A full description of the Microsimulation Screening Analysis model has been published.2
In brief, the model first simulates life histories for women in the absence of a screening programme for breast cancer and then assesses how these life histories would change as a consequence of introducing different screening policies.
The natural course of breast cancer is modelled as a progression from no breast cancer through preclinical cancer to clinical disease. Women reside in the first state (no breast cancer) before entering one of five preclinical states. There is an in situ state and four invasive states according to the tumour size (T state):
mm (T1a), >5-10
mm (T1b), >10-20 mm (T1c), and >20
mm (T2+). A cancer may be detected at screening, become clinically apparent in any one of these states, or if undiagnosed progress to the next preclinical state. The two end states of the model are death from breast cancer and death from other causes.
The model was set up using data from the Dutch screening trials at Utrecht and Nijmegen to provide estimates of the mean duration of the preclinical phase for women in different age groups and the mean duration of cancer in each of the five preclinical states. The dwelling time of a cancer in each preclinical state is assumed to follow an exponential distribution, and the rate at which cancers progress from the preclinical to the clinical state is inferred from the observed incidence and distribution of stages of clinically diagnosed cancers in the population being studied.
When modelling the performance of a screening programme, key indicators include the mean duration of the screen detectable phase, the sensitivity of the test, and the improvement in prognosis for screen detected cancers. The mean preclinical screen detectable period assumed in the model was based on data from the Dutch screening projects at Nijmegen and Utrecht and varied from 1.8 years at age 35 to 6.2 years at age 70.
The sensitivity of the screening test is assumed in the model to be the probability of detecting a cancer in the preclinical screen detectable state. For women aged over 50 it is fixed as 0.4, 0.65, 0.8, 0.9, and 0.95 for in situ disease, T1a, T1b, T1c, and T2+ tumours respectively. The improvement in prognosis for screen detected cancers was derived from the results of the Swedish breast screening trials.3
Applying model to UK population
The North West health region has a population of 4.1 million and is covered by five NHS breast screening programmes. The largest of these, the Manchester breast screening programme, has screened over 120
000 women and reported cancer detection rates similar to those elsewhere in the United Kingdom.4
The number and size of cancers detected at a first and second screen and the occurrence and size of interval cancers in this programme have been used to inform the model. Estimates of screening and diagnostic costs are based on this programme assuming that two view mammography is used at the first screen and single view mammography at subsequent screens. Treatment costs are derived from various sources, but primarily the Christie Hospital NHS Trust in Manchester. Full details of the costing, including sensitivity analysis, have been published.5
Both costs and effects are discounted at 6%.
To simulate the life histories of women with breast cancer before a screening programme is introduced the model requires information on the age, distribution of stage, and survival of women with breast cancer. Neither the prescreening distribution of stage nor stage specific survival rates before screening was introduced were available for the North West’s population. However, the prescreening stage distribution in Scotland6
and in East Anglia (J McCann, East Anglian Cancer Registry, personal communication) was similar to that of the control population in the Utrecht screening trial. We therefore assumed that the prescreening stage distribution in the North West was similar to that used in setting up the computer model. The stage distribution in women aged 50-69 at diagnosis in the Utrecht control population was: 4.6% in situ, 1.5% T1a, 6.3% T1b, 32.6% T1c, and 55% T2+. Having assumed this stage distribution, we derived stage and age specific survival rates by fitting the North West’s observed mortality for 1987 to the observed incidence for 1987. This produced an overall five year survival for women aged 50-59 and women aged 60-69 of 67% and 68% respectively. A life table describing the probability of dying from causes other than breast cancer in the North West was used to derive the number of life years gained per breast cancer death prevented.
The model was unable to simulate the detection rate and distribution of stages observed at first screening in the North West. More small cancers were observed in the North West than were predicted by the model. This discrepancy was resolved by assuming a longer screen detectable preclinical phase for small tumours. When it was assumed that small tumours (less than 10 mm) dwelt in a screen detectable phase for twice as long as that used in the initial set up, the model adequately fitted the detection rate and stage distribution observed at first screening in the North West.
This model was used to simulate the effects and costs of three screening programmes for the North West: firstly, the current UK screening policy, in which women aged between 50 and 64 are invited for screening every three years; secondly, screening every three years but extending the age of women screened from 64 to 69 years; and, finally, reducing the screening interval from three to two years while maintaining the current age range. Attendance for screening was assumed to fall by 0.5% for each year of age, from 74.2% at age 50 to 67.9% at age 70; attendance at repeat invitations was assumed to be 78% higher than among those who attended the previous invitation. Each screening programme was assumed to run for 27 years.