Rapid Responses to:
|
|
Rapid Responses: Rapid Responses published:
-
![[Read Rapid Response]](/icons/misc/sectionDOWN.gif)
Cost effectiveness of breast screening
- J Mark F Temple (11 September 1998)
-
Modelling is suspect, and results lack confidence intervals
- N E Day (21 October 1998)
|
|
|||
|
J Mark F Temple
Send response to journal:
|
We congratulate Boer et al(1) on an interesting and useful application of modelling to the important issue of breast cancer screening policy. We believe that modelling has much to offer public health professionals working in the NHS. Modelling can improve the understanding of the likely effects of potential health service interventions both on the target population and also on the total population. The paper presented the cost effectiveness of the effects of two different changes in screening policy on breast cancer mortality; a useful addition to the information available from studies demonstrating the effectiveness of screening in preventing breast cancer deaths. The real benefit of simulation modelling over randomised controlled trials is that the all cause population mortality effect of a change in screening policy can be evaluated. This wider evaluation can make a substantial contribution to the screening policy debate. Yours faithfully J Mark F Temple, Senior Registrar, David L Fone, Consultant in Public Health Medicine Nathan J Lester, Research Officer Department of Public Health, Gwent Health Authority, Mamhilad House, Pontypool, NP4 0YP. (1) Boer R, de Koning H, et al. Cost effectiveness of shortening screening interval or extending age range of NHS breast screening programme: computer simulation study. BMJ 1998;317: 376-9 |
|||
|
|
|||
|
N E Day, Director
Send response to journal:
|
The Editor Dear Sir Boer and his colleagues (1) present the results of a simulation exercise in which they investigate the cost effectiveness of changing the parameters of the national breast screening programme. In particular, they compare a two year screening interval in place of the current three year interval. Their results should not pass without comment. Firstly, their simulation model uses values for the natural history of early breast cancer and the sensitivity of breast screening taken from the Utrecht and Nijimegen programmes both of which started in the mid 1970's. The relevance for breast screening in the UK in the late 1990's is not clear, bearing in mind that there is considerable between programme variation in such basic parameters as the interval cancer rate and the screening detection rate of small (<10 mms) invasive cancers (2). Secondly, their model fits the NW England results for the second screening round badly, even though they have made some adjustment to the assumed natural history values (doubling the sojourn time of T1a and T1b cancers) to ensure that the NW England first round screening data are well fitted. Their model predicts that at the second screening test after on interval of three years, more than 60% of invasive cancers will be less than or equal to 10 mms in diameter, and 9% greater than 20 mms. The corresponding observed frequencies are 40% and 19%. Since the effect of screening on breast cancer mortality is determined largely by the reduction in T2+ cancers and the corresponding increase in T1a and T1b cancers, the effect of gross inaccuracy in modelling these two quantities is likely to be substantial. Reducing the screening interval will increase the proportion of cancers that are screen detected, and among screen detected should both increase the proportion of small cancers and decrease the proportion of large cancers. The effect on predicted mortality of poorly modelling the stage distribution of cancers detected at the second or later screen, is thus likely to be even greater, with a two year interval than a three year interval. The estimated marginal effect of reducing the screening interval will then be unreliable to an unknown extent. Thirdly, no indication whatever is given of the uncertainty surrounding the various estimates. The authors claim that the marginal cost per life year gained of shortening the screening interval from three to two years is £3545. This is a clear example of spurious precision, but more seriously is likely to be highly misleading. Given the uncertainties over the parameter values and the poor fit of the model itself, one would have expected some sensitivity analyses at the very least. Quite plausible alternative parameter values might well generate marginal cost values several times greater than the quoted estimated of £3545. In most situations the British Medical Journal quite rightly insists on uncertainty estimates, usually in the form of confidence intervals. For an article which the authors hope will make an impact on policy, omission of uncertainty bounds renders it almost valueless. One expects policy for the national breast screening programme to be evidence based. The modelling presented by Boer and his colleagues uses only a small portion of the available evidence on breast screening, and none of the evidence from randomised trials or from relevant screening in the UK. It fits the relevant parts of the NW region screening data badly, and the resulting estimates are of unknown reliability. Future policy decisions should be based on evidence directly related to the UK programme itself. Such evidence will shortly be available from the results of the multi centre randomised trial of different breast screening frequencies, undertaken in Britain under the auspices of the UKCCCR. N E Day PhD Director of the Unit 1 Boer R, de Koning H, Threlfall A, Warmerdam P, Street A, Friedman E and Woodman C. Cost effectiveness of shortening screening interval or extending age range of NHS breast screening programme: computer simulation study. British Medical Journal 1998; 317: 376-379. 2 Day N E, McCann J, Camilleri-Ferrante CC et al. Monitoring interval cancers in breast screening programmes: the East Anglian experience. J Med Screen 1995; 2: 180-5. |
|||