Molecular testing strategies for Lynch syndrome in people with colorectal cancer

All of the included studies, except Limburg et al. (2011) and Okkels et al. (2012), assessed MSI testing. There were several differences in the MSI testing procedures used in the included studies. These included variations in the number and types of markers in the panels of MSI markers used and also differences in the categorisation of test results; tumours were categorised using either 2 categories (MSI positive or negative) or 3 categories (MSI‑High [MSI‑H], MSI‑Low [MSI‑L] or microsatellite stable [MSS]). Studies also varied in the thresholds used to categorise MSI.

Sensitivity and specificity values were calculated based on a positive MSI test result for Lynch syndrome being MSI‑H alone or either MSI‑H or MSI‑L, as shown in table 1.

Secondary analyses were carried out if data allowed, with unclassified variants (variations in the sequence of MMR genes that are of unknown clinical significance) considered as positive reference standard results for Lynch syndrome (as opposed to negative reference standard results, as in primary analyses). The EAG noted that results were similar to those obtained when unclassified variants were considered as negative.

IHC for MMR proteins was carried out in all of the 10 included studies, although 2 of the studies did not have enough data to be included in the IHC analyses: the high-risk samples in Poynter et al. (2008) and Mueller et al. (2009).

The accuracy estimates from included studies are shown in table 2. The proteins targeted by the tests used and the way results were reported differed between the studies. In 7 studies (Barnetson et al. 2006; Limburg et al. 2011; Southey et al. 2005; Caldes et al. 2004; Overbeek et al. 2007; Shia et al. 2005; Hendriks et al. 2003), an overall result was given, that is, when abnormal staining of any of the MMR proteins assessed was classed as a positive IHC result. All of these 7 studies assessed MLH1, MSH2 and MSH6 proteins. Southey et al. (2005) and Overbeek et al. (2007) also assessed PMS2. So, for these 2 studies, an abnormal PMS2 result would also be included as a positive index test result.

Only 2 studies (Caldes et al. 2004; Hendriks et al. 2003) had enough data to be included in the secondary analyses (in which unclassified variants were considered as positive reference standard results for Lynch syndrome). Only sensitivity estimates could be made because Caldes et al. included people at high risk of Lynch syndrome and Hendriks et al. included people known to have Lynch syndrome. Caldes et al. showed a reduction in sensitivity (75.0%; 95% confidence interval [CI] 57.8 to 87.9) compared with the primary analyses in which unclassified variants were categorised as negative reference standard tests (96.4%; 95% CI 81.7 to 99.9). For Hendriks et al., sensitivity was only slightly reduced from 91.7% (95% CI 77.5 to 98.2) to 88.6% (95% CI 76.0 to 95.0).

No end-to-end studies meeting the inclusion criteria for the systematic review were identified.

The model included:

• a decision tree model to investigate the short-term outcomes of strategies to identify people with Lynch syndrome and

• an individual patient simulation model to assess the long-term implications of strategies to identify and manage Lynch syndrome; the model considers longer-term outcomes for both colorectal and endometrial cancer.

The decision tree started with people diagnosed with colorectal cancer (called 'probands') who could have 1 of 10 diagnostic strategies for Lynch syndrome, as described in table 3. As a result of these diagnostic strategies, probands were either diagnosed as LS‑positive, LS‑negative or LS‑assumed (if they refused genetic testing). People who were diagnosed as LS‑positive or LS‑assumed were offered 2‑yearly colonoscopies, which they could either accept or decline. People diagnosed as LS‑negative had standard colorectal cancer follow‑up and surveillance.

Decision tree models were also included for relatives of probands. Those diagnosed as LS‑positive were offered testing (which they could accept or decline). Relatives who tested positive for Lynch syndrome, or who declined testing, were offered surveillance (which they could either accept or decline). First-degree relatives of probands diagnosed as LS‑assumed were also offered surveillance. No further action was taken for the relatives of probands who did not have Lynch syndrome.

The longer-term model included outcomes relating to surveillance and treatment for both colorectal cancer and gynaecological (endometrial) cancer. Longer-term outcomes were modelled for all probands and relatives (regardless of the diagnostic path they follow) using an individual patient sampling model to simulate 240,000 patients, distributed across 24 groups, representing all combinations of the following variables:

• whether the person was a proband or relative

• whether the person had Lynch syndrome

• whether the person had been diagnosed with Lynch syndrome and accepted or declined surveillance

• sex.

Patients were simulated for 1 year at a time in the model, with the events that happened to them during that year, as well as the life years and quality-adjusted life years (QALYs) they accumulated, being determined by the health state they were in.

Estimates of test accuracy were taken from available literature identified through the diagnostic-accuracy and cost-effectiveness literature reviews. To estimate the accuracy of MSI and IHC testing, results from studies included in the clinical-effectiveness review were pooled using a multilevel mixed-effects logistic regression analysis. For MSI testing, the results from Barnetson et al. (2006), the population-based sample from Poynter et al. (2008) and Southey et al. (2005) were pooled, and for IHC testing, the results from Limburg et al. (2011) and Southey et al. (2005) were pooled.

Diagnostic-accuracy data for BRAF V600E and MLH1 promoter methylation testing were taken from Ladabaum et al. (2015). This study pooled values from studies reporting test accuracy, with included studies using various types of previous testing for Lynch syndrome (including MSI and IHC testing). Test accuracy parameters used in modelling are shown in table 4.

Estimates of parameter values relating to acceptance of tests, colorectal cancer surveillance, stage of cancer at diagnosis, gynaecological surveillance and chemoprevention were taken from identified literature, registry data and clinical expert opinion.

Costs of preliminary tumour testing, genetic tests (for both probands and relatives), and genetic counselling were sourced from the UK Genetic Testing Network (2016), Health and Social Care Unit Costs and from personal communication with providers. Further relevant costs came from NHS references costs (2014/15 and updated to 2016/17 prices), identified literature, the British national formulary (BNF 2016) and the NHS drug tariff.

Utilities associated with colorectal cancer, endometrial cancer and prophylactic hysterectomy were taken from published literature identified by systematic searches. Disutilities associated with genetic testing used in the model were as previously reported in Snowsill et al. (2014).

The key assumptions applied in the base-case analysis were:

• MSI‑L was considered a negative result.

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

• All people who accepted genetic testing had testing for all 4 MMR genes, unless they followed a strategy that used IHC, in which case they had either BRAF V600E or MLH1 promoter hypermethylation testing and only MLH1 and PMS2 were tested.

• The average number of relatives per proband was 6 (2.5 of whom were first-degree relatives).

• Surveillance colonoscopies reduced the incidence of colorectal cancer by 61%, and the incidence of metachronous colorectal cancer by 47%.

• Surveillance colonoscopies improved the proportion of people in whom colorectal cancer was diagnosed at an early stage (stage I or II) from 44.6% to 79.1%.

• Colorectal surveillance colonoscopies occurred every 2 years.

• Gynaecological surveillance reduced endometrial cancer mortality by 10%.

• People taking aspirin had a reduced incidence of colorectal and endometrial cancer that lasted for 10 years.

• Disutility was only applied to people with stage IV colorectal cancer.

• No disutility arising from prophylactic hysterectomy was assumed.

• Initial acceptance of colonoscopic surveillance was 97% for probands and relatives who tested positive for Lynch syndrome mutation, and 70% for probands and relatives who were assumed to have Lynch syndrome.

The base-case analysis included 238,175 simulated individuals and represents an annual cohort of 34,025 probands with colorectal cancer and 204,150 relatives.

Pairwise ICERs were calculated for all strategies compared with no testing (strategy 1). Only strategy 10 (universal genetic testing) had an ICER above £20,000 per QALY gained, with ICERs for strategies 2 to 9 all below £14,000 per QALY gained. Comparative (fully incremental) ICERs were also calculated for all strategies. Strategies involving MSI testing were either dominated (that is, they were less effective and more expensive than another option) or extendedly dominated (that is, a combination of other options were more effective and less expensive) by other strategies. The ICER for strategy 3 (IHC plus BRAF V600E) was £37,495 per QALY gained and the ICER for strategy 5 (IHC plus BRAF V600E and MLH1 promoter methylation) was £11,008 per QALY gained.

Subgroup analyses were carried out by restricting the age of probands, who have Lynch syndrome testing strategies, included in the model. The age groups were: under 50 years, under 60 years, under 70 years, and 70 years or over.

When the proband population was restricted to people under 50 years, all the strategies had ICERs of less than £13,000 per QALY gained compared with no testing (strategy 1). Strategies 3 (IHC plus BRAF V600E; £19,903) and 5 (IHC plus BRAF V600E and MLH1 promoter methylation; £8,090) had ICERs under £20,000 per QALY gained in the fully incremental analysis.

When the proband population was restricted to people under 60 years, all the strategies had ICERs of less than £17,000 per QALY gained compared with no testing (strategy 1). Only strategy 5 (IHC plus BRAF V600E and MLH1 promoter methylation; £9,156) had an ICER below £20,000 per QALY gained in the fully incremental analysis.

When the proband population was restricted to people under 70 years, all the strategies had ICERs of less than £20,000 per QALY gained compared with no testing (strategy 1), except for strategy 10 (universal genetic testing), which had an ICER of £20,528 per QALY gained. Only strategy 5 (IHC plus BRAF V600E and MLH1 promoter methylation; £9,912) had an ICER below £20,000 per QALY gained in the fully incremental analysis.

When the proband population was restricted to people 70 years or over, strategies 5 (IHC plus BRAF V600E and MLH1 promoter methylation), 7 (MSI plus BRAF V600E) and 9 (MSI plus BRAF V600E and MLH1 promoter methylation) had ICERs less than £20,000 per QALY gained compared with no testing (strategy 1). Strategies 5 (£18,839) and 9 (£18,766) had ICERs below £20,000 per QALY gained in the fully incremental analysis.

If both MSI‑L and MSI‑H test results were assumed to indicate Lynch syndrome (in the base-case analysis, only MSI‑H is indicative), this effectively lowered the threshold for a positive MSI test result. Only strategies involving MSI testing (strategies 6 to 9) were affected, with ICERs for testing compared with no testing increased relative to the base-case analysis. As for the base-case analysis, strategy 5 was the only strategy with an ICER below £20,000 cost per QALY gained (unchanged at £11,008) in the fully incremental analysis.

If aspirin was not included as a risk-reducing component in the model (as it was in the base-case analysis), this resulted in a marginal increase in ICER values, and strategy 5 remained the optimal strategy with an ICER of £11,659 per QALY gained in the fully incremental analysis.

In the base-case analysis, if gynaecological surveillance was accepted, it reduced the risk of mortality from endometrial cancer. Two scenarios were considered: 1 assuming that gynaecological surveillance has no benefit (but still has a cost) and another that removed gynaecological surveillance from the model (no cost and no benefit). For both scenarios, strategy 5 remained the optimal strategy and the only strategy with an ICER below £20,000 per QALY gained in the fully incremental analysis.

In the base-case analysis, the quality of life for people with colorectal cancer, except Dukes' stage D, was assumed to be similar to the general population (that is, a disutility value of 0). In this scenario analysis, increased disutility values for all colorectal cancer stages were used and values were based on Ness et al. (1999). When compared with the base-case analysis, ICER values for all strategies compared with no testing were reduced. Strategy 5 remained the optimal strategy, with an ICER of £9,775 per QALY gained in the fully incremental analysis.

If colonoscopic surveillance was assumed to have no effect on colorectal cancer incidence, ICERs for all strategies were increased compared with no testing (strategy 1), with only 3 strategies remaining, marginally, below £20,000 per QALY gained. Strategy 5 remained the only strategy with an ICER below £20,000 per QALY gained in the fully incremental analysis; however, this value increased to £19,194 per QALY gained (from £11,008 per QALY gained in the base case).

Deterministic sensitivity analyses were carried out for several parameters in the model. The ICERs for the testing strategies were sensitive to several parameters. When sensitivity and specificity values for all tumour tests (MSI, IHC, BRAF V600E and MLH1 promoter methylation) were both reduced to their lower 95% confidence interval values, the ICER for strategy 5 compared with no testing increased to £16,036 per QALY gained.

Altering diagnostic accuracy can also affect which strategy is optimal. When sensitivity was reduced for all tumour tests, strategy 4 became the optimal strategy (IHC followed by MLH1 promoter methylation). When sensitivity values were increased for all tumour tests (to their upper 95% confidence interval values), MSI testing strategies became optimal, despite MSI testing still having lower sensitivity and specificity values than IHC testing. In addition, when the cost of IHC was doubled, or the cost of MSI testing halved (both relative to base-case values), strategy 7 (MSI followed by BRAF V600E) became the optimal strategy.

Decreasing the acceptance by probands of both genetic counselling and testing after counselling (set at 90% and 92.5% respectively in the base-case analysis) to 50%, increased the ICER for strategy 5 compared with no testing to £17,767 per QALY gained (from £11,008 per QALY gained).

Increasing the incidence of colorectal cancer in people with Lynch syndrome in the model decreased the ICER for strategy 5 compared with no testing to £6,689 per QALY gained, whereas decreasing the incidence of colorectal cancer increased this value to £19,300 per QALY gained.

In the base-case analysis, people who were diagnosed as LS‑assumed because they declined genetic testing were considered as positive for Lynch syndrome. If all LS‑assumed probands, and their relatives, were instead considered to be negative for Lynch syndrome, the ICER for strategy 5 compared with no testing decreased to £5,225 per QALY gained.

Six relatives per proband were assumed in the base-case analysis. If only probands were included in the model (that is, no relatives included), the ICERs for all strategies increased, with strategy 5 compared with no testing increasing to £17,921 per QALY gained. Increasing the number of relatives per proband to 12 decreased the ICERs slightly, with strategy 5 compared with no testing decreasing to £10,068 per QALY gained.

If the costs of colonoscopy used in the base-case analysis were doubled, all ICERs for strategies compared with no testing increased; for example, for strategy 5, this increased to £16,630 per QALY gained. Reducing the acceptance of colonoscopy surveillance by people with confirmed Lynch syndrome causing mutations from 97% (as in the base-case analysis) to 70% increased the ICERs for strategies compared with no testing (for example, to £12,632 per QALY gained for strategy 5).

In the base-case analysis, disutility associated with prophylactic hysterectomy and bilateral salpingo-oophorectomy was assumed to be 0. Increasing the disutility value to 0.04 for 1 year increased the ICERs for all strategies compared with no testing, with the value for strategy 5 increasing to £14,441 per QALY gained.