Authors' replyBMJ 1996; 313 doi: https://doi.org/10.1136/bmj.313.7065.1144 (Published 02 November 1996) Cite this as: BMJ 1996;313:1144
- P D P Pharoah,
- W Hollingworth
- Senior registrar in public health Cambridge and Huntingdon Health Commission, Fulbourn Hospital, Cambridge CB1 5EF
- Health economist Health Services Research Group, Department of Community Medicine, Institute of Public Health, University of Cambridge, Cambridge CB2 2SR
EDITOR,—We agree with J Shepherd that quality of life is a relevant outcome, but we could not model this using the available data. He also points out that we did not consider the costs of some non-fatal outcomes in our model, a criticism repeated by J J V McMurray and C E Morrison. This would have little impact on our conclusions. Since the publication of our analysis, Jonsson et al have published a health economic analysis of the data from the Scandinavian study.1 They found that only 10% of the cost savings through reduced admissions was due to admissions for conditions not included in our model. Our estimated number of revascularisation procedures prevented was relative to life years and not lives saved.
Shepherd and McMurray and Morrison suggest that non-compliance will result in our model overestimating the cost effectiveness ratio, while Alan Haycox and colleagues suggest the opposite. However, the relative impact of these opposite effects in clinical practice are not known, and a model that makes no correction for non-compliance would seem a reasonable compromise.
A S Wierzbicki and T M Reynolds repeat their view that relative risk and not absolute risk should be taken into account when deciding who to treat.2 This argument was eloquently countered by Haq et al,3 with whom we agree. They also note that the cost per life year saved estimated by Jonsson et al is considerably less than our estimation (an average of £32 000 per life year saved and not the £361 000 quoted).1 Several factors account for this apparent discrepancy.
Firstly, the treatment costs considered by Jonsson et al were those occurring for the duration of the trial, but the life years saved included an estimation of life expectancy after stopping treatment. As we do not know the outcome after stopping treatment we cannot predict the impact on cost effectiveness, and given that treatment is unlikely to be stopped once started, this approach would underestimate the costs considerably. A more realistic comparison would be with the cost per life year saved over the time period of the trial, which would be £20 300.
Secondly, their drug costs were 20% less than those we used.
Finally, men comprised 80% of the population in the Scandinavian trial and only 50% of our model population. Men are at higher risk of fatal coronary heart disease and so cost effectiveness ratio in this group would be expected to be lower.
Haycox and colleagues are right to question our assumption that non-cardiovascular mortality is the same in those at high risk of cardiovascular disease as in the general population. That behavioural effects may outweigh the benefits of drug treatments is feasible but not supported by the trials, which found reductions in all cause mortality as well as cardiovascular mortality. As the concept of discounting makes Paul Masters and Arun Shetty shudder, they should perhaps reflect on the relative paucity of investment by the NHS in preventive services (to save lives tomorrow) compared with the enormous investment in acute services to treat prevalent disease (to save lives today).
These observations serve to re-emphasise our conclusion that an average cost effectiveness may cover a very wide range of values for different at risk groups, highlighting the importance of assessing absolute risk when considering which groups of patients to treat. We still need to bear in mind that treatment of all who would benefit from intervention would be prohibitively expensive for the NHS, and thus statins should be reserved for those who will benefit most.