Cost-effectiveness of diagnostic strategies for the management of abnormal uterine bleeding (heavy menstrual bleeding and post-menopausal bleeding): a decision analysis

Definition of heavy menstrual bleeding

The National Institute for Health and Care Excellence (NICE), in its 2007 guideline on the management of HMB, recommended that the condition be defined as ‘excessive menstrual blood loss which interferes with the woman’s physical, emotional, social and material quality of life, and which can occur alone or in combination with other symptoms’ (p. 30). 11 This clinical definition is the most useful one, as objective measurement, with loss of > 80 ml of blood per cycle considered definitive of HMB,12 is impractical. More applicable semi-objective measurement, using pictorial blood loss assessment of sanitary ware13,14 as a surrogate for objective measurement, has been tried but the correlation between objective and semi-objective quantification has been questioned. 13 In any case, objective quantification of menstrual loss does not correlate in many cases with a woman’s subjective complaint of HMB. 15–17

Causes of heavy menstrual bleeding

Heavy menstrual bleeding has been reported to be caused by a variety of underlying pathologies. 11 However, while many conditions have been linked to HMB, in practice most cases are attributed to fibroids, endometrial pathology or dysfunctional uterine bleeding (DUB), and subsequent treatment is dictated by the presence or absence of these conditions (Table 1).

Diagnosis of heavy menstrual bleeding

The current NICE guideline advocates full gynaecological examination and taking a full blood count to exclude the possibility of anaemia. 11 This guideline recognises the need for diagnostic tests to evaluate the uterus, namely endometrial biopsy, ultrasound scan and hysteroscopy, in specific cases. These tests are described in Table 2.

Current diagnostic pathways for heavy menstrual bleeding

It is only in the last 25 or so years that evaluation of the uterine cavity, in response to HMB symptoms, has moved on from the so-called ‘D&C’: dilatation of the cervix and curettage of the endometrium lining the uterine cavity (dilatation and curettage). This test is now used only in exceptional circumstances as it requires general anaesthesia and has been superseded by outpatient endometrial biopsy, which obtains endometrial tissue samples for histological analysis in a convenient outpatient setting without the need for anaesthesia. 20,21 Moreover, the development of high-resolution transvaginal scan (TVS) has allowed the female pelvic structure, including the uterus, to be visualised. The ‘inside’ of the uterus (i.e. the uterine cavity) cannot normally be seen without effecting distension using a fluid or gaseous medium to separate the opposing walls of the uterus. This potential limitation of TVS has been overcome by the advent of saline infusion sonography (SIS)22–24 and outpatient hysteroscopy (OPH). The latter test was previously restricted to the operating theatre owing to the relatively large diameter of the endoscopes, which required dilatation of the cervix in order to successfully instrument the uterus. Advances in instrumentation, namely miniaturisation, improved optics and digital imaging, have made direct endoscopic visualisation of the uterus a simple and acceptable outpatient investigation. 25–28 These tests described in Table 2 provide different albeit overlapping information, and diagnostic accuracy varies according to the particular pathology under scrutiny. NICE guidance from 2007 recognised that ‘. . . particular investigative methods were better for identifying certain types of pathology than others’ (p. 41). 11

Thus, the availability of different, easy-to-use, miniature and increasingly portable ‘bed-side’ tests has created uncertainty as to how best to employ them. This is particularly true with HMB, where different aetiologies need to be considered and the preceding clinical history, and more often than not the examination too, is unable to predict causation with accuracy. A rational basis for subsequent testing strategies cannot be reliably formulated and current testing is, therefore, eclectic, depending upon the vagaries of individual clinicians and availability of resources locally.

Literature review of cost-effectiveness studies for the diagnostic work-up of heavy menstrual bleeding

A systematic search was performed of the MEDLINE (from 1950 to February 2012) and EMBASE (from 1980 to February 2012) electronic bibliographic databases using the terms ‘heavy menstrual bleeding’ and ‘cost-effectiveness’ along with their MeSH terms. Three hundred and fifty articles were identified once duplicates had been removed. Three relevant economic evaluations of diagnostic tests used for evaluating HMB were identified. One evaluation took place alongside a randomised controlled trial (RCT)29 and the other two were economic modelling studies. 11,30

Cost-effectiveness was examined in a RCT conducted between 1999 and 2001 in Scotland, comparing three outpatient diagnostic tests (outpatient biopsy, ultrasound and hysteroscopy) for the evaluation of AUB in certain test combinations. 29 Women were split into high-, moderate- and low-risk groups for endometrial cancer. Resource use tended to be higher in the moderate- and low-risk women, because of the need to manage their persistent abnormal bleeding symptoms. Minimal difference in cost-effectiveness was found between investigation options in the high-risk group (post-menopausal), with the option involving hysteroscopy being marginally better than ultrasound (£88 per woman, compared with the other options). The most cost-effective investigation in the moderate-risk group was biopsy alone (saving £128–212 per woman better) and in the low-risk group ultrasound (£74–452 per woman better).

The mixed population of women with AUB, that is to say women of reproductive age with HMB and post-menopausal women with unexpected vaginal bleeding, limits clinical inferences to influence decision-making from this RCT. 29 This is because the aim of investigation of women with PMB is to exclude the possibility of endometrial cancer, whereas in pre-menopausal women it is to optimise management of benign uterine pathologies associated with HMB (i.e. selection of appropriate treatment modalities). The authors of this RCT29 highlight this themselves by stating ‘. . . in future research into the evaluation and management of AUB, postmenopausal women should be studied separately from premenopausal women with menstrual bleeding problems’ (p. 69). Furthermore, the primary end point defining ‘effectiveness’ was based upon the premise that a satisfactory diagnosis must have been reached once no further investigation had been carried out, as identified by retrospective case note review. Clearly, such an indirect assumption of effective diagnosis, while expedient, is unlikely to be a reliable or valid measure of clinical effectiveness and does not take account of patient-centred outcomes (e.g. satisfaction, reduction in bleeding, survival, etc.). Moreover, as diagnostic testing generally precedes the institution of treatments, the use of this outcome measure does not account for all treatment costs when calculating cost-effectiveness. This is important, as most women with AUB have either no identifiable pathology (‘dysfunctional uterine bleeding’) or benign pathologies (e.g. polyps or fibroids), conditions which are often amenable to less invasive, cheaper, and potentially outpatient treatments.

As well as economic data from effectiveness studies, an alternative approach to assessment of cost-effectiveness of diagnostic testing is to employ decision-analytic modelling. Two economic evaluations of diagnostic testing in HMB using decision-analytic modelling have been published. 11,30 The first of these analyses was conducted from the Dutch health-care system perspective and compared the percentage of patients treated successfully and the cost of six strategies for the evaluation of HMB: (0) hormonal treatment, (I) treatment of all patients with balloon ablation, (II) TVS and therapeutic hysteroscopy, (III) TVS, SIS and therapeutic hysteroscopy, (IV) SIS and therapeutic hysteroscopy, and (V) diagnostic hysteroscopy and therapeutic hysteroscopy. Hormonal treatment was considered to be the reference strategy with which the five strategies were compared. The study found that the strategy starting with SIS (IV) and the strategy with diagnostic hysteroscopy (V) revealed the highest number of patients treated successfully for HMB. However, the diagnostic strategy based upon initial evaluation with SIS was the most cost-effective strategy for successful treatment of HMB, especially when the prevalence of intracavity pathology [polyps or submucosal fibroids (SMFs)] was high. Study weaknesses limit to some degree the validity and stability of these findings. These included problems with construction of the decision model (limited pathologies were taken into account, e.g. diagnosis of intramural fibroids and endometrial disease were not considered). The authors used outmoded and restricted medical and surgical treatments; for example, use of long-term systemic progestogen and overlooked ambulatory outpatient-based treatment. 31 Failure rates of testing were unaccounted for, the precision and quality of data sources used for estimating test accuracy were questionable and the definition of therapeutic effectiveness was unclear. The findings of the analysis were sensitive to changes in the key assumptions limiting the robustness of clinical inferences.

The other decision-analytic model was developed to examine the cost-effectiveness of three imaging techniques, TVS, SIS and hysteroscopy, from a NHS perspective. 11 The model showed that TVS was more accurate and less costly than either SIS or hysteroscopy. For a cohort of 1000 women examined for the presence of structural abnormalities, ultrasound generated 810 correct diagnoses at a cost of £107,490 compared with 735 correct diagnoses at a cost of £145,110 using SIS and 696 correct diagnoses at a cost of £209,720 using hysteroscopy. Although the economic analysis was conducted from a NHS perspective, the general applicability of the model is limited due to its simplistic construction. Women were assumed to have one of two health states, no intrauterine pathology or any intrauterine pathology, and the outcome measure chosen was correct diagnosis. This was a pragmatic choice given the scope of the guideline11 such that it was not possible to construct a model designed to take into account the range of pathologies under consideration for HMB, and the associated range of treatment pathways. The impact on cost-effectiveness of women falsely diagnosed was not considered (the model did not follow women beyond an initial diagnosis), and so the model does not reflect the true longer-term costs and outcomes associated with each diagnostic method. Moreover, diagnosis was restricted to one test, which does not reflect contemporary practice where multiple testing is likely, either conducted simultaneously or conditional on previous test results.

The relative dearth of comprehensive diagnostic cost-effectiveness data in women with HMB reflects the complexity of care pathways (i.e. the varied outpatient tests available, the range of uterine pathologies, the relatively recent introduction of minimally invasive, ‘ambulatory’ or ‘outpatient’ treatments, and patient factors including comorbidities and preferences).

Current treatment of heavy menstrual bleeding

Defining treatment success in heavy menstrual bleeding

Menstruation is a woman’s monthly bleeding from the reproductive (vaginal) tract, as a consequence of cyclical changes in hormonal activity. It is also called menses, menstrual period or period. 11 The menstrual blood loss consists of blood and glandular tissue and fluid secretions from the inside of the uterus, which pass via the cervix into the vagina and out of the body. Menstruation is normal for women of reproductive age and so defining ‘successful treatment’ can be problematic. The primary aim of treating HMB is not to eradicate bleeding altogether, although some interventions do induce amenorrhoea, but to ameliorate bleeding symptoms to a tolerable level. As we have discussed, objective measurement of reduction in menstrual bleeding is impractical and lacks relevance. Many studies have tried to measure the impact of interventions upon patient’s quality of life and/or satisfaction with treatment outcome.

Definition of post-menopausal bleeding

The cessation of menstruation as a result of ovarian failure occurs at around the age of 52 years on average. 63 The menopause is assumed clinically once 12 months have elapsed without any further menstrual periods. There is a lack of follicular activity within the ovary and consequent absence of production of the main female sex steroid hormone and most potent oestrogen, β-oestradiol. Thus, organs and tissues responsive to oestrogen are affected, giving rise to characteristic menopausal symptoms. One such organ to be affected is the uterus and, specifically, the endometrium. Privation of oestrogen results in a quiescent, non-proliferative endometrium. The absence of ovarian activity and subsequent inactive endometrium is manifest clinically as permanent termination of menstruation and infertility. Post-menopausal bleeding is a term used in gynaecological practice to describe unexpected vaginal bleeding in a woman who has reached the menopause, taken as at least 12 months’ amenorrhoea. As described above, PMB is a common clinical problem in general practice and secondary care, associated with substantial morbidity and heavy use of health-care resources.

Urgent evaluation is recommended for PMB as in around 5% of cases the underlying cause is life-threatening endometrial malignancy. 64 Prompt diagnosis increases the chance of survival as the disease is likely to be detected at an earlier stage where curative treatment is possible. 65 Endometrial hyperplasia is considered potentially pre-malignant, especially where cytological atypia is present. The remaining causes of PMB represent a range of benign pathologies (Table 3).

Diagnosis and post-menopausal bleeding

Post-menopausal bleeding is a ‘red flag’ symptom, indicating the possibility of an underlying endometrial malignancy. Thus, the main reason for investigating PMB is to detect or exclude endometrial cancer. Current practice for the investigation of women presenting for the first time with PMB is to perform a gynaecological pelvic examination and undertake a TVS to measure the double-layer thickness of the endometrium. 66 As with the investigation of HMB, tests to evaluate the uterus include endometrial biopsy, ultrasound scan and hysteroscopy, and descriptions of these tests are provided in Table 2.

The majority of women with PMB will not have endometrial cancer. However, where it is suspected, either from the symptom alone or after testing with TVS or OPH, endometrial tissue sampling to allow histological confirmation is necessary before proceeding with treatment. This is now most commonly done by the use of endometrial biopsy (EBx), in which miniature plastic suction devices are placed into the uterine cavity in an outpatient clinic setting. This approach has largely superseded the traditional D&C. 67,68 Previous economic analysis69 did not find the routine use of EBx for all women presenting for the first time with PMB to be a cost-effective approach compared with the use of TVS. TVS is thus recommended66,69,70 as an initial test to identify higher-risk women in whom EBx could then be used in a more targeted, efficient way.

It should be noted that, in contrast to the evaluation of the uterus in HMB, SIS is not widely employed. 11 This stems from the fact that the prime purpose of diagnostic work-up of PMB is to evaluate the endometrium for serious, potentially life-threatening disease. SIS is used primarily in women of reproductive age with AUB or fertility problems after the initial TVS shows irregularity of the endometrium suggestive of a focal pathology within the uterine cavity (e.g. polyps or SMFs). However, in the diagnostic work-up of PMB, an irregular or thickened endometrium warrants an EBx and so the additional information about the uterine cavity obtained from proceeding with a SIS is not generally considered as informative as it is in the case of HMB. Furthermore, SIS is more challenging in women after the menopause because of cervical stenosis and vaginal atrophy as a consequence of chronic oestrogen deficiency.

Current diagnostic pathways for post-menopausal bleeding

While the current recommendation is to implement testing, ideally with TVS in all women presenting for the first time with PMB, an additional and often overlooked consideration relates to the role of the preliminary consultation: that is to say, history taking and clinical examination. Prevalence estimates in the UK for women presenting for the first time with PMB are of the order of 5%,64,71 although higher prevalence has been reported elsewhere. 9,72 However, while these population estimates of pre-testing probability of disease are used to inform the need for diagnostic testing, they are crude estimates. There has been recent interest in trying to individualise risk-based before instituting testing, based upon identifying the presence of certain clinical characteristics or ‘risk factors’ from the history and clinical examination. 64,73,74 Relevant individual patient characteristics include age, time since menopause, obesity, hypertension, diabetes mellitus, and reproductive factors. For example, the probability of endometrial cancer in women with PMB rises from 1% in women younger than 50 years to 23.8% in women older than 80 years and the incidence of malignancy is, regardless of age, higher in women with PMB and obesity (18%) or diabetes (21%) compared with women without one of these risk factors. 75–81

The attractiveness of integrating the clinical process with diagnostic testing is that unnecessary testing can be avoided and that the accuracy, effectiveness and, ultimately, cost-effectiveness of current management approaches can be improved upon. In light of this, we developed two multivariable prediction models using IPD to estimate the risk of endometrial cancer in patients with PMB, taking into account their clinical characteristics,73 to be used along with the established testing technologies within this economic analysis.

Literature review of cost-effectiveness of tests for the diagnostic work-up of post-menopausal bleeding

A systematic search was performed of the MEDLINE (from 1950 to February 2012) and EMBASE (from 1980 to February 2012) electronic bibliographic databases, using the terms ‘postmenopausal bleeding’ and ‘cost-effectiveness’ along with their MeSH terms. One hundred and four articles were identified once duplicates had been removed, and four of these were selected as being relevant. 29,69,70,82 One of these was the analysis performed alongside a RCT which has been appraised in the HMB section of this introduction. 29 The other three studies were based upon economic modelling. 69,70,82

The RCT which looked at cost-effectiveness compared three outpatient diagnostic tests (outpatient biopsy, ultrasound and hysteroscopy) for the evaluation of AUB in certain test combinations. 29 The effectiveness outcome measure was ‘need for no further tests’. Women were split into high-, moderate- and low-risk groups for endometrial cancer and the high-risk group comprised entirely post-menopausal women. This group of women was randomised between two investigation strategies: hysteroscopy plus biopsy or ultrasound plus biopsy. The most cost-effective combination was hysteroscopy and biopsy (£632), although ultrasound and biopsy was only £88 more expensive (£720). The study was not sufficiently powered to investigate the sensitivity and specificity of the tests for diagnosing endometrial cancer. 29 In post-menopausal women the purpose of investigation is to exclude endometrial cancer, and so, although this analysis looks at resource cost, its main limitation is that it did not examine the cost of making the correct diagnosis and the effect of this on survival.

The three economic modelling analysis were based upon three different health-care systems, with one study being from the USA,82 one from the Netherlands70 and the final one from the UK. 69

The US study82 used a decision-analytic model to evaluate six diagnostic strategies to diagnose pathology in women with peri- and post-menopausal women, with the aim of determining whether initial diagnosis with EBx or TVS minimises cost. 82 The six strategies started with EBx or TVS and added in additional testing with hysteroscopy, biopsy (if TVS was used initially) or SIS until a diagnosis was reached or bleeding was resolved. HRT was used as a treatment for 3 months if the biopsy was negative or if the TVS showed no treatable pathology with an endometrial thickness of less than 5 mm. Patients left the model if pathology was diagnosed and they had treatment or if their bleeding had resolved after 3 months of HRT. Otherwise they continued to have further sequential diagnostic tests, until a diagnosis was made or the strategy was completed. Total costs for each of the strategies were evaluated for (i) polyps, (ii) SMFs, (iii) atrophic endometrium, (iv) proliferative/secretory endometrium and (v) atypical hyperplasia/endometrial cancer. Short- and long-term patient outcomes were not included within the analyses; hence, it was a cost-minimisation study and not an evaluation of cost-effectiveness. To account for the difference in reported prevalence of endometrial hyperplasia and cancer, the strategies were performed for a range of prevalence values to determine at what level biopsy became cost-minimising. 82 They found that when the prevalence of atypical hyperplasia/endometrial cancer was between 7% and 31%, a strategy that starts with TVS and is followed by biopsy was the cheapest. Below a prevalence of 7%, however, a strategy of TVS followed by SIS, if the endometrium was thickened or bleeding did not resolve, became the cheapest. For a strategy which tests initially with EBx to be cost-efficient, the prevalence of atypical hyperplasia/endometrial cancer would have to be at least 31%. Thus, the findings of this study suggest that initial TVS investigation of patients with AUB was the cheapest method; however, they could not show whether or not it was cost-effective as survival was not used as an outcome measure. The mixed population of peri- and post-menopausal women suggests that the accuracy data used to populate the tree were not specific to post-menopausal women and, thus, the inferences cannot be reliably applied to this population.

The Dutch study70 used economic modelling to evaluate strategies for investigation of PMB. The first strategy was to have a TVS and, if the endometrium was thickened, for the patient to go on to be treated with hysterectomy. The second strategy also looked at TVS but, in this strategy, if the endometrium appeared thickened, the patient went on to have an endometrial biopsy. In the third strategy, if thickened endometrium was diagnosed by TVS, the patient underwent hysteroscopy, and in the fourth strategy patients underwent biopsy only and if the result showed atypical hyperplasia or endometrium the patient underwent hysterectomy. 70 The evaluations with TVS were examined using three cut-off levels to define an abnormal TVS: endometrial thicknesses of 3, 6 and 9 mm. Each of the four strategies was compared with a base-case strategy in which patients underwent no diagnostic testing and the outcome measure used was life expectancy. For TVS alone, using a cut-off of 9 mm was the most cost-effectives strategy at all prevalences in women aged 60. If the prevalence of endometrial cancer in the population was < 15.3%, TVS and biopsy was found to be the most cost-effective option, whereas if the prevalence was ≥ 15.3%, biopsy alone became the most cost-effective strategy. The cost-effectiveness of TVS using a high 9 mm endometrial thickness cut-off reflected the low cancer prevalence (< 15%) and the costs associated with false-positive scans associated with lower endometrial thickness thresholds. However, current guidance66 recommends, and general gynaecological practice utilises, lower endometrial thickness thresholds of 4–5 mm because of better sensitivity83,84 and fears over the implications of false-negative diagnoses.

The UK study69 was a cost-effectiveness analysis based upon an economic model which was constructed to represent clinical evaluation of women with PMB with 12 different investigation strategies, to diagnose endometrial cancer. 69 The strategies looked at EBx, TVS and OPH, used alone and in combination. TVS was assessed when endometrial thickness cut-off levels were set at 4 mm and at 5 mm. One of the 12 strategies represented women having no investigation at first presentation and only having diagnostic testing after a second episode of PMB. This strategy was used as a base case with which to compare the other strategies. The outcome measure used was cost per life-year gained. The economic model was populated using data from systematic reviews with meta-analysis and other published studies and the costs were derived from local and NHS sources. The most cost-effective strategy when compared with no treatment was using TVS with a cut-off of 5 mm for endometrial thickness, with an incremental cost-effectiveness ratio (ICER) of £11,470. When the non-dominated strategies were compared with TVS with a 5-mm cut-off, the ICERs ranged from £37,652 for the initial strategy using TVS 4 mm and £149,219 for the strategy EBx + OPH per additional life-year gained. However, these were reduced if, at the same visit, an EBx was taken following a positive result. If prevalence of endometrial cancer was 10%, EBx and TVS with a cut-off of 4 mm became more competitive options, although the additional cost over TVS with a 5-mm cut-off was still more than £20,000 per life-year gained. This study was based upon high-quality meta-analyses and focused on making a specific diagnosis (endometrial cancer) in a specific population (women with PMB).

In post-menopausal women, the primary goal of the clinician is to exclude endometrial cancer and thus prolong life. Therefore, studies that do not look at survival29,82 cannot reliably comment on the cost-effectiveness of diagnostic strategies except as a superficial assessment of the initial costs of diagnostic work-up. Despite the limitation of these studies, all of the economic analyses identified support the current recommended practice of investigating women with PMB using TVS as the initial test. 66 However, none of the studies integrated the preceding clinical process (history and examination) into the testing strategy. Important patient factors which are key in the development of endometrial cancer include obesity, parity, age, diabetes mellitus, use of hormone replacement therapy (HRT) or tamoxifen, late menopause and hypertension; these factors all increase the risk of endometrial cancer75,77,80 and some of them will also reduce life expectancy. Moreover, while the available tests themselves have not changed much over the last 10 years, advances in technology mean that the feasibility and accuracy of tests may have increased, while the relative costs of specialist equipment may have decreased. Furthermore, outpatient ‘one-stop’ testing settings have become ‘the norm’ and this method of contemporary service delivery will affect costs, satisfaction and quality of life. Therefore, any new assessment of cost-effectiveness should take the aforementioned considerations into account.

Treatment of endometrial cancer

The main aim of rapid diagnosis in cases of PMB is to diagnose endometrial cancer early and thus improve survival. Endometrial cancer can be successfully treated by surgery if it is diagnosed early and is confined to the uterus (stage 1), with 5-year survival rates of up to 90% for stage IA tumours. 85 Once the carcinoma has spread beyond the myometrium, radiotherapy and chemotherapy may be offered. 65 Women tend to present early as PMB is an alarming symptom and the majority of women will have a potentially curable stage of disease (70% present as stage 1).

Defining treatment success in post-menopausal bleeding

The prime reason for the prompt referral and immediate investigation of PMB is to detect or exclude endometrial cancer. Timely treatment can be implemented following a diagnosis of malignant endometrial disease with the aim of eradicating the disease to effect a cure. Where disease has spread beyond the confines of the uterus, treatment is palliative rather than curative. However, the timelier the diagnosis and the earlier the disease stage, the longer the survival post treatment will be. Thus, accurate diagnostic work-up allows prompt treatment, leading to increased survival. Survival is not only an appropriate, relevant measure of effectiveness but is also objectively determined. HRQL is an alternative outcome measure. In contrast to HMB, PMB is rarely heavy and persistent, and so is not as much of a physical burden. However, HRQL may be adversely impacted upon because of fear of cancer or anxiety post diagnosis. As the current evaluation of PMB investigation is focused on the diagnosis of endometrial cancer, rather than diagnoses of benign pathologies, we chose to measure effectiveness in terms of 5-year survival rather than HRQL or patient satisfaction.

Construction of decision model for the diagnosis and treatment of heavy menstrual bleeding

A clinically informed cost-effectiveness model was drawn as a decision tree using TreeAge software (TreeAge Software Inc., Williamstown, MA, USA) to reflect current service provision for the diagnostic work-up of women presenting with HMB. The tree was constructed to examine the effectiveness of different diagnostic testing strategies for women referred to secondary care by their GP. The tests evaluated were TVS, SIS, global EBx and OPH. The tree structure was informed by clinical input. As there is no consensus regarding how best to investigate women with HMB, initial investigation utilising all tests either alone or in combination were included in the model. Therefore, the tree consisted of the four tests available deployed in isolation or in various clinically relevant combinations following initial presentation. The need for any additional subsequent tests was conditional upon the preceding test result(s). This resulted in the formation of 11 clinically relevant, alternative testing strategies. In addition, two scenarios were developed where testing was dispensed with and treatment of HMB instituted immediately regardless of diagnosis. The treatments chosen were the most effective medical treatment (the LNG-IUS) and surgical treatment (hysterectomy). This allowed us to compare the various approaches to diagnostic work-up with the option of ‘no investigation’. In view of the fact that NICE guidance11 recommends the use of the LNG-IUS as first-line treatment in HMB, this arm was used as the base-case scenario to compare all other strategies against. An incremental approach was used for reporting the results. Thus, in total there were 13 different scenarios evaluated in the decision model (11 testing and two treatment-alone strategies) which are listed below:

1. LNG-IUS alone

2. hysterectomy alone

3. OPH

4. TVS

5. EBx

6. SIS

7. OPH and EBx

8. TVS and EBx

9. SIS and EBx

10. OPH and SIS

11. OPH and TVS

12. SIS, OPH and EBx

13. TVS, OPH and EBx.

Structure of the model

A series of decision trees evaluating various testing strategies for HMB were developed to represent alternative decision options and their possible consequences. The trees explicitly illustrate the patient pathway from suspected pathology underlying the clinical presentation through to the outcome of testing, distinguishing between correct and incorrect diagnosis. Then, conditional on the accuracy of the diagnostic testing strategy, the outcome of treatment for HMB was analysed at 1 year post initial presentation. Disease prevalence, diagnostic test accuracy, and treatment effectiveness along with associated costs were used to populate the relevant branches of the decision tree. The basic tree structure is illustrated in Figure 1. The 13 trees representing the diagnostic testing options for HMB are detailed in Appendix 2; however, the trees themselves are too large to display completely, so a branch of one tree has been expanded as an example and a table has been included which details the data from the remaining branches of the tree.

FIGURE 1.

Example decision tree for evaluating the cost-effectiveness of diagnostic testing in HMB.

Deterministic results were obtained using point estimates of the parameters to estimate the expected cost, outcome (satisfaction) and incremental cost-effectiveness (additional cost per extra patient satisfied). The stability of the results was then tested through sensitivity analysis.

Clinical data collection

We intended to use, where possible, IPD to populate the HMB decision tree extracted from a prospective database of over 500 women, which recorded the investigation and management of women who had presented to Birmingham Women’s Hospital (BWH) with HMB between 2004 and 2006. However, it became apparent that useful, comprehensive diagnostic and treatment data were not available. While the majority of women had undergone TVS as part of initial diagnostic work-up, further testing with OPH or EBx was sporadic (SIS was not practised) and usually undertaken in response to an abnormal TVS (i.e. limiting any assessment of accuracy data to ‘positive’ results on TVS). Furthermore, systematically collected outcome data were lacking, which precluded reliable estimation of treatment response at specific time points post therapeutic intervention. An added complexity was that the many women whose data were collected had already had multiple investigations and treatments. The scope of this project did not include time and resources for prospective data collection. However, we did use the primary IPD available to corroborate literature-derived data, especially if published data were imprecise with regard to disease prevalence and treatment choices.

All literature-derived data were obtained following systematic searches (detailed in the relevant report sections). Again, we had hoped to use IPD from published systematic quantitative reviews to facilitate estimation of test accuracy when tests were used in combination. However, published systematic reviews94–96,99 contained a small number of primary studies, of which many evaluated single tests only. Moreover, preliminary checks with study and review authors reinforced the view that pursuing test accuracy IPD (especially given the time and resources of the project) with a view to meta-analysis was futile. This was felt to be the case even assuming perfect compliance with requests for original data from primary study authors because test combination data were sparse and outdated in some instances given the length of time since publication and advances in imaging technologies. Thus, we used test accuracy data from published systematic reviews and meta-analyses where possible, followed by data derived from primary, well-conducted test accuracy studies and, finally, clinical data held at the BWH and expert clinical opinion.

For treatment data we regarded systematic quantitative reviews using IPD as the highest level of data, followed by systematic reviews of study-level data. RCTs were acknowledged as the third step down the hierarchy, followed by large comparative cohort studies and then uncontrolled observational series. Prospectively collected data from studies with large populations were considered superior to small studies and those with retrospectively collected data. When possible, we used data from a purely pre-menopausal population; however, occasionally, data came from studies of AUB incorporating both pre- and post-menopausal women. Where mixed populations of women with AUB were encountered, we aimed to stratify data by menopausal status if possible.

Results of clinical data collection

Costs

Cost values were mainly taken from Healthcare Resource Group (HRG) codes for 2009–10. 130 We used the ‘national average unit cost’ as our cost and the reported upper and lower quartile values for sensitivity analysis. The diagnostic and treatment codes for hysteroscopy include the cost of the consultation as well as any diagnostic or therapeutic procedures but the other diagnostic test codes equate to just the cost of the test. Within our decision tree, patients could undergo multiple diagnostic tests and treatments at one appointment. However, if we then used the relevant HRG code costs for each aspect we would be including the cost of the consultation multiple times. In order to give an accurate reflection of the additional costs of multiple tests including hysteroscopic treatment, we removed the cost of consultation from the diagnostic hysteroscopy costs and also for therapeutic hysteroscopy codes. We subtracted the cost of consultation and diagnostic hysteroscopy so that the value remaining was the additional cost of performing the therapeutic hysteroscopic procedure alone. We could then add up the relevant costs depending on the tests and treatments that the patients had. For example:

cost of new gynaecology consultation = £139

cost of OPH (consultation + diagnostic OPH) = £216

cost of OPH polypectomy = £263 (consultation + diagnostic OPH + polypectomy)

cost of transcervical resection of fibroid = £1344

cost of TVS (test only) = £55

cost of GnRH analogues = £226.

So the cost of the diagnostic hysteroscopy is actually £216 – £139 = £77 and the additional cost of hysteroscopic polypectomy is £263 – £216 = £47.

If a woman (assigned to the TVS diagnostic pathway) comes to clinic and has a diagnosis of ‘polyp/SMF’ made by TVS and then goes on to have removal of the lesion, the costs will equate to all the women having a consultation and a scan, half of them having an OPH polypectomy (ratio of polyps to SMFs, 50 : 50) and half of them returning at a later date for a scheduled transcervical resection of fibroid (with 70% of these women having endometrial pre-treatment with GnRH analogues) (see Chapter 2, Methods, Clinical assumptions)

However, if the HRG code was used for hysteroscopic polypectomy alone the cost would be considered as £263 or if the code for TCRF was used it would be £1344. Our derived estimate for resection of uterine polyps or SMFs of £1007.10 was felt to be reasonable and more realistic by the expert clinical panel. For the cost of LNG-IUS and GnRH analogues we used the costs reported in the British National Formulary (BNF). 131 The cost of UAE could not be identified in the HRG codes so the value used in the decision tree came from the REST study53 as reported and used by NICE in an economic analysis for treatment of fibroids in their HMB guideline. 11 Although this cost was published in 2007, the cost stated for hysterectomy was comparable with the cost stated in the HRG codes 2009–10 (£2566 vs. £2961) which are used in this economic analysis; therefore, no adjustment was made for inflation. For the cost of a GP appointment we referred to data from the Personal Social Services Research Unit (PSSRU). 132 Table 15 details all of the costs, ranges and sources of data.

Construction of decision model for the diagnosis and treatment of post-menopausal bleeding

Decision-analytic model

A decision-analytic approach was used to evaluate the outcomes of diagnostic strategies for investigating PMB. Model assumptions and input parameters were based on the earlier reported model based study by Clark et al. 69 However, in this analysis, instead of duration of survival, 5-year survival was used as the primary health outcome.

The diagnostic strategies evaluated included EBx, TVS and OPH used alone and in combination, as well as individualised strategies integrating patient characteristics with TVS and the reference case of withholding immediate investigation at initial presentation and instituting diagnostic work-up only if PMB recurred. The original analysis used D&C as the reference case. This strategy was kept in the updated analysis as an alternative reference case because it is still widely employed in the UK for investigation of PMB, albeit usually following investigative imaging. This resulted in 12 outpatient strategies for the initial clinical investigation of women with PMB for endometrial cancer. These strategies were:

• no initial evaluation

• history only

• TVS 5 mm cut-off

• selective TVS with history

• history + TVS

• EBx

• OPH

• TVS + EBx

• TVS + OPH

• EBx + OPH

• TVS + EBx + OPH

• D&C.

The constructed decision trees are detailed in Appendix 4.

The decision model133,134 was constructed to reflect current service provision. The outpatient tests available for evaluation of the uterus and endometrium have been described in Table 2. The current recommendation for investigation of PMB is to use TVS with a 5-mm endometrial thickness threshold to define for abnormality as the first-line test. 66,69,135–137 Abnormal TVS requires subsequent outpatient EBx to obtain tissue for histological analysis. 66,69,135–137 Despite these recommendations, practice remains variable for a number of reasons including available skills and clinical preference. Perhaps more importantly, technological advances, namely the miniaturisation and portability of imaging technologies, have facilitated a ‘one stop’ approach to diagnosis. This means efficient same-day testing carried out by a senior gynaecologist within a single outpatient clinic setting without the traditional dependence upon referral to other departments with the requirement for multiple appointments and delayed diagnosis. 18,138–141 Therefore, a model was developed that assessed all available tests used either alone or in combination, compared with the baseline reference option of ‘no initial testing’, that is to say awaiting re-presentation because of recurrent or persistent PMB.

The prime reason for investigating women promptly with PMB is to exclude endometrial cancer,142 which is present in between 5% and 10% of cases. Individual patient risk factors for the development of endometrial cancer have been recognised for some time and these include obesity, diabetes and advancing age. 75–81 There has been recent interest in the integration of information easily obtainable from the preceding clinical history and examination into the overall diagnostic work-up. 64,73,74 Therefore, two multivariable prediction models that had previously been developed to estimate the risk of endometrial cancer in patients with PMB, taking into account clinical characteristics,73 were integrated into the diagnostic work-up. These models were developed with data obtained from a prospective cohort study of 614 women presenting with PMB in one university hospital and seven teaching hospitals in the Netherlands. 79 These two models were used in three different diagnostic strategies for PMB to examine whether or not integration of germane clinical information with the currently recommended strategy of first-line testing with TVS66,69 improves cost-effectiveness. These integrated strategies were:73

• ‘History only’ – probability estimates based on characteristics of the women. If the probability of (pre-)malignancy exceeded 4%, EBx was performed. In this strategy TVS is not performed.

• ‘Selective TVS with history’ – recourse to TVS if the probability of endometrial cancer, based on characteristics of the women, exceeded 4%. EBx is then performed if the TVS-derived endometrial thickness exceeded 4 mm.

• ‘History and TVS’ – all women have a TVS and the probability of cancer estimate is generated based on both characteristics of the women and the TVS results. EBx is performed when the combined TVS and history probability of cancer exceeds 4%.

These models have been externally validated using two prospectively collected databases:

• Breijer database: between January 2009 and April 2011, all women presenting with PMB at the TweeSteden Hospital, Tilburg, the Maxima Medical Center, Veldhoven, and the St. Antonius Hospital, Nieuwegein, in the Netherlands, were included in a prospective study. Women with a history of hysterectomy were excluded. Age, body mass index (BMI), parity, years since menopause, HRT, presence of hypertension, diabetes, use of anticoagulants and endometrial thickness measured by TVS were recorded. If double endometrial thickness was > 4 mm, EBx was performed.

• Valentin database: between November 2002 and June 2009, all women presenting with PMB at the Skåne University Hospital, Malmö, Sweden, at the PMB clinic, were included in a prospective cohort study. Age, age at menopause, parity, HRT, weight, height, hypertension, diabetes, current use of anticoagulants and endometrial thickness measured by TVS were recorded. If double endometrial thickness was ≥ 4.5 mm, EBx was taken.

Patients from both databases were asked to contact the hospital if they had further episodes of bleeding. The patients from the Breijer database were followed up by collecting data from patient records, and, in the Valentin database, cases were matched to the national cancer register. For the purpose of the study, all patients with a thin endometrium without EBx and without recurrent bleeding were considered negative for endometrial cancer.

The ability of the models to discriminate between women with and without endometrial cancer was assessed for both the ‘patient characteristics’ (history only) and the ‘patient characteristics and TVS’ (the two TVS strategies) models. The two samples available for validation consisted of 413 and 622 non-HRT-using patients, respectively. Table 16 shows the characteristics of patients in the two databases. Age, time since menopause, anticoagulant use and BMI were significantly different between the two validation populations.

TABLE 16 - Patient characteristics in the two validation databases

SD, standard deviation.

Independent-samples t-test.

Chi-squared test.

Not normally distributed, values presented as median and interquartile range.

Kruskal–Wallis.

Characteristic	Valentin (n = 413) (SD)	Breijer (n = 622) (SD)	p-value
Age (years)	67.0 ± 12.1	62.2 ± 10.2	< 0.01a
Diabetes mellitus	64 (15.5)	85 (13.7)	0.42b
Hypertension	164 (39.7)	227 (36.5)	0.30b
Use of anticoagulants	76 (31.9)	114 (18.3)	< 0.01b
BMI (kg/m2)	27.9 ± 6.6	29.9 ± 7.9	< 0.01a
Time since menopause (years)c	15 (5–27)	6 (2–14)	< 0.01d
Nulliparity	49 (12.0)	69 (13.9)	0.40b
Endometrial thickness (mm)c	5.0 (3.1–12.1)	5.7 (2.5–10.0)	0.05d
Endometrial cancer	54 (13.1)	75 (12.1)	0.63b

Figure 2 shows the receiver operating characteristic (ROC) curves for the two prediction models in both validation data sets. The area under the ROC curve for the patient characteristics only model was 0.68 (95% CI 0.62 to 0.75) and 0.69 (95% CI 0.64 to 0.74) in the Breijer and Valentin populations, respectively, and the area under the ROC curve for the patient characteristics and TVS model was 0.87 (95% CI 0.84 to 0.90) and 0.89 (95% CI 0.85 to 0.91), respectively, in the Breijer and Valentin populations. As a reference, Figure 2c shows the ROC curve of the two models in the development database. The results show that the existing multivariable models maintained their diagnostic accuracy in two independent patient cohorts.

FIGURE 2.

Receiver operator characteristic curves of the patient characteristics only model and the patient characteristics and TVS model for PMB. (a) Validation database I. Breijer; (b) validation database II. Valentin; and (c) development database Van Doorn.

Decision Analysis TreeAge Pro 2009 Suite software (TreeAge Software Inc., Williamstown, MA, USA: www.treeage.com) was used to specify the decision model.

Data sources and modelling assumptions for decision analysis

The modelling assumptions were as previously reported in an earlier analysis. 69 The assumptions of the expert panel consulted at that time were assumed to still be valid and so they were not readdressed by the new panel of gynaecologists who had advised on the HMB model. It was assumed that the hypothetical presentation with PMB represented the first episode. No post-menopausal woman was assumed to be less than 45 years old and no other significant aetiology (e.g. other genital tract malignancy) was considered. The woman was considered to be otherwise healthy with a normal age-adjusted life expectancy. The probability of endometrial cancer in women presenting with PMB is between 5% and 10%. 64,74 A 5% prevalence of malignant disease was used for the base-case analysis.

All assumptions regarding the clinical setting and care pathways were the same as for the HMB analysis.

Diagnostic tests

Data for failure rates and estimates of diagnostic accuracy were obtained from high-quality published systematic quantitative reviews of the diagnostic literature for EBx, USS and OPH21,84,95 (Table 17). Failure rates for initial strategies utilising test combinations were estimated by the consensus panel based on the definition of a failed strategy as any test making up the strategy failing and on available failure rate data from individual tests. 21,84,95,143 Similarly, failure rates were also adjusted for tests performed in a diagnostic strategy conditional on the success of preceding tests. 64,144 As over 95% of women with endometrial cancer present with PMB,64 it was assumed that all women who were erroneously discharged following the initial presentation (i.e. false negatives) remained symptomatic. The interval to re-presentation was thus taken to be short, and all these women were then assumed to undergo reinvestigation with all outpatient tests where perfect test success and accuracy was assumed.

TABLE 17 - Estimated base-case values, ranges and data sources for clinical model input parameters for the economic analysis of PMB

EP, expert panel; NR, narrative review; PL, published literature; SR, systematic review; TPR, true-positive rate.

Range.

Variable	Baseline	Sensitivity analysis (95% CI)	Source
Failure rates
EBx	0.12	0.09 to 0.15	SR21
TVS	0.0	0.0 to 0.02	SR84
OPH	0.05	0.04 to 0.07	SR95
Ultrasound scan + OPH	0.04	0.03 to 0.06	EP
Ultrasound scan + EBx	0.12	0.09 to 0.17	EP
Ultrasound scan + EBx+ OPH	0.12	0.09 to 0.17	EP
EBx after successful OPH	0.07	0.05 to 0.10	EP
EBx after successful ultrasound scan	0.12	0.09 to 0.15	EP
Complication rates
Outpatient diagnostic procedures (EBx, TVS, OPH)	0	0	SRs21,84,95,143
TPRs
EBx	0.94	0.84 to 0.99	SR21
TVS scan 5 mm	0.97	0.94 to 0.98	SR84
OPH	0.86	0.8 to 0.89	SR95
D&C	0.96	0.82 to 1.00	EP
History only	0.99	0.97 to 1.00	Meta-regression73
History and TVS	0.99	0.96 to 1.00	Meta-regression73
Conditional TPRs
EBx if OPH positive	0.94	0.93 to 0.97	EP
EBx if ultrasound positive	0.94	0.94 to 0.95	EP
EBx if history only positive	0.94	0.93 to 0.97	Meta-regression73
EBx if history and TVS positive	0.94	0.94 to 0.95	Meta-regression73
OPH if EBx negative	0.86	0.83 to 0.87	EP
OPH if ultrasound positive	0.86	0.86 to 0.87	EP
Ultrasound scan 5 mm if EBx negative	0.97	0.80 to 0.99	EP
Ultrasound scan 5 mm if OPH negative	0.97	0.91 to 0.99	EP
FPRs
EBx	0.01	0.0 to 0.02	SR21
Ultrasound scan 5 mm	0.45	0.43 to 0.47	SR143
OPH	0.01	0.0 to 0.06	SR95
D&C	0.01	0.0 to 0.03	EP
History only	0.80	0.72 to 0.87	Meta-regression73
History and TVS	0.41	0.39 to 0.44	Meta-regression73
Probability of stage II–IV (re-presentation)	0.35	0.3 to 0.6a	EP
Prevalence	0.05	0.03 to 0.10	PL64
Surgical stage at hysterectomy (FIGO)
Probability of stage I (first presentation)	0.7	0.6 to 0.8a	FIGO85
Probability of stage II–IV (first presentation)	0.3	0.2 to 0.4a	FIGO85
Probability of stage I (re-presentation)	0.65	0.4 to 0.7a	EP
5-year survival rates
Stage I	0.89		FIGO85
Stage II–IV	0.54		FIGO85

No serious complications were assumed to be associated with any of the ambulatory procedures (ultrasound, hysteroscopy and EBx) based on evidence from systematic reviews of the available literature. 21,84,95,143 Mortality rates were assumed to be negligible for all the diagnostic tests. 21,84,95,143

Treatment

For the base-case analysis, it was assumed that all women not discharged underwent initial treatment by TAH and bilateral salpingo-oophorectomy with or without pelvic node sampling (i.e. all were fit for surgery and none had primary radical radiotherapy). All women were therefore assumed to be surgically staged. 85 There is some variation in practice in the treatment of endometrial cancer regarding the relative roles of surgery and radiotherapy/chemotherapy. 85,145 The treatment pathways in this model were based on published recommendations and reports of current practice. 65,146 All epidemiological statistics relating to endometrial cancer were taken from the FIGO results of treatments of gynaecological cancers. 85 For the base-case analysis, the cost of treating a woman correctly diagnosed with endometrial cancer on first presentation was based on the assumption that 70% of such women had localised (FIGO stage I) disease and 30% advanced (FIGO stages II–IV) disease. 85 To account for delayed diagnosis experienced by women with endometrial cancer who were erroneously discharged initially (false negatives) it was estimated that this group of women had a 5% increased probability of advanced-stage endometrial cancer (stage II–IV) in the absence of relevant data (see Table 17). Those with advanced disease (stages II, III or IV) underwent radiotherapy (adjuvant/palliative) and/or chemotherapy. 65 Women with stage IC disease or poorly differentiated (histological grade 3) stage IA or IB disease were assumed to have adjuvant radiotherapy. 65 The proportion of women undergoing additional non-surgical treatment is shown in Figure 3.

FIGURE 3.

Decision-analytic model (common pathway for further treatment of endometrial cancer following initial hysterectomy).

Standardised radiotherapy and chemotherapy regimens were assumed regardless of disease stage; radiotherapy consisted of a 5-week course of external beam radiotherapy giving a total of 25 fractions. Chemotherapy consisted of standard cytotoxic cycles. 65,146 Compliance with treatment was assumed to be 100%. The 5-year survival rates were assumed to be 89% for stage I disease and 54% for advanced (stage II–IV) disease. 85

Methods for the cost-effectiveness analysis

For each type of analysis, a decision tree was constructed using TreeAge Pro 2009. Model inputs included probabilities and costs. Effectiveness was measured in terms of positive outcomes: patient satisfaction for HMB and 5-year survival for PMB. Therefore, each branch of the tree had an effectiveness outcome of 1 for a positive outcome and 0 if otherwise. Costs were unit costs for the various tests and treatments and were therefore treated as known with certainty, while probabilities depended on data and were treated as uncertain, to be varied in probabilistic sensitivity analysis (PSA). Note that the uncertainty in overall effectiveness for a given strategy is fully accounted for in the uncertainty in the probabilities, so there is no need to vary the outcome parameters.

The model was first run using the point estimates of the branch probabilities. The results, known as ‘base case’ results, are shown in terms of mean costs and effectiveness (overall proportion of positive outcomes) for each strategy modelled. These are tabulated and shown in a cost-effectiveness plane, with the mean cost shown on the vertical axis and the mean effectiveness shown on the horizontal axis. In some cases, a further plot was made of selected strategies to show more clearly the relationship between points that were close together on the first graph.

Any strategy which has greater cost and worse effectiveness than some other strategy is said to be simply dominated. Such a strategy can be excluded from consideration. An ICER can be calculated between any two non-dominated strategies. The ICER is calculated by dividing the difference in cost by the difference in effectiveness. If the ICER is less than the maximum willingness to pay (WTP) for an additional positive outcome, then the more effective strategy can be said to be cost-effective relative to the other strategy.

There is a further reason for excluding strategies, known as extended dominance. This can apply only when there are three or more non-dominated strategies. In this case, two different strategies (incorporating a cheaper but less effective strategy and a more effective, more costly strategy relative to a third strategy) can be mixed together, with a proportion of patients getting one or other of the strategies such that the third strategy now becomes dominated (i.e. is more expensive and less effective) than the blended strategies. Suppose that A, B and C are non-dominated strategies in order of increasing cost. As they are non-dominated, they must also be in order of increasing effectiveness. Now suppose that the ICER of B over A is higher than the ICER of C over A. Then, if B is cost-effective compared with A, so also must C be cost-effective compared with A and B. In such a case, there is no value of WTP per positive outcome at which B will be the preferred strategy, and the strategy B can be excluded by extended dominance. Extended dominance can be seen on a cost-effectiveness plane. The point for strategy B will be above the straight line joining the points for strategies A and C.

Once all dominated strategies have been excluded, whether for simple or extended dominance, the remaining strategies are potentially cost-effective. Which will be preferred depends on the WTP for an additional positive outcome.

To test for the effect of uncertainty in the model inputs, two types of sensitivity analysis were used: probabilistic and deterministic.

In PSA, probability distributions are placed around the point estimate for each model parameter. If there is correlation between the uncertainties, joint or conditional distributions may be used. For the models in this report, beta distributions were used to represent the uncertainty around the branch probabilities. The beta distribution is the standard distribution for a proportion. It has two parameters, a and b, with mean a/(a + b) and variance essentially decreasing as a and b increase.

For individual parameters of the models, the information available was in the form of a point estimate and a 95% CI. In all cases, the beta distribution was selected with mean equal to the point estimate. Usually the distribution also matched the width of the CI, but there were two main exceptions to this:

• If the point estimate is either 0 or 1, then it is not possible to find a beta distribution with that mean. While it can be argued that the true mean estimate of the probability should be strictly between 0 and 1, taking any actual numerical value would risk overcompensating. Accordingly, it was decided to treat such probabilities as fixed, thereby preserving the mean but slightly underestimating the uncertainty in the model.

• If either parameter a or b is less than 2, the beta distribution gives an unreasonably high proportion of values very close to the extreme values 0 or 1. To avoid this, in such cases the values of a and b were increased to preserve the mean of the distribution but ensure that both values were at least 2. Again, this slightly underestimates the overall uncertainty.

In some cases, only a point estimate was available. In such cases, it is not appropriate to assume that the value is fixed. Instead, the widest possible uncertainty was modelled subject to the constraint that both parameters a and b should be at least 2, for the reasons given in the previous paragraph.

In the HMB analysis of SIS for detecting fibroids, point estimates of true-positive rates (TPRs) and FPRs were assumed to be the same as for TVS. In these cases, independent samples were taken from beta distributions with the same parameters, a and b.

As the costs in the model are all unit costs of specific procedures, these were treated as fixed, and the only parameters to be varied were the probabilities in the tree, which are proportions of patients expected to follow each branch.

When the models were run for PSA, 1000 replications were made, sampling from distributions for all branch probabilities simultaneously. It is generally accepted that 1000 replications are sufficient to give a clear picture of the uncertainty.

The parameters for this beta distribution used for the PSA of the HMB tree are shown in Table 19 and the parameters for the PMB tree are shown in Table 20.

For models (such as Markov models) in which there is a non-linear relationship between model inputs and outputs, it is appropriate to give a table of mean results from the PSA, as the Bayesian viewpoint is that the mean results from the PSA are the appropriate basis for decision-making. However, in the case of the models presented here, these results would be statistically equivalent to the base-case results and, therefore, there is no need to produce such a table. Results that have been shown are as follows:

A cost-effectiveness scattergraph: this shows, on a single graph, the uncertainty in the absolute expected cost and effectiveness for each option separately. For each option, the results of the 1000 replications of the model were shown each as a single point. In practice, the printed symbols used merge to form a ‘cloud’ giving the general range of uncertainty in the results. The vertical spread of this cloud reflects the uncertainty in the overall cost and the horizontal spread the uncertainty in overall effectiveness, while the centre of the cloud, where the points are most densely packed, indicates the most likely cost and effectiveness.

A cost-effectiveness acceptability frontier (CEAF): at any given WTP, the preferred option is determined by the mean outcomes, which in the case of the models here are the same as the base-case results described earlier. The CEAF shows the proportion of model replications for which this option remained the preferred strategy, as a function of the WTP for an additional positive outcome.

While the graphs described above are the only convenient ways of showing results comparing all the modelled options, they are applicable only to a decision in which exactly those options are included. For other purposes, it is helpful to look at pairwise comparisons between strategies. The results shown from a pairwise comparison are helpful to any decision problem in which both those strategies are included. Pairwise comparisons were made between successive non-dominated options (in order of increasing cost or effectiveness), and others where a dominated option was close to another option.

For pairwise comparisons, the incremental cost-effectiveness scattergraph was shown. In this type of graph, there is a single point for each of the 1000 replications of the model, showing the difference in cost and effectiveness between the two strategies. If the ‘clouds’ shown in the cost-effectiveness scattergraph for the two strategies overlap, it may be for one of two reasons. First, it may be because there is genuine uncertainty as to which is the more costly and/or more effective strategy. Second, it may be that one strategy is consistently more costly and/or more effective than the other, but this consistent difference is small compared with the uncertainty in the absolute costs and/or effectiveness. The incremental cost-effectiveness scattergraph distinguishes between these two cases and shows the relevant uncertainty for a decision-maker.

The other graph plotted for the pairwise comparisons was the cost-effectiveness acceptability curve (CEAC). This shows the proportion of model replications in which one of the strategies is cost-effective compared with the other, across a range of values of WTP per additional positive outcome.

In deterministic sensitivity analysis, one or more model inputs are varied systematically and the effect on the model outcomes is noted. This was used in the HMB analysis to test the effect of reducing the prevalence of polyps and SMFs and increasing the prevalence of DUB and also to examine the effect when the unit cost of SIS was reduced. For prevalence of the various pathologies, a Dirichlet distribution was used. This is the generalisation of the beta distribution for more than two options. Given that the prevalence data came from different sources, it was necessary to take a compromise between the effective sample sizes indicated by those sources. An effective sample size of 100 was assumed. The distribution of any prevalence parameter on its own then follows a beta distribution.

Outcomes

Direct treatment without preliminary diagnostic testing was less clinically effective than treatment instigated after diagnostic testing. The least effective approach was the base-case strategy of LNG-IUS treatment alone, followed closely by surgical treatment with hysterectomy, both approaches resulting in satisfaction rates of around 93.3–93.4%. The effectiveness of HMB management was similar across all testing strategies, ranging from 94.6% to 96.7% rates of satisfaction. The most effective strategy was combination testing with OPH and EBx.

Costs

The LNG-IUS treatment-alone base-case strategy was the cheapest, costing £1067 per woman treated for HMB in a secondary care setting, and the strategy of hysterectomy for all women bypassing the need for diagnostic work-up was the most expensive, at £3182 per woman treated, that is to say £2116 more than the approach of LNG-IUS treatment alone. The cheapest of the nine diagnostic testing strategies was the use of OPH alone, costing £1078 for every woman treated, that is to say £11 more than universal LNG-IUS treatment.

Cost-effectiveness and dominance

Only the testing strategies OPH alone and OPH combined with EBx (OPH + EBx) remain non-dominated by alternative empirical treatment or diagnostic testing strategies. It is clear from our analysis that the strategy OPH alone dominates the testing strategies SIS alone and TVS alone. The combination testing strategy TVS + OPH is excluded by extended dominance between OPH alone and OPH + EBx. The remaining seven alternative strategies are dominated by OPH + EBx. Table 22 presents the deterministic analysis restricted to the non-dominated competing strategies.

Thus, the cheapest strategy is the base-case scenario of no testing and universal treatment with the LNG- IUS alone of all women presenting to secondary care with HMB. The most effective strategy is the combination of initial testing with OPH + EBx, but this comes at a greater cost, generating an ICER of £21,500, that is to say the strategy requires an investment of £21,500 to gain an extra woman satisfied following treatment for HMB compared with investigation with OPH alone. When compared with blanket LNG-IUS treatment for all, an additional £21,859 is required to gain an additional satisfied patient when investigating with OPH + EBx. The single-test strategy of OPH is slightly less effective than the strategy of OPH with the addition of EBx, but is substantially less costly. The ICER for OPH is approximately £360, that is to say an additional financial outlay of £360 is necessary to acquire an extra woman satisfied following treatment for HMB.

Figure 4 shows the total costs and effectiveness of alternative strategies for the diagnosis and treatment of HMB in secondary care. The line presented graphically joins the non-dominated strategies (OPH alone and OPH + EBx). Any strategy plotted above this line is not considered cost-effective in relation to the non-dominated alternatives.

FIGURE 4.

Cost-effectiveness plane showing the results of deterministic analysis of the strategies for investigation of women with HMB.

It is clear that the option of direct treatment with hysterectomy is the least cost-effective strategy by some considerable margin. Replicating the figure excluding the hysterectomy reduces the scale of the y-axis, allowing closer examination of the remaining testing strategies. Figure 5 reveals that the options ‘TVS alone’ and ‘SIS alone’ are sufficiently close to the boundary of dominance that it is worth checking for the uncertainty between these dominated alternatives in addition to the non-dominated options ‘OPH’ and ‘OPH + EBx’ through sensitivity analyses.

FIGURE 5.

Cost-effectiveness plane showing the results of deterministic analysis of the strategies for investigation of women with HMB – hysterectomy excluded.

Outpatient hysteroscopy compared with levonorgestrel-releasing intrauterine system

The cost-effectiveness plane (Figure 8a) shows the modelled uncertainty in the difference in costs between OPH alone and LNG-IUS alone. It shows that OPH alone is consistently more effective than LNG-IUS alone, and is likely (but not certain) to be more costly. The CEAC (Figure 8b) shows the proportion of model replications for which OPH alone is preferred to LNG-IUS alone at any given WTP. OPH is the preferred option at any WTP over £360 per additional case satisfied, but there is considerable uncertainty when the WTP is just above this figure. However, by the time the WTP exceeds £8000 per additional case satisfied, it is almost certain that OPH is preferred to LNG-IUS.

FIGURE 8.

Cost-effectiveness plane (a) and CEAC (b): OPH-alone strategy relative to the LNG-IUS-alone strategy.

Outpatient hysteroscopy and endometrial biopsy compared with outpatient hysteroscopy alone

The graph (Figure 9a) shows the modelled uncertainty in the difference in costs between OPH + EBx and OPH alone. It shows that adding EBx to OPH increases the cost and is very likely to increase the effectiveness. The CEAC (Figure 9b) shows the proportion of model replications for which OPH + EBx is preferred to OPH alone at any given WTP per additional case satisfied. It is more likely than not that OPH + EBx is cost-effective compared with OPH above a WTP threshold of around £23,000. However, there is considerable uncertainty throughout the range of WTP values shown. About 30% of replications favour OPH alone, even if the WTP is as high as £40,000 per additional case satisfied.

FIGURE 9.

Cost-effectiveness plane (a) and CEAC (b): OPH and EBx strategy relative to the OPH-alone strategy.

Assessment of potentially cost-effective competing strategies

In addition to the non-dominated testing strategies ‘OPH alone’ and ‘OPH + EBx’, the single testing strategies TVS and SIS were sufficiently close to the boundary of dominance that we felt it prudent to explore the level of uncertainty within the model pertaining to these two dominated alternative options.

Transvaginal scan compared with levonorgestrel-releasing intrauterine system

The graph (Figure 10a) shows the modelled uncertainty in the difference in costs between TVS alone and LNG-IUS alone. It shows that TVS alone is consistently more effective than LNG-IUS alone, and is likely (but not certain) to be more costly. The CEAC (Figure 10b) shows the proportion of model replications for which TVS alone is preferred to LNG-IUS alone at any given WTP per additional case satisfied. It is more likely than not that TVS is cost-effective compared with LNG-IUS above a WTP threshold of around £1000. This is almost certainly the case at WTP thresholds beyond £9000.

FIGURE 10.

Cost-effectiveness plane (a) and CEAC (b): TVS-alone strategy relative to the LNG-IUS-alone strategy.

Saline infusion sonography compared with levonorgestrel-releasing intrauterine system

The graph (Figure 11a) shows the modelled uncertainty in the difference in costs between SIS alone and LNG-IUS alone. It shows that SIS alone is consistently more effective than LNG-IUS alone, and is likely (but not certain) to be more costly. The CEAC (Figure 11b) shows the proportion of model replications for which SIS alone is preferred to LNG-IUS alone at any given WTP per additional case satisfied. The likelihood is that SIS is cost-effective compared with LNG-IUS above a WTP threshold of around £800. This is almost certainly the case at WTP thresholds beyond £8000.

FIGURE 11.

Cost-effectiveness plane (a) and CEAC (b): SIS-alone strategy relative to the LNG-IUS-alone strategy.

Outpatient hysteroscopy compared with transvaginal scan

The graph (Figure 12a) shows the modelled uncertainty in the difference in costs between OPH alone and TVS alone. It shows that OPH alone is almost certainly more effective than TVS alone but it is unclear whether or not it is more costly. The CEAC (Figure 12b) shows the proportion of model replications for which OPH alone is preferred to TVS alone at any given WTP per additional case satisfied. The likelihood is that OPH is cost-effective compared with TVS above any WTP threshold. This is almost certainly the case at WTP thresholds beyond £9000.

FIGURE 12.

Cost-effectiveness plane (a) and CEAC (b): OPH-alone strategy relative to the TVS-alone strategy.

Outpatient hysteroscopy compared with saline infusion sonography

The graph (Figure 13a) shows the modelled uncertainty in the difference in costs between OPH alone and SIS alone. It shows that OPH alone is likely to be more effective than SIS alone, and there is considerable uncertainty as to which is more costly. The CEAC (Figure 13b) shows the proportion of model replications for which OPH alone is preferred to SIS alone at any given WTP per additional case satisfied. The likelihood is that OPH is cost-effective compared with TVS above any WTP threshold, although there is considerable uncertainty throughout. Even at a WTP of £40,000 per additional case satisfied, SIS is preferred in 20% of model replications.

FIGURE 13.

Cost-effectiveness plane (a) and CEAC (b): OPH-alone strategy relative to the SIS-alone strategy.

Assessment of the impact of reducing the cost of saline infusion sonography

The probabilistic sensitivity analyses show uncertainty around whether OPH or SIS is the most cost-effective investigative strategy. To assess this uncertainty further, deterministic sensitivity analysis was performed to reduce the cost of SIS and determine at what cost it would become more cost-effective than OPH. Table 23 details the ICER values when the cost of SIS is reduced.

In this analysis, the unit cost for SIS was reduced from £71 (base-case cost), keeping all other variables fixed. When the cost was reduced to £65, the strategy ‘SIS alone’ was no longer dominated by ‘OPH alone’. However, the modelled ICER was £76 per additional case satisfied, suggesting that OPH alone is still highly cost-effective compared with SIS alone. As the cost of SIS reduces further, the ICER increases (Figure 14). Considering an illustrative WTP of £10,000 per case satisfied, the ICER goes above this figure when the cost of SIS drops to £52. In that case, OPH is no longer cost-effective compared with SIS, and SIS becomes the preferred strategy (of the two) on cost-effectiveness grounds. It should also be noted that at a unit cost for SIS of £53 or lower, the strategy ‘SIS alone’ becomes less costly, as well as remaining more effective, than ‘LNG-IUS alone’.

FIGURE 14.

Incremental cost-effectiveness ratios between OPH and SIS when the cost of SIS is varied.

Assessment of the impact of the prevalence of focal uterine pathology

A high prevalence of intracavity focal endometrial lesions will favour OPH over the imaging technologies TVS and SIS because it is more likely to diagnose the lesions and treatment can be initiated with only a small additional cost during the diagnostic procedure (the so-called ‘see and treat’ approach). If the prevalence of endometrial polyps and SMFs is overestimated within the decision tree, OPH will falsely appear the most cost-effective. Similarly, the prevalence of DUB may be an underestimate. To assess the effect of prevalence on the analysis, the prevalence of polyps/SMFs was varied, keeping the prevalence of fibroids and endometrial disease fixed. The prevalence of DUB was changed inversely to compensate for the change in prevalence of polyp or SMF. All other variables in the model were fixed at their point estimates. Figure 15 shows the preferred option at a range of values of WTP per additional case satisfied, varying the prevalence of polyp or SMF. For example, at a prevalence of 30%, the combination of OPH and EBx is preferred if the WTP per additional case satisfied is more than about £27,000, while OPH alone is preferred if this WTP is between £2000 and £27,000. Only at a WTP below £2000 per case satisfied is LNG-IUS alone preferred. For this prevalence, other options are dominated and so not preferred at any WTP value.

FIGURE 15.

Incremental cost-effectiveness ratios between non-dominated options at a variety of prevalence values for polyps/SMFs and DUB.

Outcomes

As with the base-case analysis, the ‘no investigation’ strategies were the least effective strategies for managing women. The most effective strategy for investigating women using a multiple clinic attendance as opposed to a ‘one-stop’ approach was the combination of OPH and EBx.

Costs

Costs for the ‘no investigation’ strategies (LNG-IUS alone and hysterectomy alone) remained unchanged. The costs of the investigative strategies, however, increased due to the additional appointments required, with the costs for the investigation strategies ranging from £1204 to £1418 in this alternative analysis compared with £1078 to £1256 in the base-case analysis.

Cost-effectiveness and dominance

The strategies ‘EBx alone’, ‘TVS + EBx’, ‘OPH alone’, ‘SIS + OPH’, ‘TVS + OPH + EBx’, ‘SIS + OPH + EBx’, and ‘hysterectomy alone’ are excluded by simple dominance and the strategies ‘TVS alone’, ‘SIS + EBx’ and ‘TVS + OPH’ are excluded by extended dominance. The remaining three strategies are not dominated. In contrast to the base-case analysis the strategy SIS alone is no longer dominated, whereas ‘OPH alone’ is. Table 25 displays the non-dominated strategies from deterministic analysis.

Using SIS to investigate women in this strategy costs an additional £5070 to make an extra woman satisfied compared with not investigating and giving all women a LNG-IUS. OPH + EBx costs an additional £22,100 to the cost of SIS to gain a further satisfied patient. OPH alone does not appear as a non-dominated option in this analysis. The line on the cost-effectiveness plane in Figure 16 links the non-dominated strategies.

FIGURE 16.

Cost-effectiveness plane showing the results of deterministic analysis for strategies to investigate women presenting with HMB managed over multiple clinic appointments (hysterectomy removed).

Outcomes

Direct treatment without preliminary diagnostic testing was less effective than treatment instigated after diagnostic testing. The most effective strategy was combination testing with TVS and EBx; however, the difference between all of the diagnostic strategies was minimal, ranging from 96.28% to 96.39%.

Costs

The results presented in Table 26 show that all costs have increased compared with the base-case analysis, reflecting the increased prevalence of organic uterine pathology requiring more expensive treatments. Adopting no further treatment and persevering with the LNG-IUS treatment alone (reference strategy) was the cheapest option, costing £1355 per woman treated for HMB in a secondary care setting. The strategy of hysterectomy for all women bypassing the need for diagnostic work-up was the most expensive at £3218 per woman treated, that is to say £1863 more than the approach of LNG-IUS treatment alone. The cheapest diagnostic testing strategy was the use of OPH alone, costing £1681 for every woman treated, that is to say £326 more than continuation with LNG-IUS treatment.

Cost-effectiveness and dominance

The testing strategies OPH alone and TVS combined with EBx (OPH + EBx) remain non-dominated by alternative empirical treatment or diagnostic testing strategies. All of the remaining strategies are dominated by OPH alone except for hysterectomy, which is dominated by TVS and EBx. Table 27 presents the deterministic analysis restricted to the non-dominated competing strategies.

The line on the graph in Figure 19 joins the non-dominated strategies, LNG-IUS only, OPH alone and TVS + EBx. When LNG-IUS and hysterectomy are removed the relationship of the other strategies to the line of non-dominance is clearer (Figure 20).

FIGURE 19.

Cost-effectiveness plane showing the results of deterministic analysis for strategies to investigate women with HMB with a LNG-IUS in situ.

FIGURE 20.

Cost-effectiveness plane showing the results of deterministic analysis for strategies to investigate women with HMB with a LNG-IUS in situ (hysterectomy alone and LNG-IUS alone not shown).

Outcomes

Satisfaction rates are reduced when compared with the original analysis, as optimal surgical interventions precluding future fertility are not available for women with large fibroids or those resistant to the LNG-IUS. This is reflected by the lower satisfaction rate of 65.6% for this analysis compared with 93.33% in the original when women receive the LNG-IUS without any investigation. Satisfaction rates for the investigative strategies range from 74.19% for EBx to 86.49% for SIS + OPH. There is greater variation between the satisfaction rates in this analysis than in the base case, when values varied marginally between 94.6% and 96.7%.

Costs

It should be noted that the costs decreased when compared with the original analysis. Cost is decreased because the more expensive treatments tend to be the surgical options (EA and hysterectomy) which are contraindicated in women desiring preservation of their fertility. The cost of LNG-IUS alone has decreased to £421 from £1066 in the base-case analysis because the women identified to have large fibroids do not undergo a hysterectomy and women who are dissatisfied with the LNG-IUS cannot be offered any further treatment. The cheapest investigative strategy in this analysis is TVS alone, costing £740 per patient. The most expensive strategy is the combination of SIS, OPH and EBx, costing £1003 per patient.

Cost-effectiveness and dominance

The strategy of EBx alone is dominated by TVS alone (i.e. this is a cheaper and more effective option), which in turn is dominated by extended dominance due to a blend of LNG-IUS alone and SIS alone. The remaining strategies are dominated by either OPH alone or SIS and OPH together. Once the dominated strategies were removed, the testing strategies which remained were SIS alone, OPH alone, and SIS and OPH together. This can be more clearly appreciated in Table 29. In contrast to the base-case analysis, a combination strategy of OPH + EBx is not potentially cost-effective. Moreover, SIS alone or in combination with OPH is non-dominated, whereas in women without the need to preserve their fertility (base case), SIS and related strategies were not cost-effective.

The line on the cost-effectiveness plane (Figure 23) joins the non-dominated strategies, starting with the base case of LNG-IUS which joins to SIS, followed by OPH and then SIS + OPH. TVS lies close to this line and, therefore, when exploring the results, TVS was included to see whether or not analysing the spread of results might suggest that TVS could become cost-effective when values are varied around their point estimates.

FIGURE 23.

Total costs and effectiveness of the alternative strategies for the diagnostic work up of HMB for women wishing to preserve their fertility, excluding the strategy of hysterectomy.

Outcomes

Some form of patient evaluation in PMB, whether this be confined to obtaining germane patient information from clinical history taking alone, employing testing modalities or some combination of both approaches, was more effective than the reference case of no initial testing at all. In contrast to costs, effectiveness in terms of 5-year survival rates was similar regardless of investigative strategy adopted. A decision to undertake no testing on initial presentation, representing the least effective approach with PMB, was associated with a 5-year survival rate of 98.77%. EBx was the least effective investigative strategy with an associated 5-year survival rate of 98.84%, whereas the most effective testing strategy, ‘TVS + OPH’, was associated with a 5-year survival rate of 98.85%. Thus, only small improvements in survival were observed regardless of strategy adopted and this reflects the relatively small incidence of endometrial cancer (5%), a malignancy associated with a predominantly early stage at diagnosis amenable to curative therapy in many.

The acquisition of patient characteristics from the clinical history was more effective if used to identify higher risk women for TVS (‘selective TVS with history’) compared with universal application of these characteristics without TVS (‘history alone’). Corresponding 5-year survival rates were similar.

Costs

The reference option of ‘no initial diagnostic testing’ (diagnosis being delayed until representation with continuing PMB symptoms) was the cheapest strategy, costing £439 per woman investigated for PMB in a secondary care setting. The combination testing strategy of TVS with EBx and OPH was the most expensive at £727 per woman treated, that is to say £288 more than the approach of no initial testing. The cheapest of the 11 investigative strategies was the taking account of patient characteristics by acquiring the patient history and performing a TVS for those deemed to be high risk and then performing EBx when a thickened endometrium was diagnosed. This approach cost £537 (£98 more than no initial testing), whereas the least expensive of the diagnostic testing strategies, OPH alone, cost £550 for every woman investigated, that is to say £13 more than selective TVS with history and £111 more than the reference case – no initial testing. D&C was too expensive at £1400 to be used as a reference case for the analysis.

Cost-effectiveness and dominance

Three of the 11 diagnostic strategies, ‘selective TVS with history’, ‘history + TVS’ and ‘TVS + OPH’, remained non-dominated by alternative investigation approaches for first presentation of PMB. ‘OPH alone’ is dominated by ‘selective TVS with history’. ‘TVS with a cut-off of 5 mm’, ‘history only’ and ‘EBx alone’ are all dominated by ‘history + TVS’, and ‘TVS + EBx + OPH’, ‘TVS + EBx’ and ‘EBx + OPH’ are all dominated by ‘TVS + OPH’. There were no cases of extended dominance.

Table 31 presents the deterministic analysis restricted to the non-dominated competing strategies.

The cheapest strategy is the base-case scenario of no diagnostic work-up at initial presentation with PMB. The most effective strategy is the combination of testing at the outset with TVS and OPH, but this approach comes at a greater cost, generating an ICER of £2.69M to gain an extra woman with PMB surviving at 5 years. Other non-dominated options have ICERs of over £2M per additional survivor, compared with the next most effective strategy, and are thus most unlikely to be cost-effective.

Figure 26 shows the total costs and effectiveness of alternative strategies for the diagnostic work-up of women presenting for the first time with PMB for endometrial cancer to secondary care. The line presented graphically joins the non-dominated strategies (‘selective TVS with history’, ‘history + TVS’ and ‘TVS + OPH’) to the reference case, no initial investigation. Any strategy plotted above this line is not considered cost-effective in relation to the non-dominated alternatives.

FIGURE 26.

Total costs and effectiveness of each of the 12 alternative strategies to investigate OMB. History only, strategy selecting women for EBx based on the patient characteristics model; history + TVS, strategy selecting women for EBx based on the patient characteristics and TVS model; No work up, no initial diagnostic testing nor consideration of patient characteristics from the history, diagnostic testing being delayed until representation with continuing PMB symptoms was used as the reference case; Selective TVS with history, strategy with selective use of TVS in high-risk women only based on the patient characteristics model.

Removing the options ‘no initial work-up’ and ‘D&C’ from Figure 26 allows a clearer view of the relationship between the results for other options, as shown in Figure 27.

FIGURE 27.

Total costs and effectiveness of selected strategies to investigate PMB. History only, strategy selecting women for EBx based on the patient characteristics model; history + TVS, strategy selecting women for EBx based on the patient characteristics and TVS model; No work up, no initial diagnostic testing nor consideration of patient characteristics from the history, diagnostic testing being delayed until representation with continuing PMB symptoms was used as the reference case; Selective TVS with History, strategy with selective use of TVS in high-risk women only based on the patient characteristics model.

Selective transvaginal scan with history compared with no (initial) work-up for investigating women with post-menopausal bleeding

The graph (Figure 30a) shows the modelled uncertainty in the difference in costs between a strategy of selective TVS with history and no initial diagnostic work-up. It shows that selective TVS with history-taking is consistently more effective and between £80 and £120 more costly than the strategy of no diagnostic testing per extra woman investigated for PMB. The CEAC (Figure 30b) shows the proportion of model replications for which selective TVS with history is preferred to no initial diagnostic work-up at any given WTP per additional case surviving 5 years. Above a WTP threshold of approximately £150,000, it is probable that selective TVS with history is more cost-effective than no-workup; however, this becomes more certain above a WTP of £350,000.

FIGURE 30.

Cost-effectiveness plane (a) and CEAC (b): selective TVS with history strategy relative to the no diagnostic work-up strategy for investigating women with PMB.

History and transvaginal scan compared with selective transvaginal scan with history for investigating women with post-menopausal bleeding

The graph (Figure 31a) shows the modelled uncertainty in the difference in costs between ‘history and TVS’ and ‘selective TVS with history’. It shows that routine testing with ‘history and TVS’ is likely to be more costly and possibly slightly more effective than ‘selective TVS with history’. The CEAC (Figure 31b) shows that ‘history and TVS’ is cost-effective compared with ‘selective TVS with History’ in a small proportion of model runs, but this proportion does not reach 20% across the range of WTP values plotted.

FIGURE 31.

Cost-effectiveness plane (a) and CEAC (b): history and TVS strategy relative to the selective TVS with history strategy for investigating women with PMB.

Transvaginal scan and outpatient hysteroscopy compared with history and transvaginal scan for investigating women with post-menopausal bleeding

The graph (Figure 32a) shows the modelled uncertainty in the difference in costs between ‘TVS + OPH’ and ‘history and TVS’. It shows that routine testing with ‘TVS + OPH’ is more expensive and probably more effective than ‘history and TVS’. Figure 32b shows that it is extremely unlikely that TVS + OPH will be cost-effective compared with ‘history and TVS’ across the range of WTP values plotted.

FIGURE 32.

Cost-effectiveness plane (a) and CEAC (b): TVS + OPH strategy relative to the history and TVS strategy for investigating women with PMB.

Transvaginal scan with a 5-mm cut-off compared with selective transvaginal scan and history for investigating women with post-menopausal bleeding

‘Selective TVS with history’ was compared with ‘TVS alone’ with a 5-mm cut-off for diagnosis of a thickened endometrium because TVS is the current primary investigation for PMB. The graph (Figure 33a) shows the modelled uncertainty in the difference in costs between a strategy of using ‘TVS with a cut-off of 5 mm’ for diagnosis of a thickened endometrium and ‘selective TVS with history’. It shows that ‘TVS with a 5-mm cut-off’ is more costly than ‘selective TVS with history’ and is probably less effective. The option ‘TVS 5 mm’ is dominated by ‘selective TVS with history’ in the deterministic results. Figure 33b shows that there is a negligible probability that ‘TVS 5 mm’ will be cost-effective across the range of WTP values plotted.

FIGURE 33.

Cost-effectiveness plane (a) and CEAC (b): TVS with a 5-mm cut-off strategy relative to the selective TVS with history strategy for investigating women with PMB.

History only compared with selective transvaginal scan with history for investigating women with post-menopausal bleeding

The graph (Figure 34a) shows the modelled uncertainty in the difference in costs between ‘history only’ and ‘selective TVS with history’. It shows that routine testing with ‘history only’ is likely to be more costly than ‘selective TVS with history’ but is unlikely to be more effective. The option ‘history only’ is dominated by ‘selective TVS with history’ in the deterministic results. Figure 34b shows that there is a very low probability that ‘history only’ will be cost-effective across the range of WTP values plotted.

FIGURE 34.

Cost-effectiveness plane (a) and CEAC (b): history only strategy relative to the selective TVS with history strategy for investigating women with PMB.

History only compared with history and transvaginal scan for investigating women with post-menopausal bleeding

The graph (Figure 35a) shows the modelled uncertainty in the difference in costs between ‘history only’ and ‘history and TVS’. It shows that routine testing with ‘history only’ is probably more costly than ‘history with TVS’ and unlikely to be more effective. The option ‘history only’ is dominated by ‘history and TVS’ in the deterministic results. In just under 20% of model replications, ‘history only’ was the less costly option, but for most of these it was also less effective than ‘history and TVS’. As the WTP increases, the cost saving indicated by these points is less worth making, and the probability that ‘history only’ is cost-effective reduces (Figure 35b).

FIGURE 35.

Cost-effectiveness plane (a) and CEAC (b): history only strategy relative to the history with TVS strategy for investigating women with PMB.

Heavy menstrual bleeding

Our objective was to evaluate the cost-effectiveness of diagnostic testing in HMB, that is to say the effect of diagnostic work-up strategies on improving treatment outcomes. We chose universal treatment with the LNG-IUS without any preliminary investigation as our reference strategy to compare testing options against. This was because (i) not investigating or treating is an unacceptable alternative; (ii) the LNG-IUS is recommended by NICE as first-line treatment of HMB; and (iii) the LNG-IUS is a less invasive and more applicable initial treatment option compared with hysterectomy. We did, however, incorporate the option of bypassing investigation altogether and undertaking hysterectomy for all women with HMB because a recent report in Health Technology Assessment found hysterectomy to be a more cost-effective treatment than LNG-IUS in DUB. 2 We found that treating all women with a LNG-IUS without preliminary testing resulted in high levels of satisfaction (93%) which were increased by approximately 4% if some form of currently available diagnostic testing was undertaken to guide treatment.

Post-menopausal bleeding

A previous cost-effectiveness analysis had identified initial testing with TVS to measure endometrial thickness at a 5-mm threshold as the preferred option. 69 Within this strategy, histological testing using EBx would be restricted to those women with an abnormal TVS endometrial thickness measurement. In our updated analysis, we took account of a contemporary ‘one-stop’ clinical setting and adopted a novel approach using IPD to integrate important clinical characteristics from the history into the relevant testing options to examine whether or not this affected cost-effectiveness. Our analysis identified three potentially cost-effective testing approaches all utilising TVS: (i) ‘selective TVS with history’ where TVS was restricted to women with a > 4% chance of endometrial cancer based upon risk factors identified from the patient history; (ii) ‘history + TVS’, where historical risk factors are taken into account along with the result of a TVS; and (iii) a combination of TVS and OPH. Selective TVS based upon historical risk prediction for the diagnostic work-up of women presenting with PMB generated an ICER compared with our reference strategy of ‘no initial work-up’ of £129,000 per extra woman surviving 5 years. The ICER of combining history and TVS was £2.4M and the combination of OPH and TVS required an investment of £2.7M to acquire one additional woman with PMB living 5 years.

From ONS life tables,153 the life expectancy of a 63-year-old woman (the mean age of women diagnosed with endometrial cancer85) is approximately 22 years. This reduces to approximately 15 years when discounting at 3.5% is applied. Assuming that each extra survivor at 5 years gains a full 15 discounted QALYs, the maximum WTP at the upper NICE threshold of £30,000 per QALY would be £450,000 per additional survivor. Across the NICE threshold range of £20,000 to £30,000 per QALY gained, the option ‘selective TVS with history’ would appear to be cost-effective if each additional 5-year survivor gains 4 to 6 QALYs. This seems plausible and suggests that ‘selective TVS with history’ is the preferred option for investigating PMB at a cost acceptable to the NHS. The two other non-dominated options have ICERs over £2M per additional survivor compared with the next most effective strategy, and are thus most unlikely to be cost-effective. We drew acceptability curves and frontiers as far as a WTP of £600,000 per additional survivor to assess the stability of our findings. The certainty of selective TVS based upon information obtained from the patient history being the most cost-effective of the non-dominated strategies at a WTP investment of £129,000 was approximately 40%, but this rose to over 70% at WTP levels between circa £260,000 and £600,000.

We examined further the certainty of our findings by producing acceptability curves comparing the non-dominated strategies. Although the probabilistic sensitivity analyses showed that ‘history + TVS’ was cost-effective compared with ‘selective TVS with history’ in a small proportion of model runs, this proportion never exceeded 20% across a range of WTP values from £0 to £600,000. Similarly, the likelihood of a strategy combing TVS and OPH being cost-effective compared with ‘history + TVS’ was extremely unlikely, with WTP values up to £600,000. Interestingly, however, qualitative research has suggested that, given the choice, women want to undergo OPH in addition to strategies based upon TVS even despite the marginal additional diagnostic value. 152 Thus, economic analyses from a wider perspective, capturing patient preferences, may make a combination strategy of TVS + OPH more viable. The current recommendation is to use TVS as the first-line diagnostic test for investigating PMB. 66,69 So, despite the fact that TVS alone was dominated in our deterministic analysis by selective TVS based upon preceding patient history, we produced an acceptability curve to explore whether or not TVS alone was likely to become cost-effective over the plausible range of WTP values. Acceptability curves demonstrated that the probability of TVS alone being more cost-effective than a selective approach to TVS was negligible.

Simplifications

We aimed to develop economic models that accurately and explicitly mirrored clinical practice. Some simplification was necessary, driven by a desire to keep the extensive and comprehensive decision trees workable. One problem our analysis had to address was that of how to account for failed testing and discordance of tests within combination strategies. The accuracy of individual outpatient tests varies according to which pathology is under scrutiny. However, this does not mean that combination testing will be more cost-effective. Our model comprehensively evaluated tests used in isolation (with additional testing conditional on the test outcome) and in combination from the outset (again with any further testing being conditional on the combination testing outcome). However, where combination testing was incomplete due to tests within the combination failing, clinicians would then face a dilemma as to what to do next and this was apparent in our inability to obtain expert consensus. We debated at length how best to deal with failed tests within test combinations. In the end we decided to simplify our model with the aim of maintaining consistency and minimising potential bias. Thus, the approach we used was that any test or tests that failed within a combination testing strategy were assumed to have not allowed a diagnosis to be arrived at (i.e. the testing strategy has failed) and recourse to D&C was the result. The limitation of this approach is its clinical validity and relevance. Undoubtedly, many clinicians would consider examination under anaesthesia with D&C the ‘gold standard’ test, whereas others may pragmatically arrive at a diagnosis based upon the information provided by the successful tests within the combination strategy. However, there are a huge number of potential test combinations and failure possibilities to consider. Moreover, it was hard to arrive at any consensus from the expert panel as to what subsequent testing or management decisions would then arise. Thus, we believe that simplification of our model, such that a combination testing strategy was considered incomplete, necessitating recourse to D&C, was reasonable in order to maintain consistency and minimise bias. Moreover, the failure rates of tests are generally low and so the impact of our model simplification should be minimal.

Subtle differences between testing strategies may have been overlooked; for example, in the HMB model we assumed that women could only have one underlying pathology, whereas in practice, pathologies may, in a minority of women, coexist. The most common pathologies underlying HMB were accounted for in our model: DUB, polyps/SMFs, small and large fibroids and significant endometrial disease. Adenomyosis, a condition of the myometrium (uterine muscle) where ectopic endometrial tissue is found, can be associated with HMB but we excluded the diagnosis from our model. This omission can be justified because pain rather than HMB tends to be the presenting symptom of adenomyosis. Moreover, treatment is not affected by the suspicion of adenomyosis and indeed it only generally comes to light after histopathological examination of the uterus post hysterectomy. Thus, it is unlikely that the exclusion of adenomyosis as a potential aetiology of HMB would have had any effect on the outcome of the economic evaluation.

We modelled all relevant, widely employed testing modalities in AUB. Endometrial tissue sampling was assumed to be by outpatient EBx and inpatient D&C in failed cases. EBx can, however, be performed under direct vision by passing miniature forceps down the working channel of an outpatient hysteroscope. We chose not to include directed hysteroscopic biopsy as it is not widely used and is less likely to provide an adequate tissue sample for histological assessment compared with simpler, cheaper and universal outpatient EBx. Furthermore, OPH and SIS are accurate tests for the diagnosis of focal pathology without recourse to histological confirmation. In addition, focal endometrial disease (cancer and hyperplasia) necessitating a directed biopsy is rare, with most endometrial conditions being global hormonally induced phenomena allowing ‘blind’ sampling technologies to obtain representative, diagnostic samples to provide diagnosis with a high level of accuracy. 20,21 Thus, directed biopsy is unlikely to confer an advantage over blind EBx.

Testing with magnetic resonance imaging (MRI) was not examined because NICE recommends that it should not be used for diagnosis in women with HMB on the basis that there are no studies examining its use. 11 One study compared MRI with TVS for the identification of adenomyosis and found no significant difference between them. 155 In addition, MRI is not suitable for real-time outpatient (‘office’) testing and is expensive and rarely used in the diagnosis of HMB unless there are concerns over the nature of uterine fibroids (an infrequent situation) or the suitability for UAE is being assessed.

Our evaluation of PMB was restricted to detecting or excluding endometrial cancer and the impact of testing strategies on survival. The rationale for this approach was based upon the following considerations. In contrast to HMB, the management of PMB is aimed at identifying and treating a potentially life-threatening condition. HMB is invariably of benign origin and the aim of management is to alleviate symptoms to a woman’s satisfaction and enhance life quality by minimising restrictions on daily activities and desires. The majority of PMB is also non-malignant, but endometrial cancer and pre-cancer is present in 10–20%9 of cases and is eminently treatable if diagnosed early. Thus, survival was considered the most appropriate clinical outcome. An evaluation of optimal testing approaches for all possible causes of PMB, benign and malignant, would require a relevant clinical outcome applicable uniformly and, in the absence of HRQL data for the treatment of specific conditions underlying PMB, a unifying, all-encompassing analysis was not feasible. PMB, although an alarming symptom, is not often heavy, persistent or recurrent when associated with benign causes. Thus, the over-riding rationale for prompt diagnostic work-up is aimed at excluding serious endometrial disease rather than relieving symptoms.

Assumptions

As with all economic modelling exercises, assumptions had to be made where contentious clinical decision-making or a paucity of clinical data were encountered and these assumptions were ultimately endorsed by the expert clinical panel after extensive consultation. One such area pertained to disease prevalence, estimates of which came from a variety of sources, and this meant that, to make them sum to 1, we needed to use the prevalence of one ‘disease’, DUB, as a flexible value. The assigned value of 0.32 (32%) may not accurately reflect the true prevalence; however, published data and the expert panel considered this a reasonable estimate. As with the precision of all assumptions, they were subject to sensitivity analyses. In this particular case, the inferences of the base-case analysis were not affected with varied, plausible estimates of DUB prevalence. A similar problem was encountered for some accuracy data such that the reciprocal FPR estimates, which had originated from diverse data sources, had to be manipulated to total 1 and these were tested in sensitivity analysis.

In the HMB model, we selected the unifying outcome as satisfaction with treatment at 1 year because this is one of the most common outcomes measured in RCTs of interventions for HMB61 and is clinically relevant. Moreover, the availability of other patient-reported outcome measures, for example HRQL data (and especially their application to the specific scenarios that arose from the construction of our comprehensive decision trees), is low. Collecting such data was not possible within the time and resource constraints of this project. Approximating data from published studies was decided to be an inaccurate method for producing QALYs and so ‘cost per patient satisfied’ was used alone. In addition, the cyclical and intermittent nature of HMB makes it somewhat problematic to calculate QALYs, particularly as women are affected by the condition for only approximately 25% of the year and symptoms will naturally resolve in time once the menopause is reached. While we endeavoured through systematic searching and quality appraisal to identify the optimal data, with a minority of interventions for particular underlying diagnoses, satisfaction data were not explicitly available and, where this problem was encountered, we assumed that the outcomes reported (e.g. ‘reduced bleeding’, ‘would recommend the treatment’ or ‘cured of cancer’) were indirect measures of satisfaction following consultation with the expert panel.

Occasionally, satisfaction data were not reported at 1 year, in which case we chose the data points closest to 1. This time point, while reasonable in terms of evaluating medium-term response, may favour conservative, ‘uterine-sparing’ procedures (the LNG-IUS, EA and myomectomy) because their effectiveness reduces over time. 2 In contrast, more invasive but definitive hysterectomy is not associated with recurrence of HMB symptoms. Thus, longer-term outcome assessment may have made the option of ‘hysterectomy without diagnostic testing’ a more viable option. However, the objective and emphasis of our analysis was to estimate the cost-effectiveness of diagnostic testing. Thus, as long as the treatment options were appropriate and consistently applied according to the diagnosed pathology, we could be confident that the most cost-effective testing strategies were delineated. Our base-case analysis assumed included women to be 45 years old as this is the modal age for presentation to secondary care with HMB. Thus, they could be expected to have an average of a further 7 years of menstruation before the menopause.

We have highlighted the ability of OPH to facilitate the removal of uterine polyps in addition to simply providing a diagnosis. However, this presumed cost-efficient benefit may be blunted if uterine polyps are not causative of HMB but are simply an incidental finding. NICE were unable to find any data to link uterine polyps and HMB,11 although clinicians do assume some link. 47,156 Two systematic reviews of generally low-quality, observational studies do, however, support the notion of polyp treatment being associated with a 75–100% improvement in HMB complaints. 40,41 A RCT of polyp treatment (www.opt.bham.ac.uk), due to be published in 2014, should be able to clarify this point as data have been collected from 507 women with AUB being treated with polypectomy. 51 Furthermore, even if polyps do not cause HMB, they may negatively affect treatment (e.g. impairing the effectiveness of the LNG-IUS) and ultimately cause dissatisfaction. 157,158 This may also be a psychological effect given that most women and their clinicians are unwilling to accept conservative management of detected uterine polyps. 159 In addition to hysteroscopic surgical interventions such as polypectomy and myomectomy, other uterine-sparing therapeutic interventions, such as fitting of LNG-IUS or EA, may not be successfully completed. For the purposes of our model, however, we assumed that all treatments were successfully performed given that the objective of our economic analysis was to examine the cost-effectiveness of diagnostic testing based upon their test performance and accuracy, rather than an assessment of treatment efficacy.

Despite the need for assumptions arising in response to the ‘holes’ in our evidence base, the analytic modelling methodology allowed us to produce an extensive, comprehensive, contemporary and clinically relevant evaluation. All-embracing RCTs, assessing the wide range of diagnostic strategies defined in our models, would be a huge, impractical and ultimately futile undertaking.

Heavy menstrual bleeding

Our analysis identified two potentially cost-effective testing strategies for the investigation of women with HMB. These were initial testing with OPH alone or in combination with EBx. To adopt a strategy of OPH, £360 needs to be invested to gain one more woman satisfied at 1 year compared with a strategy of empirical treatment with a LNG-IUS. Although a testing strategy of OPH + EBx is marginally more effective, the ICER is approximately £21,000 to gain one more satisfied patient compared with OPH. We estimate that this equates to around £26,500 per QALY, which falls within the £20,000–30,000 per QALY threshold. These findings were stable during sensitivity analyses, varying model inputs including disease prevalence, test feasibility and accuracy, with OPH remaining more cost-effective than the LNG-IUS reference strategy even at relatively low WTP thresholds. In women wishing to preserve their fertility, therapy with EA and hysterectomy are contraindicated. SIS was cost-effective in this situation, with an ICER of approximately £2000, but for an additional financial outlay of £2720, testing instead with OPH produces a further satisfied patient, which is likely to be considered affordable and worthwhile by the NHS. Sensitivity analysis also showed SIS to be a cost-effective testing option along with the combination testing strategy of OPH + EBx within the context of traditional multistop pathways, although such service models are likely to diminish over time with the ongoing improvement and increasing availability of portable diagnostic health technologies. OPH remains the most cost-effective testing option if a scenario is envisaged where only women with HMB refractory to the LNG-IUS, recommended first-line treatment in general practice,11 are referred to secondary care for investigation. In this situation, we estimate that £5480 of extra funding is necessary for each additional woman satisfied. No other testing strategies fell within plausible WTP ranges.

Therefore, our data are consistent in supporting OPH as the diagnostic testing strategy of choice for women referred to secondary care with HMB, irrespective of their desire for future fertility and regardless of pre-referral treatment with the LNG-IUS. A combination strategy of OPH and EBx may be cost-effective at the upper NICE WTP threshold, in women without a desire to retain their fertility who have not undergone pre-referral treatment with the LNG-IUS or in women investigated through a traditional multivisit pathway.

Post-menopausal bleeding

Our economic analysis identified three potentially cost-effective testing approaches for the investigation of PMB, all utilising TVS. Selective TVS conditional upon patient characteristics generated an ICER compared with our reference strategy of no initial testing of £129,000 per extra woman surviving 5 years. Across the NICE threshold range of £20,000 to £30,000 per QALY gained, this option of selective TVS combined with risk factors acquired from the clinical history would appear to be cost-effective if each additional 5-year survivor gains 4 to 6 QALYs. This seems plausible and suggests that this is the preferred option for investigating PMB at a cost acceptable to the NHS. Although combination strategies of TVS with integration of patient characteristics or alternatively TVS with OPH were also non-dominated, these potentially cost-effective approaches both required an investment of over £2M for each additional woman with PMB living 5 years, which is beyond viable funding thresholds. Probabilistic sensitivity analyses showed that ‘history + TVS’ was cost-effective compared with ‘selective TVS with history’ in a small proportion of model runs, but this proportion never exceeded 20% across a range of WTP values from £0 to £600,000, supporting the case for selective TVS after considering the risk associated with adverse patient characteristics.

Heavy menstrual bleeding

Initial investigation of women presenting to secondary care with HMB who do not require preservation of their fertility appears to be a choice between OPH alone or, at WTP thresholds above £20,000 per additional woman satisfied, OPH combined with EBx.

Outpatient hysteroscopy appears to be the most cost-effective first-line diagnostic test used for the investigation of women presenting to secondary care with HMB wishing to preserve their fertility or refractory to previous medical treatment with the LNG-IUS.

A strategy of initial testing with OPH appears to be cost-effective within the context of a contemporary, ‘one-stop’ clinical service, where histological sampling, insertion of LNG-IUS and uterine polypectomy can be performed during the same visit. However, this does not appear to be the case within a more traditional multistop diagnostic and treatment service. A combination of OPH and EBx is potentially cost-effective within both ‘one-stop’ and multistop settings at WTP thresholds between £20,000 and £30,000 per QALY.

Post-menopausal bleeding

The use of transvaginal ultrasonic measurement of endometrial thickness as the first-line test in the diagnostic evaluation of PMB appears to be appropriate. However, the model-based results from this analysis, supporting a more selective use of TVS based upon risk factors for endometrial cancer, should be empirically confirmed with RCTs before current medical practice is changed.

Expanded branch from the outpatient hysteroscopy strategy for investigation of heavy menstrual bleeding

Diagnosis and treatments from the decision tree for investigation of women with heavy menstrual bleeding