50 were not considered in order to avoid the discussion of effectiveness of screening among younger women and because such schedules were expected to be of no influence on choosing the upper age limit. For estimating cost/utility ratios, simplified models for costs and for the influence of health status on utility were applied, see table 8.4 and table 8.5
Table 8.6 shows the outcomes of this simplified model for all invitation schedules that have been evaluated. Within the list of invitation schedules, some appear to be dominated by another in the sense that the other schedule is expected to yield more QALYs at a lower or equal cost. Because of the granular character of differences between invitation schedules, there is also a possibility of extended dominance, meaning that offering a fraction of the population one schedule, while the rest will be offered another schedule, is expected to lead to more QALYs gained at lower or the same cost than the schedule that is dominated. Invitation schedules that are not dominated by others are considered to be efficient. There can be several efficient schedules because this definition of efficiency does not include a preference considering the level of the cost-effectiveness ratio.(Cantor 1994) The original invitation schedule of the Dutch programme is not efficient under optimistic nor pessimistic assumptions. Neighbouring efficient schedules under optimistic assumptions are: 50(2.5)72.5 and 50(2.5)75, and under pessimistic assumptions: 50(2.5)72.5 and 50(2)72, all preferred alternatives have a higher upper age limit for the invitation schedule. Under optimistic assumptions the original Dutch invitation schedule (50(2)68) is expected to gain 4.2% fewer QALYs then efficient screening at the same cost, and under pessimistic assumptions that is 1.8%.
Both under optimistic and pessimistic assumptions, the invitation schedule 50(2)76 is efficient. The current Dutch invitation schedule of 50(2)74 is efficient under pessimistic assumptions, but dominated by 50(2.5)75 and 50(2)76 under optimistic assumptions. But even under optimistic assumptions, efficient screening at the same cost is expected to lead to only 0.2% more QALYS gained than the current invitation schedule.
In conclusion: Even though decision making was based only on considerations about the balance between favourable and unfavourable effects in order to warrant some certainty about this balance being positive, the present invitation schedule is efficient or close to efficient when point estimates are assumed for all parameters other than average sojourn time by age.