Screening women aged 65 years or over for abdominal aortic aneurysm: a modelling study and health economic evaluation

Screening men for abdominal aortic aneurysms

The NHS Abdominal Aortic Aneurysm Screening Programme (NAAASP)8 for men aged 65 years was launched in England in 2009, and similar programmes have subsequently been introduced to other parts of the UK. In NAAASP, an aortic diameter of ≥ 3.0 cm, as measured by ultrasound, is used as a diagnosis of an AAA. Men with smaller aortic diameters are reassured and not followed up further. Men with AAAs of 3.0- to 4.4-cm diameter enter a surveillance programme with annual follow-up scans, while those with AAAs of diameter 4.5–5.4 cm have follow-up scans every 3 months. Men with AAAs whose diameter is initially ≥ 5.5 cm, or which expand to that diameter during surveillance, are referred for consideration of elective surgery.

The scientific evidence supporting the implementation of NAAASP came from the results of four randomised trials9 of AAA screening that almost exclusively recruited men. These trials showed that AAA-related mortality in men could be halved by offering AAA screening along with appropriate clinical follow-up that included elective surgery when an AAA reached a threshold size. Long-term modelling based on the largest of these trials, the Multicentre Aneurysm Screening Study (MASS),10 showed that AAA screening in men aged 65 years was extremely cost-effective, with an estimated cost of £3000 per quality-adjusted life-year (QALY) gained. 11

This cost-effectiveness estimate came under scrutiny later, because initial data from NAAASP8 showed an AAA prevalence of 1.5% in men aged 65 years, rather than 4.9% as observed in the MASS trial. 10 Revision of the long-term model to reflect this lower prevalence as well as the attendance rates observed in NAAASP and updated cost estimates increased the cost per QALY. Nevertheless, NAAASP was still estimated to be highly cost-effective, at £7400 per QALY gained. 12 Indeed, provided the AAA prevalence was above 0.35%, it was estimated that screening would be cost-effective at a willingness to pay (WTP) of £20,000 per QALY. This could imply that screening women for AAAs might also be cost-effective.

Screening women for abdominal aortic aneurysms

The prevalence of AAAs in women aged 65 or 70 years may be around 0.5%. 13–15 Moreover, it is known that women have an AAA rupture rate about fourfold of that in men for a given AAA diameter,3 although their AAA growth rates are similar. 16 Women may also have worse outcomes after AAA surgery than men,17,18 for example, because of their typically shorter aneurysm necks. 19 A higher proportion of women are turned down for both elective and emergency surgery. 20 Some of these differences between women and men would probably favour systematic AAA screening in women, whereas others would not.

There are a number of reasons why the design of an optimal AAA screening programme for women might differ from that currently adopted for men. The prevalence of AAA increases with age, and women have a greater life expectancy than men, so screening women at age 70 years might be more beneficial than screening them at age 65 years. The diameter of the aorta ss typically smaller aortic in women than in men,21 and the aortic diameter that define an aneurysm could be lowered from the conventional 3.0 cm. Because AAA rupture rates are higher in women, it may be advisable to reduce the diameter threshold for considering elective surgery below the usual 5.5 cm.

There is no prospect of being able to undertake a randomised trial to answer these questions in women. First, such a trial would have to be an order of magnitude bigger than the MASS trial10 of 68,000 men and with a similar length of follow-up (13 years). 22 Second, a single trial could not address the relative merits of different designs of a screening and intervention programme. The best practical way in which these questions can be addressed is by undertaking a detailed modelling exercise, which is described here.

Scientific objectives

Screening and monitoring

We refer to an initial ultrasound scan as a ‘screening’ scan and a subsequent check-up scan as a ‘surveillance’ scan. An individual who does not respond to the initial invitation to screening is reinvited, and may either attend and be screened or fail to attend screening. Invitation, reinvitation and screening all incur costs. In a small proportion of individuals who attend screening, visualisation of the aorta will be unsuccessful; these individuals will be discharged from the screening programme. For a person who is successfully screened, there are three possibilities: (1) if the aortic size, measured using an ultrasound scan, is less than the diagnosis threshold (currently 3.0 cm), then repeat surveillance is not needed and the individual is discharged; (2) if the aortic size is greater than or equal to the diagnosis threshold and less than the intervention threshold (currently 5.5 cm), then the individual is entered into surveillance and a surveillance scan is scheduled depending on the measured AAA size (in NAAASP, this is 1 year for AAAs of diameter 3.0–4.4 cm and 3 months for AAAs of diameter 4.5–5.4 cm); and (3) if the aortic size is greater than or equal to the intervention threshold (currently 5.5 cm), then a consultation with a vascular surgeon is scheduled. The model is flexible enough to allow any of the diameter thresholds and/or surveillance times to be modified. For example, a consultation for elective surgery could be scheduled earlier by changing the intervention diameter threshold from 5.5 to 5.0 cm.

Individuals whose AAA measures less than the diagnosis threshold at any of the surveillance scans (i.e. excluding the initial screening scan) are kept in the surveillance programme and have another surveillance scan scheduled as usual. Ultrasound scans are assumed to give imprecise measurements of the underlying aortic diameter (a latent parameter that changes over time and is defined for each individual by an aortic growth model; see Modelling aortic growth and abdominal aortic aneurysm rupture).

Dropout and incidental detection

Following the screening scan, if an individual remains in the screening programme, then a future ‘dropout from surveillance’ time is scheduled to account for a certain proportion of individuals who will drop out of the screening programme over time. If individuals are not under active follow-up in the screening programme (e.g. those in the non-invited group and those in the invited group whose screening scan was normal), then an incidental detection time is scheduled. Individuals become at risk of incidental detection only once their underlying aortic diameter reaches the diagnosis threshold, and their incidental detection time is, therefore, scheduled at some time after this occurrence. However, if the diameter is decreasing over time (a rare, but possible, occurrence when simulating many individuals), then incidental detection is allowed to occur only up until the time at which the diameter of that individual’s AAA drops below the diagnosis threshold. Incidental detection and dropout times are both generated from an exponential distribution with a fixed rate.

In the case of individuals who drop out of the screening programme, a further incidental detection time is scheduled, and, for those entering the screening programme via incidental detection, a further dropout time is scheduled. Hence, individuals can repeatedly drop out and come back into surveillance, although this will be a rare phenomenon in practice.

Consultation with a vascular surgeon

If an individual has a consultation with a vascular surgeon then their aortic diameter is remeasured using a computerised tomography (CT) scan, which may give a different reading to an ultrasound scan (see Consultation scan: computerised tomography scan versus ultrasound scan). For example, CT may give a systematically higher reading of the aortic diameter than ultrasound, and may also have a different measurement error. If the measured size on the CT scan is less than the intervention threshold, then the individual is returned to surveillance – that is, a new surveillance scan is scheduled. Otherwise, either an elective operation is scheduled or surgery is deemed to be contraindicated for that individual. If elective surgery is contraindicated, no surgery is scheduled and the individual remains under a defined surveillance protocol until their aneurysm ruptures or they die from a non-AAA cause. The DES allows the non-AAA death rate to change among individuals in whom surgery is contraindicated (e.g. owing to comorbidities, the death rate in this group is generally much higher than that of the general population).

Emergency and elective surgery

If an individual’s AAA ruptures, then they will either receive emergency surgery or die before they reach the operating table (in which case their death is recorded as AAA related). Both emergency and elective surgery carry an initial operative (30-day) mortality risk and a longer-term AAA-related mortality risk to account for future complications and secondary ruptures. Surgery can be via either EVAR or open repair; the probability that an individual will undergo EVAR is a parameter of the DES. The DES is flexible enough to allow specification of operative and longer-term risks separately for EVAR and open repair, emergency and elective, together with associating different costs to each type of repair. The initial 30-day postoperative mortality is implemented in the DES as an immediate event.

The model also allows the user to specify whether or not longer-term (> 30 days) postoperative AAA-related events can occur, such as reinterventions or postoperative surveillance, which incur costs. Such events can be scheduled at the time of operation or after the occurrence of a postoperative event (to allow for recurrent events), and the rate and cost of postoperative events can depend on the type of operation (EVAR or open) and whether the operation was in the elective or emergency setting. These model extensions are not applied in the validation model described in this chapter, but are considered in the inputs to the DES for women (see Chapter 5).

Finally, the DES is also flexible enough to allow operative events (i.e. the proportion receiving EVAR, or the proportion who do not survive 30 days postoperatively) to depend on the age and AAA diameter of the individual at the time of surgery. These risk factors are incorporated using logistic regression models, with the user specifying log-odds ratios for covariates associated with each event. As above, these DES model extensions are considered in Chapter 5 in the model for women.

Death from non-abdominal aortic aneurysm causes

Rates of non-AAA deaths can be input into the DES using age-specific rates (e.g. from population mortality statistics). The model then calculates the conditional survival curve from age at screening and simulates for each pair of individuals a time of non-AAA death by sampling from a Uniform(0,1) distribution and evaluating the inverse function of the survival distribution.

Example sequences of events

Table 5 shows example sequences of events from the DES for four pairs of twins (i.e. individuals with identical baseline characteristics). Like most individuals, pairs 1 and 2 have rather short sequences of events and die of non-AAA causes – each twin dies at the same time, so the only difference between them is that more money was spent on the twin who was invited to screening. Pairs 3 and 4 were chosen because they have longer sequences of events.

For pair 3, the twin who is invited to screening attends and is found to have a 4.35-cm AAA. He has a surveillance scan 1 year later and then at 3-month intervals once his aneurysm is observed to be ≥ 4.5 cm. At 1.75 years, the ultrasound scan finds his aortic size to be 6.08 cm. This high reading is largely due to measurement error and his true aortic size is only 5.16 cm. Nevertheless, he is referred for a consultation, in which the more accurate CT scan, 71 days later, measures the aneurysm at 5.04 cm and he is, therefore, returned to surveillance. This person continues to attend his surveillance scans. About 1 year later his aneurysm is again measured to be > 5.5 cm and he has another consultation. This time the large aneurysm is confirmed by the CT scan and it is decided that he should have elective surgery. The surgery is successful and he lives for another 3 years.

His twin, who is not invited to screening, experiences a rupture, with a large aneurysm of 7.13 cm, but receives emergency surgery and survives. In the end, the two twins die at the exact same time due to non-AAA causes, so there are no gains in life-years in this example. However, the screening programme has saved overall costs by avoiding a costly emergency operation – the first twin instead has the safer and cheaper elective surgery.

Pair 4 is another example in which the screening programme works well. The twin who is invited to screening lives longer as a result of the screening. His aneurysm is detected straight away, he is referred for a consultation and he has elective surgery, which is successful. The other twin’s AAA ruptures and he dies without getting to the hospital in time to have emergency surgery.

The model for aortic growth

In the DES, the evolution of an individual’s aortic diameter over time must be taken into account as it affects many aspects of the health economic model, namely (1) when an individual can be diagnosed, (2) planned surveillance intervals, (3) when an intervention can be considered, (4) the risk of rupture, (5) the probability of receiving EVAR rather than open repair and (6) the operative mortality risk. Therefore, the evolution of the aortic diameter over time is modelled using a continuous-time linear mixed model, which allows the underlying diameter and a measured diameter (using ultrasound or CT) to be determined at any time point. Let y_ij be the aortic diameter, as measured using ultrasound, of person i at time t_ij, j = 1,. . .,n_i; so y_i0 is the baseline diameter as measured at screening. The linear mixed model is as follows:

where

Each person has two random effects: (1) their own intercept (true baseline log-diameter), b_0i, and (2) their own slope (rate of growth), b_1i, measured on the log-diameter scale. Correlation between an individual’s underlying baseline log-diameter and slope is allowed as b_0i and b_1i have a bivariate normal distribution with correlation parameter ρ. The parameters σ²₀ and σ²₁ determine the between-person variability of the intercepts and slopes, respectively, while σ²_w determines the amount of variability due to measurement error.

The linear mixed model is fitted using data from repeated ultrasound measurements of the aortic diameter from cohorts of AAA patients such as from the MASS trial10 or RESCAN studies. 35 These cohorts are restricted to the diameter range 3.0–5.5 cm. As a result, model extrapolation is used to infer true baseline diameters and growth rates for individuals outside this range.

Baseline diameter distribution and derived random effects

The baseline diameter distribution is a particularly important aspect of the DES, because it determines how many persons have aneurysms at the time at which screening would be implemented and has a great effect on how many develop aneurysms in subsequent years. The full specification of the model is that y_i0 follows a fixed baseline distribution, which we specify using external data sources (e.g. data on measured diameters from the first 700,000 men screened in NAAASP), and an individual’s random effects b_0i and b_1i are then generated conditional on their observed baseline diameter. Following evaluation of the performance of the aortic growth model, it was decided to use the following rules to generate an individual’s random-effects (full details of the reasons for these choices are given in Appendix 2).

Prevalence

The user of the DES specifies the baseline diameter distribution as an input as well as, optionally, the prevalence of AAAs, which is the probability that a baseline diameter is greater or equal to the diagnosis threshold (e.g. 3.0 cm). If provided, this optional input then reweights the baseline diameter distribution accordingly (further details of this reweighting procedure are given in Chapter 4).

Calculation of a person’s aortic diameter at any time

The aortic diameter measured at an individual’s initial screening scan is taken to be y_i0, because this ensures the correct prevalence of AAA at screening.

When an individual’s aortic diameter at time t > 0 is needed, one of the following formulas is used.

The measurement error is expressed differently in the formulas for ultrasound and CT scans. This has been done as a matter of convenience since estimates of σ_w come from the mixed-effects model that uses ultrasound measurements and is fitted on the log-diameter scale. Meanwhile, estimates of µ_CT and σ_CT will generally originate from the literature and are assumed to be additive on the diameter scale.

The model for abdominal aortic aneurysm rupture

The model for AAA rupture is the survival component of a joint longitudinal and survival model. According to this model, the hazard of rupture is:

where γ is the log-baseline hazard and α is the log-hazard ratio associated with a 1-unit increase in log-aortic diameter (the expression in the inner brackets; see The model for aortic growth). In reality, the hazard of rupture will increase with the aortic diameter, and this is the case if α is positive.

The hazard function corresponds to a Gompertz distribution with shape parameter αb_1i and rate parameter exp(γ + αb_0i). Therefore, rupture times are generated from this distribution for each pair of individuals from the time of screening.

Convergence

Differences in life-years and costs between the invited to screening and not invited to screening groups are generally small, as the prevalence of AAAs is relatively low and, therefore, a screening programme will benefit only a small proportion of the population. Thus, the model needs to be run on millions of individuals in order to obtain accurate estimates of incremental life-years. However, if interest is primarily in incremental costs and life-years (rather than absolute estimates for each group), then this approach is computationally inefficient. A pair of twins who have an aortic diameter less than the diagnosis threshold at screening should follow exactly the same life-course, as the twin who is invited to screening will be found to be normal on screening and will no longer be followed up. The only difference between the twins is that the twin invited for screening will incur extra screening costs. For this reason, if accurate estimates are required of incremental life-years, incremental costs, the ICER and INMB, then the DES can be run by selectively sampling only individuals above the diagnosis threshold. The mean incremental costs and life-years are then calculated in this subgroup and are weighted by the prevalence (proportion of individuals in the population who are above the diagnosis threshold at screening). The population below the diagnosis threshold is never sampled but has zero mean incremental life-years and mean incremental costs determined by the mean screening costs in the invited to screening group. These are weighted by (1 – prevalence) and are added to the incremental mean costs in those sampled above the threshold. An example of the convergence of the ICER using the selective sampling approach versus not using the approach is shown in Appendix 2.

Probabilistic sensitivity analysis

A PSA is conducted by running the main analysis a large number of times using a different set of parameter inputs each time to account for uncertainty in the parameters. Each of the main analyses that are performed in the PSA produces an estimate of the incremental cost and an estimate of the incremental effectiveness, and these quantities can then be viewed in a scatterplot and a cost-effectiveness acceptability curve (CEAC) can be generated.

Figure 2 shows how the functions in the DES form a hierarchy in which each function runs the function below it multiple times. At the top is a function named ‘psa’, which performs a PSA. Next is ‘processIndividuals’, which conducts the main analysis. Within this function is ‘processOnePair’, which generates a pair of twins and calculates their health-economic quantities. Embedded into this function is ‘generateEventHistory’, which generates a single sequence of events for an individual. This is run twice by ‘processOnePair’, once for the invited twin and once for the non-invited twin.

FIGURE 2.

Abdominal aortic aneurysm screening DES: hierarchy of functions.

Validating against 4-year data from the Multicentre Aneurysm Screening Study in men

To validate the DES, a model was developed based on inputs used in the original Markov model developed for men. 11 The values of the parameters used are shown in Table 27, Appendix 2, alongside distributions placed on the parameters for PSA. The baseline distribution was taken from the first 700,000 screened men in NAAASP and was then reweighted to give an AAA prevalence of 4.97%, as observed in the MASS trial. 10 A further reweighting was then undertaken to ensure that 70.8% of the individuals’ aneurysms were small (3.0–4.4 cm), 16.7% were medium (4.5–5.4 cm) and 12.5% were large (≥ 5.5 cm) at screening, as seen in the MASS trial. 10 Non-AAA death rates were taken directly from those observed in the MASS trial. 10 Costs were the same as those used originally:11 invitation (£1.31), reinvitation (£1.28), screening ultrasound scan (£19.08), surveillance ultrasound scan (£46.04), consultation for elective surgery (£309.88), elective open repair (£6908.75) and emergency open repair (£11,175.63).

The screening programme evaluated was as specified in the MASS trial10 (1-year monitoring for AAAs of 3.0- to 4.4-cm diameter, 3-month monitoring for AAAs of 4.5- to 5.4-cm diameter and consideration for elective surgery for AAAs ≥ 5.5-cm diameter). The DES was run for a population of men aged 69 years (mean age of the MASS trial10), for a mean follow-up of 4 years (with random censoring between 3 and 5 years to mimic censoring in the 4-year MASS trial10 results) and the total number of events were compared with those observed in the 4-year MASS trial10 data.

Table 6 shows the numbers of key events in the two groups over a 4-year period as estimated by the DES. The DES was run using 107 pairs of individuals, and then the estimated numbers of events that occurred was scaled to be relevant to the size of the invited and control (non-invited) groups in the MASS trial. 10 For comparison, the observed numbers in the MASS trial10 are also given, together with the expected-to-observed (E/O) ratio expressed as a percentage. The E/O ratio is within ± 20% for all events except for contraindications resulting from screen-detected AAAs. The DES was deemed accurate enough in estimating the numbers of key events in both arms, and the timing of these events also adequately followed the occurrence of the events in the trial. Further results from this validation exercise are given elsewhere, including cumulative numbers of key events over time. 46

Validating against 30-year contemporary Markov model in men

The DES was then extended to run for 30 years, with inputs updated to reflect more contemporaneous estimates for men. 12 In particular, national mortality statistics were used for non-AAA death rates, the baseline diameter distribution from NAAASP was used directly (hence giving an AAA prevalence of 1.34%) and both EVAR and open repair for elective mortality were considered along with long-term postoperative mortality. Costs were the same as previously specified:12 invitation (£1.70), reinvitation (£1.70), screening ultrasound scan (£32.20), surveillance ultrasound scan (£68.00), consultation for elective surgery (£435.25), elective open repair (£11,532.69), elective EVAR (£13,345.66) and emergency open repair (£19,984.75).

The estimated life-years and costs were compared with the previously published estimates from a Markov model. 12 Table 7 shows that the results, although not identical, provide a similar conclusion regarding the cost-effectiveness of the AAA screening programme over a 30-year period. The DES estimates a higher gain in life-years but with similar incremental costs to the 30-year Markov model and as such the estimated ICER is about £1000 less. Nevertheless, both models suggest that the programme is highly cost-effective.

Summary

This review provided an overall AAA prevalence of 0.74% for women aged ≥ 60 years, with the prevalence increasing sharply with age and current smoking. The overall prevalences of 0.43% for the 61- to 70-year age group and of 1.15% for the 71- to 80-year age group are used in the modelling (see Chapters 7 and 8) as the prevalences for women aged 65 and 75 years, respectively.

Summary

Overall, only 34% of women are suitable for standard endovascular repair (compared with 54% for men). Even with devices newer than those considered in this systematic review, the proportion of women suitable for endovascular repair according to the manufacturer’s instructions for use is likely to reach only 40%.

Summary

Overall, 34% of women with clinically relevant an AAA (usually ≥ 5.5 cm in diameter) were not offered an elective repair of their intact aneurysm, after consideration at a multidisciplinary team meeting. The non-intervention proportion in women is twice as high as in men.

Summary

Overall, the 30-day operative mortality for intact AAAs in women is almost twice as high as in men. For EVAR, the pooled operative mortality in women was 2.2% and 5.4% for open repair.

Non-intervention rates for emergency repair of ruptured abdominal aortic aneurysms in women

Earlier work by Anjum and Powell81 reported that up to 75% of women with a ruptured AAA did not receive an emergency repair, but the source data from English Hospital Episode Statistics (HES) did not permit full differentiation between patients who did and did not reach hospital alive. Later work used stratified matching by age and sex to compare non-corrective in-hospital treatment rates in the USA and England for the years 2005–2010. 82 The English data showed that non-corrective treatment was selected for about 40% women aged < 75 years, rising to over 80% for women aged ≥ 85 years; in the USA, the comparable rates were about 20% and 60%, respectively. In England, the rate of non-corrective treatment for men was far lower, about 15% for those aged < 75 years and about 40% for those aged ≥ 85 years. Data from northern Norway83 show that, for the period 2010–14, 42% of women underwent non-corrective treatment in hospital, compared with 17% of men.

These data can be supplemented by further data from the IMPROVE trial centres, where approximately 25% patients assessed by the vascular team were not considered to be candidates for emergency repair. The relative proportions of women and men considered not to be candidates for (or refusing) emergency repair were very different: 161 out of 255 (63%) and 107 out of 548 (20%) respectively. Those not considered for repair were older than those who underwent repair.

Operative mortality after emergency repair for rupture in women

There are few sources of information for cohorts including more than 100 women. A summary of the main sources (published and unpublished since 2000, often with < 100 patients) is given in Table 10. The rates in all sources are likely to depend on the proportion of patients turned down for repair, information that is rarely provided (see Non-intervention rates for emergency repair of ruptured abdominal aortic aneurysm in women). Lower mortality is likely to be reported when the proportion of patients receiving an intervention is lower (selection of the best surgical candidates).

These data all suggest that 30-day operative mortality is higher after open repair than EVAR. This also was identified in an earlier large cohort Medicare study17 (1995–2006), in which 30-day mortality in women after EVAR was 41% versus 53% after open repair. Similarly Vascunet (international registry collaboration) shows lower in-hospital mortality in women after EVAR versus open repair, 36% and 44%, respectively (Professor Maarit Venermo, University of Helsinki, 2016, personal communication).

Late mortality after emergency repair of ruptured abdominal aortic aneurysms in women

There is a paucity of information regarding survival beyond 30 days for women who have undergone repair of a ruptured AAA. Several publications in the endovascular era since 2000 have assessed mid-term survival between 1 and 5 years following the repair of a ruptured AAA. Although the results are not separated by sex, in the multivariate analyses the odds or hazard ratios reported for women range from 1.1 to 1.4, indicating a higher mortality overall in women. Further details, with a breakdown of results by sex, were requested for two of these cohorts with a strong population base, the Amsterdam region, the Netherlands87 and New England, USA (VSGNE),88 but were not available. This leaves the only data to 1 year and beyond, as shown in Table 10, with a total of only 245 women. Table 10 also shows unpublished 2- and 3-year survival data from HES and the data for the IMPROVE trial. 30

The only published longer-term follow-up comes from the earlier large Medicare data set from 1995 to 2006 with 48,865 participants (23.5% women). Among women, the 5-year survival following EVAR was 32% (95% CI 25% to 39%), compared with 19% (95% CI 18% to 21%) following open repair.

Summary

The turn-down rate or non-corrective treatment rate of women with ruptured AAAs is not usually reported but is likely to be at least 40% and rises with age. Recent data indicate that, overall, the 30-day mortality in women is about 40% but probably is lower after EVAR compared with open repair. Although longer-term data for women are scant, overall about half the patients who undergo repair are alive at 3 years. The differential mortality between EVAR and open repair observed at 30 days appears to be preserved at 1 year and data from HES suggest that this difference is maintained at 2 years, with possibly some attenuation by 3 years.

Reinvitation

No information on the proportion of women who would be reinvited to AAA screening following non-attendance could be found. Therefore, the reinvitation rate is based on data in men from NAAASP. Table 33, Appendix 4, shows the numbers invited and attending screening in the 2013/14 and 2014/15 cohorts in NAAASP. The total attendance rate was 242,674/300,667 (80.7%) in 2013/14 and 236,936/293,709 (80.7%) in 2014/15. The proportion reinvited, which affects the overall costs of the screening programme, is calculated as the number who did not attend the first appointment minus the number who declined screening, all divided by the total number offered an appointment. The proportion who are reinvited stays at a constant 23.9% across the 2 years. This is higher than the 13.6% reinvited in the MASS trial,10 which was used as the basis for the health economic evaluation in men. Uncertainty in this figure for the PSA is very low if the number of reinvitations and total number are used directly as parameters of a Beta distribution (see Table 32, Appendix 4).

In the base-case analysis, a reinvitation rate of 23.9% is used based on NAAASP data for men from 2013/14 and 2014/15.

Attendance rate

The attendance rate is an important consideration and will influence the cost-effectiveness of any AAA screening programme. Evidence regarding the potential attendance rate for screening programmes involving both men and women is summarised here.

Non-visualisation

There is no information on the proportion of women whose aortic diameter would be non-visualised at a screening session, and so this information has also been obtained from NAAASP (see Table 34, Appendix 4). Based on NAAASP data for men from 2013/14 and 2014/15 combined, the non-visualisation proportion (after attempts on two separate occasions) is very low, at 0.35%, considerably lower than the 1.21% reported in the MASS trial. 10

A non-visualisation rate of 0.35%, based on NAAASP data for men, is used in the base-case analysis.

Aortic diameter distribution

A crucial consideration when screening a population of women is the distribution of aortic diameters and, related to this, the prevalence of AAAs detected at screening (i.e. the proportion of diameters detected above the diagnosis threshold, e.g. diameter of ≥ 3.0 cm). We obtained data on the full aortic diameter distribution in women from two sources: (1) 5140 women from Uppsala and Dalarna, Sweden, aged 70 years who were screened using leading edge to leading edge diameter measurements between 2007 and 2009,14 and (2) 570 women from Viborg, Denmark, aged 67 years who were screened using outer-to-outer wall diameter measurements in 2015 (Professor Jes Lindholt, personal communication).

A third source of information comes from the first 700,000 men screened in NAAASP, using inner to inner wall diameter measurements. Although not directly relevant, this large source of information may still be useful if it is suitably reweighted (see Prevalence of abdominal aortic aneurysms) so that it has the same prevalence of AAAs as seen in women. Table 12a compares each data source in terms of the proportion of screened individuals by 0.5-cm categories of aortic diameter. These results show that the prevalence of AAAs ≥ 3.0 cm is higher in the population of men in NAAASP than in the population of women screened in Sweden and Denmark, as is the prevalence of aortic diameters ≥ 2.0 cm (18.3% in NAAASP vs. 8.4% and 14.0% in Sweden and Denmark, respectively). This latter size range is important as it is aortic diameters of this size that are allowed to grow and potentially rupture within the DES model (see Chapter 2, Modelling aortic growth and abdominal aortic aneurysm rupture). The mean aortic diameters are 1.79 cm and 1.66 cm in NAAASP men and Swedish women, respectively. The mean diameter could not be accurately calculated in the Danish women as aortic diameter was available only in 0.5 cm categories. In the Swedish study, the standard deviation (SD) of the distribution is 0.26 cm, resulting in an aortic diameter of 2.5 cm being 3.2 SDs above the mean (or 51% higher) and one of 3.0 cm being 5.2 SDs above the mean (or 81% higher). In comparison, in men (NAAASP), 2.5 cm is 2.0 SDs above the mean (40% higher) and 3.0 cm is 3.4 SDs above the mean (68% higher). Outer–outer diameter measurements are expected to be larger than leading edge–leading edge or inner–inner measurements, which makes it even more surprising that the prevalence of AAAs was 0% in the Viborg study.

Prevalence of abdominal aortic aneurysms

The estimates obtained from the systematic review of AAA prevalence in women (see Chapter 3, Current prevalence of screen-detected abdominal aortic aneurysms in women) are used to inform the proportion of the aortic diameter distribution, that is, ≥ 3.0 cm. The pooled prevalence estimate from this systematic review was found to be 0.74% (95% CI 0.53% to 1.03%) overall and 0.43% (95% CI 0.23% to 0.80%) in 60- to 69-year-old women. This is higher than the estimate found in either the Swedish or Danish studies, but lower than that seen in NAAASP. To use this information from the systematic review, each of the aortic distributions described is reweighted. This has the effect of shifting the distribution in order for the desired prevalence to be achieved. A linear reweighting approach is taken using the following algorithm.

Let p_old be the prevalence of AAAs calculated in the aortic diameter distribution being considered and p_new be the prevalence that we wish to recalibrate the distribution to (e.g. 0.43% for 60- to 69-year-old women). Each aortic diameter size x (accurate to 1 mm) has an associated probability weight w(x) indicating the proportion of individuals in the distribution who were screened with that diameter. The weights sum up to 1. It follows that:

Given the desired prevalence, p_new, calculate new weights, w*(x), as follows:

where f(x) = a + bx is a linear function of x. The conditions that must be satisfied are:

and

A pair of simultaneous equations can, therefore, be obtained to give the solutions:

and

After reweighting, some of the new weights may be negative. If this occurs, these are set to zero and then a further reweighting step is performed to ensure the weights above the diagnosis threshold (e.g. 3.0 cm) sum to the desired prevalence. Applying this algorithm to the aortic diameter distributions shown in Table 12a, using the estimated prevalence of 0.43% found from the systematic review for 60- to 69-year-old women (see Chapter 3, Current prevalence of screen-detected abdominal aortic aneurysms in women), gives the distributions shown in Table 12b. Note that this approach could not be used with the Viborg data as there were no cases of AAAs reported in this screening study (and, hence, the slope parameter b is infinite). It can be seen that the reweighted Swedish and NAAASP distributions are different, with a higher proportion of ≥ 2.0-cm aortic diameters in NAAASP. Furthermore, no AAAs of diameters of ≥ 5.5 cm were found in the Swedish study. This would have an important impact in the modelling, giving rise to no AAAs that are immediately referred for elective surgery.

Growth modelling

A mixed-effects model was used to model the longitudinal AAA diameter trajectories for each woman in each of the 11 studies. A model was fitted separately within each study assuming a linear relationship between log-AAA diameter and time since entry into the study (see Chapter 2, Modelling aortic growth and abdominal aortic aneurysm rupture, for the rationale to using this model). The model allowed a separate intercept and slope parameter for each individual through the use of random effects. For individual i with measurement j at time t_ij years after study entry, the AAA diameter y_ij (cm) is modelled as:

where (21)ϵij∼N(0,σw2), (22)β=(β1β0) and (23)G=(σ02ρσ0σ1ρσ0σ1σ12).

β₀ and β₁ represent the mean intercept and slope of the AAA diameter trajectories on the log-scale, respectively, while b_0i and b_1i allow for individual variation about the intercept and slope (random effects). The random effects for each patient are correlated and come from a bivariate normal distribution. Parameter estimates obtained from this model for each study are shown in Table 36, Appendix 4, and a forest plot for the average factor increase in AAA diameter per year (exp β₁) is shown in Figure 6. In a second stage, study-specific estimates are pooled via multivariate random-effects meta-analysis (overall estimates shown in Table 36, Appendix 4). On average, AAA diameter increases by 5% per year, but with considerable heterogeneity between both studies and people. The average and distribution of AAA growth rates in women and men are in fact similar. 35

FIGURE 6.

Mean proportionate increase in AAA diameter per year, from RESCAN35 studies.

As described in Chapter 2, Modelling aortic growth and abdominal aortic aneurysm rupture, the approach taken in the DES is first to sample baseline diameters from our chosen distribution (see Aortic diameter distribution) and then generate random effects for each individual conditional on their baseline diameter, using the parameter estimates obtained from the overall linear mixed-effects model.

Table 37, Appendix 4, shows the estimated proportion of individuals predicted to cross the intervention threshold (5.5 cm) within 5 and 10 years, in the absence of any deaths. For comparison, the empirical rates estimated in the 11 RESCAN35 studies for women in the absence of any competing risks (e.g. deaths) are also shown. The data from the 11 RESCAN35 studies are naively pooled to estimate the empirical rates. The predicted rates lie within the 95% CIs for the observed rates in all size/threshold categories.

Rupture rates

Rupture data were available in only 6 out of the 11 RESCAN studies35 that provided growth data (Edinburgh and Leeds did not record rupture information and the Propranolol, PIVOTAL and Swedish studies did not have rupture events in both men and women; see Table 35, Appendix 4). Characteristics of these six studies are shown in Table 38, Appendix 4.

A joint growth and rupture model was fitted to the data separately within each study before pooling estimates using multivariate random-effects meta-analysis. As ruptures were rare, we used data from both men and women and allowed for sex differences in the baseline AAA diameter and rate of rupture by including sex as a covariate in both the longitudinal (growth) and survival (rupture) submodels. As described in Growth modelling, a linear relationship between log (diameter) and time was assumed to model the growth of an aneurysm. The hazard of rupture was related to an individual’s current predicted (log) AAA diameter, m_i(t), and their sex as follows:

Pooled estimates obtained from the rupture submodel are shown in Table 39, Appendix 4, together with an estimate of between-study heterogeneity as given by the I² statistic. The association between the risk of rupture and AAA diameter is depicted in Figure 7 for women, predicted from each study-specific model and from the pooled estimates. For comparison purposes, the empirical observed rates of rupture by 0.5 cm categories are also shown. The pooled model trajectory gives a reasonable fit to the overall data. Study-specific estimates can be seen to vary considerably. The pooled rate of rupture reaches 1 per 100 person-years at a predicted diameter of 4.2 cm.

FIGURE 7.

Comparison of observed rates of rupture (with 95% CIs) and those predicted from study-specific models and a pooled model, from RESCAN studies. 35

Dropout

Data on the rate at which women drop out from regular surveillance were kindly made available from two AAA surveillance programmes that have recruited both men and women with incidentally detected AAAs: (1) the Leicester AAA surveillance programme (81 women and 353 men between September 2004 and September 2015; Professor Matthew J Bown, personal communication) and (2) the Imperial College AAA surveillance programme (28 women and 97 men recruited in 2010; Professor Janet T Powell, personal communication). Dropout information from Leicester was available from January 2014 onwards and, hence, only individuals still in the surveillance programme from 2014 onwards were considered (n = 389). Dropout was defined as any of the following reasons for leaving surveillance: (1) discharged, (2) patient cancelled, (3) moved location, (4) referral for other surgery and (5) other. For the Imperial College AAA surveillance programme, information on dropouts was collected from 2010 to 2015 and included the following reasons: (1) did not attend and (2) moved away. Only year of scan was recorded and, hence, follow-up time was an integer defining year of last scan minus year of first scan. In this study, 30 individuals had only one scan recorded and were excluded from the analysis. Table 13 shows the rate of dropout for women and men from the two screening programmes along with the estimated hazard ratio between men and women from a Cox regression model. There was no evidence from either of the screening programmes of a differential dropout rate between women and men.

A further source of data on dropout rates in men is NAAASP. NAAASP includes 13,271 men who were under surveillance (11,136 screen detected, 2135 self-referrals) up to 4 April 2016. Follow-up was defined as the date of first scan to the date of last scan or status update date, whichever came later. Of these individuals, 2537 in whom only one scan was recorded were excluded from these analyses. Dropout was defined as any of the following reasons for leaving the programme: (1) appointment missed, (2) declined, (3) non-visualised, (4) out of cohort, (5) surveillance ceased or (6) temporarily ineligible. These additional data are also summarised in Table 13. The dropout rate was estimated to be 5.5 per 100 person-years. A sensitivity analysis including the 2537 individuals with only one scan (giving them a very small follow-up time) gave a very similar estimated rate (5.6 per 100 person-years). There was also little evidence of a difference in dropout rates between self-referred and screen-detected individuals (p = 0.081).

Incidental detection

Data from electronic hospital records of women aged ≥ 65 years undergoing CT scanning were obtained from the University Hospital of South Manchester in 2014; 2494 women underwent an abdominal CT during this period and 65 AAAs were identified. Of these, 53 were newly identified AAAs, but only seven were referred on to vascular surgeons to be followed up with surveillance or elective surgery. The population (women aged ≥ 65 years) of the referral catchment area for the university hospital is approximately 24,500. Assuming that 181 (0.74%) of these women have an aneurysm (see Chapter 3, Current prevalence of screen-detected abdominal aortic aneurysms in women), this would indicate an incidental detection rate to a surveillance programme of approximately 7/181 = 3.9 per 100 person-years for women aged ≥ 65 years with an AAA. This is similar to the rate of 4.6 per 100 person-years used in the most recent health economic model for men. 12

Further data come from a study conducted in Canterbury, New Zealand,36 in which 167 new incidental AAAs were detected in men and women from CT scans over a period of 4.25 years. About one-quarter of all detected AAAs (incidental and known) were in women. Assuming this proportion also applies to the incidental AAAs and that 97% of AAAs were in individuals aged ≥ 65 years, there would be approximately 40 AAAs detected in women aged ≥ 65 years. From census data, the 2006 population of women aged ≥ 65 years for the catchment area (Canterbury, West Coast and Timaru regions of South Island, New Zealand) was approximately 43,500. Assuming that 321 (0.74%) of these women had an aneurysm (see Chapter 3, Current prevalence of screen-detected abdominal aortic aneurysm in women), this would indicate an incidental detection rate of approximately 40/(321 × 4.25) = 2.93 per 100 person-years for women aged ≥ 65 years with an AAA. This is also quite similar to the rate of 4.6 per 100 person-years used in the most recent health economic model for men. 12

Delay from ≥ 5.5-cm scan to consultation

Data from NAAASP for the years 2013/14 and 2014/15 indicate that 981 men in total were referred to vascular services, of whom 947 (97%) received a consultation. The mean time from referral scan to consultation was 10.6 days, much lower than the mean delay of 71 days observed in the MASS trial. 10

A time delay from referral screen to consultation of 10.6 days based on NAAASP is used in the modelling.

Consultation scan: computerised tomography scan versus ultrasound scan

At consultation, an AAA is confirmed (or otherwise) using a CT scan. The measurement of the AAA diameter made with this CT scan may be systematically higher than that seen on an ultrasound scan. Evidence for this comes from the RESCAN collaboration35 in which four studies measured diameters using both ultrasound and CT scans. CT measurements were, on average, significantly larger than ultrasound measurements {Leeds 3.91 mm [standard error (SE) 0.33 mm], PIVOTAL 1.75 mm (SE 0.21 mm), Galdakao 1.77 mm (SE 0.10 mm), Stirling 2.46 mm (SE 0.27 mm)}. A pooled estimate from these studies suggests an average increase of 2.44 mm.

There is also evidence that the measurement error for a CT scan may be different from that based on an ultrasound scan. A paper by Singh et al. 95 suggests that a CT measurement of AAA diameters has interobserver ‘variability’ of 5.2 mm (defined as 1.96 multiplied by the SD of interobserver differences). This equates to a CT measurement error SD of 1.9 mm. This is lower than the estimated ultrasound measurement error SD from the RESCAN35 model for a large AAA, approximately 55exp(–2.96) = 2.9 mm for a 5.5-cm aneurysm (see Table 36, Appendix 4).

In the modelling, the mean observed diameter from a CT scan is assumed to be 2.44 mm higher than that obtained from an ultrasound scan, with a measurement error SE of 1.9 mm.

Decision at consultation: proportion returned to surveillance

The DES programmed for men (see Chapter 2) used the observed CT scan diameter at consultation to decide whether or not the individual should be returned to surveillance, with those with AAAs of measured diameter < 5.5 cm returned to surveillance. Based on the CT measurements, 13.7% of consultations resulted in an individual being returned to surveillance. This is a much higher rate than the 36 out of 947 men (3.8%) who were ‘inappropriate referrals’ (AAA diameter of < 5.5 cm, other or not stated) in NAAASP data. Nevertheless, in our modelling, the proportion of women who are returned to surveillance after a consultation is derived from the proportion of CT measurements that are < 5.5 cm.

Decision at consultation: non-intervention rate in women not returned to surveillance

Women may refuse surgical intervention, or may be turned down because of contraindications. Information on the non-intervention rate in women not returned to surveillance (i.e. the proportion turned down for elective surgery or refusing an operation) comes from four hospitals in the UK (see Chapter 3, Proportion of women versus men not offered an intervention). The overall non-intervention rate is 34% (95% CI 29% to 40%), with no between-study heterogeneity (I² = 0%).

The proportion of individuals elective surgery in whom surgery is deemed to be contraindicated is based on the pooled estimate from four UK hospitals of 0.3423. The PSA is based on a normal(–0.653, 0.135²) distribution for the logit pooled probability.

Decision at consultation: proportion who will receive elective surgery

The proportion of women who receive elective surgery is defined in the model based on the remaining population who are not turned down, refuse surgery or are returned to surveillance.

Delay from consultation to elective surgery

Among 827 individuals in whom surgery took place in NAAASP, the mean time from referral to surgery was 81.4 days. Assuming that the mean time from consultation to referral was 10.6 days, this would imply a mean time from consultation to surgery of 70.8 days, slightly higher than the mean delay of 59 days observed in the MASS trial. 10 In our modelling, a time delay from consultation to surgery of 70.8 days is used for everyone for whom surgery is planned, based on NAAASP data.

National Vascular Registry

The submission of data to the NVR by vascular units is voluntary, but it is generally thought to be about 90% complete. 26 Data are entered into the NVR by surgeons at the time of surgery and/or at the time of discharge from hospital. The registry covers all types of vascular surgery, including elective and emergency AAA repairs. Under a data sharing agreement with the Healthcare Quality Improvement Partnership, individual-level data were obtained for all AAA repairs reported to the NVR from 1 January 2010 to 31 December 2014. The initial year was set at 2010 to focus on recent practice and because this is when EVAR became reliably recorded in the NVR. NVR provides data on AAA size and in-hospital mortality (rather than 30-day mortality). For men, incidentally detected and screen-detected AAAs are sometimes (but not always) distinguished; for women, it is assumed that all AAAs have been incidentally detected as no systematic screening was in place in the UK during the period covered by the data extract.

Hospital Episode Statistics

Summary tabular data were made available from HES28 for the same time period as the NVR data extract (1 January 2010 to 31 December 2014). To comply with confidentiality requirements, cells in the tables with values of 5 or below were either merged with neighbouring categories or supressed. Identifying operations as AAA repairs is more difficult in HES than the NVR, as one admission may generate multiple hospital episodes each recorded separately in HES. Data on both 30-day and in-hospital mortality can be extracted from HES.

Vascunet

Vascunet is an international register of vascular surgical procedures, and includes data principally from mainland Europe but also some from the UK and Australasia. 27 Submission of data to this register is performed much less routinely than for the NVR, and Vascunet should be regarded as far from complete. Nevertheless, it provides an interesting comparator as it includes data from outside the UK. Summarised tabular data from Vascunet were obtained for the years 2010–13.

Use of National Vascular Registry data as a source of parameter estimates in women

As the NVR provides the most detailed data, it is the principal source of evidence we use for surgical parameters for women. This section describes the principles employed.

The NVR allows modelling of individual data with respect to sex (men vs. women), age and AAA diameter. Most of the information is in the form of proportions (p), for example, the proportion of patients receiving EVAR, or the proportion of patients undergoing EVAR who die in hospital. We use logistic regression models, including all possible sex interactions, to provide estimates for women in two ways, as described here. The SEs from these regressions are used to represent parameter uncertainty in the PSA via correlated normal distributions. The data in the NVR on whether the AAA was detected by screening or incidentally are available only for men (as there was no screening for women during this time period) and, even for men, this information is only about 35% complete. As we are primarily concerned with estimates for women, we do not include this variable in the logistic regression models.

First, we consider just the overall proportion for women derived from the simple logistic regression:

where sex is coded as ‘0’ for women and ‘1’ for men. The parameter a_F is the log-odds for women and a_M is the log-odds ratio comparing men with women. We use expit(a_F) = exp(a_F)/[1 + exp(a_F)] as the estimated probability for women.

In a second analysis, we use the more detailed logistic regression:

Here, b_1F is the change in the log-odds per year of age and b_1M is the difference in this log-odds between men and women. Similarly, b_2F is the change in log-odds per cm increase in AAA diameter and b_2M is the difference in this log-odds between men and women. Subtracting the values of age (80 years) and AAA diameter (6.0 cm) reduces the correlations between parameter estimates; the intercept a_F now refers to a woman aged 80 years with an AAA diameter of 6.0 cm. For example, we use Equation 27 to estimate the relevant proportion for a woman aged 71.2 years with an AAA of diameter 6.2 cm, as expit (a_F – 8.8 × b_1F + 0.2 × b_2F).

In this second analysis, we include all the terms in the logistic regression (Equation 27), whether or not they are statistically significant. We use linear terms for age and AAA diameter. We do not model trends according to calendar time, as the purpose here is to use relevant recent evidence; any extrapolation to the future would likely be very unreliable. One slight disadvantage of the second model is that the few patients with missing values of age or AAA diameter in the NVR have to be omitted.

Proportion receiving endovascular aneurysm repair for elective abdominal aortic aneurysm surgery

Open AAA repair and EVAR have different immediate mortality rates and different subsequent rates of reinterventions and AAA-related mortality,96 so the proportion of women receiving each type of operation needs to be estimated.

Data for women from the NVR and HES are shown in Table 14a. The NVR reports fewer operations than HES for the same period (around 80% of the HES total), reflecting under-reporting in the NVR. The overall proportion of EVAR operations is 58.6% in the NVR and very similar to the HES at 60.6%. These overall rates conceal strong trends: the use of EVAR increases with age and decreases with AAA diameter (Figure 8).

FIGURE 8.

Proportion of women receiving elective EVAR in the NVR (with 95% CIs). (a) By age, and a superimposed logistic regression fit (for women with an AAA of diameter 6.2 cm, the mean in the NVR data); and (b) by AAA diameter, and a superimposed logistic regression fit (for women aged 76.8 years, the mean in the NVR data).

For comparison, the overall use of EVAR in men in the NVR was 63.5% and there were similar trends according to age and aneurysm diameter as in women. 97 Data from Vascunet give the overall proportion of women receiving EVAR as 56.1%, similar to the figures from the NVR and HES (see Table 14a).

We use the data from the NVR as described in Use of NVR data as a source of parameter estimates in women to provide estimates of the proportion of women receiving EVAR in the model (see Table 40, Appendix 5). In the base-case analysis, the overall proportion is simply that observed in the NVR (i.e. 58.6%).

Proportion who are morphologically suitable to receive endovascular aneurysm repair for elective abdominal aortic aneurysm surgery

In a sensitivity analysis, we instead use the proportion of women whose AAA was considered morphologically suitable for elective EVAR. This may be different from the proportion of women who receive EVAR in practice (i.e. in the NVR or HES). An estimate of this is provided by the systematic review of the literature described in Chapter 3, Suitability of women versus men for standard endovascular repair.

Across the five studies included, the pooled estimate of suitability for EVAR according to the manufacturers’ instructions for use (IFU) was 34% (95% CI 25% to 44%). This is substantially less than the NVR estimate of 59% for the proportion of women receiving EVAR. Assuming that only those within IFU receive EVAR, the proportion of 34% could be used in place of 59%, although the consequent effects on postoperative mortality for both EVAR and open repair are unknown. More recent data, which could not be included in the systematic review (see Chapter 3, Suitability of women versus men for standard endovascular repair) but assess the use of newer endografts, suggest that 40% of women are eligible for EVAR within the IFU.

Elective endovascular aneurysm repair operative mortality

Overall in-hospital or 30-day postoperative mortality rates from the NVR and HES in women undergoing elective EVAR are shown in Table 14b. There were too few deaths to show any convincing trends according to age or AAA diameter in women. The overall figures for in-hospital mortality from the NVR and HES are very similar, 1.8% and 1.6%, respectively. From the HES data, 30-day mortality is somewhat greater than in-hospital mortality (2.3% vs. 1.6%).

For comparison, the overall in-hospital mortality in the NVR for men was 0.7%, lower than in women, with evidence of increasing mortality with age. 97 Data from Vascunet give an overall value of 2.0% for 30-day mortality in women, slightly lower than the figure of 2.3% from HES (see Table 14b). In the systematic literature review (see Chapter 3, Thirty-day operative mortality in women versus men), the overall estimate of 30-day mortality for women after elective EVAR was 2.2% (95% CI 1.9% to 2.7%), similar to the figure from HES.

For the modelling, we adjust the NVR in-hospital mortality to reflect the (greater) 30-day mortality. Thus, we use the NVR data to estimate the log-odds of in-hospital mortality (see Table 40, Appendix 5) according to Equation 26 or 27 in Use of National Vascular Registry data as a source of parameter estimates in women, but then add the log-odds ratio corresponding to the 30-day mortality compared with the in-hospital mortality in HES (namely log-odds of 2.3% vs. 1.6% = 0.370) before transforming back to the probability scale. Working on the log-odds scale ensures that probabilities cannot exceed 1. For the base-case analysis, this gives an overall 30-day mortality estimate of 2.4%. In the PSA, we ignore the fact that the difference between 30-day and in-hospital mortality from HES is estimated with error.

Elective open abdominal aortic aneurysm repair operative mortality

Overall, postoperative mortality rates for women following elective open AAA repair are shown in Table 14c. In-hospital mortality rates are much higher than those after elective EVAR: 6.9% and 6.0% in the NVR and HES, respectively. In HES, as for elective EVAR, the 30-day mortality rate is slightly higher than the in-hospital mortality rate, 7.0% vs. 6.0%. The in-hospital mortality rate in the NVR increased with age, but not convincingly with AAA diameter (Figure 9).

FIGURE 9.

In-hospital mortality rate in women receiving elective open AAA repair in the NVR (with 95% CIs). (a) By age, and a superimposed logistic regression fit (for women with an AAA of diameter 6.2 cm, the mean in the NVR data); and (b) by AAA diameter, and a superimposed logistic regression fit (for women with AAA at age 76.8 years, the mean in the NVR data).

Compared with women, the overall in-hospital mortality was lower in men: 4.0% and 3.8% in the NVR and HES, respectively. 97 The Vascunet data show an overall 30-day mortality rate of 6.6% for women, again quite similar to HES (see Table 14c). In the systematic literature review (see Chapter 3, Thirty-day operative mortality in women versus men), the overall estimate of 30-day mortality for women after elective open repair was 5.4% (95% CI 4.2% to 6.9%), somewhat lower than the figure from HES.

To estimate the 30-day mortality rate for women (see Table 40, Appendix 5), we follow the procedures in Use of National Vascular Registry data as a source of parameter estimates in women along with a similar conversion from in-hospital to 30-day mortality (as described at the end of Elective endovascular aneurysm repair operative mortality). For the base-case analysis, this gives an overall 30-day mortality estimate of 8.1%.

Reintervention rate after successful elective surgery

A ‘successful operation’ is taken to mean that the patient is alive 30 days after the operation. The NVR does not provide information on reintervention rates after the initial hospitalisation for AAA repair. In principle, such reinterventions can be extracted from HES data, but the correct linking of subsequent hospitalisations for individuals that are related to the initial AAA repair (as opposed to other related or unrelated conditions) is fraught with difficulty. Moreover, the length of follow-up available in the 2010–14 HES data is limited to a maximum of 5 years. Thus, we base our estimates on the long-term follow-up (up to 15 years) of the EVAR-1 trial38 of 1252 patients with a large AAA (diameter of ≥ 5.5 cm) randomised to either open AAA repair or EVAR.

There are some drawbacks of the EVAR-1 trial38 data for our purpose. The first is that about 90% of the patients in the trial were men. The second is that the trial patients were restricted to those deemed both fit for open repair and anatomically suitable for EVAR, whereas the groups receiving open repair or EVAR in practice include additional patients. Furthermore, rather than analysing the trial by randomised group from the date of randomisation, we present the data by operation received from the date of operation, omitting patients who did not receive an operation; this makes only a slight difference for the EVAR-1 trial38 since 93% of patients received their randomly allocated surgical intervention, and the median delay between randomisation and surgery was only 40 days. Patients in whom EVAR was converted to open AAA repair in the initial admission are classified as open repairs. We include all reinterventions, excluding the first 30 days following the operation, whether they are first or subsequent ones, and express them as a rate per 100 person-years. We also note that there was strict adherence to the IFU in the EVAR-1 trial,38 and for both EVAR and open repairs reintervention rates rise where morphology is outside the IFU.

Reinterventions are taken to include the following AAA-related conditions: added stent, staple or ligation, type I–III endoleaks, embolisation of endoleak, sclerosis, conversion to open repair, aneurysmal extension above or below original graft, thrombosis of graft limb, graft infection, incisional hernia, false femoral aneurysm, fem-fem graft, FEVAR, axillo bi-fem, distal limb procedure/revascularisation, reoperation of open repair, replacement stent graft and amputation. Reinterventions for laparotomy-related complications were not initially included in the EVAR-1 trial. 38

Based on these definitions, the analysis is based on 1172 patients (1065 men and 107 women) who survived 30 days after their operation, rather than the 1252 originally randomised in EVAR-1. 38 The number of reinterventions occurring in these patients was 262, over a period of up to 15 years, constituting 9321 person-years of observation. The rate of these reinterventions over time is depicted in Figure 25, Appendix 5. The rate of reinterventions is much higher after EVAR than after open AAA repair. The rates can be adequately represented by exponential distributions (constant hazard over time) within periods of 31–120 days and > 120 days after the operation; Weibull distributions did not provide a better fit to the data.

There was substantial evidence that the rates of reintervention differed between women and men, and that this sex effect differed for EVAR and open AAA repair (see Figure 26, Appendix 5); the p-value for including the main effect of sex and its interaction with operation type was 0.006. Reintervention rates were higher in women than in men after EVAR, but lower in women than in men after open AAA repair. Thus, we use the reintervention rates for women alone as parameters in our modelling (see Table 15). For example, the rate of reinterventions for days 31–120 after successful EVAR is estimated as 3 per 15 woman-years or 20.3 per 100 woman-years. For the PSA, we use a Gamma(3,15) distribution to reflect the number of reinterventions and woman-years in the EVAR-1 trial data. 38 There were no reinterventions after 30 days after open AAA repair in 388 woman-years of observation (Table 15); we combine these two periods after 30 days, and apply a zero rate in the base-case analysis. In a sensitivity analysis, we use data from the DREAM76 and OVER77 trials in men to estimate an alternative reintervention rate.

Long-term abdominal aortic aneurysm-related mortality rate after successful elective repair

For similar reasons as for reinterventions, we use the long-term EVAR-1 trial data38 in preference to HES–ONS data. The latter has limited follow-up available and it is doubtful that AAA-related mortality can be reliably defined based on death certification. For example, many of the deaths occurring within 30 days of an AAA operation are not categorised as AAA related in the HES–ONS data set. We also note that data to 14 years for all-cause mortality from the DREAM trial76 are similar to those from the EVAR-1 trial. 38

Following the same principles as for reinterventions, the rates of AAA-related deaths in the EVAR-1 trial38 after successful AAA repair are shown in Figure 27, Appendix 5; AAA-related deaths include all those within 30 days of any AAA surgery. For AAA-related deaths occurring > 30 days after operation, an exponential model fit to the data was reasonable. There was strong evidence of an increased hazard for females (hazard ratio 2.72, 95% CI 1.35 to 5.46; p = 0.005). Therefore, we use AAA-related mortality rates for women alone in our modelling (see Table 15). For example, the AAA-related mortality after successful EVAR is estimated as 8 per 444.7 woman-years, or 1.8 per 100 woman-years. For the PSA, we use a Gamma(8,444.7) distribution to reflect the number of deaths and woman-years in the EVAR-1 trial data. 38

Proportion operated on after an abdominal aortic aneurysm rupture

Many patients with an AAA rupture die before getting to hospital or the operating theatre, are turned down or refuse AAA repair. So, the proportion of patients with an AAA rupture receiving an operation is an important parameter that crucially influences the survival rate after an AAA rupture. However, estimates of this parameter for women are not easy to obtain.

We have taken data from the literature and from recruitment to the randomised trials of EVAR versus open repair for ruptured AAAs (such as the IMPROVE trial30) to provide relevant estimates (see Chapter 3, Mortality following ruptured abdominal aortic aneurysm in women). The conclusion is the overall proportion of women with a ruptured AAA who receive an emergency repair is low, at around 25%. We use this figure in our modelling, but allow considerable uncertainty (95% uncertainty interval 15–35%) in the PSA.

Proportion receiving endovascular aneurysm repair for an abdominal aortic aneurysm rupture

The numbers of women in the NVR and HES receiving open repair or EVAR for ruptured AAAs are shown in Table 16a. The proportion of operations identified in HES that are also reported in the NVR is lower than for elective operations (around 70%). This may underlie the more substantial difference in the reported proportions receiving EVAR: 16.8% in the NVR compared with 22.4% in HES. There is an increasing rate of EVAR use with age (Figure 10a).

FIGURE 10.

Emergency surgery data in the NVR. (a) Proportion of women receiving EVAR for an AAA rupture by age in the NVR, and a superimposed logistic regression fit; and (b) in-hospital mortality rate in women receiving open repair for an AAA rupture by age in the NVR, and a superimposed logistic regression fit.

For comparison, the overall proportion of men receiving EVAR for emergency operations in the NVR was 19.4%;97 an increasing trend with age was again evident. Data from Vascunet give the overall proportion of women receiving EVAR as 23.5% (see Table 16a), but do not distinguish urgent and emergency cases.

In our analysis, we use the data from the NVR, first, because of the potential coding problems with HES and, second, because individual-level data were available for incorporating the influence of age. We use the same methods as before, described in Use of National Vascular Registry data as a source of parameter estimates in women, but ignore any effect of AAA diameter. Parameter estimates from the logistic regressions are given in Table 41, Appendix 5; in the base-case analysis, the overall proportion for women is simply that observed in the NVR (i.e. 16.8%).

Emergency endovascular aneurysm repair operative mortality

The in-hospital and 30-day mortality rates from the NVR and HES are shown in Table 16b. The overall in-hospital mortality rate from the NVR is 25.0%, substantially greater than the 12.7% reported in HES. This may reflect mortality events being missed in HES when patients are transferred to another hospital (e.g. for rehabilitation or long-term nursing care) and subsequently dying. The transfer of care results in the end of a HES episode. There were increasing mortality rates with increasing age. As for elective operations, the 30-day mortality rate determined by HES–ONS linked data is greater than the in-hospital mortality rate (19.7% vs. 12.7%). Vascunet provided very incomplete data for 30-day mortality after urgent and emergency operations (which are combined in Vascunet), in contrast to the data for 30-day mortality after elective operations shown earlier. So we report the Vascunet in-hospital mortality data, which are more complete, giving a figure of 20.9% after EVAR (see Table 16b). The literature review (see Chapter 3, Mortality following ruptured abdominal aortic aneurysms in women) suggested a 30-day mortality rate of 32%. Some of the differences may relate to how EVAR converted to open repair (which has very high mortality) and is categorised in the different studies and to the differential use of anaesthesia types in different countries.

For comparison, the in-hospital mortality rate for men in the NVR is 20.7%,97 slightly lower than the 25.0% for women.

In our modelling for women, again we use the NVR data for in-hospital mortality and make an adjustment to reflect 30-day mortality based on HES (as in Elective endovascular aneurysm repair operative mortality). For the base-case analysis, this yields an overall 30-day mortality rate of 35.9%. Parameter estimates from the logistic regressions are given in Table 41, Appendix 5.

Emergency open abdominal aortic aneurysm repair operative mortality

The corresponding mortality rates for women after open AAA repair are shown in Table 16c. The overall in-hospital mortality rate is 39.8% in the NVR, compared with 33.6% in HES. An increasing mortality rate with age was again evident (see Figure 10b). Again, in HES–ONS, the 30-day mortality was greater than the in-hospital mortality (37.8% vs. 33.6%, respectively). Vascunet data give the in-hospital mortality rate as 34.3%, which is similar to HES, but does not distinguish emergency and urgent cases. The literature review (see Chapter 3, Thirty-day operative mortality in women versus men) suggested a 30-day mortality rate of 51%.

For comparison, the in-hospital mortality rate for men in the NVR is 36.9%,97 slightly lower than the 39.8% for women.

We use the same methods as before to provide estimates from the logistic regressions in Table 41, Appendix 5. For the base-case analysis, this yields an overall 30-day mortality rate of 44.2%.

Reintervention rates after successful emergency surgery

Again, a ‘successful operation’ is taken to mean that the patient is alive 30 days after the operation. Obtaining information on reintervention rates after emergency AAA operations, especially for women, is difficult. For similar reasons as for elective operations, we based estimates on the IMPROVE trial,30 the largest and longest trial of EVAR versus open repair for ruptured AAAs. The trial randomised patients with a ruptured AAA either to a policy of EVAR if possible, compared with open repair. We have access to the provisional unpublished 3-year follow-up data from the IMPROVE trial,30 and are able to report reinterventions by operation received, for the group of patients with confirmed ruptured AAAs. The time between randomisation and operation (if received) in IMPROVE30 is very short (median 0.7 hours), so we use time since randomisation as the time scale.

The IMPROVE trial data30 were limited by the fact that only about 20% of the patients were women, and that the available follow-up extends only to 3 years. In the trial, about 50% of the ruptured AAA patients had died within 3 years. Because of the available follow-up, we consider the period after 30 days after the operation as one period, include all reinterventions, whether they are first or subsequent ones, and express this as a rate per 100 person-years. The reinterventions included are as listed in Reintervention rate in successful elective surgery.

The reinterventions data from the IMPROVE trial30 are summarised in the upper two rows of Table 17. Because there are possible differences in rates between men and women, we use the data for women alone despite the small numbers. Thus, the rate of reinterventions after 30 days after EVAR is estimated as 15.8 per 100 woman-years and correspondingly after open repair as 2.3 per 100 woman-years.

Long-term abdominal aortic aneurysm-related mortality rate after successful emergency surgery

For similar reasons as before (see Long-term abdominal aortic aneurysm-related mortality rate after successful elective repair), we use the IMPROVE trial30 data to estimate the long-term AAA-related mortality rates after emergency surgery in women (lower two rows of Table 17). For the period after 30 days after emergency EVAR, the rate is estimated as 0; correspondingly, after emergency open repair, the rate is estimated as 1.2 per 100 woman-years.

Screening costs

Screening costs were taken from NAAASP (Professor Jonothan Earnshaw, Gloucestershire Hospitals NHS Foundation Trust, 2012, personal communication) and updated to reflect general health service inflation to 2014/15 prices. 101 The cost of screening women was assumed to be the same as in the programme for men (Table 18).

Pre-surgical consultation costs

In previous modelling, the cost of a pre-surgical consultation was based on data from the MASS trial98 and subsequently uprated for general NHS inflation. 35 This estimate was from data collected from a subsample of the full trial population. On average, 1.6 consultations were conducted before elective surgery. Unit costs came from the finance departments of centres involved in the trial. In the current modelling, the number of consultations was assumed to be the same as observed in the MASS trial,98 but the unit cost was updated using contemporary estimates from the Department of Health and Social Care’s NHS Reference Costs 2014 to 2015. 39 The unit cost comprised a weighted mean of face-to-face consultant-led outpatient visits for vascular surgery, cardiothoracic surgery and cardiac surgery specialties. The new estimate of the cost of pre-surgical consultations was £328.64, rather than £435.25 as used before. 35

Costs of elective and emergency abdominal aortic aneurysm repair

A similar approach was adopted to estimate contemporary costs for women undergoing surgical AAA repair, given the infeasibility of conducting detailed microcosting. Cost estimates were taken from UK-based randomised trials (EVAR-138 and IMPROVE24) focusing on women-specific data. These were updated using registries to provide robust data on the general AAA repair population and reflect potential trends in hospital LOS. LOS data were available from HES28 and the NVR26,97 between 2010 and 2014. Hospital stay constitutes the largest component of resource use, with significant differences between men and women, and includes that incurred by renal dialysis in the primary admission. Unlike previous modelling, which had limited emergency surgery to only open repair, an estimate of the cost of emergency EVAR was required, given evidence of its increased use in this setting from both the NVR and HES.

For elective repair costs, the EVAR-1 trial38 was again utilised. EVAR-1 recruited patients between 1999 and 2004 in 38 UK centres. 38 Women-specific elective AAA repair costs were obtained from the EVAR-1 trial38 investigators in 2014/15 prices and updated using LOS data from HES. These costs related to the primary admission. HES LOS data were preferred to the NVR as HES is a more complete database of AAA repairs, and additional analysis using more accurate coding was possible.

The components of total cost comprising mean vascular ward and critical care stay were removed and replaced with women-specific mean LOS data from HES, multiplied by unit costs obtained from NHS Reference Costs 2014 to 2015. 39 Elective AAA repair LOS observed in HES was significantly lower than in the EVAR-1 trial38 for EVAR repair, particularly the general vascular ward stay; however, open repair hospital LOS was similar. Updated costs are shown in Tables 19a and b.

Emergency AAA repair cost data for women were obtained from the IMPROVE trial30 investigators in 2011/12 prices for the primary admission. The trial recruited patients between 2009 and 2013 in 29 UK centres and one Canadian centre. Data provided were restricted to those patients with a confirmed AAA rupture and according to treatment received rather than randomised group. In the trial, an ‘EVAR where possible’ strategy was adopted and analysed by intention to treat, so that patients in that group did not always receive EVAR.

Components of total cost comprising general vascular ward and critical care stay were removed and the remaining costs were inflated for general NHS inflation to 2014/15 prices using published indices. 101 Women-specific HES data were used to update LOS and were multiplied by unit costs from NHS Reference Costs 2014 to 2015. 39 Critical care LOS observed in HES was lower for both EVAR and open repair, although mean stay on vascular ward was higher. Updated costs are shown in Table 19c and d.

For comparison, HES data on AAA repair for men were utilised to update costs from the EVAR-138 and IMPROVE24 trials using the same approach. Elective AAA repair, both EVAR and open repair, was estimated to be less costly for men than for women (EVAR £12,993, open repair £11,712), largely due to a lower observed LOS. For emergency repair, costs for men were higher than for women (EVAR £18,045, open repair £17,995) because of longer critical care stays. This could be related to the higher mortality rate among women undergoing AAA repair, reducing LOS, although the pattern of general ward stay between men and women was dissimilar.

Surveillance costs

Post-surgery surveillance resource use was based on expert opinion [one vascular surgeon (MJB) and one vascular biologist (JTP) on the study team] of additional imaging performed in UK clinical practice. All unit costs were obtained from NHS Reference Costs 2014 to 2015. 39 For open repair, it was assumed that patients received one 6-week follow-up consultation. The cost of this consultation was assumed to be the same unit cost as pre-surgical consultation (£196.79). For EVAR, patients were assumed to have annual surveillance for their lifetime, consisting of one consultation (£196.79) and one ultrasound scan. A weighted mean of unit costs of an ultrasound scan with duration of ≥ 20 minutes (RD42Z-RD43Z) was obtained (£61.37).

Reintervention costs

Reintervention costs were incorporated into the model explicitly using data from the EVAR-1 trial. 38 The cost, reflecting mean resource use, of a reintervention during the 10-year follow-up of the EVAR-1 trial38 was estimated for women only for EVAR and open repair. The costs were £7546 and £8986, respectively. These costs were assumed to be the same for reinterventions occurring after elective and emergency repairs.

Sensitivity analyses for costs

Because of the nature of the costing exercise, which produced surgical cost estimates with components combined from different sources (both randomised trial data and observational data), a formal estimate of the associated stochastic precision could not be computed. Therefore, imprecision in unit costs was included in the PSA conducted by representing a 95% uncertainty interval from 20% lower to 25% higher costs as a symmetrical normal distribution for log-costs. The impact of changes in costs was also explored in DSAs by varying the costs of screening, surveillance and surgical operation by –20% or + 25% (see Costs).

Quality of life in the population

There is limited evidence that an AAA-screened population has a lower health-related QoL than the general population. 10 For the purpose of calculating QALYs, the life-years accrued of all women, screened and unscreened, in the model were adjusted using UK population EuroQol-5 Dimensions (EQ-5D) utility survey data, specific to women. 102 The QoL weights used were as follows: 0.81 for ages 55–64 years, 0.78 for ages 65–74 years and 0.71 for ages of ≥ 75 years.

Quality of life after surgery

An additional consideration relates to the QoL of those who undergo surgery. Trials of elective AAA repair indicate that there is a reduction in QoL following repair, but that this is transient. 103 In the EVAR-1 trial,99 mean EQ-5D utility (range 0–1) score in the EVAR arm was 0.74, 0.73, 0.71 and 0.74 at baseline and 1-month, 3-month and 12-month follow-up, respectively. Using a visual analogue scale [(VAS); scale 0–100] the corresponding figures were 70.82, 70.20, 69.69 and 71.29, respectively. In the open repair arm, mean EQ-5D utility score was 0.74, 0.67, 0.73 and 0.75 at baseline and 1-month, 3-month and 12-month follow-up, respectively (corresponding figures for the VAS were 70.78, 64.09, 71.36 and 72.53, respectively). Given the focus on a whole-screened population, these small differences in QoL are not likely to have a material bearing on results, and it is more important to reflect age-related differences in QoL.

Given the nature of emergency repair and the patient’s condition at randomisation and baseline, comparative utility scores are harder to acquire. However, at the 12-month compared with the 3-month follow-up in the IMPROVE trial,30 mean EQ-5D utility score was 0.01 higher for EVAR and 0.04 higher for open repair. Again, it was considered that these differences were not material in terms of long-term modelling for a whole population.

Non-abdominal aortic aneurysm mortality

Mortality not related to an AAA is a competing risk, in that AAA screening will be less effective when such competing mortality is higher. Age-specific non-AAA mortality rates for women were estimated using two data sets: (1) overall life tables and (2) rates of death by age and cause. Life tables for 2012–14 were available from the ONS for the UK population. 40 Overall annual mortality risks were adjusted by subtracting the AAA-specific death rates for women among the UK population, using ONS cause-of-death data. 104

Previous modelling exercises of AAA screening based on the MASS trial98 have noted that the non-AAA mortality after being turned down for elective surgery was higher than that in the general population. 11,12 This is because comorbidities are a major reason for surgery being contraindicated. In previous modelling for men, the corresponding non-AAA mortality rate was taken directly from data in the MASS trial. 98 However, such comorbidities occur at the same rate in both the invited and non-invited groups, although in the latter they may be largely unobserved. Hence, including an increased non-AAA mortality rate after being turned down for elective surgery, which occurs mainly in the group invited to screening, unfairly biases the results against screening. With this understanding, the current modelling for women does not include a similar increase in the rate of non-AAA mortality. This provides fair estimates of incremental costs, life-years and QALYs (i.e. the differences between the invited and non-invited groups). However, it will very slightly overestimate absolute life-years and QALYs, and maybe absolute costs as well, because the women following contraindication to elective surgery are on average being assumed to survive for longer than they may do in reality.

Numbers of key events

Table 21 shows the numbers of women in the non-invited and invited to screening groups that experience key events from the base-case run of the DES model over 30 years. About three-quarters of all elective operations in the invited to screening group occur through incidental detection, indicating that many AAAs are not initially detected at age 65 years, owing to non-attendance or an aneurysm that has yet to develop. Elective operations following screen detection of an AAA occur predominantly in the 68- to 78-year age group (Figure 11). In total, 86% of the population die by the age of 95 years in the non-invited group and 0.83% die of AAA-related causes. Screening prevents approximately 2500 AAA deaths in this population of 10 million women, with the percentage who die of AAA-related causes in the invited to screening group being reduced to 0.80%. The relative risk reduction is 3.0%; 4100 women need to be invited to screening to save one death from an AAA.

FIGURE 11.

Cumulative elective operations, emergency operations and AAA deaths in the base-case analysis ages 65–95 years among 10 million women. (a) Elective operations; (b) emergency operations; and (c) AAA deaths.

The difference in numbers of emergency operations and AAA deaths between the invited and non-invited groups accrues gradually over the 30-year period after initial screening (see Figure 11). No effect is evident within the first 5 years as only a small proportion of women are initially over the diameter threshold for elective intervention. This is in contrast to the MASS trial10 of screening in men, in which the benefit of screening was apparent at an earlier stage. In the non-invited group, 0.92% of women have a ruptured AAA over the course of 30 years, compared with 0.88% of the invited group (see Table 21).

Health economic outputs

Estimates of mean life-years, QALYs and costs per woman invited to screening for the base-case are given in Table 22. The group not invited to screening has an average life expectancy from age 65 years of 20.5429 years, which increases by 0.00285 years for the group invited to screening. The gain of 0.00285 life-years equals 1.04 days of life; this average figure reflects the very many who have no change in life expectancy through screening, the few who gain (some substantially) because rupture of the AAA is prevented and the very few who lose by dying in elective surgery.

For health economic analyses, life-years and costs are both discounted at 3.5% per annum. The increase in QALYs associated with screening is 0.00110 years per woman invited to screening. Mean discounted costs associated with screening and AAA-related events increase by £33.99 from £49.56 to £83.55. Overall, this gives an ICER of £22,000 per life-year gained and £31,000 per QALY gained for the base-case screening strategy versus no screening. Therefore, the INMB [calculated as (net discounted QALYs × WTP) – net discounted costs] is negative for both £20,000 and £30,000 per QALY thresholds (see Table 22). This suggests that a screening programme for women based on the current NAAASP implementation is unlikely to be cost-effective.

Uncertainty from the PSA in the estimated incremental QALYs and incremental costs is shown in Figure 12a. The scatterplot lies in the north-east quadrant, where invitation to screening is both more effective and more expensive. However, the majority of the points lie above the threshold of £20,000 per QALY gained and about half lie above a £30,000 threshold (shown in Figure 12a by the diagonal lines that pass through the origin). There is considerable uncertainty in the incremental discounted QALYs. In particular, the prevalence of AAAs is highly correlated with the estimated incremental QALYs and, as this quantity is not precisely known, it is a key driver behind the large amount of uncertainty. We express uncertainty on the INMB scale: an INMB of –£12.03 (95% uncertainty interval –£27.88 to £22.12) per woman invited is estimated if a QALY is valued at £20,000, and an INMB of –£1.05 (–£23.76 to £54.79) per woman invited if a QALY is valued at £30,000. A CEAC (Figure 12b) indicates that the screening programme, as implemented, is unlikely to be considered cost-effective: there is a < 20% probability that the programme would be cost-effective at a threshold of £20,000 per QALY.

FIGURE 12.

Cost-effectiveness from the PSA of the base-case model. (a) Results on the cost-effectiveness plane; the green and blue lines indicate WTP thresholds of £20,000 and £30,000 per QALY, respectively; and (b) CEAC.

Baseline aortic diameter distribution and prevalence

We first consider how changes in the aortic diameter distribution and prevalence affect the cost-effectiveness of the screening programme. Although the base-case model is derived from the aortic size distribution from 700,000 men in NAAASP, a sensitivity analysis is based on the Uppsala study14 aortic size distribution, undertaken in 5140 women (see Chapter 4, Screening). After reweighting both distributions to give the desired prevalence (0.43% ≥ 3.0 cm), the Uppsala study14 had a much higher proportion of medium and large AAAs (0.074% vs. 0.005% in the reweighted NAAASP distribution; see Table 12b). In addition, we investigate the effect of halving the prevalence, from 0.43% to 0.21%, and doubling the prevalence, from 0.43% to 0.86%, from the base-case model.

A notable change in the ICER is seen when changing the baseline aortic diameter distribution from the (weighted) NAAASP distribution to the (weighted) Uppsala distribution (see Table 22). There is a small increase in costs, due to an increase in elective operations, but the mean difference in life-years almost doubles owing to timely elective operations taking place and the subsequent reduction in ruptures. The effect is a screening programme that is more cost-effective, with the ICER reduced to £18,000 per QALY gained and an INMB gain of £4.29 per woman invited if a QALY is valued at £20,000. This effect becomes even clearer if the prevalence of AAAs is doubled to 0.86%, with the ICER decreasing to £13,000 per QALY gained, and an INMB gain of £23.55 per woman invited if a QALY is valued at £20,000. As expected, the ICER increases substantially if the prevalence is halved, to £65,000 per QALY gained.

Dropout and incidental detection rates

We next consider the effect of a change in the dropout and incidental detections rates. First, the dropout and the incidental detection rates are halved, from 5.5 to 2.7 per 100 person-years and from 2.9 to 1.5 per 100 person-years, respectively. Following this, the dropout and the incidental detection rates are doubled, from 5.5 to 10.9 per 100 person-years and from 2.9 to 5.9 per 100 person-years, respectively.

Halving the dropout rate ensures that a larger number of individuals stay in the surveillance programme, giving a greater chance of preventing a rupture via an elective operation should their AAA grow large enough. This increases the cost of the programme, but a greater number of life-years are gained owing to the increase in AAA treatment. In addition, life-years are lost in the non-invited group owing to the halving of the incidental detection rate. This results in a more cost-effective programme than in the base-case, with an ICER of £24,000 per QALY gained (see Table 22). Conversely, an increase in the dropout and incidental detection rates results in a greater ICER of £51,000 per QALY gained, attributable to a greater relative reduction in the incremental life-years than in the reduction in the incremental costs.

Parameters affecting elective operations

We now consider the effects age and AAA size have on an individual’s suitability for elective EVAR surgery and their operative mortality rates, and whether or not changes in these parameters affect the cost-effectiveness of the screening programme. First, the percentage receiving elective EVAR and the elective operative mortality rates for EVAR and open repair are allowed to depend on age and AAA diameter (see Chapter 5). Next, the percentage receiving elective EVAR and the elective operative mortality rates are based on the systematic reviews (see Chapter 3). Finally, the elective open AAA repair operative mortality, estimated from the NVR26 and HES,28 is decreased from 8.1% to 5%; this might be regarded as a potentially attainable target after a performance improvement programme.

These changes all have a similar, but small, effect on the cost-effectiveness estimates, with ICERs in the range £28,000–30,000 (see Table 23). Although the incremental costs are similar or slightly lower than in the base case, the incremental QALYs are slightly higher. A reduction in the elective open repair operative mortality to 5% has little effect on the cost-effectiveness estimates when compared with the base case, which suggests that cost-effectiveness is largely unaffected by even quite substantial changes in elective open AAA repair mortality.

Parameters affecting emergency operations

As earlier, we first allow the percentage receiving emergency EVAR, and emergency operative mortality rates for EVAR and open repair, to depend on age. Second, the percentage receiving emergency EVAR, and emergency operative mortality rates are based on the literature reviews (see Chapter 3, Mortality following ruptured abdominal aortic aneurysms in women).

Allowing emergency operations to depend on age has limited overall effect on the cost-effectiveness, with an ICER of £31,000 per QALY gained. The effect of the systematic review point estimates also has little effect on the cost-effectiveness estimates, resulting in an ICER of £30,000 per QALY gained (see Table 23).

Reintervention rates following successful abdominal aortic aneurysm repair

Uncertainties in the reintervention rates estimated as zero in the base-case analysis were not included in the PSA. We consider complications following both elective and emergency operations, by increasing the reintervention rate after successful elective open repair from 0.0 to 4.4 (31–120 days) and 2.9 (> 120 days) per 100 person-years (based on men in the DREAM76 and OVER77 trials), and increasing the long-term AAA mortality rate after successful emergency EVAR repair from 0.0 to 0.985 per 100 person-years (based on men in the IMPROVE trial30). This has almost no effect on the cost-effectiveness of the screening programme for women (see Table 23). This highlights the fact that overall cost-effectiveness is relatively insensitive to changes in rates of events that affect only a small proportion of the population (e.g. those with an AAA who have undergone and survived an elective or emergency operation). Therefore, although there were no long-term data for women about reinterventions after successful emergency surgery, it is reassuring that this parameter is unlikely to have any substantial effect on overall cost-effectiveness estimates.

Costs

Finally, we consider the effect of costs, by means of combinations of alterations to the unit costs in different stages of the screening programme. First, we lower the costs of screening, surveillance and consultation by 20%. This is followed by the increase in screening, surveillance and consultation costs of 25%, a symmetrical increase on a log-scale. Next, we consider opposing decreases and increases in costs of elective and emergency surgeries: a 20% lower cost of elective surgery and 25% higher cost of emergency surgery, then a 25% increase in the cost of elective surgery and 20% decrease in the cost of emergency surgery.

The decrease and increase in costs associated with screening and surveillance result in an expected increase and decrease in the incremental costs with no change in the incremental life-years. This has the effect of decreasing and increasing the ICER by approximately 20% to £26,000 and £38,000 per QALY gained, respectively (see Table 23). A smaller effect on the ICER was seen when the elective and emergency surgery costs were varied. Decreasing elective surgery costs and increasing emergency surgery costs makes the screening programme more cost-effective, but only slightly, with the ICER decreasing to £30,000 per QALY gained. Similarly, increasing elective surgery costs and decreasing emergency surgery costs increases the ICER to only £32,000 per QALY gained.

Reference case

This includes the effects of age and AAA diameter on parameters for elective operations, and the effect of age on parameters for emergency operations, but does not make any other changes. Neither of these sensitivity analyses in Chapter 7 changed the cost-effectiveness estimates very much compared with the base case. Therefore, their combination is also similar to the base case (Table 24 Reference case columns), with an ICER of £30,000 per QALY gained. The numbers of key events are provided for this reference case in Table 25 (Reference case columns), which are, again, similar to the base case (see Table 21).

Age at screening

We consider raising the screening age from 65 years to 70 or 75 years. The AAA prevalence at age 65 years was 0.43% (95% CI 0.23% to 0.80%), based on the systematic review for women aged < 70 years (see Chapter 3, Current prevalence of screen-detected abdominal aortic aneurysms in women), and increased to 1.15% (95% CI 0.59% to 2.24%) for women aged 75 years, based on the 70- to 79-year age group in the systematic review. Interpolating linearly on a logit scale between these two estimates gives an AAA prevalence of 0.70% (95% CI 0.37% to 1.34%) at age 70 years. The attendance rate at screening was 72.7% at age 65 years, and this is estimated to decrease to 67.6% at age 70 years and to 62.3% at age 75 years (see Chapter 4, Screening). Both of these factors are taken into account when adjusting the screening age.

As shown in Table 24, offering screening at age 70 or 75 years increased the gain in QALYs per woman invited to screening by factors of about 1.4 and 2.2, respectively, compared with the reference case. The costs per woman invited decreased in both groups compared with the reference case, with the incremental cost being larger, especially for age 75 years screening. The estimated ICERs were £24,000 per QALY gained for age 70 years screening and £18,000 for age 75 years screening. The more favourable ICER for age 75 years screening gives rise to a positive INMB at a threshold of £20,000 per QALY.

The improved cost-effectiveness at higher ages illustrates the importance of AAA prevalence. This offsets the disadvantages incurred through a lower life expectancy (16.4 years at age 70 years and 12.6 years at age 75 years, compared with 20.5 years at age 65 years) resulting from the increased mortality rates from non-AAA causes, as well as the increased mortality from elective operations at higher ages (see Chapter 5). However, two points should be noted. First, the non-intervention rate for elective operations used in the model is not age dependent, as there was insufficient evidence to quantify this in the systematic review (see Proportion of women versus men not offered an intervention). This is not clinically realistic, and it is appreciated that non-intervention rates probably increase markedly with age, especially for women in whom EVAR is not suitable (as open repair operations in elderly women have the greatest risk). This implies that the results for screening, at age 75 years in particular, may be overly optimistic. Second, by screening at a higher age, AAA deaths at younger ages are not prevented. From the unscreened group in the base-case analysis (see Figure 11c), one can see that about 5% of all AAA deaths occur between the ages of 65 and 70 years, and about 15% between the ages of 65 and 75 years. These earlier deaths are also associated with the greatest number of life-years lost. Although this does not affect the cost-effectiveness of screening at higher ages, it reduces the benefit of screening at the population level.

Abdominal aortic aneurysm diameter for considering surgery

We consider reducing the threshold of the AAA diameter for considering elective surgery to 5.0 or 4.5 cm. The surveillance intervals were also altered in these scenarios, so that a 3-month interval applied for the 1.0 cm range below the threshold (i.e. 4.0–4.9 cm when the threshold is 5.0 cm, and 3.5–4.4 cm when the threshold is 4.5 cm). This is in keeping with the use of 3-month intervals for 4.5- to 5.4-cm AAAs in both the base case and reference case in which the threshold is 5.5 cm.

Using lower thresholds increased the incremental life-years and QALYs compared with the reference case (see Table 24). Both options slightly increased the incremental costs because of a greater number of elective operations, especially in the invited group. The consequence is that for a 5.0-cm threshold the ICER was £28,000 per QALY gained, while, for a 4.5-cm threshold, the ICER was £27,000, both slightly lower than the reference case.

Aortic diameter defining an abdominal aortic aneurysm

We consider reducing the aortic diameter for defining an AAA in women to 2.5 cm. In both the Swedish and Danish data, (see Chapter 4, Screening), 2.5 cm is roughly 1.5 times the average aortic diameter in women (which is sometimes suggested as an appropriate definition of an AAA105) and roughly 3 SDs above the mean. Based on an AAA prevalence of 0.43% for an aortic diameter of ≥ 3.0 cm at age 65 years, the AAA prevalence becomes 1.44% for an aortic diameter of ≥ 2.5 cm. For the 2.5- to 2.9-cm group, which are now entered into the surveillance programme, we consider two choices of surveillance intervals: (1) 1 year (as in the 3.0–4.4 cm group) or (2) 5 years (as has been suggested for men with subaneurysmal aortic diameters106).

Both options similarly increased the life-years and QALYs gained (see Table 24), but only by about one-third compared with the reference case. However, they also increased the incremental costs, through more surveillance scans in the invited group, especially in the first option, which employed 1-year surveillance intervals, and more elective operations. As a result, the cost-effectiveness estimates were more favourable than in the reference case, with ICERs of £27,000 and £25,000 in the two cases, respectively. However, the programme would still not be considered cost-effective at a threshold of £20,000 per QALY.

Surveillance intervals and rescreening

Finally, we consider changing the surveillance intervals. First, we increase the surveillance intervals for the smallest AAAs (3.0–3.9 cm) to 2 years, an option that improved cost-effectiveness slightly for men. 35 Second, we consider a programme in which all women are rescreened every 5 years from age 65 years upwards. In essence, this includes everyone with an aortic diameter of < 3.0 cm into a monitoring programme with a surveillance interval of 5 years.

In this first option, unsurprisingly, the number of life-years and QALYs gained is very slightly lower, but the incremental cost is also slightly lower than the reference case through fewer surveillance scans (see Table 24, penultimate column). This yields an ICER of £30,000 per QALY gained, hardly changed compared with the reference case. The second option has dramatic effects on both the life-years and QALYs gained, which are approximately doubled compared with the reference case, and the incremental cost, which is increased more than three-fold (see Table 24). The result is not favourable in terms of cost-effectiveness, with an estimated ICER of £54,000 per QALY gained.

Combined options considered

The results in Table 24 suggest that a favourable combined option might be obtained by screening at age 70 years, employing an intervention threshold of 5.0 cm and a diagnosis threshold of 2.5 cm. These options considered singly gave ICERs of £24,000, £28,000 and £25,000, respectively. Combining them is presented as option 1 in Table 26, and shows increases in both QALYs gained and in incremental costs compared with both the reference case and the options considered singly (see Table 24). However, the ICER is estimated as £23,000 per QALY, which represents only a slight improvement in cost-effectiveness.

Given that option 1 did not reduce the ICER to below £20,000 per QALY, all the other additional combinations were also investigated, being presented as options 2–7 in Table 26. A number of conclusions can be drawn from the total 12 options considered: (1) many of the options yield very similar ICERs, with, for example, 9 out of the 12 between £22,500 and £25,000 per QALY; (2) the combinations employing screening at age 65 years and a diagnosis threshold of 3.0 cm are clearly not the best options; (3) screening at age 70 years is almost uniformly more cost-effective than screening at age 65 years; (4) a diagnosis threshold of 2.5 cm is uniformly more cost-effective than one of 3.0 cm; and (5) no option gave an ICER below £20,000 per QALY.

The best choice of intervention threshold is moot. Except when screening at age 65 years using a 3.0-cm diagnosis threshold, all three intervention thresholds (4.5 cm, 5.0 cm and 5.5 cm) yield very similar ICERs. The lower thresholds give rise to greater QALYs gained but also to greater incremental costs from an increased number of elective operations. However, since the increases in QALYs and costs are proportionally similar, the ICERs remain similar.

Given the uncertainty in the ICERs, it is not possible to be definitive about which combined option is ‘best’ in terms of cost-effectiveness. However, option 1 gave the lowest ICER estimate and the next to largest gain in life-years and QALYs. We now examine this option in more detail.

Detailed results for option 1

The numbers of key events for this option, which is the best in terms of overall cost-effectiveness, are shown in Table 25 (right-hand side). Compared with the reference case, there are fewer AAA ruptures, emergency operations and AAA deaths due to screening at the higher age of 70 years. The relative reductions in each of these outcomes in the screening group are 12%, 12% and 8%, compared with 4%, 4% and 3% in the reference case, respectively. However, there are more elective operations in this screening option than in the reference case, owing to the lower intervention threshold, with a relative increase of 55% compared with 21%.

The cumulative numbers of elective operations, emergency operations and AAA deaths for option 1 are shown in Figure 13. Compared with the corresponding figures for the base-case (see Figure 11), the separation of the lines for the invited and non-invited groups is greater. For emergency operations and AAA deaths, the separation gradually increases over the whole of the 25-year time course from age 70 to 95 years, with no suggestion, for example, of a marked benefit in the early years after screening. A marked early benefit might be anticipated only if there were a substantial number of women discovered at screening who were near or over the AAA diameter threshold for considering surgery.

FIGURE 13.

Cumulative elective operations, emergency operations, and AAA deaths among 10 million women in combined screening option 1. (a) Elective operations; (b) emergency operations; and (c) AAA deaths.

A PSA was carried out for this case, using 1000 runs for different parameter values representing their uncertainty distributions, each run including 500,000 pairs of women. This involved using the variance–covariance matrix from the logistic regressions that quantified the relations between age and AAA diameter on parameters for elective and emergency operations (see Chapter 5). Random draws of parameter values were taken from the relevant multivariate normal distributions. A normal distribution for the uncertainty of the altered AAA prevalence estimate (on a logit scale) was used, together with a Beta distribution for the altered attendance rate. The uncertainty in other parameters was as in the base-case analysis (see Table 11).

The distribution of incremental costs and QALYs is shown on the cost-effectiveness plane in Figure 14a. As for the base-case (see Figure 12a), there is a wide spread for incremental QALYs and a lesser spread for incremental costs. The centre of the distribution lies between the lines representing the thresholds of £20,000 and £30,000 per QALY. The INMB at a threshold of £20,000 per QALY is –£5.08 (95% uncertainty interval –£31.53 to £69.98), whereas at a threshold of £30,000 it is £14.91 (95% uncertainty interval –£25.18 to £135.16). The cost-effectiveness estimate is more favourable than that for the base-case analysis, there being about a 40% chance that this screening programme would be cost-effective at a threshold of £20,000 per QALY, as indicated in the CEAC shown in Figure 14b.

FIGURE 14.

Cost-effectiveness from the PSA of combined screening option 1. (a) Results on the cost-effectiveness plane (the green and blue lines indicate WTP thresholds of £20,000 and £30,000 per QALY, respectively) and (b) CEAC.

Discrete event simulation model

The DES model developed for this project was novel in a number of respects. First, in contrast to the previous multistate Markov model, it modelled the progress of individuals rather than groups. Aortic diameter expansion could, thus, be more precisely represented, allowing for the substantial heterogeneity between people in growth rates; avoiding the use of arbitrarily defined categories for aortic diameters and the awkward, but necessary, introduction of time-dependent transition rates in the multistate model;35 and better accounting for uncertainty. Moreover, and importantly in the context of investigating AAA screening for women, the modelling structure allowed the investigation of different screening options, which would not be feasible in a single multistate model. The downside of individual modelling is the computational requirements, as enough individuals have to be modelled to ensure that the results obtained are stable and reliable. The computational demands also increase substantially when undertaking PSAs. This problem was ameliorated to an important extent by appreciating that only those with an AAA had to be modelled in detail in order to obtain precise estimates of incremental effects.

The model developed also has potential for further use outside the present work. For example, with parameters reflecting those for men rather than women, it can be used to investigate alternative screening options for men, such as the choice of intervention threshold or the inclusion of subaneurysmal men (aortic diameters of 2.5–2.9 cm) within the surveillance programme. This could suggest ways in which NAAASP could be improved. The model could also be used for investigating selective screening, for example, of women smokers, if sufficiently robust data on the parameters for this group could be obtained. The model also has the potential to be used in other contexts, for example, internationally, by specifying relevant parameter inputs. In practice, the problem is less in adapting the model, but more in the likely unavailability of data on relevant parameters in new contexts and the assumptions that might have to be made.

Prevalence and definition of abdominal aortic aneurysms in women

The systematic review of AAA prevalence was of key importance to this project, but a number of limitations became apparent. First, the number of data in women was rather limited and complicated by studies including different age groups being undertaken in different calendar years (so any secular trends might influence prevalence estimates) and using different screening approaches (population based, self-referred or physician initiated, and free, self-purchased or reimbursed) and ultrasound measurement techniques (inner to inner aortic diameter, outer to outer, leading edge to leading edge or unstated), although nearly all used 3.0-cm diameter as the minimum diameter for definition of an AAA. Second, except for age and smoking status, rather few characteristics of the women invited (or screened) were available (e.g. body size or diabetes mellitus, which may influence baseline arterial diameters and AAA growth rates, respectively). However, it was clear that the prevalence of AAAs was much higher in current smokers. In the future, it is possible that AAA prevalence in women may increase because of the historical increase of smoking in women some decades ago.

The usual definition of an AAA in men is either a widest diameter > 50% greater than the suprarenal aortic diameter or the widely used pragmatic definition of ≥ 3.0 cm. Women have smaller-diameter aortas than men. 21 Most of the screening studies did not hold individual participant data, but these were available from two modestly sized studies (a Swedish study14 of 5140 women aged 70 years and a Danish study of 570 women aged 67 years), which we used to provide an estimate of the AAA diameter distribution in women. Analysis of these data indicated that 2.5 cm might be a better minimum threshold diameter for the definition of an AAA in women. The definition of an AAA in women had considerable impact on both prevalence and cost-effectiveness estimates.

Data sources analysed

As AAAs have traditionally been considered a male-dominated disorder, female-specific data or inputs required for the DES model were not readily available and had to be gathered for this project. Estimates of over 40 female-specific parameters were required (see Tables 1–3). We were fortunate to have access to several databases that could be explored, including those from the RESCAN study,35 EVAR-138 and other trials of EVAR versus open repair in the elective setting, and the IMPROVE trial30 for emergency repairs.

Other female-specific input parameters for the model came from exploration of local data sources (e.g. women dropping out of hospital-based surveillance programmes) or comparative reviews (e.g. proportion of women at any age accepting invitation to screening, and incidental detection rates). In addition, UK registries were explored for data about aneurysm repairs, particularly the NVR database and HES: analysis of such data indicates that EVAR is used preferentially in the older groups and that AAA repair in those aged > 80 years is now common. However, sadly, they showed that the mortality following elective repair in women was unacceptably high, particularly for open repair, for which the figure was 8%. We also obtained international data from the Vascunet collaboration. 112

Based on the individual data in the RESCAN study,35 the AAA rupture rate in women increased by about 30-fold as the AAA diameter increased from 3.0 to 5.5 cm, and the AAA rupture risk was about four-fold higher in women than men at the same AAA diameter. This might naively suggest that an AAA diameter threshold of 4.5 cm for considering elective surgery in women would correspond to the same balance of risks and benefits as the choice of a diameter of 5.5 cm in men. However, because of the worse elective surgery outcomes in women, there is a trade-off that comes into play, and this is reflected in the cost-effectiveness analyses.

The analysis of the individual data for women in the NVR26 was important in a number of respects. First, it substantiated in these recent UK data the overall higher mortalities in women than men for elective operations, both EVAR and open AAA repair, as found in the systematic review. Second, it showed that the proportion of patients actually receiving elective EVAR was lower in women than men, which paralleled the difference in the proportion suitable for EVAR found in the systematic review. Third, it provided reliable estimates of these parameters for emergency surgery for ruptured AAAs, while the literature review undertaken for emergency surgery was less detailed and the data sparser than for elective surgery. Fourth, it allowed the dependence of these parameters on age and AAA diameter for elective operations, and on age for emergency operations, to be modelled. This was especially important when considering screening options that changed the age at screening or intervention threshold; using overall figures in these analyses would have been unrealistic.

In summary, the derivation of the female-specific inputs for the model used a very wide array of sources and resources; it was a major task, taking over 1 year to complete.

Base-case and sensitivity analyses

The base-case cost-effectiveness analysis, which adopted the screening options used for men in NAAASP, showed that the estimated life-years gained per woman invited was very small: 0.00285 life-years or 1.04 life-days. A small average life-years gain is expected in population screening, as the vast majority of those invited are screened as normal and have no change in life expectancy. Nevertheless, the extremely small gain in life-years in this base-case analysis is the main reason for the unfavourable cost-effectiveness results. Using standard discounting for both costs and life-years, the ICER per QALY gained was estimated as £31,000. This is above the threshold of £20,000 generally used by NICE as a basis for accepting health interventions for use in the NHS. Moreover, there was considerable uncertainty in this cost-effectiveness estimate: in the PSA, the INMB estimate of –£12.03 (at a threshold of £20,000 per QALY) had a 95% uncertainty interval of –£27.88 to £22.12. The probability that AAA screening is cost-effective at this WTP threshold is < 20%.

The sensitivity analyses undertaken did not change this conclusion. These also underlined the pivotal role of AAA prevalence in determining the ICER. When the AAA prevalence was doubled, from 0.43% to 0.86%, the estimated ICER fell below £20,000 per QALY gained. A prevalence of 0.86% is beyond the upper limit (0.80%) of the 95% CI for women aged 65 years derived from the systematic review. So only an extreme change in prevalence, beyond what is likely to be compatible with the evidence, could lead to a conclusion that AAA screening is cost-effective using standard criteria. However, the cost-effectiveness is also sensitive to the exact shape of the distribution of aortic diameters (not just the prevalence), as shown by replacing NAAASP-based distribution with one based on the Uppsala14 distribution. This emphasises the need for better contemporary data on the distribution of aortic diameters in women at ages relevant to screening.

Screening options in women

Screening women for AAAs might become more cost-effective if one moved away from the options adopted in NAAASP for men. This was the purpose of the scenario analyses undertaken. As expected, screening became more cost-effective if offered at age 70 or 75 years. Perhaps surprisingly, it was more cost-effective at age 75 years than at age 70 years. Thus, the effect of increasing prevalence as age increased, from 0.43% at age 65 years to 0.70% at age 70 years and 1.15% at age 75 years, outweighed the more limited life expectancy remaining at older ages. Nevertheless, we adopted a screening age of 70 years rather than age 75 years in the final combined scenario investigated, for two reasons. First, the estimated non-intervention rate for elective surgery applied in the model was not dependent on age; this rate might become unrealistically low at higher ages as surgeons are likely to turn down many elderly women for elective surgery, especially open AAA repair, for which the risks are substantial. Second, the estimates for attendance at screening and AAA prevalence become more imprecise and potentially unreliable at older ages.

Lowering the threshold for defining an AAA to 2.5 cm, lengthening surveillance intervals somewhat for the smallest AAAs, or lowering the threshold for considering elective surgery to 5.0 or 4.5 cm, made AAA screening for women slightly more cost-effective, but these changes considered individually did not bring the ICER down below £20,000 per QALY gained.

The remaining possibility of finding a cost-effective screening option for women was to combine the alternative screening options. The best one investigated combined invitation to screening at age 70 years, an aortic diameter of 2.5 cm defining an AAA (with a 5-year surveillance interval for AAAs with a diameter of 2.5–2.9 cm), and consideration for elective surgery at an AAA diameter of 5.0 cm. The estimated QALY gain was greater than the base-case analysis (0.00200 vs. 0.00110 per woman invited). Combined with a greater incremental cost than the base case, principally due to an increased number of elective operations, the ICER was estimated as £23,000 per QALY gained. In the PSA, the corresponding INMB estimate of –£5.08, at a threshold of £20,000 per QALY, also had a substantial 95% uncertainty interval of –£31.53 to £69.98.

Conclusion

The conclusion of these analyses is that the accepted criteria for an AAA screening programme in women are not currently met with respect to either clinical effectiveness (low operative mortality rates) or cost-effectiveness. We also did not find a combination of screening options for women that would make population AAA screening cost-effective for the NHS at a WTP threshold of £20,000 per QALY.

Recruitment via abdominal aortic aneurysm patient forum

The Leicester Vascular Surgery Unit and the Leicester NAAASP run an annual patient education forum for men with small AAAs. The aim of this group is to provide information and advice to men with small AAAs and their partners in order to prepare them for the clinical decision-making process around surgery, and to provide general health advice. As part of this group, in June 2015, the issue of whether or not to screen women for AAAs was specifically discussed with the 42 men in attendance. All these men and their partners were asked if they would be willing to attend a meeting to discuss the issue of AAA screening for women in more detail. Four men and two women subsequently attended a PPI meeting in July 2015 (see Patient and public involvement meeting 24 July 2015).

Recruitment via media

Because of the poor representation of women in the group recruited via our male patient forum, it was decided that a more representative PPI group should be recruited. Initial efforts were based on social media (website, micro-blogging applications, Biomedical Research Unit newsletters) but generated no responses. Feedback from our existing Biomedical Research Unit PPI group and the PPI group recruited via our patient forum indicated that this was likely to be due to the limited use of social media in the target group. To address this, it was decided to engage traditional media outlets in a call for recruitment. Through direct contact with the Leicester British Broadcasting Corporation (BBC) office, the project was featured on the regional television news (BBC East Midlands Today) and on local radio (BBC Radio Leicester). The radio interview (two parts) is available on the project website (www.screeningaaawomen.com/; accessed 1 March 2017). In response to these activities, expressions of interest were received from 15 women, 11 of whom subsequently attended and formed the ongoing PPI group for the project.

Patient and public involvement meeting 24 July 2015

The initial PPI meeting for the project was held on 24 July 2015, facilitated by Matthew J Bown. The PPI group at this time consisted of the four men and two women recruited from the Leicester AAA patient forum. One of the women was the wife of one of the men, but otherwise the group members were unrelated.

At this meeting, a presentation was given to those attending detailing the background to the project. This included the evidence for screening in men and why this evidence was not available for women. Broad strategies regarding how this evidence may be gathered were discussed, including how to determine if screening in women would be cost-effective. The concept of asking lay people to assist with the research project was introduced, using the example of asking how women might respond to being asked to attend for screening.

The group were specifically asked the following questions:

1. Whether or not women would want to attend screening for AAAs?

2. Is it acceptable to invite only women in high-risk groups?

3. Do women want to be screened?

4. What would prevent them from attending?

5. Would women want an operation to correct an AAA if it was offered, bearing in mind that the operation is more risky in women than in men?

It was acknowledged by the group that only two women were present and all members had personal knowledge of the disease. This could lead to the answers not being generalisable to the wider population of women.

Following a general discussion on all of these points, the group felt that women were more health conscious and may be more inclined to go for screening if offered. However, there were also mixed feelings on whether or not women would want to be screened given that they have previously undergone other forms of screening (they might be ‘fed up’ with screening). Some felt that the detection of an AAA by screening may cause much anxiety among women, possibly because of the higher risks of treatment (surgery). The group stated that people generally do not want to know too much about their health, especially when there may not be a simple treatment for a condition detected by screening.

The group was also asked to consider:

1. What would the public want us to find out from this research project?

2. Is saving lives all that matters?

3. Should financial, personal, or psychological cost be a consideration in AAA screening?

There was a suggestion from the group that screening might be better added to well woman general practitioner (GP) visits (and health checks for men). Overall, the group felt that if providing screening saved lives at minimal psychological cost, it would be worthwhile. Financial and personal costs were not thought to be an issue (but the group stated that this was because AAA screening is so cheap and easy).

The question was raised regarding what is happening in other countries and why can we not simply see if their programmes work. It was pointed out that the UK is the only country with a national programme for men and there are not programmes for women anywhere.

Attendees were asked how they find out information on their health in general and AAAs in particular. Everyone agreed that websites were of some use, as was e-mail, but not many people in this age bracket would use social media to find out health information.

The general feeling was that there is no media coverage of AAAs and, until you are asked to go for a screening or are diagnosed with an AAA, you do not know what an aneurysm is. It was suggested that this be addressed by local media coverage (e.g. local radio or television, church groups, local newspapers).

Patient and public involvement meeting 12 January 2016

This meeting was the first meeting of the project-specific PPI group that had been recruited through the radio and television engagement programme. Eleven women attended the meeting, none of whom had previously been diagnosed with an AAA. Age information was not formally collected. One woman had a strong family history of AAAs (two first-degree relatives) and one woman’s husband had previously undergone an AAA repair. The majority (nine women) had family members who had been affected by AAAs.

The same information was presented to this group as had been presented to the previously convened PPI group and the group were asked the same set of questions.

The majority of the group thought that women would want to attend for AAA screening if invited. However, they recognised that this may be a biased response as they all had an interest in this area owing to having affected family members. The group thought that women were generally more accepting of screening because they had been used to being screening for other diseases during their lives. They noted that for the majority of those women screened the reassurance of a negative scan was very important and well worth the financial cost of screening and the cost to the individual.

The concept of screening only high-risk groups, using the example of tobacco smoking, was the most contentious area of discussion. Overall, the group thought that this would be unacceptable to the majority of women, largely because they knew of anecdotal cases in which women had been diagnosed with an AAA despite being non-smokers. Other high-risk groups, such as those with a family history of disease, were discussed. The group were surprised that the NHS does not record family history of diseases in any systematic manner and that this would be unavailable as a method to select women for screening.

The women in attendance were asked about age groups likely to attend for screening. It was proposed that women would probably be invited for screening between the ages of 65 and 75 years. The group wanted screening after age 75 years to still be done if the participant wanted this, although they did recognise that this may not be financially viable.

Barriers to screening were discussed. The group thought the location of screening (especially given the current model of community screening) was unlikely to have a negative influence on attendance. One barrier to screening was the lack of knowledge among NHS staff of the possibility of AAAs in women and that a reluctance of NHS staff to refer for screening, or to exclude women from screening, may have a negative effect on uptake.

The group were asked about whether or not they would want to undergo AAA repair if this were indicated, particularly with the knowledge that women have higher perioperative risk than men. The women thought that, providing the overall risks were considered, most women would want to undergo AAA repair. The effect of age on perioperative risk was raised by members of the group, who also suggested that older women may not want screening as they would not want to know or undergo surgery if diagnosed with an AAA.

Given the failure to engage a suitable group of women for PPI via social media, the use of social media for PPI and for patient information was discussed. The groups acknowledged that very few people in this age bracket would use social media to find out health information. There was an acknowledgement that public awareness of AAAs needs to be improved. The group identified the common scenario in which individuals know nothing about AAAs until the time of diagnosis. The best way of communicating with the likely target groups was thought to be through face-to-face meetings, traditional media (local radio/television) and community groups such as churches.

Patient and public involvement meeting 15 August 2016

At the time this meeting was held, the project was in a data gathering/processing phase and minimal PPI relevant updates had been made to project outputs. As one of the aims of the PPI activities for the project was to set up a PPI group specific to screening women for AAAs, this meeting was used to develop the discussion in this area. In order to provide a framework for this discussion, the main focus of the meeting was to discuss a separate study, the Female Aneurysm screening STudy (FAST). FAST is a NIHR Research for Patient Benefit (RfPB)-funded pilot of AAA screening for high-risk women with the main aims of determining screening attendance and disease prevalence.

The FAST was presented to the group. In brief, FAST will be based in Leicestershire, Rutland and parts of Northamptonshire, and will replicate the male screening programme processes. It will use the same invitation process and information sheets as the male programme. Women will be invited based on their risk of an AAA, with smokers, ex-smokers and those with a history of coronary artery disease forming the three groups being assessed. Information on whom to invite will be taken from GP records, but these records do not show women who have a first-degree relative with an AAA. However, accuracy of GP records will need to be ascertained, as extracting data from GP records is costly.

The group thought FAST represented the next step in establishing an evidence base for AAA screening in women, but were disappointed that the study was focused on high-risk groups. They felt that, as AAA screening was simple and cheap, it should be offered more widely. The group reiterated the positive psychological effects of a negative scan.

Patient and public involvement meeting 20 March 2017

In preparation for this final meeting, the main issues arising from the previous meetings were summarised. These were surrounding the acceptability of AAA screening, the positive effects of a negative screen and targeted screening. The following themes formed the basis for discussion:

• The PPI group’s perception of AAA screening is that it is highly acceptable and there is a good likelihood that the uptake among women would be at least equivalent to that seen in men, or higher. Very few physical or logistical barriers to the uptake of screening exist, but the lack of public knowledge regarding AAAs is a significant area of need that should be addressed, preferably through traditional media outlets.

• One of the main positive aspects of screening for AAAs in women may be the psychological benefit of a negative screening scan. Although no objective evidence for this effect exists (and, therefore, is a potential avenue for future investigation), our PPI group felt that this was important. Furthermore, one of the unifying motivations for the women to attend the PPI group was that they all wanted to be screened for AAA themselves to obtain this reassurance.

• Targeted screening, particularly if the target group is smokers, is a contentious issue for the public. Given the simplicity and low cost of AAA screening and positive psychological effect of a negative screen, our PPI activities suggest that there is a public perception that there is no requirement to target screening. Key to improving public understanding of AAA screening and/or the acceptability of targeted screening at high-risk groups will be to provide information regarding the scale of AAA screening and the resultant effect this has on the overall cost, and cost-effectiveness of AAA screening.

Preliminary discussion of these themes did not raise any additional points. Following this discussion, the group was presented with the Plain English summary of the project results. The group provided feedback on the presentation of results and edited the Plain English summary (these edits are incorporated into the final version presented in this report). The group thought that the individual costs and the overall costs of screening were important. Comparison of costs for screening with other common NHS interventions would provide a good reference point for the public. After giving consent, all members present underwent an aortic screening scan to get direct experience of the procedure.

Following discussion of the project’s results and having undergone screening, the three themes were discussed again. In addition, the group were asked to give opinions on what future research should be performed and what further information the public would want.

All three themes were confirmed by the PPI group. Given the project’s results presented, some women present thought that targeted screening may be better than no screening at all for women but questioned the methods for identification of high-risk groups. Many of the women in the PPI group had an interest in AAA because of a family history or personal knowledge of disease. There was a majority view that screening based on family history of disease would be important. The group recognised the deficiencies in clinical systems for recording this type of data. The suggestion was made that when an AAA is diagnosed, NHS information systems should alert the relatives of the patient. The issues of confidentiality preventing such a process were discussed. The PPI group felt that alternatives may be to (1) improve public awareness of the increased risk associated with family history of AAAs and (2) provide information for patients with an AAA to encourage them to tell their relatives of the diagnosis. The importance of improved QoL was deemed to be extremely important for the women present. The group thought that the positive effects of a negative screening scan should be investigated as a research priority going forward.