A systematic review and economic evaluation of diagnostic strategies for Lynch syndrome

Diagnosis/testing

Currently, the Amsterdam criteria (AC) II and Revised Bethesda criteria, as seen in Table 2, may be used to assist with diagnosis of LS. In 1989, the AC were proposed in order to provide uniform family material required for international collaborative research studies. In 1999, these criteria were revised to include extracolonic tumours. 1 However, with the development of techniques to investigate tumours, such as MSI and MMR immunohistochemistry (IHC), in 1997 the Bethesda guidelines were developed to aid selection of tumours for testing and subsequently identifying individuals with LS. These guidelines were revised in 2004. It should be noted that all AC must be met whereas only one Bethesda criterion is necessary.

The Bethesda criteria include MSI-high (MSI-H). This refers to MSI testing where the National Cancer Institute (NCI) has recommended a panel of five markers, known as Bethesda (or NCI) markers, which include two mononucleotides (BAT25 and BAT26) and three dinucleotide repeats (D2S123, D5S346 and D17S250). Tumours with no instability in any of the markers are considered to be microsatellite stable (MSS). When one reference marker is mutated, a tumour is considered to be MSI-low (MSI-L), and if two or more markers are altered, it is considered to be MSI-H. 12 In some cases an additional panel of five markers is used; if 3 out of 10 show instability then it is classified as MSI-H, and if two or fewer, MSI-L.

Unfortunately, there are limitations to MSI testing due to MLH1 silencing commonly occurring in non-hereditary cancers. Thus, MSI is found in approximately 15% of sporadic CRC cases (i.e. CRC with no apparent hereditary component),1 and according to Umar and colleagues, as many as 50% of suspected cases of LS are not confirmed by a genetic defect (that is, mutation in one of the known MMR genes). 12 Hence, the Bethesda criteria have been criticised as being insensitive and non-specific, because strictly applied they would result in approximately 25% of all CRC being tested. In turn, this has stimulated the development of additional tests for the diagnosis of LS, as presented in Table 3.

Current evidence supports genetic testing for LS to include:13

1. evaluation of tumour tissue for MSI through molecular MSI testing and/or IHC of the four MMR proteins (MLH1, MSH2, MSH6 and PMS2)

2. molecular genetic testing of the tumour for MLH1 gene methylation and/or somatic BRAF V600E mutation to help identify those tumours more likely to be sporadic than hereditary, as the presence of a BRAF V600E mutation makes LS very unlikely1

3. molecular genetic testing of the MMR genes to identify a constitutional (germline) mutation when findings are consistent with LS.

Prognosis

Colorectal tumours in LS appear to evolve through the adenoma–carcinoma sequence. However, this progression is accelerated compared with CRC in sporadic and other familial settings, i.e. 2–3 years as opposed to 8–10 years. 12,14 Furthermore, adenomas in LS often occur in younger individuals and tend to be larger and more severely dysplastic than in sporadic cases. 14 That said, recent studies have confirmed early suspicions that patients with CRC from LS families survive longer than sporadic CRC patients with same-stage tumours. 15 The reasons for the favourable survival rate with CRC in this syndrome remain unclear, but are likely related to a reduced propensity to metastasise. Explanations include that immunological host defence mechanisms may be more active in tumours of the MSI Pathology Research International 3 phenotype, and that the relatively high mutational load that occurs in tumours with defective DNA repair systems is detrimental to their survival. 14

Furthermore, there is definite evidence for a genotype–phenotype correlation in LS; for example, one study found that MSH6 mutation carriers had markedly lower cancer risks overall than MLH1 or MSH2 mutation carriers. 2 Carriers of a MMR gene mutation have a very high risk of developing CRC (25–70%) and endometrial cancer (EC) (30–70%) and an increased risk of developing other tumours. 5

Surveillance

As LS is a hereditary condition, identification of family members carrying a MMR gene defect is desirable, as colonoscopic surveillance, and possibly prophylactic and/or altered surgical management, may be offered to high-risk individuals.

Given that screening for a mutation is time-consuming and expensive – largely because four genes may have to be analysed and their mutational spectra are wide (Vasen 20071) – the British Society of Gastroenterology (BSG) and the Association of Coloproctology of Great Britain and Ireland (ACPGBI)9 recommend that individuals with a substantially elevated personal risk of gastrointestinal malignancy be offered surveillance on the basis of one or more of the following criteria:

• a family history (FH) consistent with an autosomal dominant cancer syndrome

• pathognomonic features of a characterised polyposis syndrome personally or in a close relative

• the presence of a constitutional (‘germline’) pathogenic mutation in a CRC susceptibility gene

• molecular features of a familial syndrome in a CRC arising in a first-degree relative (FDR).

Individuals fulfilling at least one of the above criteria should be referred to a NHS regional genetics centre for assessment, genetic counselling and mutation analysis of relevant genes, where appropriate.

Vasen and colleagues (2007)1 highlight a study in which 10-year surveillance of 22 LS families reduced the development of CRC by 60% and also decreased mortality. 10,16 Appropriately targeted surveillance also means that those without a gene defect may be spared intensified surveillance, which is costly and carries not insignificant risks of morbidity and mortality. 1

If LS has been identified, large bowel surveillance is recommended by the BSG and ACPGBI for probands and family members as follows:9

Total colonic surveillance (at least biennial) should commence at age 25 years. Surveillance colonoscopy every 18 months may be appropriate because of the occurrence of interval cancers in some series. Surveillance should continue to age 70–75 years or until co-morbidity makes it clinically inappropriate. If a causative mutation is identified in a relative and the consultand is a non-carrier, surveillance should cease and measures to counter general population risk should be applied.

Reproduced from Gut, Cairns SR, Scholefield JH, Steele RJ, Dunlop MG, Thomas HJ, Evans GD, et al. , Volume 59, pp. 666–89, 2010 with permission from BMJ Publishing Group Ltd.

• Families fulfilling Amsterdam criteria, but without evidence of DNA mismatch repair gene defects (following negative analysis of constitutional DNA and negative tumour analysis by microsatellite instability testing/immunohistochemistry), require less frequent colonoscopic surveillance.

• Gastrointestinal surveillance should cease for people tested negative by an accredited genetics laboratory for a characterised pathogenic germ-line mutation shown to be present in the family, unless there was a significant, coincidental finding on prior colonoscopy.

• Reproduced from Gut, Cairns SR, Scholefield JH, Steele RJ, Dunlop MG, Thomas HJ, Evans GD, et al. , Volume 59, pp. 666–89, 2010 with permission from BMJ Publishing Group Ltd.

  The evidence for upper gastrointestinal surveillance in all of these disorders is weak, but limited evidence suggests it may be beneficial.

Debate continues regarding the appropriate age for and frequency of surveillance, but the above criteria are in agreement with further published data. 1 However, the situation becomes more complex when the proband does not have a detectable DNA alteration associated with LS, or when an alteration with an unclear significance is identified. 4 Vasen and colleagues (2007)1 suggest that this is the case for approximately 30% of families meeting the AC I, for whom a less intensive surveillance protocol may be recommended (i.e. colonoscopy at 3–5 year intervals, starting 5–10 years before the first diagnosis of CRC or at > 45 years).

Surgical management

Several studies have shown that patients with LS have an increased risk of developing multiple (synchronous and metachronous) CRCs. 1 The type of surgery received, i.e. total or subtotal colectomy, depends on the location of the tumour and the stage of the cancer. Studies have shown that adenomas in patients with LS are located mainly in the proximal colon (ascending and transverse);14 therefore, a subtotal colectomy is favoured, which involves removal of most of the colon, leaving a small amount to be reattached to the rectum. Clinicians may also discuss prophylactic colectomy as a reasonable option in mutation carriers for whom colonoscopy is painful or difficult, or for a patient with adenomas that cannot be removed easily; however, this remains controversial.

Chemotherapy

At least three chemotherapeutic agents have been proven to be effective in the treatment of CRC – 5-FU (also known as fluorouracil) with or without leucovorin (also known as folinic acid), oxaliplatin and irinotecan – although experimental and clinical studies suggest that MSI-H tumours are resistant to 5-FU-based chemotherapy. Therefore, according to Vasen and colleagues (2007),1 prospective clinical trials are needed before definitive recommendations can be given.

Epidemiological studies have demonstrated that non-steroidal anti-inflammatory drugs (e.g. aspirin) reduce the risk of CRC. 14 A recent study showed that a daily dose of aspirin reduced the incidence of CRC in carriers of LS after 56 months’ follow-up. 17 The mechanisms by which aspirin prevents the development of cancer are unknown, though some have suggested that aspirin may be proapoptotic in the early stages of CRC development. Importantly, the Colorectal Adenoma/Carcinoma Prevention Programme 2 (CAPP2) trial of aspirin prophylaxis in LS has demonstrated that aspirin treatment for up to 3 years reduces, a decade later, the overall incidence of LS-associated cancers, including CRC, by 63%. 17 A further dosage determination trial (CAPP3) is therefore planned in LS patients worldwide (www.capp3.org) and highlights the importance of identifying individuals and families with LS.

Immunohistochemistry

In families with an increased probability of a MMR gene mutation, IHC analysis for MMR proteins MSH2, MLH1 and MSH6 in tumour tissue may be used as the first step to confirm the presence of MMR deficiency. Pathogenic mutations in MMR proteins frequently lead to the absence of a detectable gene product, or expression of the protein in an abnormal location, for example in the cytoplasm rather than the cell nucleus. Therefore, when tumour tissue from patients suspected of having LS is stained for MMR proteins, a negative or less intense nuclear staining may be visible as compared with the surrounding normal colonic tissue used as a positive control. 4,18

The advantage of IHC, as opposed to MSI, is that abnormal staining of a specific MMR protein is related to the underlying gene defect and can therefore direct further genetic mutation analysis. 1 IHC is a well-established technique widely available in cell pathology laboratories; however, when used to analyse MMR proteins in the setting of LS diagnosis, it must be performed to an adequate standard. Hence, at a workshop in 2006 it was decided that MMR IHC should only be available within the NHS via a laboratory accredited to Clinical Pathology Accreditation standards, obliged to participate in the UK National External Quality Assessment Scheme Immunocytochemistry (NEQAS ICC) for MMR proteins. 19 This workshop made a number of recommendations, including that MMR IHC should be performed for all four main MMR proteins, in part to address the issue of tissue fixation artefact. Care must also be taken in histopathological interpretation of MMR IHC that an adequate and representative tissue sample has been analysed.

Although MMR IHC can give useful and informative results, its sensitivity is limited by a number of factors, for example tissue fixation, the variety and different performance characteristics of primary antibodies and the fact that some pathogenic mutations may result in catalytically inactive but antigenically intact proteins. 15,20–23 Hence, there is a place for MSI analysis in cases with a high prior probability of LS, but with apparently normal expression of the MMR proteins. 1

A particular issue with IHC is that approximately 15% of sporadic colon cancers lose expression of MLH1 because of somatic hypermethylation of the gene’s promoter. Therefore, whereas abnormal expression of MSH2, MSH6 or PMS2 is in itself reasonably good evidence that a tumour was due to LS, loss of MLH1 in itself is not. Other evidence must be used in interpretation in these circumstances, and thus testing for BRAF V600E and/or MLH1 promoter methylation may also be performed. The presence of the BRAF V600E mutation indicates a sporadic rather than LS-associated CRC, but the absence of BRAF V600E does not distinguish between sporadic tumours and those caused by LS. Similarly, MLH1 promoter methylation is highly correlated with a sporadic origin for a tumour, but is not absolutely conclusive, because individuals and families are described with constitutional MLH1 promoter methylation defects. 24

Microsatellite instability testing

Microsatellite instability refers to the variety of patterns of microsatellite repeats observed when DNA is amplified from a tumour with defective MMR compared with DNA amplified from surrounding normal colonic tissue. Repetitive mono- or dinucleotide DNA sequences (microsatellites) are particularly vulnerable to defective MMR. 4 MSI is prevalent in tumours from patients with MMR mutations, and in patients meeting either AC. 4 Therefore, microsatellite analysis is commonly used as the first diagnostic screening test for LS. 18

Microsatellite instability testing involves amplification of a standardised panel of DNA markers (Bethesda/National Institutes of Health markers), although laboratories may use 10 or more markers and, more recently, a commercially available kit based on five mononucleotide markers has become popular as mononucleotide microsatellites may be the most sensitive markers for use in detecting MSI. 4 The process involves microdissection of tumour tissue, followed by extraction of DNA which is then amplified and run on a DNA fragment length analyser. Using such microsatellite markers, additional peaks in tumour tissue DNA in comparison with normal tissue DNA indicate MSI. 25 Instability in 30% or more of the markers is considered MSI-H, less than 30% MSI-L and no shifts or additional peaks MSS. However, if instability is observed at any mononucleotide markers, MSI may be diagnosed. For this reason, MSI testing is moving to a smaller panel of mononucleotide markers, making the process more efficient and cheaper.

As for any molecular pathological analysis, tissue to be selected for MSI analysis must be first assessed by a histopathologist, prior to some degree of microdissection, which aids in maximising sensitivity. There is debate regarding the relative costs of MMR IHC and MSI testing, but NHS service laboratory costings indicate there is little to choose between the two. MSI may be more reproducible and can be performed with smaller amounts of tissue. 26 As there is not yet a UK NEQAS scheme for MSI, the reproducibility of MSI compared with IHC is not established.

BRAF V600E and methylation testing

The presence of MSI in the tumour by itself is not sufficient to diagnose LS because 10–15% of sporadic CRCs exhibit MSI. 25 MSI in non-LS tumours is usually caused by hypermethylation of the MLH1 gene. This acquired epigenetic inactivation of MLH1 is typically associated with mutations in the BRAF gene (specifically the V600E mutation), which has been described in ≈ 35% of sporadic MSI-H CRCs. 25 Therefore, identification of hypermethylation of MLH1 and/or BRAF V600E is an indication that a patient does not have the LS germline mutation.

Ideally, tests would be performed together as the presence of the BRAF V600E mutation theoretically reduces the chance of LS as the cause of that tumour; however, because any test has a finite false negative (FN) rate, it is still a possibility. Additionally, if MLH1 promoter methylation is present but the BRAF V600E mutation is not, this would highlight the small possibility that the patient may have LS due to a constitutional MLH1 methylation defect. It is also possible that he or she could have an inherited MLH1 genetic mutation and could have acquired MLH1 promoter methylation as the ‘second hit’ in the tumour. In these cases, loss of heterozygosity of chromosome 3p (where MLH1 is located) is observed. 25

Constitutional genetic testing

Multiple methods have been used for constitutional genetic testing (tests for mutations that affect all cells in the body and have been there since conception) in LS, in order to find inherited or, if de novo, potentially inheritable MMR gene mutations. The method(s) used should ideally be able to detect any possible mutation associated with LS, for example nonsense, missense and frameshift mutations, genomic deletions, duplications and rearrangements, as explained in Tables 4 and 5. 4

Measuring the accuracy of diagnostic tests for Lynch syndrome

One aspect of the evaluation of new tests is measuring their accuracy by calculating their sensitivity and specificity. This requires specification of the best available method of identifying the target condition of interest, known as the reference standard. Most mutations causing LS are point mutations or small insertions or deletions, suitably detected by DNA sequencing. However, some LS-associated mutations are deletions/duplications of exons in MLH1 and MSH2. These are more difficult to detect and, currently, the most appropriate technology available is multiplex ligation-dependent probe amplification (MLPA), which is a multiplex polymerase chain reaction (PCR) method able to simultaneously detect copy number changes across multiple DNA sequences within one sample. Therefore, the ideal reference standard is considered to be sequencing plus MLPA.

Population

• All newly diagnosed patients under the age of 50 years with CRC.

• Participants considered to be at high risk of LS, i.e. those fulfilling AC II or Bethesda criteria.

• Individuals with personal cancer history or FH indicators.

Intervention

Tumour-based tests for evidence of mutations in the genes encoding the MLH1, MSH2, MSH6 and PMS2 DNA MMR enzymes. These tests include MSI, IHC, BRAF and methylation.

Comparators

Genetic testing by sequencing followed by MLPA is considered the gold standard.

Design

An evidence synthesis by systematic review to determine the accuracy of tumour-based tests.

Health-care setting

Primary and secondary care settings.

Test outcomes

The outcomes of interest include measures of:

• diagnostic test accuracy

• test failure rate

• discordant test results.

Cost considerations

The cost analysis will be based on the UK NHS setting and will be from an NHS and Personal Social Services (PSS) perspective.

The costs for consideration include:

• cost of equipment, any additional tests (pre screening), reagents and consumables, participation in NEQAS

• staff and training of staff

• maintenance of equipment

• costs associated with surgeon time and the management of operating theatre time

• medical costs arising from ongoing care following test results, including those associated with clinical genetics, surgery, time spent in hospital and treatment of cancer.

Identification of studies

The search used clusters for LS and HNPCC, joined together using the Boolean connector OR for sensitivity. The following databases were searched: MEDLINE, MEDLINE In-Process & Other Non-Indexed Citations, EMBASE, PsycINFO, Health Management Information Consortium (HMIC) (all via Ovid), The Cochrane Library (all), Cumulative Index to Nursing and Allied Health Literature (CINAHL) (via EBSCOhost), Applied Social Sciences Index and Abstracts (ASSIA) [via Cambridge Scientific Abstracts (CSA)] and Web of Science [via Institute for Scientific Information (ISI)]. The search was limited to human-only populations and to the English language, but did not use any methodological search filters. The National Research Register (NRR), Current Controlled Trials, ClinicalTrials.gov, the Food and Drug Administration (FDA) website, the European Medicines Agency (EMEA) website and Google were also searched. The search is recorded in Appendix 1.

Searches were deduplicated and managed using EndNote X5 (Thomson Reuters, CA, USA). Relevant studies were then identified in two stages. Titles and abstracts returned by the search strategy were examined independently by two researchers (TJH and HC) and screened for possible inclusion. Disagreements were resolved by discussion. Full texts of the studies which could not be excluded were obtained. Two researchers (TJH and HC) examined these independently for inclusion or exclusion, and disagreements were again resolved by discussion.

Inclusion and exclusion criteria

Data extraction strategy

Data were extracted by one reviewer (TJH) using a standardised data extraction form and checked by a second reviewer (HC). Disagreements were resolved by discussion, with involvement of a third reviewer if necessary. Appendix 2 shows the blank data extraction forms used.

Critical appraisal strategy

The methodological quality of the studies was assessed according to criteria specified by the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool for test accuracy studies. 29

Quality was assessed by one reviewer and judgements were checked by a second. Any disagreement was resolved by discussion, with involvement of a third reviewer as necessary. The two instruments are summarised below. Results were tabulated and the relevant aspects described in the data extraction forms.

Methods of data synthesis

Details of the extracted data and quality assessment for each individual study are presented in structured tables and as a narrative description. Any possible effects of study quality on the effectiveness data are discussed. Data on test accuracy are presented as sensitivity and specificity, where available.

In most of the studies, the accuracy of the interventions has been evaluated against the reference (gold) standard of constitutional genetic testing and thus, for the purpose of this assessment of test accuracy, a genetic definition of LS is assumed. The results are generally reported as follows:

• Sensitivity: true positive (TP)/(TP + FN). This is the probability of detecting LS in someone with LS.

• Specificity: true negative (TN)/[false positive (FP) + TN). This is the probability of not detecting LS in someone without LS.

• Positive predictive value (PPV): TP/(TP + FP). This is the probability of someone with a positive result actually having LS.

• Negative predictive value (NPV): TN/(TN + FN). This is the probability of someone with a negative test result actually not having LS.

• Accuracy or concordance with reference standard: (TP + TN)/(TP + TN + FP + FN). This is the percentage of test results correctly identified by the test, i.e. the rate of agreement with the reference standard.

• Discordance: cases of disagreement between the reference and index test.

Number of studies identified

The electronic searches retrieved a total of 3713 titles and abstracts. A total of 3640 papers were excluded, based on screening of title and abstract. As a relevant technology assessment (TA) was retrieved in the search [Bonis and colleagues (2007)4], rather than duplicate effort, we included studies dated from 2005 onwards, and provide a summary of findings by the previous TA. Full text of the remaining 73 papers was requested for more in-depth screening, to give a total of 10 published papers included in the review. The process of study selection is shown in Figure 1.

FIGURE 1.

Summary of study selection. a, It is unclear whether or not two of the included studies are from the same population.

Number of excluded studies

Papers were excluded for at least one of the following reasons: duplicate publication; narrative review; and publication (systematic review or individual primary study) not considering the relevant intervention, population, comparison or outcomes. The bibliographic details of the 73 studies retrieved as full papers and subsequently excluded, along with the reasons for their exclusion, are detailed in Appendix 3.

Number and description of included studies

Assessment of test accuracy

Individual results for this review from 2005 to 2012 are presented alongside a narrative description below. Summary tables are displayed at the end of this section (see Tables 23 and 24).

Summary of test accuracy studies

Nine primary research studies and one technology appraisal were identified for inclusion in this review. Four studies investigated MSI testing and eight investigated IHC. As tests to identify LS are a rapidly evolving area, and in the knowledge that a TA was already in existence which had identified studies up to 2005, only studies from 2005 onwards were included in this review.

Overall, the quality of the studies was mixed and, as many studies were not directly comparable owing to variations in design, the data could not be pooled. The results are summarised in Tables 23 and 24.

Methods

Searches of electronic databases were devised by a trained information specialist (CC) and applied to the following databases:

• MEDLINE (via Ovid)

• MEDLINE In-Process & Other Non-Indexed Citations (via Ovid)

• EMBASE (via Ovid)

• PsycINFO (via Ovid)

• HMIC (via Ovid)

• EconLit (via EBSCOhost)

• CINAHL (via EBSCOhost)

• Web of Science (via ISI)

• The Cochrane Library

• NRR

• Web of Science Proceedings

• Current Controlled Trials

• ClinicalTrials.gov

• FDA website

• EMEA website.

A search filter was used to identify economic evaluations, as used for identifying studies for inclusion in the NHS Economic Evaluation Database (NHS EED). No publication date limit was imposed. Searching was limited to the English language and to human-only populations.

Study selection criteria and procedures

Titles and abstracts obtained from searching were screened by four reviewers (NH, CH, RM and TS) using inclusion criteria described in Table 25, with disagreements resolved by discussion between two reviewers (NH and TS). An inclusion criterion relating to guidelines was included to assist in the understanding of LS. Full papers of citations that met these criteria were obtained and assessed for inclusion in the review by two reviewers (NH and TS), with disagreements resolved by discussion by the same two reviewers.

Results

The flow diagram of papers is summarised in Figure 2. In summary, 2036 citations were identified, 227 of which were ordered in full. Three of these could not be retrieved but the information available from the titles and abstracts did not suggest that they would have a high probability of remaining included studies. Of the 224 which were retrieved, 7 were duplicates and 119 were excluded. Of the remaining 98, 1 was an opinion piece, 55 were guidelines only, 3 were previous reviews of cost-effectiveness studies, 5 were purely effectiveness models (kept for their insight into modelling this problem) and the remaining 34 included papers represented cost-effectiveness studies of some form. An additional three cost-effectiveness papers, one of which was an update of a previously identified paper, and one additional review were identified during the update search. Of the final 37 cost-effectiveness papers, 16 looked at the short-term cost-effectiveness of identifying LS, 5 looked only at the long-term cost-effectiveness of management of LS and 16 examined the long-term impact on cost-effectiveness of both the strategies to identify and manage LS. Of the 16 sources that only looked at short-term cost-effectiveness, there were 15 distinct studies (one of the papers was an abstract of a study written up in full in another paper). Of the five that looked purely at management of LS, only four were distinct studies (as one paper was an abstract of preliminary results which were reported in full in another paper). Of the 16 sources that looked at the cost-effectiveness of both the strategies to identify and manage LS, there were two papers on the same study, one abstract of another paper and a commentary on a paper, making the total number of distinct studies 13. The most common reasons for exclusion were on the basis of a study design or population not relevant to the review.

FIGURE 2.

Flow diagram of the search, retrieval and inclusion of articles in the systematic review of evidence on the economic evaluations to identify and manage LS.

We did not formally review the papers that included guidelines, but they were generally consistent in their suggestions for the management of LS, i.e. routine surveillance for CRC by 1- to 2-yearly colonoscopy. As our criteria for guidance were quite inclusive, we included a wide range of sources for this, from different study groups including working groups and HTA groups, among others.

Conclusions

Despite the large number of studies identified by this review, there was still the need for a new model to be developed to address our decision problem.

Firstly, the majority of studies identified by this review addressed either diagnosis or management of LS only, and were further divided by their disease focus (predominantly CRC or EC). Therefore, each of these studies could only attempt to answer part of our study question.

The remaining studies that looked at both diagnosis and management were hindered by the advancements in understanding of LS, and populations and country settings not relevant to our analysis. On the basis of population and the parameter used for CRC risk, the studies could therefore be narrowed down to two that would be directly comparable with our model: Ladabaum and colleagues (2011)48 [plus Wang and colleagues (2012)57] and Mvundura and colleagues (2010). 54 Neither of these were UK-based studies and therefore the results could not necessarily be translated to the NHS.

The wide range of diagnostic strategies across all the studies did not make it clear which tests or combinations thereof would be most cost-effective, particularly in a NHS setting. This therefore justified further modelling.

There had also been little investigation into the modelling of more than one cancer in the long term; in those studies that did investigate more than one cancer, the methods were not clear in terms of whether or not the modelling of interactions between them was actually appropriate. Additionally, adherence to genetic counselling was also something rarely touched upon in detail in the studies. Both of these concerns provided other areas where our model could add value to those that had come before.

Our review agreed with the conclusions of those reviews that had been conducted previously, and this supported our justification for conducting further research into this area.

Diagnostic model

The diagnosis of LS is performed for probands with newly diagnosed CRC and their relatives. It is assumed that all diagnosis occurs at time zero.

We consider the following strategies to identify LS in probands:

1. No genetic testing, subdivided:

   1. 1(1) no testing

   2. 1(2) AC only.

2. IHC four-panel test for MLH1, MSH2, MSH6 and PMS2, followed by genetic testing if IHC result abnormal.

3. IHC four-panel test, followed by BRAF testing for abnormal MLH1 results. Genetic testing is done for any other (not MLH1) abnormal IHC result or for a negative BRAF test (negative for V600E).

4. MSI testing, followed by genetic testing for MSI result.

5. MSI testing, followed by BRAF testing for MSI result, followed by genetic testing for negative BRAF test.

6. MSI testing, followed by BRAF testing for MSI result, followed by IHC testing for negative BRAF test. Genetic testing occurs regardless of IHC result.

7. IHC testing, followed by genetic testing if result abnormal. For normal IHC results: MSI testing, followed by BRAF testing for MSI result followed by genetic testing for negative BRAF test.

8. Universal genetic testing.

There are three possible diagnoses that result from each of these testing strategies: LS mutation positive, LS assumed and LS negative. LS mutation positive occurs when a proband receives a positive genetic test result and LS negative occurs when a proband is ruled out by one of the tests prior to genetic testing, or by FH when genetic testing is either declined or uninformative. ‘LS assumed’ can occur when genetic testing is either uninformative or simply not done (the proband declines testing) and is informed by the proband’s FH. Probands who test mutation positive or LS assumed are offered LS management. It is assumed that only a proportion of probands diagnosed as LS positive or LS assumed accept an offer of LS surveillance.

The diagnosis of LS in relatives of a newly diagnosed CRC proband directly depends on the diagnosis of the proband. Relatives of probands who test mutation positive can be identified as LS positive with a predictive genetic test. FDRs of probands diagnosed as LS assumed cannot be tested, but instead are also assumed to have LS and are offered LS management. Relatives of a higher degree are assumed to be LS negative, as are relatives of probands diagnosed LS negative.

The main outcomes for each testing strategy are:

• the overall sensitivity and specificity for probands alone

• the overall sensitivity and specificity for probands and relatives combined

• the numbers of probands and relatives according to true LS status, diagnosis and management strategy.

The accuracy of individual tests, given by sensitivity and specificity, with regard to either LS in general or to particular genes, were taken from published literature, particularly the Evaluation of Genomic Applications in Practice and Prevention (EGAPP) Working Group Review. 65 Owing to a lack of evidence, the sensitivities and specificities of individual tests are treated as independent of tests performed previously, except for the BRAF test. In common with other cost-effectiveness models [Mvundura and colleagues (2010)54 and Ladabaum and colleagues (2011)48], we assume that the test accuracies apply to all qualities of tumour/blood sample and that tests are always available and successfully produce a result. The only known exception to this is the IHC test, where the failure rate is specifically included in the sensitivity.

There is little published evidence on the rate of acceptance of testing. The following assumptions were made: that the rate of acceptance of a test was independent of any previous tests, and acceptance of one genetic test implied acceptance of all genetic testing. The rate of acceptance of IHC and BRAF was assumed to be the same as for MSI, which was the only tumour-based test for which an estimate was available. The rate of acceptance of genetic counselling allowed for the rate of acceptance of FH screening. For patients who decline genetic counselling and for strategies where genetic testing is not offered, rate of acceptance of FH is estimated separately.

The number of probands aged < 50 years per year in England, 1699, was taken from the Office for National Statistics (ONS). 119 Of these, 8.4%120 were assumed to have LS. These probands were then subdivided into the number with each gene associated with LS.

The number of relatives per proband is set to five in the base case to balance the values from two main UK sources: Barrow and colleagues121 and unpublished data supplied by Ian Frayling (Cardiff University, 2012). Based on two published studies114,120 and these unpublished data, the proportion of relatives that are first degree is estimated at 42%. Next, we estimate that the proportion of relatives that test positive for LS is 44%, based on a random-effects meta-analysis.

The costs of the preliminary tumour tests were obtained either directly from laboratories in the UK or from experts. Costs for genetic tests (for probands and relatives) are taken directly from genetic testing laboratories, identified from the UK Genetic Testing Network (UKGTN). As genetic testing becomes routine, these costs are expected to decrease. We performed a sensitivity analysis to reflect this. The cost of genetic counselling was also included.

The psychological impacts of testing for LS and prophylactic TAHBSO were incorporated into overall health-related quality of life (HRQoL), using data from the literature. All disutilities were assumed to apply for 4 months.

• A disutility of 0.04 was applied to people declining testing.

• A disutility of 0.02 was applied to people accepting testing and subsequently being diagnosed with LS.

• A disutility was applied to people offered TAHBSO and a further disutility was applied if TAHBSO was declined.

Survival model

Uncertainty

Cost-effectiveness analyses often employ probabilistic sensitivity analysis, whereby uncertainty in many of the parameters is simultaneously considered. We did not use probabilistic sensitivity analysis. Instead, we investigated uncertainty as follows. First, we performed the following scenario analyses.

• EC and prophylactic TAHBSO were excluded to assess their impact on the cost-effectiveness results.

• The maximum age of probands was increased from the base case of 50 years to 60 years, and separately to 70 years.

• BRAF testing, which was assumed in some testing strategies in the base case, was replaced by methylation testing.

Second, the sensitivity of cost-effectiveness to individual parameters was assessed either by doubling and halving the parameter value (e.g. the cost of all genetic tests), or by selecting a different plausible data source (e.g. for the disutilities due to CRC).

Diagnostic testing model

Here, we consider two groups: the newly diagnosed CRC patients aged under 50 years (the probands), and their biological relatives. These groups are connected, as shown in Figure 3. This section of the model was built as a decision tree and has no time component. It therefore assumes that diagnosis occurs instantaneously, though the diagnosis of LS may actually take up to several months or even years; laboratories report waiting times of around 20 working days for tumour-based tests, 40 working days for gene sequencing for probands and 10–20 working days for predictive genetic testing for relatives. 122–126 This also does not account for the time taken to collect a tissue/blood sample, transport the sample and arrange counselling sessions, nor when in a proband’s treatment this process is begun. According to our clinical experts (Carole Brewer and Ian Daniels, Royal Devon and Exeter NHS Trust), it is also not standard practice to have LS results available before the proband’s treatment is given. Therefore, to emulate standard practice, where the proband’s LS status is not yet known, we assume that he or she receives standard treatment for CRC. In most cases this is surgical resection with the possibility of chemotherapy or radiotherapy, depending on the stage of the cancer.

FIGURE 3.

Overview decision tree for diagnosing LS.

Diagnostic strategies for probands

Probands follow one of nine diagnostic strategies, chosen on the basis of available tests, existent cost-effectiveness models, expert advice and requirements of the project scope. As the only conclusive test for LS is constitutional (‘germline’) genetic testing, this is used in every testing strategy. We use three well-documented preliminary tumour-based tests: IHC, MSI and BRAF. There is little consistency in the literature about the usage of these tests, either in sequence or individually, so we include a variety of testing strategies. Although FH, for example assessed by the AC, is no longer recommended as a selection tool for the diagnosis of LS, we include it as one of the testing strategies for comparison with the genetic testing strategies. 13 We compare the following eight strategies:

1. No systematic testing to identify LS

   1. No testing to identify LS

   2. AC

2. IHC four-panel test for MLH1, MSH2, MSH6 and PMS2, followed by genetic testing if IHC result abnormal. 54

3. IHC four-panel test, followed by BRAF testing for abnormal MLH1 results. Genetic testing is done for any other (not MLH1) abnormal IHC result or for a negative BRAF test (negative for V600E). 54

4. MSI testing followed by genetic testing for MSI result. 54

5. MSI testing, followed by BRAF testing for MSI result, followed by genetic testing for negative BRAF test (advised by Ian Frayling).

6. MSI testing, followed by BRAF testing for MSI result, followed by IHC testing for negative BRAF test. Genetic testing occurs regardless of IHC result (advised by Ian Frayling).

7. IHC testing, followed by genetic testing if result abnormal. For normal IHC results: MSI testing, followed by BRAF testing for MSI result followed by genetic testing for negative BRAF test (advised by Ian Frayling).

8. Universal genetic testing (i.e. as first and only test for all) (asked for in the project scope).

Figures 4–11 show further details of these eight strategies.

FIGURE 4.

Strategy 1: no testing.

FIGURE 5.

Strategy 2: IHC test followed by genetic testing.

FIGURE 6.

Strategy 3: IHC test followed by BRAF testing and genetic testing.

FIGURE 7.

Strategy 4: MSI test followed by genetic testing.

FIGURE 8.

Strategy 5: MSI test followed by BRAF test and genetic testing.

FIGURE 9.

Strategy 6: MSI test followed by BRAF, IHC and genetic testing.

FIGURE 10.

Strategy 7: IHC test followed by MSI, BRAF and/or genetic testing.

FIGURE 11.

Strategy 8: universal genetic testing.

There are three possible diagnoses given by these testing strategies: LS mutation positive, LS assumed and LS negative. LS mutation positive occurs when a proband receives a positive genetic test result and LS negative occurs when a proband is ruled out by one of the tests prior to genetic testing, or by FH when genetic testing is either declined or uninformative.

‘LS assumed’ can occur when genetic testing is either uninformative or simply not done (the proband declines testing). For example, we assume that a negative genetic test result is uninformative for probands, meaning that although the test did not detect LS, it does not rule it out. This is because there is still debate over which genes and mutations cause LS, some of which will not be identified by current genetic tests. Our experts agree that this is a plausible interpretation as current practice does not necessarily rule out LS on the basis of a genetic test, especially with regard to LS management strategies. The outcome allows us to reflect clinical practice, whereby management strategies are offered to both LS mutation-positive and LS-assumed probands (regular surveillance) and with regard to strategies identifying LS in relatives of both probands diagnosed as LS mutation positive and LS assumed. In practice, the decision to diagnose a proband as LS assumed is informed by FH and we model it similarly. As our probands already fulfil the Revised Bethesda guidelines, we model this based on the AC II. We also use this approach to decide how many of the probands who tested negative for mutations in MLH1, MSH2 and MSH6 should be put forward for PMS2 testing. We do this to reflect the smaller number of PMS2 tests carried out in current practice, compared with testing for MLH1, MSH2 and MSH6. This occurs because testing PMS2 is technically challenging and hence more expensive, and there are correspondingly fewer laboratories capable of testing PMS2 (in the UK there is currently only one126). Using the AC II, instead of a fixed proportion, allows the flexibility of a changing population; the proportion of patients at diagnosis of CRC with a FH will not be the same as the proportion of patients with a FH after tumour testing.

Figures 5–11 show that we split genetic testing into simultaneous testing for MLH1, MSH2 and MSH6 (abbreviated to M126), and a separate test for PMS2 where necessary. This strategy has been adopted in current practice where available, and is likely to become standard as three-gene sequencing becomes more widely available, according to our expert advisors. The only strategy where this differs slightly is strategy 3, where probands with a MLH1 (with or without a PMS2) abnormal IHC result, followed by a BRAF test where V600E is not found, have a single genetic test for MLH1. This is to reflect the phenomenon that a mutation in the MLH1 gene can cause abnormal MLH1 with abnormal PMS2 results in IHC; PMS2 mutations can only cause abnormal IHC results for PMS2; MSH2 mutations can cause abnormal MSH2 and MSH2 with MSH6 IHC results; and MSH6 mutations can only cause abnormal MSH6 IHC results. 127 Those with an abnormal IHC result in any of the other genes follow the genetic testing route of the other strategies, M126 followed by PMS2. In all other strategies, we do not attempt to identify the gene responsible before the genetic testing process, as there can be discrepancies in the quality and interpretation of tumour testing (particularly IHC) between laboratories. 128 It is also assumed that if a proband complies with some genetic testing, he or she agrees to all of it.

It is assumed that only a proportion of probands diagnosed as LS positive or LS assumed accept an offer of LS surveillance.

In Figures 5–10, IHC refers to IHC testing associated with all four known LS genes (MLH1, MSH2, MSH6 and PMS2). Though MSI is simply the initialised form of ‘microsatellite instability’, in our case we use it to mean the associated test for microsatellite instability rather than the state itself. In the diagrams, all positive test results indicate (do not rule out) LS. In other words, the test results reflect the LS diagnosis; so, for example, a positive BRAF test result for V600E is indicated as a negative test result for LS.

In the strategies using MSI, test sensitivities are higher for MLH1 and MSH2 than for MSH6 and PMS2, to reflect the known reduced sensitivity associated with the last two genes. 65 Thus, the MSI section of the model is split into MLH1 with MSH2 (M12) and MSH6 with PMS2 (M6P2), with results reported for both individual genes and overall (as it is actually only one test). The individual gene results are then incorporated into the prevalence at the next stage of the strategy by combining them appropriately. For example, in the case of strategies using MSI, the prevalence after MSI is split between M12 and M6P2 (given the information we have for the sensitivity and specificity of MSI), but as the genetic test for M126 only looks at three of those genes, the input prevalence for the M126 genetic test must only include the prevalences of mutations in MLH1, MSH2 and MSH6. The prevalence for P2 is carried through in the negative population so that it is later used in PMS2 genetic testing.

Testing outcomes for probands

The primary outputs from the short-term model for each testing strategy that lead into the survival (i.e. long-term) section of the model are:

• number of probands with LS receiving LS surveillance

• number of probands with LS not receiving LS surveillance (probands will receive some surveillance in line with BSG guidelines9); these are split into those identified as LS positive but who declined surveillance, and those who were diagnosed LS negative

• number of probands without LS receiving LS surveillance

• number of probands without LS who do not receive LS surveillance (probands will receive some surveillance in line with BSG guidelines9); these are split into those identified as LS positive but who declined surveillance, and those who were diagnosed LS negative.

Other outcomes include overall sensitivity and specificity of each diagnostic strategy. Probands who are diagnosed as LS (either mutation positive or assumed), but who refuse surveillance may still be offered prophylactic surgery for metachronous CRC.

Diagnostic strategies for relatives

Testing of relatives depends on whether the proband was diagnosed as LS positive, LS assumed or LS negative, as follows.

Relatives of probands diagnosed as Lynch syndrome mutation positive

The decision tree for relatives of these probands is shown in Figure 12.

FIGURE 12.

Decision tree for relatives of probands diagnosed LS mutation positive.

Relatives of probands who receive a positive genetic test result are offered a predictive genetic test which targets the particular gene (MLH1, MSH2, MSH6 or PMS2) found to be mutated in the proband. Relatives who accept are then diagnosed as LS mutation positive or LS negative depending on their test result. We assume that the targeted testing for relatives is 100% accurate, so we do not include an option for inconclusive tests. This assumption is confirmed as current best practice by our expert advisors and is also made in other models. 54 Additionally, on advice from our clinical experts, the small numbers of relatives at high risk of LS, but who decline targeted genetic testing, are diagnosed as LS assumed and are offered LS surveillance. Relatives diagnosed as LS mutation positive or LS assumed are offered risk-reducing surgery if they develop CRC, regardless of their surveillance status.

Relatives of probands diagnosed as Lynch syndrome assumed

The decision tree for relatives of these probands is shown in Figure 13.

FIGURE 13.

Decision tree for relatives of probands diagnosed LS assumed.

As agreed by our advisors, FDRs of probands who are diagnosed as LS assumed are also assumed to have LS and are therefore at sufficiently high risk of LS to be offered LS surveillance, as targeted genetic testing for relatives cannot be done without a specific mutation for which to test. Higher-degree relatives (e.g. cousins, grandchildren, nieces and nephews, etc.) of this group are assumed to be LS negative, and therefore do not warrant being offered LS surveillance.

Relatives of probands diagnosed as Lynch syndrome negative

All relatives of probands who are diagnosed LS negative are assumed to be LS negative also, and these relatives are subject to no LS tests or LS surveillance (Figure 14).

FIGURE 14.

Decision tree for relatives of probands diagnosed as LS negative.

Testing outcomes for relatives

The primary short-term model outputs are:

• number of relatives with LS receiving LS surveillance

• number of relatives with LS not receiving LS surveillance (split into those identified as LS positive but who declined surveillance, and those who were diagnosed LS negative)

• number of relatives without LS receiving LS surveillance

• number of relatives without LS who do not receive surveillance (split into those identified as LS positive but who declined surveillance, and those who were diagnosed LS negative).

The sensitivity and specificity for each testing strategy for relatives is recorded.

Long-term management model

All patients continue from the diagnostic section of the model to the survival section (this includes all probands and their relatives, irrespective of their adherence and diagnosis), in which a number of outcomes are simulated.

Outcomes are simulated for 24 patient groups, which are all combinations of:

• male/female

• proband/relative

• truly LS positive/negative

• LS diagnosed and surveillance accepted/LS diagnosed and surveillance declined/LS not diagnosed.

The survival section of the model is an individual sampling model. Each patient is simulated for 1 year at a time and starts each year in a particular health state. A cycle length of 1 year was chosen because it was judged reasonable to assume constant hazard from various risks for a year, and indeed most parameters informing the model are based on annual measurements of risk.

During each year there are a number of competing risks (Table 39) to which the patient may be exposed, and these are all assumed to have a constant hazard rate throughout the year (with the exception of colonoscopies, which have a different mechanism; see Morbidity and mortality of Lynch syndrome surveillance colonoscopy). For each competing risk, a time to event is drawn randomly according to the hazard rate of that event. If none of the times is within a year, then no event has occurred and the patient commences the next year in the same state. If at least one of the times is within a year, then the earliest of the times determines which event occurs and the state is updated accordingly for the next year. Figure 15 gives a graphical representation of the events and states in the model, in the form of an influence diagram.

FIGURE 15.

Influence diagram of management model. (M)CRC, (metachronous) CRC.