Testing strategies for Lynch syndrome in people with endometrial cancer: systematic reviews and economic evaluation

Purpose of the decision to be made

Lynch syndrome is an inherited genetic condition. It is caused by mutations in genes that are involved in repairing errors that occur in deoxyribonucleic acid (DNA) when cells replicate. When mutations occur in these genes, DNA errors are not repaired. Over time, this can lead to uncontrolled cell growth. Lynch syndrome is associated with an increased risk of cancers, including colorectal, endometrial, gastric, pancreatic and kidney cancers. There is 50 : 50 chance that a person with Lynch syndrome will pass it to their children.

Recently, the National Institute for Health and Care Excellence (NICE) has recommended that people who are diagnosed with colorectal cancer (CRC) are tested for Lynch syndrome. 1 Routine testing for Lynch syndrome among people with endometrial cancer is not currently conducted. Detection of Lynch syndrome might lead to reductions in the risk of developing cancer for both the individual and their family members (through surveillance and risk-reducing strategies, such as chemoprevention) and to earlier treatment of cancers. 2,3

The external assessment group (EAG) assessed the accuracy of immunohistochemistry (IHC)-based and microsatellite instability (MSI)-based testing strategies to identify people who are at high risk of Lynch syndrome, and assessed the clinical effectiveness and cost-effectiveness of testing for Lynch syndrome among people who have endometrial cancer and their biological relatives. This will inform the NICE Diagnostics Advisory Committee guidance on whether or not testing for Lynch syndrome in people who have endometrial cancer represents a cost-effective use of NHS resources.

Population: people with endometrial cancer

Endometrial cancer (cancer that develops from the lining of the uterus) is the most common gynaecological cancer in the Western world. 5 Each year in the UK, there are approximately 9300 new cases of endometrial cancer and 2200 endometrial cancer-related deaths. 6,7 The incidence of endometrial cancer generally increases with age, reaching a peak of 97.3 per 100,000 population between the ages of 75 and 79 years. 6,7 The most recent estimates suggest that people with endometrial cancers have a 1-year survival rate of 89.6% and a 5-year survival rate of 75.7%. 8 Risk factors for the development of endometrial cancer include obesity, nulliparity, early age at menarche, use of hormone-replacement therapy and Lynch syndrome. 9–11

Target condition: Lynch syndrome

Lynch syndrome, formally called hereditary non-polyposis colorectal cancer (HNPCC), is a cancer-predisposition syndrome. It is estimated that there are approximately 175,000 people with Lynch syndrome in the UK. 12

Lynch syndrome is usually caused by mutations to any one of four DNA mismatch repair (MMR) genes: mutL homologue 1 (MLH1), mutS homologue 2 (MSH2), mutS homologue 6 (MSH6) or postmeiotic segregation increased 2 (PMS2). 13 A small proportion of Lynch syndrome cases are caused by deletions to the epithelial cellular adhesion molecule (EPCAM) gene, which leads to epigenetic silencing of MSH2. 13 MMR genes encode proteins that are involved in recognising and repairing errors that occur in DNA during cell division. Mutations in MMR genes prevent DNA errors from being corrected. This can lead to uncontrolled cell growth and the development of cancer. A range of cancers have been associated with Lynch syndrome, the most common of which are endometrial and colorectal. 14 Lynch syndrome accounts for 2–9% of endometrial cancers. 15,16 By the age of 75 years, approximately 57% of people with Lynch syndrome will have endometrial cancer. 14 The type and prevalence of cancer appears to vary according to which of the genes are affected. 14

Lynch syndrome has an autosomal dominant inheritance pattern, meaning that a person has a 50% chance of passing the mutated gene(s) onto their children.

Immunohistochemistry

Immunohistochemistry, in this case, uses antibodies to look for the expression of four MMR proteins (MLH1, MSH2, MSH6 and PMS2). An absence of staining for any of the proteins suggests a genetic mutation. IHC testing identifies which MMR gene is potentially affected. If MLH1 has an abnormal expression, an additional test (MLH1 promoter hypermethylation testing) can be conducted (see MLH1promoter hypermethylation testing). IHC can detect non-functional, but antibody-binding, MLH1 proteins (which would be incorrectly classified as normal);17 therefore, this may lead to a false negative result.

Microsatellite instability testing

Microsatellites are short repeats of DNA sequences. These repeats are prone to acquiring errors. When the MMR genes are not functioning, these errors are not corrected. Mutations in MMR genes lead to variations in the size of these repeats. This is called MSI. MSI testing is used to determine whether or not there are differences in the repeat numbers between tumour and non-tumour regions in a person being tested. Various markers have been described. 18 The Bethesda guidelines19 identify five markers (BAT25, BAT26, DS123, D17S250 and D5S346) for MSI for Lynch syndrome. Typically, three classifications are derived from this approach:

1. MSI-high – two or more markers show instability/> 30% of markers show instability.

2. MSI-low – one marker shows instability/< 30% of markers show instability.

3. MSI-stable – zero markers show instability [also known as microsatellite stable (MSS)].

Additional testing can be conducted to help rule out sporadic epigenetic silencing of MLH1, which might present as Lynch syndrome (see MLH1promoter hypermethylation testing).

MLH1 promoter hypermethylation testing

Hypermethylation is an epigenetic process that stops a protein being produced by a gene. MLH1 promoter hypermethylation testing is initially conducted on tumours. The test is undertaken following IHC or MSI testing, usually on patients with a MSI-high result or IHC loss in the MLH1 protein. A positive result on this test suggests that the tumour is sporadic and not a result of Lynch syndrome. However, there is some evidence that constitutional epimutations of MLH1 in normal tissue may be a cause of Lynch syndrome in a small number of cases. 20

Identification and selection of studies

Extraction and study quality

Methods of analysis/synthesis

In the gold-standard study design for assessing test accuracy, an entire sample of participants receives both the index test and the reference standard. This allows direct, unbiased comparisons of the agreement between the two tests. For reasons such as cost and practicality, in many test accuracy studies only a subsample of participants receive both tests, that is individuals who are index-test positive (at higher risk for the disease or condition) receive the reference standard, whereas individuals who are index-test negative do not receive the reference standard. Although this approach accurately reflects how tests are used in clinical practice, it leads to partial verification bias (also called detection bias or workup bias); data are missing and the true diagnostic status of participants who are negative on the index is not known. Partial verification can lead to overestimation of sensitivity and underestimation (or overestimation) of specificity. 36 Inaccurate test accuracy metrics can have an impact on clinical practice in relation to referral decisions and costs.

In this report, test accuracy results are divided into ‘complete’ test accuracy studies (in which all participants receive both the index test and the reference standard) and ‘partial’ test accuracy studies (in which only participants who are index-test positive receive the reference standard). For ‘complete’ test accuracy studies, we present results on all available test accuracy metrics, that is true positives, false positives, true negatives, false negatives, sensitivity, specificity, positive predictive values (PPVs) and negative predictive values (NPVs). For ‘partial’ test accuracy studies, we present results for only those test accuracy metrics that relate to participants who have received both the index test and the reference standard, that is true positives, false positives and PPVs. Furthermore, as there is a risk that the likelihood that someone will receive the reference standard is associated with disease status (e.g. individuals who truly have a disease may be more likely to get the reference standard than those who do not have the disease), which biases PPV upwards, we included only studies in which at least 95% of women who were eligible for germline testing (i.e. those who were index-test positive) received it. The sensitivity, specificity, PPVs and NPVs presented in this report were all calculated by the review authors and based on the true positive, false positive, false negative and true negative values that were reported in individual papers. Confidence intervals (CIs) were calculated using Wilson’s continuity correction. 37

Test accuracy results are presented for testing strategies 1–10, comparing the index tests with the eligible reference standards. Test accuracy was not assessed for strategy 11, as this approach does not include an index test. For studies that included an initial test followed by MLH1 promoter hypermethylation testing, we have analysed data at each stage of the process: (1a) IHC alone, then; (1b) IHC plus MLH1 promoter hypermethylation testing; and (2a) MSI-based testing alone, then; (2b) MSI-based testing plus MLH1 promoter hypermethylation testing. For IHC results, we have reported results together and separately for each protein. For MSI results, we have reported the panel used as per the papers, and provided a narrative summary of results on microsatellite instability-low (MSI-L) and microsatellite instability-high (MSI-H) patients. A subgroup analysis was not conducted for the different combinations of microsatellite markers because of the small number of studies and the wide range of panels used. The main analysis assumed that MSI-L was a negative test result. Owing to insufficient data, we did not conduct subgroup analyses of test accuracy by (1) age (≤ vs. > 70 years) or (2) people who have had a prior Lynch syndrome-related cancer (as defined in NHS England’s National Genomic Test Directory: Testing Criteria for Rare and Inherited Disease23). A narrative summary of the evidence is presented because meta-analysis was not possible as a result of heterogeneity.

Variants of uncertain clinical significance on germline testing are not considered to have Lynch syndrome in our test accuracy analysis. The EAG has recorded how many of these there are for a scenario analysis in the economic modelling, considering either all or none as having Lynch syndrome. In practice, patients with a negative germline test result (with no somatic cause of the tumour identified), but a positive index test, may be considered to have Lynch-like syndrome (also known as putative or cryptic Lynch syndrome) and undergo further investigation or surveillance. In particular, further investigation is undertaken if there is family history of Lynch syndrome. Because of this, the EAG descriptively recorded the characteristics of these cases such as family history, IHC results and discordant cases between the two index tests. This provides contextual information about the possibility of Lynch-like syndrome, and variants of uncertain clinical significance. However, for the reporting of test accuracy data, germline testing using sequencing with or without MLPA was considered the primary reference standard. We included studies using other diagnostic tests outlined in the Association for Clinical Genomic Science best-practice guidelines33 for genetic testing and diagnosis of Lynch syndrome, that is array-based comparative genomic hybridisation, and long-range polymerase chain reaction (PCR). The uncertainty around the effectiveness of germline testing to diagnose all cases of Lynch syndrome (see above regarding Lynch-like syndrome) is a potential weakness of the reference standard and a limitation of this review. As a subanalysis, for studies that report extra steps to the reference standard (e.g. sequencing of tumours or incorporating family history data), we recorded the additional tests that were used. Owing to the small number of studies using alternative tests, we did not compare the results of these multistage reference standards with the results of germline testing for MLH1, MSH2, MSH6, and PMS2 using sequencing with or without MLPA.

Quality assessment strategy for test accuracy studies

Quality assessment of eligible test accuracy studies was undertaken with a tailored QUADAS-2 tool. Methodological quality was assessed by two independent reviewers. Disagreements were resolved by consensus or a third reviewer.

Modifications to tailor the form of the QUADAS-2 tool to the research question in terms of the risk-of-bias assessment are outlined in Appendix 2 (the tailored QUADAS-2 form and guidance notes). No additional questions were added to the patient selection domain, the reference standard domain, flow and timing domain or any of the applicability sections. One question was added to the index test domain to assess whether or not quality assurance measures were in place.

Identification and selection of studies

Extraction and study quality

Methods of analysis/synthesis

No studies were identified that met the inclusion criteria, and so no data synthesis was undertaken.

Key question 3

What is the cost-effectiveness of testing for Lynch syndrome among people diagnosed with endometrial cancer using IHC- and MSI-based strategies, compared with the current pathway for the diagnosis of Lynch syndrome?

Review of existing cost-effectiveness models

Model structure for independent economic assessment

A de novo economic model was developed. The model structure reflected the decision problem: to determine the costs and benefits associated with implementing a policy to offer genetic testing to identify Lynch syndrome in women newly diagnosed with endometrial cancer; to offer testing to relatives of those thereby identified as having Lynch syndrome; and to offer interventions to those identified as having Lynch syndrome (probands and cascadees) aimed at reducing the risk of them developing (further) Lynch syndrome cancers, and improving outcomes if they do.

This decision problem can be analysed in two stages. The first stage is to determine what the costs and consequences are of the initial and cascade testing strategy being considered. This stage results in estimates of the total number of individuals with Lynch syndrome identified (probands and cascadees), together with the costs incurred in identifying them. The second stage involves estimating the incremental impact of being identified with Lynch syndrome compared with not knowing this. The impact occurs as a result of various risk reduction and surveillance interventions that can be offered once it is known that a person has Lynch syndrome. The costs and consequences of these interventions need to be modelled from the point when they are offered over the lifetime of a recipient.

We adopted a modular approach involving two submodels, one for each of these stages. The first stage was modelled with a decision tree structure, as testing strategies naturally lend themselves to this approach. The second stage was modelled with a Markov cohort model structure, to analyse the lifetime incidence of CRC and endometrial cancer from the point when an individual is identified with Lynch syndrome until their death (from CRC, endometrial cancer or another cause). The outputs from this Markov model were the (mean and variance) lifetime discounted costs and QALYs resulting from risk reduction measures, surveillance and cancer. These were calculated for a range of ages at which Lynch syndrome might be identified, as a table. This table became an input for the decision tree model, hence integrating the two submodels in a unified model.

Construction of the model involved consulting the previous HTA report undertaken by Snowsill et al. 12 comparing diagnostic strategies to identify Lynch syndrome in people with CRC. This also comprised two separate stages: a diagnostic stage and a management stage. The first stage used a decision tree structure to estimate the number of probands and their relatives who would be diagnosed with Lynch syndrome, and the resource use and costs involved. The second stage used an individual patient-level model to simulate the long-term costs and benefits (life-years and QALYs accrued) associated with management and surveillance, and prophylactic treatment for probands and relatives with Lynch syndrome. In addition, data and the modelling approach used by Snowsill et al. 43 in their cost-effectiveness analysis of reflex testing for Lynch syndrome in women with endometrial cancer were drawn on, as this was the model identified as being the closest to address the current decision problem under review.

The resultant model constitutes an initial diagnostic section, a decision tree model built in Microsoft Excel^® (Microsoft Corporation, Redmond, WA, USA), and a subsequent Markov cohort state transition model, in R software package (The R Foundation for Statistical Computing, Vienna, Austria), to estimate the long-term benefits accrued through risk reduction and surveillance measures for both CRC and endometrial cancer as a result of Lynch syndrome identification and cascade testing of relatives.

The diagnostic pathway in the decision tree component of the model is assumed to take place within 1 year, with no discounting applied to costs. The Markov model covers a lifetime time horizon (until death or age 100 years) with annual cycles in which costs and QALYs are discounted at a rate of 3.5% per year. Both models are conducted from an NHS and Personal Social Services (PSS) perspective.

The model will now be described in greater detail.

Diagnostic decision tree

This section of the model estimates the number of endometrial cancer probands and their relatives diagnosed with Lynch syndrome using the 11 strategies for inclusion in this review against the comparative strategy of no reflex testing. Figure 12 shows an overview of the testing pathway modelled for endometrial cancer probands undergoing one of the available strategies and Figure 13 shows an overview of the testing and management pathway for relatives of probands identified with Lynch syndrome or who are assumed to have Lynch syndrome.

FIGURE 12.

Overview of diagnostic model for probands. EC, endometrial cancer; LS, Lynch syndrome.

FIGURE 13.

Overview of testing and management pathway for relatives of probands. LS, Lynch syndrome.

Probands with endometrial cancer enter the model and are assigned to one of the 11 diagnostic strategies under assessment. Their path through the model is dependent on the result of the index test (combination of tests in the strategy) that they receive. Those with a positive index result are offered confirmatory germline testing. This is via a process of accepting genetic counselling and then accepting the genetic test. The proband can choose to accept or decline counselling, and those who accept counselling may then either accept or decline genetic testing. For probands who do consent to germline testing, Lynch syndrome status is confirmed.

Probands with a positive index test result and positive germline result are diagnosed with Lynch syndrome. Those with a positive index result and negative germline result are considered Lynch syndrome negative, but management of this group of individuals is subject to further investigation, as described in detail below. Probands with a positive index result who decline germline testing are assumed Lynch syndrome mutation negative, except for a specified proportion who are assumed Lynch syndrome, based on clinical suspicion.

Probands with a negative index result are not offered any further testing and are diagnosed with sporadic endometrial cancer.

In the final strategy, no index testing is performed, but probands proceed straight to genetic testing. In this case, genetic counselling and testing are offered directly to all endometrial cancer probands.

Diagnostic strategies for probands

The strategies modelled in the diagnostic component are as follows:

1. MSI testing followed by germline testing for Lynch syndrome-related mutations.

2. MSI testing followed by MLH1 promoter hypermethylation testing, followed by germline testing for Lynch syndrome-related mutations.

3. IHC MMR testing followed by germline testing for Lynch syndrome-related mutations.

4. IHC MMR testing followed by MLH1 promoter hypermethylation testing, followed by germline testing for Lynch syndrome-related mutations.

5. MSI followed by IHC then germline testing for Lynch syndrome-related mutations.

6. MSI followed by IHC plus MLH1 promoter hypermethylation testing then germline testing for Lynch syndrome-related mutations.

7. IHC followed by MSI then germline testing for Lynch syndrome-related mutations.

8. IHC followed by MSI plus MLH1 promoter hypermethylation testing then germline testing for Lynch syndrome-related mutations.

9. MSI and IHC done simultaneously, then germline testing for Lynch syndrome-related mutations.

10. MSI and IHC done simultaneously plus MLH1 promoter hypermethylation testing, then germline testing for Lynch syndrome-related mutations.

11. germline testing for Lynch syndrome-related mutations.

These strategies were compared with no testing for Lynch syndrome-related mutations and a fully incremental analysis was performed to report outcomes as ICERs based on cost per QALY.

Outcomes of diagnostic model for probands

Probands who test positive for a pathogenic mutation at germline testing are diagnosed with Lynch syndrome and offered Lynch syndrome surveillance for CRC and risk-reducing interventions, as appropriate. This is subject to the individual accepting these management options. Cascade testing is also triggered by Lynch syndrome-positive identification of the proband, whereby systematic testing of biologically at-risk relatives is undertaken. Output from the model is the number of probands with Lynch syndrome receiving Lynch syndrome surveillance and the number of probands with Lynch syndrome not receiving Lynch syndrome surveillance. As endometrial cancer probands are considered not to be at risk of further endometrial cancer, only female relatives of endometrial cancer probands who are diagnosed with Lynch syndrome are offered risk-reducing interventions for endometrial cancer.

Probands who test negative for a pathogenic mutation on index testing are diagnosed with sporadic endometrial cancer and continue with standard endometrial cancer management. They are not offered surveillance, nor is cascade testing pursued with their relatives.

Probands who decline germline testing after positive index test results are assumed a Lynch syndrome status based on clinical suspicion. Those who are assumed to not have Lynch syndrome are not offered surveillance or onward testing for their relatives. For those assumed to have Lynch syndrome (Lynch syndrome assumed), surveillance and risk reduction are offered, as well as surveillance and risk reduction for their first-degree relatives.

Probands with positive index test results on tumour tissue and negative germline results are considered Lynch syndrome negative, but, for a proportion of these, the clinical suspicion of Lynch syndrome remains. Similarly, despite negative results for currently identified pathogenic mutations for Lynch syndrome, germline testing may detect other mutation variances on these genes. These VUSs may or may not be later identified as pathogenic for Lynch syndrome, in which case status and management can be upgraded or downgraded accordingly. In these cases, it is assumed that further testing occurs on tumour tissue (somatic analysis) to either confirm sporadic cause of tumour or establish that the VUS is non-pathogenic for Lynch syndrome and management is then downgraded to that of non-Lynch syndrome individuals. Identification of new pathogenic variants is an alternative outcome of further testing, in which case individuals are modelled as being offered surveillance, as per Lynch syndrome assumed.

Probands who decline germline testing following a positive index test result are further categorised into ‘assumed non-Lynch syndrome’ or ‘Lynch syndrome assumed’, and managed accordingly.

Diagnostic strategies for relatives

Relatives follow strategy 11, straight to germline testing. This is also subject, in the model, to their acceptance of genetic counselling and acceptance of genetic testing following this counselling.

Outcomes of diagnostic model for relatives

Relatives who test positive for a pathogenic mutation at germline testing are diagnosed with Lynch syndrome and offered Lynch syndrome surveillance for CRC and risk-reducing interventions, as appropriate. This is subject to the individual accepting these management options.

Relatives who test negative for a pathogenic mutation at germline testing are not diagnosed with Lynch syndrome and no further surveillance measures are offered.

First-degree relatives who decline germline testing are diagnosed Lynch syndrome assumed and offered surveillance for CRC. Second-degree relatives and more distant relatives are subject to no further action.

Outcomes of diagnostic model summary

• Number of probands with Lynch syndrome receiving Lynch syndrome surveillance (true positive accepting).

• Number of probands without Lynch syndrome receiving Lynch syndrome surveillance (false positive accepting).

• Number of probands without Lynch syndrome who do not receive Lynch syndrome surveillance (delineated as those identified as Lynch syndrome positive who decline surveillance and those diagnosed as Lynch syndrome negative) (false positive declining and true negative not offered).

• Number of VUSs and Lynch syndrome assumed diagnoses.

Long-term outcomes model

We estimated the benefits of cascade testing by developing cohort state-transition models that simulate the incidence and mortality associated with Lynch syndrome-related cancers. We use these models to predict the benefit of being identified with Lynch syndrome through cascade testing by simulating incidence and mortality with, and without, surveillance and risk reduction measures, which we assume are adopted once Lynch syndrome has been identified. The cohort that is modelled consists of a group of individuals identical in terms of age at which they were identified as having Lynch syndrome, sex and previous Lynch syndrome cancer history (the model is repeated for a wide range of cohorts to provide the information needed for the decision tree model; this is described further in Figure 14).

FIGURE 14.

Overview of long-term model diagram. EC, endometrial cancer.

The model has five states: cancer free, CRC, endometrial cancer, both CRC and endometrial cancer, and dead. The endometrial cancer state comprises 10 ‘tunnel states’ reflecting time since incidence of endometrial cancer. These are known as tunnel states because a person in this state must move to the next state in the sequence at the end of the cycle (unless they move to death). The cohort can be of any age from 0 to 100 years and be male or female, and can start in any state. The state for women who have both endometrial cancer and CRC, therefore, has four substates, each with 10 tunnel states. For this decision problem, we simulate cohorts who are cancer free or recently diagnosed with endometrial cancer, male or female and aged in annual increments between 25 and 74 years. This gives 200 cohorts in total. We do not model outcomes for those without Lynch syndrome, on the assumption that they experience no long-term costs or benefits from Lynch syndrome testing.

For the comparator, we assume that, as the person is unaware of their Lynch syndrome status, no surveillance or risk reduction measures are offered. We model age-related annual incidence of CRC and endometrial cancer. For CRC, we further assume that incidence is gene dependent. In line with Snowsill et al. ,43 we assume that this incidence has a log-normal distribution. Previous work in this field has drawn on data on individuals with Lynch syndrome who benefit from colonoscopic surveillance. We follow that work in assuming that, based on Järvinen et al. ,44 surveillance reduces incidence with a hazard ratio of 0.387. We apply this to the log-normal distribution to derive the incidence rates illustrated in Figures 15 and 16.

FIGURE 15.

Modelled cumulative incidence of CRC in females with Lynch syndrome, assuming no surveillance.

FIGURE 16.

Modelled cumulative incidence of CRC in males with Lynch syndrome, assuming no surveillance.

For endometrial cancer, we sourced incidence data from the Prospective Lynch Syndrome Database (PLSD). 45 This database reported gene-based risk of cancer based on 6350 individuals with Lynch syndrome. Risks are reported at ages 25, 40, 50, 60, 70 and 75 years. We fitted a piecewise linear model to these data. The cumulative lifetime incidence of endometrial cancer in the absence of preventative measures implied by this assumption is illustrated in Figure 17.

FIGURE 17.

Modelled cumulative incidence of EC in females with Lynch syndrome. EC, endometrial cancer.

For CRC, we assumed that the proportion presenting with stages I to IV were 18.8%, 48.8%, 21.3% and 11.3%, respectively. We assumed a one-off cost of treatment, dependent on age and stage at diagnosis (Table 3).

TABLE 3 - One-off whole-diease treatment costs of CRC by age and stage

Age (years)	Cost (£)
	Stage I	Stage II	Stage III	Stage IV
0–49	8754.12	8740.53	14,489.51	11,704.91
50–59	5712.39	7015.84	9691.73	8443.68
60–69	4623.22	5351.77	7259.39	6508.89
70–79	3177.62	3454.61	4485.25	4365.04
≥ 80	1379.75	1545.95	1560.59	806.95

We assumed that CRC mortality is stage dependent, with transition probabilities of 0.009, 0.035, 0.098 and 0.543 for stages I to IV, respectively. 43

For endometrial cancer, we assumed a one-off treatment cost of £6510, in line with previous work. 12 We drew on Cancer Research UK-reported statistics on endometrial cancer mortality,46 and assumed that these were the same for those with Lynch syndrome as for those without. We assumed that those who have one Lynch syndrome risk cancer are at the same risk of developing the second one as if they were cancer free, conditional on not having died from the first cancer. We also applied an age-dependent transition probability for mortality from other causes. All those still alive in the model were assigned an age-dependent quality-of-life utility weighting using accepted methodology by Ara and Brazier,47 except that those with CRC stage IV were assigned a utility of 0.178, as modelled by Snowsill et al. 12

With these assumptions, we ran the cohort model separately for a number of cohorts defined as having the same age at identification, sex and cancer history. For each cohort, we estimated the mean lifetime costs and QALYs incurred.

We then assumed that the following risk reduction and surveillance methods were offered when an individual is identified as having Lynch syndrome.

Chemoprophylaxis

We assume that, once identified with Lynch syndrome, individuals take aspirin as indicated in the Colorectal Adenoma/carcinoma Prevention Programme 2 (CAPP2) trial48 and, based on the results of that trial, their probability of developing cancer each year is reduced by a factor of 0.56 (applied equally to endometrial cancer and CRC risk).

Colorectal cancer surveillance

We assume that individuals known to have Lynch syndrome have biennial colonoscopies from age 25 years (or age at identification of Lynch syndrome if later) until age 74 years. We assume the cost of colonoscopy is £325. 49 We assume that 100,000 colonoscopies result in 8.3 deaths, 40 perforations, and 55 bleeding events necessitating hospital treatment (of which 40 are mild, 10 are moderate and five are severe). This increases the average cost of colonoscopy by £2.89. We assume that this surveillance affects both incidence and stage at presentation. For stage at presentation, the assumed proportions for those participating in surveillance are 68.6%, 10.5%, 12.8% and 8.1% for stages I to IV, respectively.

Surgical prophylaxis to prevent endometrial cancer

We assume that women with Lynch syndrome can opt for hysterectomy and oophorectomy (H-BSO), and that this eliminates their risk of endometrial cancer. We make the assumption that the uptake of this increases with age, based on consideration of the average age at diagnosis of probands and subsequent ages of their identified relatives, and when women might feel ready based on completion of family, menopause, etc. The cost of this is assumed to be £3428. This assumption is illustrated in Figure 18.

FIGURE 18.

Uptake of surgical prophylaxis.

Gynaecological surveillance

We assume that women who have not had surgical prophylaxis undergo annual surveillance to detect endometrial cancer. The cost is £39, plus an additional cost of £473.41 for those requiring referral for invasive surveillance. We assume that this referral occurs in 10% of cases. This does not affect the incidence of endometrial cancer, but reduces mortality by 10.2%.

Parameters

Parameter input values for both diagnostic and long-term components of the model were sourced from literature obtained during the clinical effectiveness and cost-effectiveness systematic literature review process, with the best available evidence used to inform the base case. When suitable input parameters were not obtained, targeted searches were undertaken and individual publications critiqued. Additional information was also provided by clinical experts in the field. Discussion and critique of the sources of each parameter are detailed in Chapter 6, Discussion of model input parameters.

The model runs in 1-year cycles. The starting population is of the same age and sex, in the same state, (i.e. cancer free or recent diagnosis of endometrial cancer).

Each year:

• Transition occurs from all states to the death state based on annual mortality rates for all causes other than CRC or endometrial cancer. Death from the respective cancer state is accounted for in further transition from CRC, based on stage, and from endometrial cancer based on length of time they have spent in the endometrial cancer state. Transitions from all states to death are based on all-cause mortality.

• Further transitions from CRC or CRC plus endometrial cancer to death based on stage (CRC) or dwell time in state (endometrial cancer).

• Survivors in the endometrial cancer or endometrial cancer plus CRC states at the end of each cycle move on to the next tunnel state or remain in the final tunnel state prior to death. All those in the endometrial cancer or endometrial cancer plus CRC state who survive move to the next tunnel state (or stay in tunnel state 10). A quality-of-life score is assigned to the average number of individuals inhabiting each state at the start and end of each cycle.

Treatment costs of the respective cancers are assigned to an individual on entry to the cancer state and applied to the first year only (as a single, whole-disease cost). The average of the number of individuals in each state at the start and end of the cycle is assigned a quality-of-life score based on their age.

Those who move to a cancer state during the cycle are assigned treatment costs (all treatment costs are assumed to occur in the first year in the state). The model is run twice for each cohort: once assuming no Lynch syndrome-ameliorating measures (e.g. screening, prophylactic drugs), and once assuming measures are applied (as the model starts at the age at which an individual would be identified with Lynch syndrome were they to undergo genetic testing). These measures affect transition probabilities such as incidence and mortality, thereby capturing the benefit of the measures. Costs are also captured for those eligible for such measures. Colonoscopy is costed every other cycle. The number of women undergoing surgical prophylaxis is estimated from the number of women in the cancer-free or CRC states, by applying a proportion based on age, as described previously. It is assumed that the costs of aspirin, as a cheap over-the-counter medication, are not borne by the NHS.

The outputs from the model were the incremental costs and QALYs resulting from the addition of Lynch syndrome cancer-ameliorating measures. These were calculated separately by sex, for those cancer-free and those recently diagnosed with endometrial cancer, and for ages 25–74 years in 1-year intervals. These results provide an estimate of the benefit of Lynch syndrome cancer-ameliorating measures, and how these benefits vary by age and sex. To further illustrate how benefits arise in the model, results were extracted on the numbers in, and moving between, each state. These allowed the calculation of life-years gained, cancers avoided and cancer deaths prevented.

To allow these results to inform the decision tree model, we assumed that cascadees were equally likely to be any age between 25 and 74 years, and that the mean age of probands was 49 years. From this, we were able to define an output from the model as the average of the incremental results across all ages for cascadees, and the incremental results for women aged 49 years recently diagnosed with endometrial cancer for the probands. These results were used as the pay-offs for the terminal node in the decision tree model, so that the costs and QALYs per strategy could be calculated.

Quality assurance

Modelling of the independent economic assessment was conducted by two health economists, with primary development of each of the two components of the model done independently, and then checked by the second. Internal review by a senior health economist was also undertaken, with code review and cross-checking of input parameters to ensure that they originated from the described source. Furthermore, the reviewer constructed an alternative version of the diagnostic model in TreeAge (TreeAge Software, Inc., Williamstown, MA, USA) (rather than Excel) so cross-checking of results could also be carried out.

Probabilistic sensitivity analysis

A probabilistic sensitivity analysis (PSA) was used to determine the impact of joint parameter uncertainty. Model parameters were assigned a distribution reflecting the amount and pattern of variation, and cost-effectiveness results were calculated by simultaneously selecting random values from each distribution. This process was repeated 10,000 times, with simulations plotted on an incremental cost-effectiveness plane, with each point representing uncertainty in the incremental mean costs and QALYs between the strategies being compared. The results from these simulations were used to obtain cost-effectiveness acceptability curves (CEACs), which illustrate the effect of sampling uncertainty, and present the probability that the testing strategy is optimal at a range of willingness-to-pay (WTP) threshold values.

To propagate uncertainty across the decision tree and lifetime cohort models, we first carried out Monte Carlo simulation for the lifetime model with distributions assigned to all stochastic parameters. This produced an output set that could be used as an input table for the pay-off nodes for probands and relatives with Lynch syndrome in the decision tree model when it was run stochastically, producing PSA outputs that reflected joint uncertainty across the two models.

Sensitivity analysis

Univariate one-way sensitivity analysis was used to explore the impact of varying one parameter at a time, while keeping all other inputs constant, to assess the robustness of the model. We varied parameter values using upper and lower limits and presented results in the form of a tornado plot.

Scenario analyses

Alternative analyses were conducted for the following scenarios:

1. Strategy-level test accuracy obtained from the Proportion of Endometrial Tumours Associated with Lynch Syndrome (PETALS) study (Dr Neil AJ Ryan, University of Manchester, 11 November 2019, personal communication).

2. Costs of testing obtained from the PETALS study (Dr Neil AJ Ryan, personal communication).

3. Strategy-level test accuracy obtained from the PETALS study (Dr Neil AJ Ryan, personal communication) and costs of testing obtained from Ryan et al. 50

4. Disutility increment as a result of having cancer increased from the value modelled in the base case.

5. Gynaecological surveillance excluded.

6. Three-year colonoscopy surveillance.

7. Excluding benefit from aspirin.

8. Excluding hazard ratio reducing incidence of CRC as a result of surveillance.

Assumptions in base case

• Microsatellite instability-high results are treated as a positive indicator of Lynch syndrome, whereas MSI-L results are treated as a negative indicator of Lynch syndrome.

• The sensitivity of MSI and IHC testing did not depend on which MMR gene is mutated.

• The average number of relatives per proband was six (2.5 of whom were first-degree relatives).

• Colorectal surveillance colonoscopies occurred every 2 years, starting age 25 years and stopping at age 75 years.

• Surveillance colonoscopies are effective immediately on commencement of surveillance and ineffective immediately after discontinuation (i.e. no lag time).

• Disutility is applied only to people with stage IV CRC.

• Endometrial cancer is not modelled for women without Lynch syndrome-causing mutations.

• Treatment for endometrial cancer is assumed to be total abdominal H-BSO with/without chemotherapy with/without radiotherapy.

• Survival of probands with endometrial cancer is not affected by Lynch syndrome status.

• Surveillance for endometrial cancer comprises an annual review with their general practitioner (GP), with 10% of women attending referred for invasive gynaecological surveillance, consisting of gynaecological examination, transvaginal ultrasonography, endometrial biopsy and cancer antigen-125 (CA-125) testing.

• Gynaecological surveillance reduced the risk of mortality from endometrial cancer by 10.2%.

• No disutility arising from prophylactic hysterectomy was assumed.

• Prophylactic hysterectomy (total abdominal H-BSO) eliminates risk of endometrial cancer.

• Prophylactic hysterectomy (total abdominal H-BSO) is offered to all female relatives, with no age restrictions.

Search results

The study selection process for the clinical effectiveness review is illustrated in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram in Figure 19. The search identified 6259 records through database and other searches. Following duplicate removal we screened 3308 records. One additional unpublished study, the PETALS study, was provided by NICE and included for key question 1 (Dr Neil AJ Ryan, personal communication). A total of 2981 studies were excluded by their titles and abstracts, leaving 327 full texts to be assessed for eligibility for inclusion in the review. Of these, 282 papers were subsequently excluded, leaving 45 papers (including the unpublished PETALS study). 15,16,51–92 All 45 papers were relevant for key question 1: the test accuracy of MSI- and IHC-based strategies for determining Lynch syndrome in people with endometrial cancer. The most common reasons for exclusion of test accuracy studies at this stage were that there was no eligible reference standard in the studies and that too little information was included to enable quality appraisal. The full list of excluded studies with reasons for exclusion can be found in Appendix 3.

FIGURE 19.

The PRISMA flow diagram showing the study selection process for the clinical effectiveness review.

For key question 2, on the clinical effectiveness benefits and harms of testing for Lynch syndrome among people who have endometrial cancer, and/or their relatives, the search identified 29 studies that were potentially eligible for this review. We carried out the full-text assessment of the 29 records against the predefined inclusion criteria as stated in Table 2. No studies were identified that were relevant for key question 2 on the clinical effectiveness benefits and harms of testing for Lynch syndrome among people who have endometrial cancer, and/or their relatives. The most common reason for exclusion of clinical effectiveness studies at this stage was study design [not randomised controlled trials (RCTs)]. The full list of excluded studies with reasons for exclusion can be found in Appendix 3.

Population

The 45 included papers included approximately 10,600 participants, with sample sizes ranging from 1285 to 1459 participants. 56 The results of five studies were reported in more than one paper. 53,56,57,60,65,68,70,76,87,89 These papers have been reported together (i.e. two papers are combined into one study) throughout this report. Only two studies took place in the UK [one published51 and the PETALS study (Dr Neil AJ Ryan, personal communication)], with the majority taking place in the USA (15/45; 33%) and Europe (11/45; 24%). However, ethnicity was largely unreported [32/45 (71%) did not report ethnicity]. Several studies included age as an inclusion criterion, often limiting patients to ≤ 50 years for inclusion in a study. 16,51,54,79 In the remaining studies, ages ranged from 17 to 100 years. 15,63 Only 24% of studies (10/41) were in unselected populations, meaning that all patients in particular settings were included over the study time period, without any restrictions by age, cancer histology or family history. Two studies have been classified as unselected populations but limited to all adults (all those aged > 18 years). 55,81

A total of 44% (18/41) of studies reported on patients who had previous or concurrent cancers. The number of patients included across the studies who had a history of cancer ranged from 0 to 100. 58,74,80,85 This range can be explained by studies using a history of cancer as an inclusion or exclusion criterion. For studies not using cancer history as an inclusion or exclusion criterion, the proportion ranged from 0.8% to 22.4%. 54,77,90 The types of cancer reported were ovarian, pancreatic, colon, endometrial, urinary tract, brain, breast, skin, bladder, cervical and gastric cancers.

Index tests

Nine studies (11 papers) included IHC only,52,56,59,65,69,70,76,77,84,88,90 three studies included MSI-based testing only,15,64,74 28 studies (31 papers, including the unpublished PETALS study) included both tests16,51,53–55,57,58,60–63,66–68,71–73,75,78–83,85–87,89,91,92 and 24 studies (29 papers, including the unpublished PETALS study) included MLH1 promoter hypermethylation testing in combination with IHC or MSI testing (MSI and MLH1 promoter hypermethylation testing, n = 2; IHC and MLH1 promoter hypermethylation testing, n = 6; MSI, IHC and MLH1 promoter hypermethylation testing, n = 16). 15,16,55–61,63–65,67–72,76–78,81,83,84,86,87,89,92

Comparator and reference standard

The reference standards considered appropriate in this review were sequencing in combination with MLPA, long-range PCR or targeted array comparative genomic hybridisation. Of the 33 studies (36 papers) that included a reference standard, 21 studies (24 papers, including the unpublished PETALS study) included sequencing in combination with an additional method deemed appropriate by this review to detect larger structural changes. 15,16,51,52,55,57–59,61–63,65–68,77,78,81–83,85,88–90 Two studies (three papers) reported only sequencing, and did not report any details on the method of sequencing. 60,70,92 One study did not mention sequencing, but used array comparative genomic hybridisation, PCR and MLPA in combination. 84 Two studies (three papers) did not report clearly the methods of germline testing. 56,69,76 Six of the included studies used an additional reference standard test prior to sequencing that was not an eligible reference standard in this review. Two studies used single-strand conformational variance. 64,74 The studies by Berends et al. ,54 Rubio et al. 82 and Mercado et al. 73 used denaturing gel electrophoresis, and the study reported across two papers by Baldinu et al. 53 and Strazzullo et al. 87 used denaturing high-performance liquid chromatography.

Outcomes

Data on the number of Lynch syndrome diagnoses among women with endometrial cancer were reported in 32 studies (including the unpublished PETALS study). 15,16,51–59,61–64,66,67,69,70,77,78,81–85,88–90,92 Four studies provided head-to-head test accuracy data for IHC- and MSI-based testing. 16,54,58,82 Complete test accuracy data were provided by five studies for IHC,16,54,58,82,90 four studies for MSI-based testing16,54,58,82 and four studies for IHC, MSI-based testing and MLH1 promoter hypermethylation testing. 16,58,81,83 An additional nine studies provided partial test accuracy data (true positives, false positives and PPVs) in which only women who tested positive on index tests were considered for germline testing. 16,54,58,63,66,78,82,90,92

Concordance between IHC and MSI-based testing was assessed in 23 studies. 15,16,51,54,55,58,61–63,67,68,71,72,75,78–82,85–87,91

Setting

The majority of the participants in the included studies were recruited from hospitals [26/41 (63%)]. 15,51,52,57–68,70,77–79,81–84,88,89,92 Other studies took place in cancer registries [5/41 (12%)], cancer and radiation centres/clinics [6/41 (15%)], medical centres [2/41 (5%)] and tissue biobank repositories [1/41 (2%)]. In one study, the setting was not reported. 71

Study design

All the studies in this review had a cohort design. A total of 39% of studies (16/41) were prospective cohort studies, 46% (19/41) were retrospective and 12% (6/41) had both prospective and retrospective elements. One study had a mixed design, comprising both a prospective cohort study looking at MMR assessments and a cross-sectional study comparing clinical and pathological features between Lynch syndrome and Lynch-like syndrome groups. 60,70

Quality Assessment of Diagnostic Accuracy Studies-2

In the proposed testing strategies 1–10 (below), only women who tested positive on the index tests would be offered germline testing. Some studies report results from implementing the strategies of interest; however, these are partial test accuracy studies because data on true negatives and false negatives are not available. It is not, and it would not be, possible to calculate sensitivity, specificity or NPVs from these studies because of a lack of follow-up of women who were negative on the index tests. Studies in which all patients receive the reference standard provide sensitivity, specificity, PPVs and NPVs, and have been defined here as full test accuracy data studies. A total of 41 studies (45 papers) were identified, of which seven provided full test accuracy data (as all participants received both the index test and reference standard),16,54,55,58,82,83,90 26 studies (29 papers, including the PETALS study) provided partial test accuracy data (as only a subsample of participants received both the index test and reference standard)15,51–53,55–57,59–64,66–70,73,74,76–78,84,85,88,89,92 and 23 studies (including the PETALS study) provided data on concordance. 16,51,54,55,58,61,63,67,68,71,72,75,78–80,82,86,87,91

The studies providing test accuracy information were appraised using the QUADAS-2 tool and the seven complete test accuracy studies are presented prior to, and separately from, the partial test accuracy studies. Studies reporting on concordance were appraised using the quality appraisal tool for studies of diagnostic reliability (QAREL). Sixteen studies (16 papers, including the PETALS study) reported both test accuracy and concordance and were appraised using both tools. 15,16,51,54,55,58,61–64,67,68,78,82,85

Quality appraisal tool for studies of diagnostic reliability

Twenty-three studies (including the unpublished PETALS study) provided data on concordance. 15,16,51,54,55,58,61–63,67,68,71,72,75,78–80,82,85–87,91 These studies were appraised using the QAREL. Two of the questions in the QAREL were deemed not applicable to the studies. Question 7, ‘were raters blinded to additional cues that were not part of the test?’, was not applicable, as this is covered by question 6 on clinical information. Question 9, ‘was the time interval between repeated measurements compatible with the stability (or theoretical stability) of the variable being measured?’, was also judged as not applicable following guidance from clinical advisors.

Quality considerations in the included concordance studies are shown in Table 6. In general, the quality of the included studies was poor, with only one study (the unpublished PETALS study) having > 50% of the answers meeting the desired criteria in the questions. In particular, the representativeness of the sample was problematic in 78% of studies (18/23). 15,16,51,54,55,61,62,67,68,71,72,78–80,82,85–87 The studies were not comparable to clinical practice in the UK, with populations selected based on age, type of endometrial cancer and presence of synchronous/metachronous cancers (question 1). Only 13% of studies (3/23, including the PETALS study) were deemed representative. 55,61 Similarly, there were concerns regarding the representativeness of the raters performing the tests (question 2). In 87% of studies (20/23), there was not enough information reported to determine whether or not tests were conducted/interpreted by individuals who have undertaken the appropriate training and in laboratories that are participating in quality assurance programmes (e.g. UK National External Quality Assessment Scheme, Nordic immunohistochemical Quality Control, Clinical Laboratory Improvement Amendments). 15,16,51,54,55,58,62,67,68,71,72,75,78–80,82,85–87

There was a consistent lack of reporting regarding blinding across the studies. In 83% of studies (19/23), it was unclear whether or not raters were blinded to the findings of other raters (question 3); in 90% of studies (21/23), it was unclear whether or not raters were blinded to their own findings (question 4); in 65% of studies (11/17; six studies were concordance only studies with no reference standard, so this question was not applicable), it was unclear whether or not raters were blinded to the results from the reference standard (question 5); and, in 96% of studies (22/23, including the PETALS study), it was unclear whether or not raters were blinded to a patient’s clinical information (question 6). 16,51,54,55,58,61,63,67,68,71,72,75,78–80,82,87,91

In addition, there was a lack of reporting on how the tests were undertaken, meaning that it could not be determined whether or not the order of the testing varied [question 8; 19/23, including the PETALS study (83%)],15,16,51,54,55,58,61–63,67,68,71,72,78,80,82,85,87 or if tests had been conducted according to best-practice guidelines/via laboratories that are participating in quality assurance programmes [question 10; 21/23 (91%)]. 15,16,51,54,55,58,61–63,67,68,71,72,75,78–80,82,85–87 The majority of studies [21/23, including the PETALS study (91%)] reported raw data, but did not use any appropriate statistical measures (such as Bland–Altman plots or intraclass correlations, or between categorical/ordinal data with kappas). 15,16,51,54,55,58,61–63,67,68,71,75,78–80,82,85,87,91

Number of Lynch syndrome diagnoses

Thirty-three studies provided data on Lynch syndrome diagnoses,15,16,51–59,61–64,66,67,69,70,73,77,78,81–85,88–90,92 including the unpublished PETALS study (Dr Neil AJ Ryan, personal communication). Full details of the number of women identified with Lynch syndrome and VUSs/Lynch-like syndrome, and the types and frequencies of mutations, are reported in Appendix 4. Across all 33 studies, 349 cases of Lynch syndrome were identified from 7367 women tested. The reported prevalence of Lynch syndrome ranged from 0% (0 out of 140 women tested, in a clinical experience study from the USA that included all women undergoing hysterectomies at two hospitals) to 62%. 73 The prevalence of Lynch syndrome was typically lower in studies that recruited unselected samples of women (median 3.2%, range 0–5.3%) than in studies of selected samples of women (median 7.5%, range 0.9–62%). The prevalence of Lynch syndrome in two UK studies was (confidential information has been removed) [(confidential information has been removed) women tested, including (confidential information has been removed) women with known Lynch syndrome] in an unselected sample of women (PETALS study) and 8.5% (3 out of 35 women tested) in a selected sample of women aged < 50 years. 51 The types and frequencies of MMR gene mutations varied between studies. Combining data from all studies, variants in MSH2 gene mutations were the most common (38.6% of Lynch syndrome cases), followed by MSH6 (30.4% of Lynch syndrome cases), MLH1 (23.6% of Lynch syndrome cases) and PMS2 (7.3% of Lynch syndrome cases). One study did not report which of the MMR genes were mutated,83 and 10 studies did not assess all four MMR genes. 16,53,54,64,74,78,82,85,88,92 MHL1 and MSH2 were not assessed in one study,64 MSH6 was not assessed in three studies53,74,85 and PMS2 was not assessed in 10 studies. 16,53,54,64,74,78,82,85,88,92 Combining data from studies of unselected samples of women, variants in MSH6 were the most common (39.1% of Lynch syndrome cases), followed by MSH2 (32.2% of Lynch syndrome cases), MLH1 (19.5% of Lynch syndrome cases) and PMS2 (9.2% of Lynch syndrome cases). Combining data from studies of selected samples of women, variants in MSH2 were the most common (42.1% of Lynch syndrome cases), followed by MSH6 (25.7% of Lynch syndrome cases), MLH1 (25.4% of Lynch syndrome cases) and PMS2 (6.8% of Lynch syndrome cases).

Eighty-nine VUSs were reported in 10 studies (including the PETALS study), ranging from 2–15 cases per study. 15,16,54–58,61,63,82,90 In one study (confidential information has been removed) of the VUSs were identified in women who were (confidential information has been removed). Nine women were reported to have Lynch-like syndrome from two studies, ranging from 3 to 6 cases per study. 59,70

Head-to-head studies

Four studies provided head-to-head test accuracy data for IHC- and MSI-based testing, although the numbers of included tumours were not identical for each of the tests because of insufficient tumour tissue being available and test failures. 16,54,58,82 Three studies had a larger number of results for IHC than for MSI: 102 versus 83,58 99 versus 9516 and 94 versus 83. 82 One study had a larger number of results for MSI than for IHC: 57 versus 51. 54 All four studies comprised selected samples of women. Two studies excluded women aged > 50 years,16,54 one study excluded women with recurrent or synchronous cancers58 and one study excluded women (1) without a personal/family history of Lynch syndrome or (2) who were aged > 50 years. 82 Two studies included an ineligible reference standard (conformational-sensitive gel electrophoresis/denaturing gradient gel electrophoresis) as part of their diagnostic process. 54,82 Three of the studies were judged to be at high risk of bias. 54,58,82 The remaining study was judged to have an unclear risk of bias, as insufficient information was presented on which to make an assessment. 16 All four studies were rated as having high applicability concerns (see Risk of bias for complete test accuracy studies and Applicability of study findings for complete test accuracy studies for further details).

For IHC, there were 28 true positives, 78 false positives, 235 true negatives and 5 false negatives; point estimates ranged from 66.7% to 100% for sensitivity, 60.9% to 83.3% for specificity, 14.3% to 37.5% for PPVs and 95.2% to 100% for NPVs. For MSI testing, there were 21 true positives, 57 false positives, 232 true negatives and 8 false negatives; point estimates ranged from 41.7% to 100% for sensitivity, 69.2% to 89.9% for specificity, 20% to 33.3% for PPVs and 88.7% to 100% for NPVs. There were no statistically significant differences (on the basis of CIs) between MSI and IHC on any of the four tests’ accuracy metrics.

Immunohistochemistry- and microsatellite instability-based testing, with MLH1 promoter hypermethylation testing

Four studies provided test accuracy data for IHC- and MSI-based testing, where a lack of expression on IHC without MLH1 methylation or MSI-H (two or more unstable markers) test was considered index-test positive. 16,58,81,83 Full details are reported in Table 7. The circumstances under which MLH1 promoter hypermethylation testing was conducted varied in the studies. In two studies, methylation testing was conducted in women who had tumours that were categorised as MSI-H or had IHC loss (MLH1 or MLH1/PMS2);16,83 in one study, methylation testing was conducted in women who had IHC MLH1 loss only. 58 In the remaining paper, the circumstances under which MLH1 promoter hypermethylation testing was conducted were not reported. 81 Three studies comprised selected samples of women,16,58,83 and one study comprised an unselected sample of women. 81 One study excluded women aged > 50 years;16 one study excluded women with recurrent or synchronous cancers;58 and one study included an unselected sample of women, but did not report data on women with uninformative MMR results or without prior tumour testing. 83 Each study used a different panel of MSI markers. There were 85 true positives, 290 false positives, 475 true negatives and 4 false negatives. Two studies reported the gene variants in Lynch syndrome cases. 16,58 The most commonly affected gene was MSH2 [9/15 cases of Lynch syndrome (60%)], followed by MSH6 [4/15 cases of Lynch syndrome (26.7%)], MLH1 [2/15 cases of Lynch syndrome (13.3%)] and PMS2 [0/15 cases of Lynch syndrome (0%)]. PMS2 was assessed in only one study. 58 In two studies, 25 VUSs were identified (median 12.5; 11–14 cases per study). 16,58 One study did not report VUSs. 83 In the remaining study, 25 VUSs were identified, but the study did not report whether or not the participants had undergone index testing. 81 Point estimates ranged from 90.5% to 100% for sensitivity, 6.6% to 92.3% for specificity, 18.3% to 56.3% for PPVs and 75.0% to 100% for NPVs. In the study with an unselected sample of women,81 there were 19 true positives, 32 false positives, 312 true negatives and 2 false negatives. Comparing CIs, there was no statistically significant difference in sensitivity, specificity, PPVs or NPVs between the studies with selected samples and those with unselected samples.

Two studies reported the results of MLH1 promoter hypermethylation testing. 16,58 Twelve out of 13 tumours (92.3%)16 and 12 out of 15 tumours (80%) were hypermethylated. 58

A secondary analysis of test accuracy in which VUSs were considered germline positive was possible for two studies. 16,58 Estimates of test accuracy were as follows: sensitivity, 100.0% (95% CI 80.0% to 100.0%); specificity, 86.3% (95% CI 75.8% to 92.9%); PPV, 66.7% (95% CI 47.1% to 82.1%); and NPV, 100.0% (95% CI 92.8 to 100.0%);58 and sensitivity, 100.0% (95% CI 80.0% to 100.0%); specificity, 78.8% (95% CI 67.9% to 86.8%); PPV, 54.1% (95% CI 37.1 to 70.2%); and NPV, 100.0% (95% CI 92.8 to 100.0%). 16 These were similar to estimates in which VUSs were consider to be germline negative, with the exception of PPV for Chao et al. ,58 which was higher when VUSs were considered to be germline positive [66.7% (95% CI 47.1% to 82.1%) vs. 20.0% (95% CI 8.4% to 39.1%) for germline positive vs. germline negative, respectively].

Immunohistochemistry alone

Five studies provided test accuracy data for IHC. 16,54,58,82,90 Full details are provided in Table 8. All five studies comprised selected samples of women. Two studies excluded women aged > 50 years;16,54 one study excluded women with recurrent or synchronous cancers;58 one study excluded women (1) without a personal/family history of Lynch syndrome-related cancer or (2) who were aged > 50 years;82 and one study excluded women who were (1) aged > 50 years, (2) without a personal/family history of Lynch syndrome-related cancer or (3) did not have loss of expression of any MMR protein on IHC testing. 90 Four studies assessed all four MMR proteins54,58,82,90 and one study assessed MHL1, MSH2 and MSH6. 16 There were 69 true positives, 193 false positives, 243 true negatives and 6 false negatives in the five included studies. The most commonly affected gene was MSH2 [34/69 cases of Lynch syndrome (49.3%)], followed by MSH6 [18/69 cases of Lynch syndrome (26.1%)], MLH1 [14/69 cases of Lynch syndrome (20.3%)] and PMS2 [3/69 cases of Lynch syndrome (4.3%)]. PMS2 was assessed in only two studies. 58,90 In total, 33 VUSs were identified in the five studies (median 4; 3–11 cases per study). The point estimates ranged from 66.7% to 100% for sensitivity, from 6.5% to 83.3% for specificity, from 14.3% to 37.5% for PPV and from 88.9% to 100% for NPV (see Figure 23). With the exception of PPV in the study by Tian et al. ,90 CIs between studies overlapped for each of the test accuracy metrics. Test accuracy estimates for the single study that employed only MLH1, MSH2 and MSH6 were within the ranges reported by the studies using all four MMR proteins.

Of the five studies, one presented data in sufficient detail to estimate test accuracy by individual proteins. 16 This study reported on MLH1, MSH2 and MSH6 in 99 patients. True positives were determined by the individual protein test’s ability to detect any germline mutation, not necessarily the corresponding mutation. There was wide variation in the test accuracy between the different proteins. Sensitivity was 11.1% (95% CI 0.6% to 49.3%) for MLH1, 66.7% (95% CI 30.9% to 91.0%) for MSH6 and 77.8% (95% CI 40.2% to 96.1%) for MSH2. Specificity was 87.8% (95% CI 79.2% to 93.2%) for MLH1, 95.6% (95% CI 88.4% to 98.6%) for MSH6 and 95.7% (95% CI 88.6% to 98.6%) for MSH2. PPV was 7.7% (95% CI 0.4% to 37.9%) for MLH1, 60.0% (95% CI 27.4% to 86.3%) for MSH6 and 63.6% (95% CI 31.6% to 87.6%) for MSH2. NPV was 90.7% (82.0% to 95.6%) for MLH1, 96.6% (95% CI 89.8% to 99.1%) for MSH6 and 97.6% (95% CI 91.0% to 99.6%) for MSH2. The wide variations between the test accuracy for MLH1 and the other proteins may be accounted for by the difference in the number of false positives. There were 12 false positives when testing using MLH1, compared with only four for MSH2 and MSH6.

In three of the studies, information on the IHC result by individual protein was presented only for those with a germline mutation,54,82,90 and, in the remaining study, eight IHC cases were not reported and there were discrepancies between values reported in the text and table. 58

A secondary analysis of test accuracy in which VUSs were considered germline positive was possible for four studies. 16,54,58,82 Estimates of test accuracy were as follows: sensitivity, 100.0% (95% CI 59.8% to 100.0%); specificity, 62.8% (95% CI 46.7% to 76.6%); PPV, 33.3% (95% CI 16.4% to 55.3%); and NPV, 100.0% (95% CI 84.5% to 100.0%);54 sensitivity, 90.0% (95% CI 66.9% to 98.2%); specificity, 75.6% (95% CI 64.7% to 84.1%); PPV, 47.4% (95% CI 31.3% to 64.0%); and NPV, 96.9% (95% CI 88.2% to 99.5%);58 sensitivity, 100.0% (95% CI 80.0% to 100.0%); specificity, 83.5% (95% CI 73.1% to 90.6%); PPV, 60.6% (95% CI 42.2% to 76.6%); and NPV, 100.0% (95% CI 93.1% to 100.0%);16 and sensitivity, 82.4% (55.8% to 95.3%); specificity, 75.3% (64.0% to 84.1%); PPV, 42.4% (26.0% to 60.6%); and NPV, 95.1% (85.4% to 98.7%). 82 These were similar to estimates in which VUSs were considered to be germline negative.

Microsatellite instability-based testing alone

Four studies provided test accuracy data for MSI-based testing. 16,54,58,82 Full details are provided in Table 9. All four studies comprised selected samples of women. Two studies excluded women aged > 50 years;16,54 one study excluded women with recurrent or synchronous cancers,58 and one study excluded women (1) without a personal/family history of Lynch syndrome or (2) who were aged > 50 years. 82 Three different panels of markers were used in the four studies; only two studies used the same panel of markers. 16,82 Using MSI-H (two or more unstable markers) as a cut-off point, there were 21 true positives, 57 false positive, 232 true negatives and 8 false negatives in the four included studies. The most commonly affected gene was MSH2 [13/21 cases of Lynch syndrome (61.9%)], followed by MSH6 [4/21 cases of Lynch syndrome (19%)] and MLH1 [4/21 cases of Lynch syndrome (19%)]. PMS2 was assessed in only one study;58 there were no cases of Lynch syndrome with a PMS2 mutation. In total, 29 VUSs were identified in the four studies (median 7; 3–12 cases per study). Point estimates ranged from 41.7% to 100% for sensitivity, from 69.2% to 89.9% for specificity, from 20% to 89.9% for PPV and from 88.7% to 100% for NPV. One of the included studies reported data that allowed us to calculate test accuracy using MSI-H or MSI-L (one or more unstable marker) as a cut-off point. 82 There were 5 true positives, 17 false positives, 54 true negatives and 7 false negatives. 82 The most commonly affected gene was MSH2 [3/5 cases of Lynch syndrome (60.0%)], followed by MSH6 [1/5 cases of Lynch syndrome (20%)] and MLH1 [1/5 cases of Lynch syndrome (20%)]. PMS2 was not assessed. Three VUSs were identified. Test accuracy metrics were similar to those reported using MSI-H as a cut-off point: sensitivity was 41.7%, specificity was 76.1%, PPV was 22.7% and NPV was 88.5%. Using a cut-off point of one or more stable markers changed the status of one index test result from true negative to false positive.

A secondary analysis of test accuracy in which VUSs were considered germline positive was possible for four studies. 16,54,58,82 Estimates of test accuracy were as follows: sensitivity, 87.5% (95% CI 46.7% to 99.3%); specificity, 69.4% (95% CI 54.4% to 81.3%); PPV, 31.8% (95% CI 14.7% to 54.9%); and NPV, 97.1% (95% CI 83.4% to 99.9%);54 sensitivity, 100.0% (95% CI 75.9% to 100.0%); specificity, 91.0% (95% CI 80.9% to 96.3%); PPV, 72.7% (95% CI 49.6% to 88.4%); and NPV, 100.0% (95% CI 92.6% to 100.0%);58 sensitivity, 100.0% (95% CI 79.1% to 100.0%); specificity, 80.3% (95% CI 69.2% to 88.2%); PPV, 55.9% (95% CI 38.1% to 72.4%); and NPV, 100.0% (95% CI 92.6% to 100.0%);16 and sensitivity, 53.3% (95% CI 27.4% to 77.7%); specificity, 76.5% (95% CI 64.4% to 85.6%); PPV, 33.3% (95% CI 16.4% to 55.3%); and NPV, 88.1% (95% CI 76.5% to 94.7%). 82 These were similar to estimates in which VUSs were considered to be germline negative.

Concordance between immunohistochemistry- and microsatellite instability-based testing

Twenty-three studies, including the unpublished PETALS study (Dr Neil AJ Ryan, personal communication), provided data on concordance between IHC- and MSI-based testing. 15,16,51,54,55,58,61–63,67,68,71,72,75,78–80,82,85–87,91 Twenty studies provided complete concordance data (agreement/disagreement between IHC positive/negative and IHC negative), and three studies provided partial concordance data (IHC conducted only for MSI-H tumours,87 MSI conducted only for women with IHC loss79 and IHC conducted only for women with MSS results15). Full details of concordance are reported in Appendix 5. In the studies providing complete concordance data, there was a high level of agreement between the results of the tests (median agreement 91.8%, with the lowest level of agreement being 68.2% and the highest level of agreement being 100%) and a low level of disagreement (median disagreement 9.8%, with the lowest level of disagreement being 0% and the highest level of disagreement being 31.8%), median kappa 0.84 (range 0.32–0.97). Kappa values were calculated by the reviewers.

Few studies examined characteristics of discordant cases. Four studies reported that MLH1 promoter hypermethylation was common in discordant cases: 50% (1/2 cases),71 75% (3/4 cases),72 80% (4/5 cases)55 and 83% (10/12 cases). 86 Seven of the 23 concordance studies reported on the characteristics of discordant cases of MSI and IHC testing. 15,16,51,55,63,72,85 In two of these seven studies, it was possible to determine germline results for the discordant cases. 15,55 Bruegl et al. 55 found 5.1% disagreement, with seven out of 197 discordant cases. Of these seven, only one was found to have a germline mutation and this was in the MSH6 variant. Likewise, Hampel et al. 15 found that the only discordant case with a germline mutation was in the MSH6 variant. By contrast, Lu et al. 16 found that, of the five discordant cases, all were germline mutation negative.

Across three studies,16,55,72 20–57% (4/7, 1/5 and 2/6) of discordant results were due to MLH1 promoter hypermethylation, suggestive of epigenetic changes rather than Lynch syndrome.

For one study,51 discordance was associated with the classification of MSI-L cases. When MSI-L cases were grouped with MSS cases there were two discordant cases, whereas when MSI-H or MSI-L cases were grouped together and compared with MSS cases, there were no cases of discordance between MSI and IHC testing results. 51

It was possible to calculate the average age for discordant cases in three studies. 15,51,85 In Anagnostopoulos et al. ,51 discordant cases (n = 2) had a median age of 39.5 years, which was lower than the overall median of 48 years in the sample. Although Shin et al. 85 and Hampel et al. 15 found no real difference in age between discordant cases and the whole sample, Shin et al. 85 found two discordant cases with a mean age of 55 years at diagnosis of endometrial cancer and a mean age of 52.5 years at diagnosis of CRC, compared with the overall sample mean age of 52.5 years at endometrial cancer and 54.5 years at CRC. 85 Hampel et al. 15 found a mean age of 60.5 years in discordant cases, compared with the overall mean of 60.9 years in the whole sample.

One study85 reported on the comorbidities of other cancers in discordant cases. All cases in the study had a history of both endometrial cancer and CRC. They found that one of the two discordant cases also had a history of bladder cancer. Likewise, this was the only study to discuss family history in relation to discordant cases, and noted that both cases met the Amsterdam II criteria. 93

Testing pathways under review: partial test accuracy studies

In the proposed testing strategies 1–10, only women who test positive on the index tests would be offered germline testing. Some studies report results from implementing the strategies of interest; these are partial test accuracy studies because data on true negatives and false negatives are not available. It is not possible to calculate sensitivity, specificity or NPVs from these studies because of a lack of follow-up of women who were negative on the index tests. Studies in which full test accuracy could be extracted/calculated have already been reported (see Assessment of test accuracy), so here we report results from any studies (full or partial test accuracy) that report the numbers of true positive and false positive results for each strategy. Full details of all the strategies are provided in Table 10.

There is a risk that the likelihood that someone receives the reference standard is associated with disease status, for example individuals who truly have a disease may be more likely to get the reference standard than those who do not have the disease. This biases the PPV upwards. Therefore, we included only studies in which at least 95% of women who were eligible for germline testing (i.e. those who were index-test positive) received it.

Strategy 1: microsatellite instability testing alone

Eight studies, including the unpublished PETALS study (Dr Neil AJ Ryan, personal communication), provided test accuracy data for this strategy. 16,54,58,66,73,78,82

Six studies comprised selected samples of women,16,54,58,73,78,82 one study provided insufficient information on which to make an assessment of sample selection type66 and the PETALS study comprised an unselected sample of women (Dr Neil AJ Ryan, personal communication). Two studies excluded women aged > 50 years,16,54 one study excluded women with recurrent or synchronous cancers,58 one study included only women with a family history of endometrial cancer78 and one study excluded women (1) without a personal/family history of Lynch syndrome-related cancer or (2) who were aged > 50 years. 73,82

There were (confidential information has been removed) true positives and (confidential information has been removed) false positives out of (confidential information has been removed) women tested. The most commonly affected gene was MSH2 [(confidential information has been removed) cases of Lynch syndrome (confidential information has been removed)], followed by MSH6 [(confidential information has been removed) cases of Lynch syndrome (confidential information has been removed)], MLH1 [(confidential information has been removed) cases of Lynch syndrome (confidential information has been removed)] and PMS2 [(confidential information has been removed) cases of Lynch syndrome (confidential information has been removed)]. PMS2 was assessed in only four out of the eight studies (including the PETALS study). 58,66,73 In total, (confidential information has been removed) VUSs were identified in the seven studies [median (confidential information has been removed); (confidential information has been removed) cases per study]. Point estimates for PPVs ranged from (confidential information has been removed) to (confidential information has been removed).

Strategy 2: microsatellite instability testing with MLH1 promoter hypermethylation testing

No studies were identified that examined this strategy.

Strategy 3: immunohistochemistry-based testing alone

Six studies provided test accuracy data for this strategy. 16,54,58,73,82,90 All six studies comprised selected samples of women. Two studies excluded women aged > 50 years;16,54 one study excluded women with recurrent or synchronous cancers;58 one study excluded women (1) without a personal/family history of Lynch syndrome-related cancer or (2) who were aged > 50 years;82 and one study excluded women who (1) were aged > 50 years, (2) were without a personal/family history of Lynch syndrome-related cancer or (3) did not have loss of expression of any MMR protein on IHC testing. 73,90 In the five studies in which MMR genes were considered together, there were 69 true positives and 193 false positives, out of 552 women tested. The most commonly affected gene was MSH2 [34/69 cases of Lynch syndrome (49.3%)], followed by MSH6 [18/69 cases of Lynch syndrome (26.1%)], MLH1 [14/69 cases of Lynch syndrome (20.3%)] and PMS2 [3/69 cases of Lynch syndrome (3%)]. PMS2 was assessed in only three out of the six studies. 58,73,90 In total, 22 VUSs were identified in the seven studies (median 3; 2–11 cases per study). In the single study in which MMR genes were considered separately, the most commonly affected gene was MSH2 [40/80 cases of Lynch syndrome (50%)], followed by MLH1 [31/80 cases of Lynch syndrome (38.8%)] and MSH6 [9/81 cases of Lynch syndrome (11.2%)]. 73 Point estimates for PPVs ranged from 12.2% to 37.5% in the studies that reported on all four MMR genes,16,54,58,82,90 and from 77.4% to 84.6% in the study that reported each gene separately. 73

Strategy 4: immunohistochemistry testing with MLH1 promoter hypermethylation testing

Three studies provided test accuracy data for this strategy, including the PETALS study (Dr Neil AJ Ryan, personal communication). 16,78 Two studies comprised selected samples of women. 16,78 One study excluded women aged > 50 years,16 and one study included only women with a family history of endometrial cancer. 78 The PETALS study was conducted in an unselected sample of women with endometrial cancer (Dr Neil AJ Ryan, personal communication). There were (confidential information has been removed) true positives and (confidential information has been removed) false positives out of (confidential information has been removed) women tested. The most commonly affected gene was MSH2 [(confidential information has been removed) cases of Lynch syndrome, (confidential information has been removed)] followed by MSH6 [(confidential information has been removed) cases of Lynch syndrome (confidential information has been removed)], MLH1 [(confidential information has been removed) cases of Lynch syndrome (confidential information has been removed)] and PMS2 [(confidential information has been removed) cases of Lynch syndrome (confidential information has been removed)]. Only the PETALS study assessed PMS2 (Dr Neil AJ Ryan, personal communication). In total, (confidential information has been removed) VUSs were identified in the three studies [median (confidential information has been removed); (confidential information has been removed) cases per study]. Point estimates for PPVs ranged from (confidential information has been removed) to (confidential information has been removed). All three studies reported the results of MLH1 promoter hypermethylation testing. (Confidential information has been removed) out of (confidential information has been removed) tumours (confidential information has been removed),78 (confidential information has been removed) out of (confidential information has been removed) tumours (confidential information has been removed)16 and (confidential information has been removed) (PETALS study, Dr Neil AJ Ryan, personal communication) were hypermethylated.

Strategy 5: microsatellite instability testing followed by immunohistochemistry testing

No studies were identified that examined this strategy.

Strategy 6: microsatellite instability followed by immunohistochemistry testing with MLH1 promoter hypermethylation testing

No studies were identified that examined this strategy.

Strategy 7: immunohistochemistry followed by microsatellite instability testing

No studies were identified that examined this strategy.

Strategy 8: immunohistochemistry testing followed by microsatellite instability testing with MLH1 promoter hypermethylation testing

No studies were identified that examined this strategy.

Strategy 9: microsatellite instability and immunohistochemistry testing

No studies were identified that examined this strategy.

Strategy 10: microsatellite instability and immunohistochemistry testing with MLH1 promoter hypermethylation testing

Six studies provided test accuracy data for this strategy. 16,58,63,78,83,92 All six studies comprised selected samples of women. One study excluded women with recurrent or synchronous cancers;58 one study excluded women who were not considered suitable candidates for surgery, who had had prior retroperitoneal surgery or prior pelvic or abdominal radiation therapy, or who were pregnant;63 one study excluded women aged > 50 years;16 one study included only women with a family history of endometrial cancer;78 one study included an unselected sample of women, but did not report data on women with uninformative MMR results or without prior tumour testing;83 and one study included only women who answered questions about family/personal history of cancer and who had tumour and normal tissue available for analysis. 92 Four panels of markers were used in the six studies; the studies by Berends et al. 54 and Ollikainen et al. 78 used the same five-marker panel, and the studies by Lu et al. 16 and Rubio et al. 82 used the same six-marker panel. There were 94 true positives and 311 false positives out of 1627 women tested. Five studies reported the affected genes. 16,58,63,78,92 The most commonly affected gene was MSH2 [19/43 cases of Lynch syndrome (44.2%)], followed by MSH6 [16/43 cases of Lynch syndrome (37.2%)], MLH1 [5/43 cases of Lynch syndrome (11.6%)] and PMS2 [3 out of 43 cases of Lynch syndrome (7.0%)]. Only two studies assessed PMS2. 58,63 Five studies reported details of VUSs. 16,58,63,78,92 In total, 18 VUSs were identified (median 2; 0–14 cases per study). Point estimates for PPVs ranged from 18.3% to 43.1%. Five studies reported the results of MLH1 promoter hypermethylation testing. 16,58,63,78,92 From 14.3% to 92.3% of tumours were hypermethylated (368/516 tumours).

Strategy 11: germline testing only

Nine studies provided data on germline-only testing, whereby women were offered the reference standard(s), irrespective of the result of index tests. 16,54,58,62,74,81–83,90 Lynch syndrome was identified in 166 out of 1375 (12.1%) women tested (median 9; 5–51 cases of Lynch syndrome per study). In total, 47 VUSs were identified (median 3; 0–15 cases per study).

Full details on strategies 1–11 are provided in Table 10.

Complete test accuracy studies

Partial test accuracy

Prevalence

For the health economic model, we incorporated data from the nine studies (reported in 11 papers) that assessed prevalence of Lynch syndrome in unselected samples of women with endometrial cancer; these studies were the most applicable to our population of interest, that is they did not limit on the basis of age or prior cancers. The median prevalence of Lynch syndrome in these papers [including the PETALS study (Dr Neil AJ Ryan, personal communication)] was 3.2% (range 0–5.3%). 15,52,55,56,59–61,70,76,88 This was used in our base case as overall prevalence of Lynch syndrome in endometrial cancer.

Ryan et al. 100 also systematically reviewed the evidence on the prevalence of Lynch syndrome among endometrial cancer patients. Few studies undertook germline testing in all women with endometrial cancer, so Ryan et al. 100 took a stepwise approach to estimation. They conducted a series of meta-analyses of test positivity of IHC (MLH1 specific and across all proteins), MSI and methylation. They combined data from these separate meta-analyses on overlapping, but differing, populations to estimate what proportion of women would be referred for germline analysis using a combination of these tests. They then meta-analysed the proportion of women who were positive for Lynch syndrome in germline testing in a population that was an approximation to the testing strategy positive population. They combined these analyses to estimate that 3% of women with endometrial cancer have Lynch syndrome. This approach enabled the combination of data from a large number of studies, but made assumptions about the equivalence of different populations, and was inclusive of studies that did not exactly represent the population or test of interest.

Both reviews suggest a figure for overall prevalence of around the 3% level (confidential information has been removed) [(confidential information has been removed) out of (confidential information has been removed) women tested, including (confidential information has been removed) women with known Lynch syndrome] present in their sample.

A higher base-case prevalence of 3.91% obtained through random-effects meta-analysis of results from 15 studies was used by Snowsill et al. 43 However, when studies at risk of bias as a result of high dropout rates (≥ 10%) were excluded (n = 7), the estimated prevalence obtained was reduced to 3.0%, nearer to the figure used in our base case.

When we varied our approach from using studies with unselected endometrial cancer probands to using studies with selection criteria, prevalence estimates increased to a median of 6.5% (range 0.9–36.1%). In the systematic review by Ryan et al. ,100 a subgroup analysis of studies that did not use a tumour triage stage, but proceeded directly to germline testing, also found a higher proportion of Lynch syndrome carriers, of 6%. We have therefore conducted a sensitivity analysis using an increased overall prevalence figure of 6.5%.

Diagnostic accuracy

Initially, we attempted to identify the best test accuracy estimate per strategy from the systematic review. However, we did not use this approach in the economic model, because of issues of inconsistency described below. Instead, we used data from Lu et al. 16 to ensure consistency across strategies and to aid comparison between strategies. We undertook sensitivity analyses using estimates from the PETALS study and Snowsill et al. 43

Resnick et al.101

Resnick et al. 101 used a decision tree illustrative model structure to assess the cost-effectiveness of screening strategies for diagnosing Lynch syndrome among newly diagnosed endometrial cancer patients. The model depicted the clinical pathway that endometrial cancer patients would take while being screened for Lynch syndrome. The model started with a hypothetical cohort of 40,000 women expected to have endometrial cancer who underwent a screening strategy: Amsterdam criteria93 (full gene sequencing for women with endometrial cancer who met the revised Amsterdam criteria), sequence all (full gene sequencing for all women with endometrial cancer), sequence for all women aged < 60 years with endometrial cancer and IHC/single-gene strategy (IHC for all women with endometrial cancer after gene sequencing). After testing, women were categorised as Lynch positive or Lynch negative. With the IHC and sequencing testing strategy, women were categorised as MLH1 ≥ 60 years of age, MLH1 < 60 years of age, MSH6 or MSH2. Women with MSH6 deletion were considered to be Lynch positive, and women with MSH6 normal were categorised as MSH2 deletion (Lynch positive) or MSH2 normal (Lynch negative).

Clinical as well as cost information was required to populate the model, and this was obtained from published sources. Clinical information included the probability of fulfilling the Amsterdam criteria,93 people who do fulfil the Amsterdam criteria and have Lynch syndrome, all women with Lynch syndrome, women with Lynch syndrome stratified by age (< 60 years and ≥ 60 years) and people with normal IHC results. Resource use and costs were required for genetic consultation, full genetic sequencing, IHC and MLH1, MSH2 and MSH6 sequencing. All costs included in the model were reported in 2008 US dollars. The analysis was conducted from a third-party payer perspective, with the results presented in terms of an ICER, expressed as cost per additional Lynch syndrome case detected. Authors undertook a scenario analysis around the cost of full gene sequencing.

The base-case deterministic results showed that IHC/single-gene strategy, when compared with the Amsterdam testing strategy, had an ICER of approximately US$13,800 per Lynch syndrome case detected. The results of the scenario analysis showed that the ICER was sensitive to the cost of the full gene sequencing. Authors acknowledged and discussed the limitations of the economic analysis, then concluded that the testing strategy IHC and sequencing was the most cost-effective for identifying Lynch syndrome in women diagnosed with endometrial cancer.

The economic analysis provides a useful starting point to assess the cost-effectiveness of different testing strategies to detect Lynch syndrome in women diagnosed with endometrial cancer. Although the decision tree structure was appropriate to address the research question, the analysis was limited, as the ‘downstream’ costs and benefits associated with identifying women Lynch syndrome were not captured in the economic model. Thus, the impact of identifying these additional cases remains unanswered. In addition, the authors acknowledged that the testing strategy genotyping for the screening of MMR deficiency was not included in the economic analysis. In general, the economic evaluation was transparent and adhered to the reporting guidelines for undertaking economic analyses. Future model-based analyses could build on this simplistic model to capture the impact of including testing and treating of women with Lynch syndrome in a single cost-effectiveness analysis.

Kwon et al.102

Kwon et al. 102 used a Markov Monte Carlo simulation model to assess the cost-effectiveness of different testing strategies to identify Lynch syndrome in women diagnosed with endometrial cancer. The model started with a hypothetical cohort of women who had received treatment for endometrial cancer and were now receiving one of the following testing strategies endometrial cancer: aged < 50 years with at least one first-degree relative with a Lynch syndrome-associated cancer, endometrial cancer at < 50 years (IHC triage), endometrial cancer at < 60 years (IHC triage), endometrial cancer at any age with at least one first-degree relative with a Lynch syndrome-associated cancer (IHC triage), all endometrial cancers and any age (IHC triage), compared with Amsterdam II criteria. 93 Authors have outlined and justified why they have excluded testing strategies that include MSI.

The model was populated with clinical parameters, as well as information about resource use and costs. Clinical parameters included the prevalence, sensitivity and specificity of each testing strategy; the lifetime risk of CRC; and the 5-year mortality from CRC, all of which were obtained from the published literature. Resource use and costs included the costs for genetic counselling, gene sequencing, IHC for MMR proteins, colonoscopy and CRC treatment costs. Details of resource use and costs were provided, and references reported. All costs were reported in US dollars and reported in 2010 prices. Several simplifying assumptions were made to have a workable model structure.

The base-case analysis was undertaken from the societal perspective, with costs incurred and benefits accrued discounted based on 3% per annum. The economic analysis concluded at a lifetime time horizon, with the results reported as an ICER expressed as cost per life-year gained. It should be noted that the costs included in the analysis did not accurately reflect the viewpoint of the analysis. To our knowledge, only costs incurred by the health-care provider were included in the economic analysis, thereby reflecting a narrower perspective (third-party provider).

Deterministic results showed that IHC triage of women of any age with at least one first-degree relative with Lynch syndrome-associated cancer, when compared with women aged < 50 years and with at least one first-degree relative with Lynch syndrome-associated cancer, had an ICER of approximately US$9100 per life-year gained. Furthermore, results were reported on the number of women who would undergo IHC and, subsequently, the women diagnosed with Lynch syndrome and those who went on to develop CRC. Sensitivity and scenario analyses results showed that the ICER was robust to changes made to the model input parameters. Under the current model structure, model inputs and assumptions led the authors to conclude that IHC triage of women of any age, with at least one first-degree relative with Lynch syndrome-associated cancer, was the most cost-effective testing strategy when compared with using the Amsterdam II criteria. 93

This economic analysis adds to the existing literature about which screening strategies provide good value for money in diagnosing Lynch syndrome in women with endometrial cancer. However, there were several concerns related to this analysis. First, it is unclear about the patient pathway following testing, as no illustrative structures have been presented in the main document or online supplementary material. Second, it is unclear what assumptions are being made about the CRC mortality rate derived from the 5-year mortality obtained from the published literature. Third, care should be taken when interpreting the deterministic results, as the analysis was undertaken from the societal perspective, but the costs included in the analysis did not reflect this viewpoint.

Bruegl et al.103

Bruegl et al. 103 undertook an economic analysis alongside a clinical study to assess the cost-effectiveness of universal tissue testing (IHC for all and MLH1 methylation analysis when indicated) versus the Society of Gynaecologic Oncology (SGO) 5–10% clinical criteria105 (n = 97) for identifying Lynch syndrome in a cohort (n = 412 cases) of unselected women with endometrial cancer. Two approaches were used to assess the cost-effectiveness. First, the direct costs associated with identifying patients with probable Lynch syndrome and, second, the direct costs associated with identifying cases with probable Lynch syndrome among women with endometrial cancer, as well as their potentially affected first-degree relatives.

The analysis was conducted from a third-party payer perspective, with all costs reported in US dollars, in 2012 prices. The economic analyses included hospital and health-care professional costs associated with identifying women with probable Lynch syndrome. Costs included initial genetic counselling and follow-up visits; IHC for MLH1, MSH2, MSH6 and PMS2; MLH1 promoter hypermethylation assay for tumours with loss of MLH1; and single germline-mutation testing.

Under the SGO 5–10% clinical criteria,105 97 women would undergo further evaluation, of which 15 would be diagnosed with probable Lynch syndrome, resulting in a cost of approximately US$6100 per probable Lynch syndrome case diagnosed. Including screening for probable Lynch syndrome and their first-degree relatives under the SGO 5–10% clinical criteria105 strategy would cost approximately US$6300 per probable Lynch syndrome case diagnosed. This is based on the average number of first-degree relatives (5.3 relatives) and the estimated germline mutation rates among probable Lynch syndrome endometrial cancer patients’ first-degree relatives eligible for single-site gene mutation analysis. Under the universal tumour testing strategy, 43 women with probable Lynch syndrome would be identified, resulting in a cost of approximately US$5900 per probable Lynch syndrome case diagnosed. Including universal tumour screening for probable Lynch syndrome and screening their first-degree relatives would cost approximately US$6500 per probable Lynch syndrome case diagnosed. This is based on the average number of first-degree relatives (5.5 relatives) and the estimated germline mutation rates among probable Lynch syndrome endometrial cancer patients’ first-degree relatives eligible for single-site gene mutation analysis.

The authors concluded that, under the existing SGO 5–10% clinical criteria105 to identify Lynch syndrome in women diagnosed with endometrial cancer, this strategy is likely to miss some cases, when compared with using a universal tumour-testing strategy (IHC for DNA MMR proteins and PCR-based MLH1 methylation analysis for tumours with loss of MLH1).

The economic analysis presented here is conducted alongside a clinical trial to assess the clinical effectiveness and cost-effectiveness of different strategies that can be used to identify and diagnose women (and their first-degree relatives) with Lynch syndrome. Although the analysis adds to the existing literature, there are some concerns regarding the transferability and robustness of these results. First, as acknowledged by the authors, all potentially relevant strategies have not been included in the analysis; this is common in clinical trials. Second, the authors assumed that there is a 100% genetic counselling referral rate for endometrial cancer patients meeting the SGO 5–10% criteria,105 but referral rates are likely to be between 17% and 48%. Third, it is assumed that all patients meeting the SGO 5–10% criteria, or with tumour testing suggestive of Lynch syndrome, will accept germline counselling and/or germline testing, but this is not likely to be 100%. Fourth, the resource quantity used to derive costs is unclear, which limits the transparency about how costs were derived. Finally, the authors did not conduct sensitivity analyses to address uncertainty in the economic analysis.

Goverde et al.104

Authors undertook an economic analysis of a population-based cohort of endometrial cancer patients aged ≤ 70 years undergoing routine screening for Lynch syndrome. The economic analysis compared routine screening for Lynch syndrome by analysis of MSI and IHC for MLH1, MSH2, MSH6 and PMS2 protein expression in endometrial cancer patients up to the age of 70 years with screening for Lynch syndrome in endometrial cancer patients using an age cut-off point.

The analysis required clinical and cost information. Clinical parameters included acceptance of prophylactic gynaecological surgery, complication rate following colonoscopy, lifetime risk of developing CRC for Lynch syndrome carriers and reduction in CRC risks by Lynch syndrome surveillance. Resource use and costs included MSI analysis, genetic counselling and germline mutation analysis, IHC and MLH1 promoter hypermethylation testing, which were derived using microcosting methodology. All costs were reported in euros, in 2013 prices.

The analysis was undertaken from the third-party payer perspective, with costs incurred and benefits accrued discounted based on 3% per annum. The economic analysis concluded at a lifetime time horizon, with the results reported as an ICER, expressed as cost per life-year gained. Base-case deterministic results showed that routine screening of endometrial cancer patients up to the age of 70 years for Lynch syndrome by analysis of MSI, IHC and MLH1 promoter hypermethylation testing was cost-effective when compared with screening up to the age of 50 years, with an ICER of approximately €5300 per life-year gained. Sensitivity analysis results showed that economic analysis was sensitive to the life-years gained per female relative. The authors concluded that routine screening by analysis of MSI, IHC and MLH1 promoter hypermethylation testing for Lynch syndrome in people diagnosed with endometrial cancer up to the age of 70 years was the most cost-effective strategy, compared with an age cut-off point of 50 years.

The economic analysis builds on the current cost-effectiveness evidence of different strategies to detect Lynch syndrome in women diagnosed with endometrial cancer. In comparison to previous analyses, this analysis included the costs and benefits for first-degree relatives of probands.

Snowsill et al.43

Authors conducted an economic analysis by using a decision tree structure with Markov nodes to assess the cost-effectiveness of different testing strategies (MSI with methylation, direct mutation testing, IHC with methylation, MSI alone, IHC and a no-testing strategy) to identify Lynch syndrome in women treated for endometrial cancer. Authors clearly provided an illustrative model structure that depicted the patient pathway for endometrial cancer survivors undergoing screening for Lynch syndrome. The model started with a hypothetical cohort of women undergoing one of the screening strategies. In general, women were diagnosed as actually Lynch syndrome, actually sporadic Lynch syndrome or probable Lynch syndrome. Following diagnosis, women received CRC surveillance.

The model required clinical information (natural history, epidemiology, HRQoL, diagnostic accuracy, preventative effectiveness and utility values) and resource use and cost information (testing strategies, events and outcomes) for women undergoing screening for Lynch syndrome.

The analysis was conducted from the NHS and PSS perspective, with all costs reported in Great British pounds, in 2016/17 prices. All costs incurred and benefits accrued were discounted based on a 3.5% per annum rate. The analysis was conducted over a lifetime time horizon, with the results presented in terms of an ICER, expressed as cost per QALY. An ICER at or below the £20,000 per QALY WTP threshold was cost-effective. Several one-way sensitivity analyses, including PSAs, were undertaken based on the cost per QALY.

The base-case deterministic results showed that the IHC with methylation strategy was the most cost-effective strategy, with an ICER of approximately £14,200 per QALY. The IHC-alone strategy yielded the most QALYs and was most costly, but the results did not reach cost-effectiveness when compared with IHC with methylation, with an ICER of approximately £129,000 per QALY. The PSA results showed that there was a 0.36 probability that IHC with methylation was the most-cost-effective strategy at a WTP threshold of £20,000 per QALY. One-way sensitivity analysis results showed that the ICER was sensitive to the age of the proband and the effectiveness of colonoscopy in reducing CRC incidence. Scenario analysis results showed that, by using the effectiveness of colonoscopic surveillance to reduce the CRC incidence derived from information obtained from Arrigoni et al. ,106 none of the testing strategies was cost-effective.

The economic analysis builds on the existing cost-effectiveness evidence in this disease area, by including the diagnosis of Lynch syndrome and the benefit to probands of CRC screening. This analysis could have been improved by reporting the results in terms of the natural units, in addition to reporting the results in terms of cost per QALY alone. In addition, the model was sensitive to some model input parameters. Specific attention in the form of systematic reviews around these key parameters with a detailed critique between sources would improve the transparency in the selection of model inputs.

Quality assessment of the modelling methods and economic analyses

Full details of the quality appraisal of included economic studies can be found in Appendix 6.

Structure

The structures of the models included were of judged to be of satisfactory quality. Studies clearly stated their decision problem or research questions, the viewpoint of their analysis, and the objectives of the models and economic analyses, which were coherent with the decision problem. Only one study43 provided extensive detail about pre-model analyses conducted to estimate the prevalence of Lynch syndrome in endometrial cancer patients, test performance on the sensitivity and specificity of the different testing strategies, and incidence of developing other ‘downstream’ cancers. When appropriate, all studies were conducted over a lifetime time horizon and included discounting costs incurred and the benefits accrued using appropriate rates.

Most studies that conducted a model-based analysis clearly showed the illustrative model structures, which depicted the clinical pathway for endometrial cancer patients undergoing screening for Lynch syndrome. Earlier models were simplistic, but were adequate to address the decision problem, and only included screening and diagnosis of Lynch syndrome. Subsequent models were more complex, and, in general, their model structures followed the screening → diagnosis → surveillance → treatment pathway. In general, authors assessed testing strategies that included IHC (with/without MLH1 methylation) followed by germline testing, MSI (with/without methylation) followed by germline testing, direct mutation testing (using the SGO 5–10% clinical criteria,105 Amsterdam II criteria,93 Bethesda guidelines19) or a no-testing strategy; confirmatory diagnosis by use of germline testing was included in all analyses. Studies that included risk-reducing interventions considered surveillance as a means to reduce the incidence of CRC. Other risk reduction interventions (e.g. surgery, chemoprevention and aspirin) were not included. Authors have alluded to this as a limitation and have provided reasonable justification for not including these interventions. Goverde et al. 104 included costs associated with gynaecologic surveillance for relatives.

Data

All studies required clinical as well as cost information to undertake the economic analyses. The methods used to identify relevant information were clearly stated. References were provided for all inputs, but authors were not clear about the choices made between sources of information, especially when more than one source was available. In addition, it was not clear if quality appraisal of these studies was undertaken. Information to populate the economic models was obtained mainly from published sources and supplemented with information from unpublished sources, which included clinical expert opinion. To our knowledge, no study undertook systematic reviews to identify studies reporting key inputs.

Studies clearly reported clinical (natural history, mortality, diagnostic accuracy for each testing strategy, preventative effectiveness and utility values) information and resource use and costs (testing strategy, colonoscopic surveillance for probands and relatives, treatment of CRC, genetic counselling and germline mutation analysis for relatives, and prophylactic surgical treatment for relatives) information required. Natural history information was required for the prevalence of Lynch syndrome, mutation status, lifetime risk of CRC, endometrial cancer mortality and CRC mortality. The prevalence of Lynch syndrome was required in all studies. Prevalence was reported by age of the proband101,102 and overall prevalence. 43,103,104 All studies reported the references for individual studies, but only Snowsill et al. 43 elaborated on the methods used to estimate the prevalence. The distribution of gene mutation status was reported in all studies. In general, studies reported gene mutation status for the overall population,102–104 older/younger than 60 years of age101 and by a given age. 43 The lifetime risk of CRC was required in three studies. 43,102,104 Both Kwon et al. 102 and Goverde et al. 104 provided estimates for lifetime risk of CRC, for which information was obtained from the literature. However, Kwon et al. 102 provided lifetime risk of CRC by mutation status and in the absence or presence of screening. Goverde et al. 104 provided estimates for Lynch syndrome carriers only. These two studies did not elaborate on the methods used to combine/pool the results from individual studies that reported lifetime risk of developing CRC. Conversely, Snowsill et al. 43 provided details about the methods used to estimate the lifetime risk of CRC. All economic analyses undertaken over a lifetime time horizon included mortality. People were subjected to endometrial cancer mortality, CRC mortality and age- and sex-specific mortality according to their respective locations. Snowsill et al. 43 derived transition probabilities for the risk of CRC mortality for people with/people without Lynch syndrome and by stage of the cancer. However, it was unclear if stage-specific risks of CRC mortality were derived in other analyses.

Information was required about the performance (sensitivity and specificity) of the different testing strategies included in the economic analyses. Derivation of sensitivity and specificity varied across studies, with most studies obtaining information from the literature, but authors have provided little information about how the evidence was appraised or synthesised. One study43 clearly stated the methodology used to derive pooled estimates, where appropriate. In addition, it was unclear what assumptions were made when combining the test accuracy of individual tests to form a testing strategy.

In all studies, the effectiveness of Lynch syndrome screening was based on cases of Lynch syndrome diagnosed. In addition, studies included the health benefit to women with Lynch syndrome and their first-degree relatives. 43,102,104 Economic analyses that included colonoscopic surveillance estimated the effectiveness/impact of surveillance on the incidence of CRC and mortality. 43,102,104

One study43 reported their results in terms of QALYs. Snowsill et al. 43 elaborated on the assumptions made, with justification about how utility values were estimated. First, baseline utility values were estimated from age- and sex-specific population values. Second, it was assumed that there was no disutility associated with people with stages I–III CRC. People with stage IV CRC had their utility scaled by 0.79, as opposed to 1.00 for stages I–III CRC. Finally, it was assumed that genetic counselling or testing had no impact on QALYs.

All studies reported the perspective of the analysis, but, in one study,102 these costs did not reflect the viewpoint stated. Resource use and costs were required for the costs of screening tests/strategies, genetic consultation and testing, CRC screening and treatment of CRC. All studies reported the sources of costs but, in some studies, it was difficult to decipher the resource use that was used to estimate unit costs.

Uncertainty

Most analyses43,101,102,104 included a one-way sensitivity analysis or scenario analysis by varying key input parameters to reflect lower and upper limits, or by making changes to input parameters if multiple sources of information were available to assess the impact on the base-case ICER, and/or to determine the key drivers of the economic model. It was unclear in some analyses if the sensitivity analysis was exhaustive, as no tornado plots were reported. Results were reported for all sensitivity and scenario analyses. Authors reported which input parameters were the most influential. To our knowledge, ‘best-case’ and ‘worst-case’ analyses were not undertaken. Snowsill et al. 43 explored heterogeneity and undertook a PSA. In addition, no economic analysis undertook a value-of-information assessment.

Assumptions

Authors clearly stated the assumptions made to have an executable model. In general, the assumptions made appeared to be feasible, with others being strong in some instances. There was little overlap between studies about the assumptions made. This may be because of the heterogeneous nature between the economic analyses. As expected, as model complexity increased, so did the number of assumptions. Details of the assumptions made for each study are reported in Appendix 4.

Diagnostic model

Cost-effectiveness results

Table 14 summarises the base-case cost-effectiveness results (prior to rounding). IHC with MLH1 methylation is the most cost-effective strategy, with germline testing direct incurring the highest costs per QALY, compared with no testing. All 11 strategies are considered cost-effective at a WTP threshold of £20,000 per QALY.

Number of people identified as having Lynch syndrome

The number of probands and relatives with Lynch syndrome identified for the 11 strategies are illustrated in Figure 21. This shows that very similar numbers of true positive Lynch syndrome individuals are identified across all testing strategies except for strategies 2 and 6 (MSI with MLH1 promoter hypermethylation testing and MSI followed by IHC with MLH1 promoter hypermethylation testing). This is expected, as sensitivity of these testing strategies are the lowest of all the strategies, at 62.5%. This is a result of our assumptions on test accuracy, which are uncertain.

FIGURE 21.

Number of probands and relatives with Lynch syndrome identified by each strategy. FN, false negative; FP, false positive; TP true positive.

Diagnostic performance is diminished across all strategies, as some of the relatives identified decline the offer of predictive testing. In addition, if probands receive a positive diagnosis but no causative mutation is found (i.e. Lynch syndrome-assumed diagnosis), their first-degree relatives are treated as having Lynch syndrome, but second-degree relatives and beyond are treated as not having Lynch syndrome.

Long-term clinical outcomes

Results from long term-modelling of cancer outcomes

Cumulative incidence of identifying Lynch syndrome

The graphs in Figures 22–24 show model predictions of CRC and endometrial cancer incidence, by gene, from birth, illustrating our assumptions for incidence of the respective cancers in Lynch syndrome individuals without diagnosis/intervention. This is used to simulate outcomes for these individuals with and without interventions as a result of being identified as having Lynch syndrome.

FIGURE 22.

Cumulative incidence of CRC among females with Lynch syndrome.

FIGURE 23.

Cumulative incidence of CRC among males with Lynch syndrome.

FIGURE 24.

Cumulative incidence of EC among females with Lynch syndrome. EC, endometrial cancer.

Figure 25 shows the predicted improvement in life expectancy when a person of a given age who has not previously had a Lynch syndrome-related cancer is identified with Lynch syndrome (through cascade testing) and measures are initiated to reduce their risks. For women, being identified at age 30 years through cascade testing results in an extra 6.7 years of life, falling to 0.9 years if the woman is identified at age 70 years. For men, the equivalent predicted gains are 7.4 years, falling to 0.7 years.

FIGURE 25.

Predicted improvement in life expectancy with age at identification.

Despite the magnitude of benefits in terms of life-years gained declining as age of identification rises, this graph demonstrates that some degree of benefit is maintained through identification at any point across the life course until at least age 70 years, as modelled.

Figure 26 shows the benefits of identifying Lynch syndrome in a cohort of women of the same age in terms of cases of CRC and endometrial cancer avoided, and deaths from CRC or endometrial cancer averted, when Lynch syndrome is identified. Results are presented per 100 women identified. Figure 27 shows the CRC cases prevented and deaths averted per 100 men identified.

FIGURE 26.

Benefits of identifying 100 women with Lynch syndrome in a cohort of the same age. EC, endometrial cancer.

FIGURE 27.

Benefits of identifying 100 men with Lynch syndrome in a cohort of the same age.

The number of CRC cases and CRC deaths prevented when 100 women of a given age who have Lynch syndrome, and have recently presented with endometrial cancer, benefit from CRC surveillance and risk reduction are presented in Figure 28. This declines with age as the relative risk of dying from other causes increases.

FIGURE 28.

Number of CRC cases and CRC deaths prevented by identification of 100 endometrial cancer probands.

Additional outcomes

Predicted lifetime QALY gains by age of Lynch identification

Figure 29 shows the QALY gains predicted by the lifetime cancer model, as a function of the age at which a person is identified with Lynch syndrome. As expected, these decrease with age, as the number of life-years that can be gained falls as the age at which cancer would present increases. QALY gains are similar for the three groups, except for younger men, who gain greater benefit from CRC protection as a result of their increased risk.

FIGURE 29.

The QALY gains predicted by the lifetime cancer model as a function of the age at which a person is identified as having Lynch syndrome.

Disaggregated costs

If a person is identified with Lynch syndrome, they will incur additional costs owing to protective measures such as surveillance and prophylactic surgery. At the same time, they may incur reduced Lynch syndrome-related cancer treatment costs if the protective measures are effective. Figure 30 shows how the take-up of such measures, from the age at which a person is identified with Lynch syndrome, affects total costs. The costs for female relatives is significantly higher, largely because of the costs of prophylactic surgery to prevent endometrial cancer, which is not incurred by men or by women who have been diagnosed with endometrial cancer.

FIGURE 30.

Cost impact of Lynch syndrome management across probands and male and female relatives.

Subgroup analyses

Potential subgroup analyses of reflex testing in endometrial cancer probands aged < 70 years and probands who had previously had CRC but who did not already have a Lynch syndrome status assigned were not feasible; therefore, no results are presented.

Scenario analyses

Deterministic/probabilistic sensitivity analyses

One-way sensitivity analysis results

Deterministic sensitivity analysis results were conducted by varying key model input parameters used in the base-case analysis to assess the impact on the cost per QALY, and presented in the form of tornado plots. Figure 31 shows the tornado plot for the comparison between IHC with MLH1 methylation and a no-testing strategy. We chose this comparison because, in the base-case analysis, the incremental results showed that IHC with MLH1 methylation was the most cost-effective strategy. In addition, IHC with MLH1 methylation had the most cost-effective ICER (approximately £9460 per QALY) when each testing strategy was compared with a no-testing strategy. The sensitivity analysis results show which parameter is the key driver of the cost-effectiveness. These results show that varying the percentage of relatives accepting counselling was the most influential parameter. Decreasing the number of relatives who accept counselling by 50% led to an increase in the ICER. Likewise, increasing the percentage of relatives who accept counselling led to a decrease in the ICER. In the model, these relatives receive the germline tests if appropriate, but do not incur the costs of genetic counselling. Moreover, as expected, if there was a decrease in the prevalence of Lynch syndrome among women with endometrial cancer, the ICER increased to approximately £13,640 per QALY. Similarly, if the prevalence was increased to 6.4%, this resulted in an ICER of approximately £7350 per QALY. Based on the parameters varied, the ICER changed slightly, but remained below current WTP thresholds.

FIGURE 31.

Tornado plot for the impact of a ± 50% change in individual parameters on the ICER per QALY gained.

Probabilistic sensitivity analysis results

We report the PSA results that were generated by assigning distributions to key input parameters and randomly sampling from these distributions over 10,000 simulations to derive any uncertainty in the costs and outcomes. The PSA results for IHC with methylation, compared with no testing, produced an incremental cost of £600 and an incremental QALY of 0.0517, giving an ICER of £11,600 per QALY gained. Exact results have been obtained from TreeAge but were rounded by the authors. We chose this comparison because this strategy was shown to be the most cost-effective in the base-case analysis, and, across all scenario analyses, the results remained robust. Including the combined uncertainty across the parameters included in the PSA showed that the expected mean costs and QALYs yielded in the base case were underestimated, which resulted in an ICER greater than that in the deterministic results.

The probabilistic results are presented in the form of an incremental scatterplot and its corresponding CEACs. Figure 32 presents the results of the 10,000 runs of the Monte Carlo simulations; the scatterplot shows that there is some variation in the incremental costs and QALYs. Figure 33 shows the probabilistic results presented in the form of a CEAC, which shows the probability that an intervention is cost-effective at different WTP thresholds per QALY. At a WTP threshold of £20,000 per QALY, IHC has a 0.93 probability of being cost-effective when compared with no testing.

FIGURE 32.

Incremental cost-effectiveness scatterplot for the comparison between IHC with MLH1 versus no screening.

FIGURE 33.

The CEAC for the IHC with methylation strategy at different WTP thresholds.

MEDLINE (via Ovid)

Database: MEDLINE(R) ALL (via Ovid).

Date range searched: 1946 to 6 August 2019.

Date searched: 7 August 2019.

EMBASE (via Ovid)

Database: EMBASE Classic plus EMBASE.

Date range searched: 1947 to 6 August 2019.

Date searched: 7 August 2019.

Cochrane Database of Systematic Reviews and Cochrane Central Register of Controlled Trials (both via Wiley Online Library)

Date searched: 8 August 2019.

Date range searched: inception to 8 August 2019.

Database of Abstracts of Reviews of Effects and Health Technology Assessment Database (both via Centre for Reviews and Dissemination)

Date searched: 8 August 2019.

Date range searched: inception to 8 August 2019.

Science Citation Index and Conference Proceedings Citation Index – Science (via Web of Science)

Date searched: 8 August 2019.

Date range searched: inception to 8 August 2019.