Gynaecological cancer surveillance for women with Lynch syndrome: systematic review and cost-effectiveness evaluation

Significance for patients

An individual with LS may develop several clinical and pathological features, including cancers such as EC and OC, early onset of cancer, multiple independent cancers. 19 The value of a surveillance regimen for women at risk of EC or OC has yet to be established. 20 The Manchester International Consensus Group concluded that ‘further research is required to establish the value of gynaecological cancer surveillance in LS’. 6

The risks of endometrial and ovarian cancer affect people with female reproductive organs, and this may include transgender men as well as cisgender women. We are not aware of any research beyond isolated case reports in LS involving transgender men, and it is possible that evidence that is collected (believed to relate entirely to cisgender women) may be less reliable when applied to transgender men who have used exogenous sex hormones, GnRH analogues or anti-oestrogens. In this monograph, we may at times refer to women – in this context we mean all people with female reproductive organs, which could include transgender men. Our preference is to avoid the noun ‘females’ for people of female sex, as this term can be seen as dehumanising.

Significance for the NHS

The majority of individuals with LS caused by path_MLH1 or path_MSH2 will go on to develop at least one cancer in their lifetime, and LS accounts for around 3% of CRC and ECs. 21,22 In 2017, there were nearly 35,000 CRC diagnoses in England and over 7000 cases of EC. 23 This suggests that over 1000 cancers each year are being diagnosed in people with LS, which may be preventable through risk management strategies.

Clinical guidelines and why this research is needed now

As a result of publication of recent National Institute for Health and Care Excellence (NICE) guidance on identifying LS,17,18 more women will have a diagnosis of LS and need to manage their gynaecological cancer risk. Evidence is needed to determine which interventions are effective and cost-effective to reduce the morbidity and mortality from gynaecological cancer and to contribute to the NHS Long Term Plan goal of improving early diagnosis of cancer. 24 Surveillance is a preferred option for patients managing cancer risk, but good-quality estimates of its effectiveness and cost-effectiveness have not been produced.

Care pathways

Two main interventions are available to manage the gynaecological cancer risk in LS: risk-reducing surgery (removal of the uterus and ovaries) and surveillance. Some patients may have surveillance initially and then have risk-reducing surgery. In addition, chemoprevention with aspirin4 and hormone therapy25 may be considered, as well as lifestyle changes.

Gynaecological surveillance can identify precancerous lesions in the uterus, for which there are fertility-sparing treatments. Surveillance may also be able to identify OC in the early stages, where management options could maintain fertility or allow egg harvesting.

Gynaecological surveillance has previously been estimated to cost the NHS £473 per year for a woman with LS. 9 Later-stage gynaecological cancers can be costly to treat; for example, stage 3 OC costs twice as much to treat as stage 1 cancer. 26

Surgery is widely offered, since its effectiveness is well documented,7 typically at 40–45 years when most women have completed their families. 27 This prevents women from becoming pregnant and artificially brings on menopause, which can detrimentally affect health-related quality of life (HRQoL) and long-term health unless managed with HRT. In some cases, women do not have risk-reducing surgery because of technical difficulties due to past surgery for CRC, high anaesthetic risk due to medical comorbidities or patient preferences. 28

Some NHS providers offer surveillance for gynaecological cancer in women with LS not undergoing risk-reducing surgery; others do not because of a lack of evidence-based guidelines and resource constraints. Some women opt to pay privately for surveillance, but not all women can afford this service.

Population

Studies must have been based on an adult population (age ≥ 18 years) at risk of EC and/or OC (individuals born with and retaining a uterus and/or ovaries, including trans men, and intersex individuals at risk of EC and/or OC) and with confirmed or suspected LS.

Prior to screening being conducted, it was clarified that studies based on a wider population should be included if subgroup data were provided for the eligible population. In these cases, only the relevant subgroup data were included. This clarification was an omission from the published protocol.

Intervention

Studies that evaluated any gynaecological surveillance strategy either alone or in combination with a comparator were included. The target conditions of surveillance had to be EC and/or OC. Surveillance strategies included (but were not limited to): hysteroscopy and directed biopsy, undirected biopsy, ultrasound (transvaginal or transabdominal) and cancer antigen-125 testing. Studies were not excluded based on the surveillance schedule.

Comparators

For controlled study designs, eligible comparators included no surveillance, alternative surveillance programmes, surgical or hormonal prevention. In diagnostic test accuracy (DTA) studies, any eligible surveillance test result could be used as a reference standard for another test but, primarily, cancer diagnosis using histology was the preferred reference standard.

Outcomes

To be eligible for inclusion, studies needed to evaluate at least one of the following outcomes: all-cause mortality, cancer-specific mortality, cancer survival, cancer treatment response, fertility, cancer stage, cancer detection rates (in symptomatic and asymptomatic individuals), interval cancer rates, incidental medically important findings, diagnostic accuracy, test failure rates, adverse events (including pain/discomfort), risk factors impacting adverse events or HRQoL.

Study design

Eligible study designs included randomised and non-randomised controlled trials (RCTs; for all review questions), prospective and retrospective comparative and non-comparative observational designs (review questions related to cancer detection and harms), DTA studies, including any study from which 2 × 2 data could be ascertained (for the diagnostics test accuracy question only), surveys, interviews and studies with visual analogue scale (VAS) or Likert-type scales (for the questions related to harms). Case reports, opinion pieces, editorials and studies that only published in abstract form were excluded.

Few comparative studies were found to address the clinical effectiveness questions, so a protocol amendment was made to also present relevant clinical effectiveness data (mortality, survival and stage of cancers) from non-comparative cohort studies.

Summary of included studies

Of the 30 included studies, 10 were comparative cohort studies (Table 2) and 20 were single-arm cohort studies (Table 3). The comparative studies were those that included data from more than one cohort (either completely separate or overlapping due to the use of two periods) who received different surveillance programmes, or where one group received surveillance and at least one other was an eligible comparator. Cohort studies providing some separate data for different tests within a surveillance programme, but for the same individuals, were not considered to be comparative studies. In these cases, data were extracted separately for the different tests.

Potential overlap between included studies

Two reviewers (HC and SB) initially evaluated 59 articles and abstracts to group together those that clearly represented the same study. Conference abstracts that did not appear to report data connected to any of the included full articles were subsequently excluded (n = 15; see Figure 1 and Table 25, Appendix 2).

The design of the included studies meant that the likelihood of participant overlap between the included studies was high. Two reviewers (HC and NM) evaluated the studies for potential overlap, paying particular attention to studies conducted in the same country. This evaluation concluded that none of the 30 studies clearly overlapped sufficiently to be deemed the same study, but there were several studies based in the UK and the Netherlands, where overlap between participants was probable or possible. Furthermore, the study by Møller et al. (2017) pooled data from several countries and would, therefore, include participants from other included studies. 8,53

Meta-analyses were not conducted for this review, but potential double-counting of participants could impact upon narrative syntheses and was considered.

Critical Appraisal Skills Programme checklist for cohort studies

Risk of bias ratings for all 30 included studies, according to the CASP checklist for cohort studies,31 are given in Table 4.

It is important to consider whether the included studies had sufficiently long follow-up periods to assess the primary outcomes of interest. For cancer detection rates and data on adverse events, follow-up periods were likely to be sufficient. However, 12 studies reported data on mortality and/or survival (see Mortality and survival) and none of these reported a follow-up period of 10 years or more: three had a mean/median follow-up period of < 5 years,42,60,61 two had a follow-up period between 5 and 10 years,62,66 and it was also likely that the study by Møller et al. (2017) had a mean follow-up period between 5 and 10 years, but this is assumed (mean observation years were given for each mutation and ranged from 8.0 years for MLH1 to 4.3 years for PMS2). 8,53 The remaining six studies providing mortality and/or survival data did not clearly report the follow-up period. 37–39,43,44,59,69–71,73 However, with the exception of Woolderink et al. (2018),69–71 these studies provided other information indicating that follow-up was unlikely to be long enough to assess mortality/survival.

Risk Of bias In Non-randomized Studies – of Interventions

For the 10 studies providing comparative data, additional ROB ratings using the ROBINS-I tool were made (Table 5). It is important to remember that ratings refer to the level of ROB and not the amount of bias. For each study, ROBINS-I ratings were made in consideration of all outcomes that were provided for both surveillance and an eligible comparator; where ROB differed according to outcome, this was noted and the highest ROB rating was recorded.

For all included comparative studies, selection to groups was based on participant preference (likely influenced by confounding factors) or was dictated by comparison of different periods. 42–44,73 Four of the five studies comparing surveillance with RRS were at critical ROB due to confounding, as baseline characteristics of the participants in each group were not considered or accounted for, including factors which could impact upon key outcomes (e.g. previous cancers, family history of gynaecological cancer). 38–40,60,61,67 Additionally, the meaningfulness of comparisons between surveillance and RRS is inherently limited, particularly with regard to cancer detection rates: except for cancers detected during surgery, it is expected that few post-surgical endometrial or ovarian cancers would occur (depending on the exact nature of the surgical procedure) and, conversely, a negative impact on fertility would be expected due to surgery. These critical ROB ratings were, therefore, applicable to data on detection rates (fertility data were not reported).

Detection of malignancies and premalignancies

Data are summarised in Table 7.

Other medical findings

Eleven studies reported other (incidental) medical findings during surveillance (see Table 7). 40,41,43,44,47–52,54–58,60–62 Of the three comparative studies providing these data,40,43,44,60,61 only Helder-Woolderink et al. (2013) reported these data for more than one study group (i.e. the different surveillance time periods). 43,44 In this study, simple hyperplasia was detected more frequently in the earlier period (before biopsy became routine) and ovarian cysts were detected more frequently in the later period. 43,44 However, numbers were small and no conclusions should be drawn from this. Eikenboom et al. (2021) and Renkonen-Sinisalo et al. (2007) only reported data on other medical findings during surveillance and not for surgical comparators, and so effectively provided single-arm data. 40,60,61

During surveillance, the detection of hyperplasia (either simple or complex) without atypia was reported in eight studies. Two studies reported detecting complex hyperplasia without atypia [with an incidence of 3.6% in Nebgen et al. (2014)54–58 and 1.7% in Renkonen-Sinisalo et al. (2007)]. 60,61 Seven studies reported detecting simple hyperplasia, with incidence rates (where calculable) ranging from 0.6% in Renkonen-Sinisalo et al. (2007)60,61 to 6.8% in Helder-Woolderink et al. (2013). 43,44 Eikenboom et al. (2021) reported an incidence of endometrial hyperplasia of 7.2% but it was not clear whether this referred to simple or complex hyperplasia, or indeed whether or not atypia was present in any of the cases (see Detection of malignancies and premalignancies). 40

Six studies reported finding endometrial and/or endocervical polyps during surveillance and three reported the number of participants with polyps (rather than the number of polyps): Elmasry et al. (2009)41 and Rijcken et al. (2003)62 reported incidence rates of polyps (not clear if endometrial or endocervical) of 8.0% and 4.9%, respectively, and Manchanda et al. (2012) reported incidence rates of endometrial polyps as 14.6% and endocervical polyps as 4.9%. 52

In Ketabi et al. (2014), an additional gynaecological cancer which was outside of the scope of this review was found (fallopian tube carcinoma), and additionally a mucinous borderline OC was reported (see Detection of malignancies and premalignancies). 47,48 It was unclear whether these findings were detected with surveillance or during the interval between surveillance visits. Other incidental findings reported across the studies were fibroids/myoma/adenomyosis (reported in three studies), ovarian cysts (two studies), atrophy (two studies) and disturbed proliferative endometrium (one study; see Table 7).

Diagnostic test accuracy

Five studies provided DTA data (either sufficient data to complete a 2 × 2 table or reported both sensitivity and specificity values) for specific tests used in gynaecological surveillance programmes. 34,35,50–52,63,64 In four of these studies, the target condition was EC,34,35,50–52 and in one study the target condition was OC. 63,64

The QUADAS-2 ROB ratings for these studies are reported in the section Risk of bias in test accuracy data. 32 Only one of the studies provided test accuracy data for more than one test. 52 Little between-test comparison of DTA was made in the study report. 52 It was therefore decided not to deviate from protocol by assessing this study using the QUADAS-C for head-to-head test accuracy studies published in 2021. 77 Instead, any differences in the ROB for each test are discussed in the section Risk of bias in test accuracy data.

Diagnostic test accuracy data from these studies are provided in Diagnostic test accuracy data, and should be considered alongside the ROB assessments in Risk of bias in test accuracy data.

Test failure rates

Five of the included studies reported test failures for at least one test used as part of a surveillance programme. 34,35,41,50,62,72 These are given in Table 11.

Test failures for hysteroscopy were reported as 26.1% and 8.8% with a 4-mm hysteroscope41 and a 3-mm hysteroscope,50 respectively. However, because of small sample sizes and other differences between studies (e.g. different samples of patients, different individuals and teams performing the procedures), no inferences or generalisations should be made regarding test failures and the devices/tests themselves. Test failure rates were reported for endometrial biopsy in four studies (see Table 11) and ranged from 10.3% in Lécuru et al. (2008)50 to 24.0% in Elmasry et al. (2009). 41 All of these studies used a Pipelle device; differences between-test failure rates may be due to the small sample sizes in these studies.

Failure of TVUS was reported in two studies [4.0% in Elmasry et al. (2009) and 0.6% in Rijcken et al. (2003)]. 41,62 The remaining two studies reported test failures for tampon sampling72 and uterine cavity washing (with MSI analysis). 34,35 In the Bats et al. (2014) study, it was reported that all procedures were successfully carried out. 34,35 Conversely, in the Woolderink et al. (2020) study it was reported that although tampons were successfully collected, in all cases there was a failure to collect endometrial cells. 72 The authors suggested that this may be due to the nature of their study sample; endometrial cells may not have been shed because all participants were healthy at the time of sampling and no cancers or any other abnormalities were found.

When looking within studies at the test failure data for different tests, Elmasry et al. 2009 found lower rates of test failure with TVUS (4.0%) than with either hysteroscopy (26.1%) or endometrial biopsy (24.0%). 41 This is consistent with the data from Rijcken et al. (2003), where lower rates of test failures were reported with TVUS (0.6%) than with endometrial biopsy (11.8%),62 and Lécuru et al. (2008),50 where similar rates of test failures were reported for hysteroscopy (8.8%) and endometrial biopsy (10.3%).

Clearly, in the feasibility study by Woolderink et al. (2020), test failure rates were much higher for tampon sampling (100.0% failure to collect endometrial cells) than with endometrial biopsy (20.0% test failures). 72 However, this might be interpreted differently if it were to be established that endometrial cells are consistently collected in individuals with endometrial abnormalities (in which case the absence of such cells may not be considered a test failure but a negative test result). Further research, in a larger LS sample, would be needed to clarify this.

Mortality and survival

Data on mortality and survival were sought from comparative end-to-end studies (i.e. studies that compared surveillance with a control and thus provided comparative data on the clinical effectiveness outcomes). Ideally comparative RCTs were sought, but these were not found.

Seven comparative cohort studies provided data on mortality and/or survival. 38,39,42–44,60,61,66,69–71,73 However, these data were sparse [see Mortality and survival (time-to-event data)], so we amended the inclusion criteria and additionally present mortality or survival data from the single-arm studies.

All data relating to mortality or survival reported in the included studies are presented in Table 12.

Cancer prognosis

None of the included studies provided data on cancer prognosis (e.g. clinically estimated prognosis in months) for women with LS who were undergoing gynaecological surveillance.

Stage of cancer at diagnosis could, however, be used as a proxy for prognosis. Indeed, one included study did report that FIGO stage 3 and 4 disease was associated with poorer prognosis (defined as percent survival over time in months) than FIGO stage 1 and 2 disease, although numbers with advanced disease were small and did not reach statistical significance. 65 Data on stage of cancer at diagnosis are described in the section Stage at diagnosis.

Treatment response

While many of the included studies provided some information on the number of women who had surgical treatment following detection of gynaecological cancer (or premalignancies), only three single-arm studies provided data for more than one treatment and reported the response to the treatments. 37,52,54–58 However, even data in these three studies was extremely limited. None of the studies provided end-to-end data; no comparative studies reported treatment response data, so there was no comparison of the eventual treatment responses of individuals who had surveillance with individuals in control groups.

In Dove-Edwin et al. (2002) it was merely stated that one participant received a hysterectomy with BSO for interval EC and OC, while another received a hysterectomy for interval EC, and that both of these individuals were well at the time the study was reported (i.e. no other data were reported about this). 37 In another study, the treatment of five participants with surveillance-detected endometrial abnormalities was described. 54–58 Three individuals with CAH (one of whom was thought to have EC at the time of biopsy and another of whom was thought to not have atypia at the time of biopsy) were successfully surgically treated with hysterectomy. The other two individuals had complex hyperplasia without atypia and were treated with contraceptives [one with oral contraceptives and the other with a levonorgestrel-releasing intrauterine system (LNG-IUS)]. After 6 months, an endometrial biopsy in the participant treated with oral contraceptives displayed no signs of hyperplasia. Similarly, for the participant treated with the LNG-IUS, there was no sign of hyperplasia at surveillance visits 10.3 months and 14.5 months later. 54–58

Similarly, Manchanda et al. (2012) reported successful treatment in a small subsample of individuals with EC or AEH detected during surveillance. 52 Three participants received a hysterectomy because of EC (one who also had AEH) but response to treatment is not explicitly reported (although surgery is likely to be successful, spread to other organs detected at a later date cannot be completely ruled out). One further participant with AEH was reported to have received treatment with a LNG-IUS and outpatient hysteroscopy and endometrial sampling did not find EC or atypia 4 months later.

These limited data suggest that individuals with premalignancies detected during surveillance can be successfully treated with contraceptives but, in the absence of direct within-study comparator groups (e.g. comparison with no surveillance), we cannot elucidate whether the surveillance had any impact upon the treatment response. The two individuals with interval gynaecological cancers reported in Dove-Edwin et al. (2002)37 were successfully treated with surgery and three individuals with surveillance-detected EC reported in Manchanda et al. (2012)52 also received surgical treatment (success not reported). All premalignancies were detected during surveillance and successfully treated either surgically or with contraceptives.

Fertility

Five included studies provided baseline parity data (see Participant characteristics at baseline). 41,45,46,72,74,75 One study provided parity data at different time points but did not clearly indicate whether new pregnancies occurred during the study period. 45,46 Another study reported that 87.5% of participants undergoing surveillance and 88.1% of participants who had opted for RRS (some who had previously been on surveillance) had one or more previous delivery. 74 However, this study was a cross-sectional survey, and it was not clear whether any of the deliveries occurred during surveillance. 74 Two studies investigated parity (and/or menopause status) as a subgroup for adverse events data. 45,46,54–58 These data do not elucidate the impact of surveillance on fertility.

One of the studies reported survey data (in 261 individuals with LS at risk of gynaecological cancer) on the number of births, number of pregnancies, use of assisted reproduction technology, decisions about starting families earlier due to LS, and reasons why participants could not have children (or their ideal number of children). 75 However, these data were not provided separately for individuals who had undergone surveillance so cannot contribute meaningfully to this review.

The single-arm study by Rijcken et al. (2003) reported that, over 10 years of surveillance, there were six births (four participants delivered one child and one delivered two children). 62 However, without a comparator, the impact of surveillance on fertility cannot be assessed. None of the studies, therefore, provided clear data that would enable the impact of surveillance on fertility to be assessed (e.g. by directly comparing different surveillance programmes or by comparing individuals undergoing surveillance with those who did not attend surveillance – noting that these data would not be expected for surgical comparators where fertility is necessarily impacted).

It should be noted, however, that one of the included studies reported that three individuals did not start surveillance (and did not enter the study) because they were trying to conceive and were concerned about the impact of hysteroscopy on fertility. 41 Of the 25 individuals who did enter the study and were undergoing surveillance, 2 nulliparous participants did not consent to hysteroscopy (but did consent to TVUS and to endometrial biopsy) because they were concerned about risk of infection and potential impact upon fertility. This highlights the need to consider the safety and acceptability of surveillance techniques for individuals who are actively trying to conceive (with some techniques being more invasive than others) but does not provide data to clarify whether certain tests have a greater impact upon fertility in the LS population.

Quality of life and mental health outcomes

One single-arm study provided data on QoL. 68 This study provided data at baseline, 3 and 6 months [using the Short Form 36 Health Survey (SF-36)] in individuals with LS undergoing a surveillance programme that could include outpatient hysteroscopy, endometrial biopsy, TVUS and/or CA-125.

The study also provided data on anxiety and depression [using the Hospital Anxiety and Depression Scale (HADS)]. It was decided that, for completeness, these data would also be presented here. The data regarding anxiety, depression and QoL (n = 15) are, therefore, presented together in Table 13. 68 For all HADS and SF-36 subscales there was a trend towards a reduction in scores from baseline over 6 months of surveillance, indicating a reduction in depression and anxiety, and an improvement in QoL for individuals with LS once gynaecological surveillance is started. However, these trends did not reach statistical significance. It is possible that this is due to the follow-up period only being 6 months (improvements in these types of outcomes may take longer to emerge).

Additionally, the single-arm design of the study precludes any conclusion that surveillance itself may be driving any reductions in depression and anxiety or improvements in QoL.

Stage at diagnosis

The stage at which malignancies are diagnosed can impact upon prognosis, treatment and end-point clinical outcomes (mortality and survival); detection at an earlier stage is expected to lead to better outcomes. Cancer stage at diagnosis can, therefore, be considered a proxy for prognostic data.

Data on cancer stage at diagnosis were extracted from both single-arm and comparative studies and are provided in Table 7. The study by Stuckless et al. (2013) included a surveillance group and two control groups who did not receive surveillance, but did not provide data on stage of cancers for the groups who did not have surveillance. 66 Woolderink et al. (2018) reported the stage at which OCs were detected for those undergoing surveillance and a no-surveillance control. 69–71 The OCs detected with surveillance appear to be generally lower stage than those detected without surveillance, but as this is based on one study with only 11 patients with OC in the surveillance group, this should be interpreted with caution. 69–71 Similar data were not available for EC.

For single-arm data, an evaluation of the stage of cancers detected at surveillance versus those detected in the interval between surveillance visits can illustrate whether surveillance is likely to miss earlier stage cancers. Three single-arm studies provided data on the stage of cancers detected during surveillance as well as the stage of interval cancers. 47,48,52,65 Across these three studies, data were very sparse and there was no obvious indication that there was a difference in cancer stage between surveillance detected and interval detected OC or EC (see Table 7). Additionally, although one of the comparative studies provided EC staging data for two different surveillance periods, the data were too sparse (low number of cases) to be meaningful. 42

Studies comparing surveillance with RRS (following surveillance) are more akin to single-arm studies; the cancers detected during RRS among those previously undergoing surveillance are similar to interval cancers. Dueñas et al. (2020),38,39 Eikenboom et al. (2021)40 and Tzortzatos et al. (2015)67 reported the stage of cancers detected with RRS and with surveillance. Again, data were sparse, but the cancers detected during surgery were generally as low or lower stage than those detected with surveillance. There are insufficient data to draw any conclusions from this. One further study provided cancer staging data for surveillance and a surgical comparator that included those undergoing surgical treatment (i.e. with cancer previously detected). 60,61 This study lacks meaning in this context because surgery group participants would not have been eligible for surveillance.

Rates and severity of adverse events

Seven of the included studies directly provided either (1) rates of adverse events or (2) continuous data using rating scales, with regards to gynaecological surveillance for individuals with LS. 41,45,46,52,54–58,72,74,75 All seven studies were either single-arm cohort designs41,45,46,52,54–58,72 or provided survey data for either a single sample75 or for two subsamples. 74 With the exception of Manchanda et al. (2012),52 all these studies provided data on pain; two on the need for pain relief,45,46,75 one on infections,75 one on vasovagal reactions,41 and one on perforations. 52

An additional study provided data relevant to pain but did not report the proportion of participants experiencing pain, or continuous data using a pain rating scale 43,44 Instead, this study reported some limited data on anxiety about pain. 43,44 Four of the studies that did provide direct data on pain also provided limited data on concerns and worries related to surveillance and the impact of these concerns on surveillance discontinuation or on opting for prophylactic surgery. 41,45,46,72,74

None of the included studies reported data on discomfort or discomfort severity (rather than pain) or on serious infections. Wood et al. (2008) reported data on QoL and on anxiety and depression, but these data were provided as clinical effectiveness outcomes, rather than as data pertaining to harms due to surveillance procedures and are therefore presented in the section Quality of life and mental health outcomes. 68 No other adverse events were reported in the included studies, but this does not mean that other harms did not occur.

Risk factors impacting adverse events

Four of the included studies investigated whether particular subgroups were more likely to experience adverse events or to experience them more severely. 41,45,46,54–58,72

Kwon et al. (2008)

Kwon et al. 79 modelled effectiveness and cost-effectiveness through Markov microsimulation. Five treatment strategies evaluating combinations of screening and risk-reducing or prophylactic surgery were included. These strategies were no prevention; prophylactic surgery at 30 years; prophylactic surgery at 40 years; combined annual screening from 30 years and prophylactic surgery at 40 years; and annual screening with endometrial biopsy, CA-125 and TVUS from 30 years. A time horizon of 40 years, from 30 to 70 years of age was assumed. Base case data included lifetime risk, incidental diagnosis during surgery, age at diagnosis, probability of symptoms, sensitivity and specificity of screening, cancer stage distribution, lifetime risk of peritoneal carcinoma and perioperative mortality. Cost estimates were based on existing literature and national registers and statistics from official governing or healthcare bodies. Benefits and QALY weights were derived from existing literature. Evidence on effectiveness for each intervention is presented as an incremental cost-effectiveness ratio (ICER).

Yang et al. (2011)

Yang et al. 80 built a decision analytic model in TreeAge (TreeAge Software LLC, Williamstown, MA, USA) to evaluate the effectiveness and cost-effectiveness of gynaecological cancer management. The model included age-specific cancer risks in HNPCC, diagnosis, surgery and survival by cancer stage, surgical mortality, and efficacy, sensitivity and specificity of screening. The decision tree contained three arms, one for each of the following management strategies: annual examination, annual surveillance (including endometrial biopsy, CA-125 and TVUS as in Kwon et al. 79); and prophylactic surgery at 30 years. Health benefits and utility estimates were identified from the existing literature and the Surveillance, Epidemiology and End Results database and applied against discounted life expectancy to generate QALYs. Costs were obtained from University of California San Francisco Patient Financial Services and departmental billing offices, alongside existing literature. Where screening costs were unavailable they were estimated based on a cost-to-charge ratio of 0.48 per hospital charge. Cost-effectiveness was evidenced by the average cost-effectiveness, or cost per QALY, for each strategy, but we have calculated ICERs as only these are appropriate for decision-making. 83 The model was an extension of a prior publication by Chen et al. (2007). 84

Wright et al. (2021)

Wright et al. 81 analysed the cost-effectiveness of genotype-specific surveillance and prevention of gynaecologic cancers through a Markov state transition cohort-level model. Women began in the healthy state at 25 years of age, progressing through health states following an annual cycle until 75 years, intervention or death. Alternate strategies were examined by varying type and timing of risk-reducing surgery and age of surveillance initiation. Overall 12 different strategies were evaluated: hysterectomy and BSO at 35 years, surveillance from 30 years; hysterectomy and BSO at 4 years, surveillance from 30 years; hysterectomy and BSO at 50 years, surveillance from 30 years; surveillance alone from 30 years; hysterectomy and BSO at 40 years, surveillance from 35 years; hysterectomy and BSO at 50 years, surveillance from 35 years; surveillance alone from 35 years; hysterectomy and BSO alone at 35 years; hysterectomy and BSO alone at 40 years; hysterectomy and BSO alone at 50 years; no intervention; two-stage approach – hysterectomy and bilateral salpingectomy at 40 years, oophorectomy at 50 years. Alongside costs and utilities, the model included base parameters for mortality, surveillance sensitivity and specificity, surgical complication, gynaecological cancer risk by gene, and cancer stage distribution for no intervention, surveillance, surgery and survival. Health benefits and utilities were obtained from existing literature and included a half-cycle correction. Costs were derived from Medicare data and published analyses. They were adjusted for inflation using the Medical Consumer Price Index. The effectiveness for each intervention was assessed by calculating an ICER for each gene mutation.

Kwon et al. (2008)

Kwon et al. (2008)79 found the reference strategy of no prevention to be the least costly, but it achieved the lowest QALYs. Annual screening (without risk-reducing surgery) was the most expensive strategy and was dominated by the three strategies which included risk-reducing surgery. Risk-reducing surgery alone was more expensive when conducted at 4 years than 30 years, but it also led to more QALYs and risk-reducing surgery at 30 years was extendedly dominated by no prevention and risk-reducing surgery at 40 years. The combined strategy of surveillance from 30 years until risk-reducing surgery at 40 years provided the most QALYs, but this was at a very high cost, meaning that it was not cost-effective.

Threshold analysis revealed that when utility from risk-reducing surgery rises from its base case value of 0.86–0.88 or higher, the QALYs of risk-reducing surgery at years exceed those of risk-reducing surgery at 40 years.

Overall Kwon et al. found risk-reducing surgery alone to be the cost-effective strategy (cost-effectiveness threshold of $50,000 per QALY gained). 79 Varying the age at which surveillance was initiated or risk-reducing surgery was performed could not make the combined strategy approach cost-effectiveness.

Although an analysis was not conducted by the authors in which risk-reducing surgery is not an option, we have calculated that the ICER for annual surveillance (starting at 30 years) compared with no prevention would be around $85,000 per QALY gained, which is not cost-effective.

Yang et al. (2011)

Yang et al. (2011)80 used average cost-effectiveness ratios rather than ICERs, so many of their results are not useful for decision-makers. They also did not include a strategy of no prevention. Nevertheless, they found that risk-reducing surgery dominated annual gynaecological examination and annual gynaecological surveillance.

Wright et al. (2021)

Wright et al. (2021)81 stratified their economic evaluation according to LS genotype. For path_MLH1 and path_MSH6, they identified the economically optimal approach was two-stage RRS, that is hysterectomy and bilateral salpingectomy at 40 years and bilateral oophorectomy at 50 years. For path_MSH2, the economically optimal approach was risk-reducing hysterectomy and BSO (HBSO) at 40 years, while for path_PMS2 the optimal approach was risk-reducing HBSO at 50 years.

For patients of all genotypes, risk-reducing surgery was dominant over no intervention and over strategies including surveillance.

Numerous one-way sensitivity analyses were conducted and reported:

• When the utility value associated with surgical menopause was varied from its base case value of 0.90 in the range 0.86–0.95, RRS remained the economically optimal approach for all genotypes, although the precise surgery which was optimal (HBSO at 35, 40 or 50 years, or two-stage surgery) did change in a predictable pattern.

• When the utility value associated with hysterectomy and with postmenopausal HBSO was varied from its base case value of 1 in the range 0.97–1, the economically optimal strategies were unchanged. When it was varied in the range 0.95–0.96, however, HBSO at 40 years became the optimal strategy for path_MLH1 and no intervention became the optimal strategy for path_PMS2.

• If risk-reducing HBSO can reduce the risk of OC, this does not affect the economically optimal strategy, at least with a risk reduction of up to 25%.

• If bilateral oophorectomy can increase the rate of general mortality, this does not affect the optimal strategy over a range of hazard ratios to 1.24. For a range of hazard ratios between 1.32 and 1.4 the optimal strategy for path_MSH2 changed to the two-stage approach.

• In path_MSH6, if the lifetime risk of OC was in the range 23.2–33.9%, the economically optimal strategy became surveillance from 3 years until risk-reducing HBSO at 35 years.

Probabilistic sensitivity analysis was conducted and confirmed the robustness of findings to parameter uncertainty. The two-stage approach remained optimal in 84.2% of iterations for path_MLH1 and in 71.0% of iterations in path_MSH6. For path_MSH2 HBSO at 40 years was optimal in 86.2% of iterations, the two-stage in 12.2% and HBSO at 35 years in 1.6%. Finally for path_PMS2 HBSO at 50 years remained optimal in 91.6% of iterations.

The authors did not consider a population in which RRS is not an option, but they have reported costs and QALYs for all strategies, so the review authors have been able to investigate this question:

• For path_MLH1 and path_MSH6, surveillance from 35 years would be cost-effective (ICER ~$9000/QALY vs. no intervention) and would dominate surveillance from 30 years.

• For path_MSH2, surveillance from 30 years would be cost-effective (ICER ~$8000/QALY vs. surveillance from 35 years) and would dominate no intervention.

• For path_PMS2, no intervention dominated the surveillance strategies.

Endometrial cancer

Fourteen studies88,95–107 were identified providing utility values for EC. Most had a sample size below 500 (93%), with over half (57%) including fewer than 100 participants. Participants were predominantly aged between 50 and 70 years (71%) and came from Europe (36%) or North America (29%). Disease stage was largely unreported, with only four studies clear in their evaluation of early-stage EC. No study reported utility in the context of gynaecological surveillance. Five studies evaluated EC utility in the context of gynaecological surgery including hysterectomy (n = 2), HBSO (n = 2) and minimally invasive surgery (n = 1). Utility was most often measured by the EQ-5D (n = 10), followed by standard gamble (n = 2), time trade-off (n = 1) and Short Form 6 Dimensions (SF-6D; n = 1). These health state valuations were largely made by patients themselves in evaluating their own health (79%, n = 11), although alternative populations including patients and public (n = 2), and the public alone (n = 1) were also used. Follow-up period was often unclear in EC studies (64%, n = 9). Where stated, most considered only baseline cancer (21%), though one study extended follow-up to 3 months and the final study up to 6 months.

Ovarian cancer

We identified 32 studies88–94,99–101,105,107–127 reporting OC utilities. No study on OC reported a sample size greater than 1000, with most (59%) using samples of 500 or less. Age was not consistently reported, being not reported in 41% of studies. Again, 50–70 years was the most common (25%), although evaluating OC at multiple ages across the lifespan was also popular (31%). Study location was reported in only half of the included studies (53%), of which half (n = 8) considered participants from North America. Disease stage was often not reported, undefined or considered multiple stages (53%). No study evaluated early-stage OC, nine studies (28%) evaluated stage IV or advanced cancer and six studies (19%) considered recurrent OC. While not solely reporting surveillance or surgical intervention, other than cytoreductive surgery (n = 1), OC studies did consider treatment through chemotherapy and medication (63%). Utilities were most commonly measured through the EQ-5D (50%), followed by time trade-off (19%) and the converted FACT (19%) while the standard gamble and SF-6D were each used in two (6%) studies. Utilities were evaluated by patients (72%), both patients and the public (22%), the public alone (3%) or patients and clinicians (3%). Three studies reported aspects of data from the ICON7 trial. 110,120,123 Again, most studies did not clearly report a follow-up period (69%). Where follow-up was reported, the last time points were baseline (n = 4), 6 months (n = 3) and over 1 year (n = 3).

Gynaecological surveillance

Four studies28,88,115,128 reporting utilities for gynaecological surveillance were identified. Two studies115,128 considered screening or surveillance for OC, so the surveillance modality was aimed towards detecting OC. Havrilesky et al. (2009)115 were identifying utilities for OC screening (not in a high-risk population) and they explicitly defined the test as either a blood test or TVUS, while van Roosmalen et al. (2002)128 focused on ovarian surveillance for BRCA1/BRCA2 mutation carriers but did not specify the surveillance modality. Sun et al. (2019)28 and Kuppermann et al. (2013)88 sought to identify preferences for management of cancer risk in LS, but while Sun et al. (2019)28 were explicit about the form of surveillance, Kuppermann et al. (2013)88 gave no details of surveillance. Health states were described by vignettes in two studies,28,88,115 while in the remaining study128 it was unclear how health states were described. In all four studies participants were valuing hypothetical health states rather than their own health. Health state valuation was by time trade-off in three studies,88,115,128 while in the remaining study,28 the authors describe their approach as a modified standard gamble, in which the downside risk is developing cancer over an individual’s lifetime rather than dying immediately.

Gynaecological surgery

A total of 23 studies published in 24 papers were included that evaluated gynaecological surgery. 28,86–90,104,114,128–143

Two studies28,88 specifically investigated risk-reducing gynaecological surgery for LS, while another four studies89,90,114,128 investigated risk-reducing gynaecological surgery for hereditary breast OC (pathogenic variants in BRCA1/BRCA2). In these studies, participants gave values to hypothetical disease states using the time trade-off method, except for one study28 where a modified standard gamble was used.

The remaining studies86,87,104,129–143 considered various benign gynaecological conditions and obtained utility values directly from patients, generally using the EQ-5D, although three studies135,139,142 used time trade-off, one study104 used standard gamble and one study143 used the 15D measure.

Endometrial cancer

Utility values for EC ranged from 0.5698 to 1.00. 99,107 The lowest utility value of 0.56 was observed immediately 1-week post-laparotomy surgery at mean utility 0.56. 98 The greatest utility values related to EC were observed in primary disease at 0.999,99 and in patients participating in a UK breast cancer trial yielding utility 0.913. 104

In general, utilities valued by both the public and patients were lower than those classified only by patients. Utilities are not observed to consistently differ by geographic location in cross-study comparison; however, within-study assessment utilities as greater in Canada and the USA than the UK population. 105 Additionally, utilities are not seen to dramatically alter by medical facility with only minimal variation observed. 88 There is evidence of a relationship between EC utility and body mass index (BMI) with greater utility identified in individuals with BMI below 35 kg/m². Furthermore, findings suggest utility associated to EC will decrease as stage and so severity of the cancer increases, whether comparing primary with advanced disease,99 or stage I with stage I–III EC. 102,103

More invasive surgery for EC (e.g. total abdominal hysterectomy or laparotomy) is seen to result in a more dramatic short-term reduction in utility than minimally invasive or laparoscopic surgery. 96,98 However, in the longer term at 3 or 6 months post surgery, this difference is no longer observable, given the dramatic improvement in utility for invasive surgery, with Bijen et al. (2011)96 even implying greater long-term results from invasive abdominal hysterectomy than the less invasive laparoscopic hysterectomy. Further details of EC study results are presented in Table 27, Appendix 5.

Ovarian cancer

Ovarian cancer utility estimates ranged from 0.16115 to 0.977,126 the lower bound being for end-stage OC and the upper in a cluster of patients undergoing chemotherapy. Again within-study comparison by geographical location suggests that utilities are valued higher in North America than in the UK. 105 Utilities based on conversion from FACT-O scores show slight variation dependent on transformation algorithm. The Dobrez algorithm consistently yields the greatest utility values, followed by the Cheung algorithm with the lowest value produced by linear, ordinary least squares and/or Tobit transformation. 91,118

As in EC, utility decreases as cancer stage and so severity increases,99,105,113,115 with the lowest utility across all studies observed at 0.16 in end-stage OC. 115 While these findings are not echoed in the median values identified by Hildebrandt et al. (2014),99 inspection of the range provides evidence of lower utilities in advanced disease. Furthermore, utility is lower following progressive disease than before its onset. 94,111 Treatment through chemotherapy and medication are shown to improve utility for individuals with OC. 108–110,118,120,123 However, this was not observed across all included studies. Treatment by chemotherapy medication docetaxel and carboplatin were unable to prevent a reduction in utility from randomisation to end of study in patients with platinum-sensitive recurrent OC. 116 Likewise, this lack of long-term reduction in utility from screening to post-progression was observed in platinum-sensitive recurrent patients with OC undergoing niraparib maintenance therapy or placebo both with and without BRCA mutations. 92 In comparison with patients receiving chemotherapy, patients under surveillance displayed lower utility regardless or adverse events, treatment efficacy and emotional well-being. 119

Women with BRCA1/2 display lower utility values at the older age of 33–50 years, with mean 0.58, than when aged 20–32 years at mean 0.84. 89 The three studies on BRCA1/2 by Grann et al. also reveal a difference in utility dependent on who the health state is valued by. 89,90,114 Patients and the public, mutation carriers and individuals with no known personal risk are consistent in their utility estimates; however, health professionals present a lower estimate than other individuals. Underestimation from physicians is not present across all OC studies. 119 Additionally, one study ascertaining utility values from patients with advanced OC and healthy female volunteers suggested inconsistency between the groups with patients presenting higher utilities than volunteers for all medication strategies regardless of response. 124 Further details of OC studies are presented in Table 28, Appendix 5.

Gynaecological surveillance

Gynaecological surveillance utilities ranged from 0.7028 to 1.00. 128 The highest levels of utility were found for the health state of screening for breast and OC in patients with BRCA1/2 valued via the time trade-off method. 128 The lowest utility (median 0.70) was obtained by Sun et al. (2019)28 for annual gynaecological surveillance with endometrial biopsy, TVUS and CA-125, although this study did not use a method that gives utility values suitable for QALY calculation.

Havrilesky et al. (2009)115 estimated median utility values of 0.97 for all health states relating to OC screening (blood test or TVUS with or without false positive result), but mean utilities varied more substantially. Unexpectedly, the mean utility was higher for TVUS with a false positive result than for TVUS without a false positive result, which may suggest a problem with using the time trade-off method to evaluate temporary health states.

Additional utility detail from studies on gynaecological surveillance can be found in Table 29, Appendix 5.

Gynaecological surgery

Utilities related to gynaecological surgery ranged from weighted health state 0.03 on day 0 of hysterectomy with GA,144 to utility of 0.99 in presymptomatic BRCA1/2 mutation carriers undergoing prophylactic oophorectomy alongside breast cancer screening. 128 When not confounded by anaesthesia, the lowest reported utility related to gynaecological surgery was observed in premenopausal women with LS undergoing prophylactic HBSO at median 0.40. 28 Utility in the context of prophylactic HBSO was greater in postmenopausal than premenopausal women. 28

For benign gynaecological conditions, while utility is seen to drop immediately following gynaecological surgery (hysterectomy, oophorectomy, endometrial ablation, laparotomy), in the longer term QoL recovers and utility reaches a level better than in the preoperative state. This initial drop is observed in the weeks following surgery, with an improvement in utility compared to preoperative state identified around 1–3 months after surgery. 131,140,144 Conversely, in patients with noncancerous pelvic populations one study identifies an immediate improvement in utility from immediately before to immediately after both hysterectomy and uterus-preserving surgery. 139 Clarification of the term ‘immediately’ is however not provided by the authors. Where data are not collected in the first weeks following surgery, no initial drop in utility is observed, rather only an improvement as a result of surgery. 127,132–138,141,143,145 Improved utility is most evident in the first 3 months post surgery with more of a plateau observed after around 3–6 months. 98,132,134,136,140,145 One study found that while remaining above preoperative levels, there is evidence of a slightly longer-term reduction in utility at around 15 months post randomisation,132 while another study identifies this reduction at 7 years. 139 This finding is, however, not consistent across all included studies, with sustained improvements in utility observed to extend beyond 1 year elsewhere. 134,136

Little difference in outcome utility is observed between inpatient and outpatient total laparoscopic hysterectomies,131 or when comparing vaginal and abdominal hysterectomy to their laparoscopic alternatives. 136 Failure or complication in vaginal hysterectomy is seen to reduce utility below that at baseline. 138 Meanwhile, in uterine preservation surgery, failure results in only a negligible reduction in utility, while a complication still results in an overall improvement in utility compared to that at baseline. 138 When asked to value, by time trade-off, the utility of total abdominal HBSO, greater utility was expected from undergoing rather than forgoing the surgical intervention. 88 Additionally the study found higher utility values from assessing the surgical decision from the sibling state (participants imagine having sibling with CRC and positive test for LS) than the proband state (participants imagine themselves having CRC and suspected LS).

In the BRCA1/2 population, as would be expected owing to the additional burden, combined oophorectomy and mastectomy yielded a lower utility value than oophorectomy alone,89,90 or oophorectomy accompanied by breast cancer screening. 128 This was consistent across patients of different age groups, with high risk, and when positive for breast cancer. This higher utility in oophorectomy alone was also reflected in valuations by ‘other individuals’ in the public, however health professionals did not make a distinction between oophorectomy alone and combined oophorectomy and mastectomy in their utility estimates for the BRCA1/2 population. In considering prophylactic oophorectomy alone, BRCA1/2 mutation carriers placed a slightly higher utility value on the procedure than a control group of women with personal or family history but no known personal risk of breast/OC. 114 This was also observed in a second study by the same authors, however while the public underestimated compared with BRCA1/2 carriers aged 20–32 years, they greatly overstated utility related to prophylactic oophorectomy compared to carriers aged 33–50 years. 89

Additional utility detail from studies on gynaecological surgery can be found in Table 30, Appendix 5.

Metachronous cancer

Although metachronous CRC does occur in LS (and is reported in approximately 1 in 10 with LS who are diagnosed with CRC), the model does not at present incorporate metachronous CRC.

Survival

As with other LS-associated cancers, survival from CRC is better among individuals with LS than among individuals with no diagnosis of LS. This could be caused by a number of factors, since CRC tends to be diagnosed at an earlier age and stage in LS (and these are presumably somewhat affected by surveillance), and CRCs in LS may be less likely to metastasise due to immune differences.

Xu et al. (2021)162 have conducted an analysis of CRC survival in LS, which attempts to isolate the impact of LS. After diagnosing 47 patients who had CRC with LS via reflex testing, they used propensity score matching to identify two controls (patients with CRC not diagnosed with LS) for each patient with LS. After this, they conducted survival analysis using Kaplan–Meier curves and evaluated the risk ratios for surviving to 5 years. At 5 years, overall survival was 0.976 in LS and 0.826 in sporadic CRC; in a multivariate analysis the risk ratio for death by 5 years was 0.106 (95% CI 0.025 to 0.446) for LS compared with sporadic CRC.

As a hazard ratio is more useful for modelling, we estimated the unadjusted risk ratio and hazard ratios to be 0.138 and 0.127, respectively. We assumed that the effect of adjustment on the log scale would be approximately equal for the risk ratio and hazard ratio, and therefore estimated a hazard ratio for overall survival of 0.098, lnβLS~N(−2.33,0.88), with the uncertainty estimated by inflating the standard error in the adjusted log risk ratio by 20%.

For sporadic CRC, we fitted an exponential survival model with an unshared frailty component and terms related to stage (I, II, III, IV), sex and age group (15–44, 45–54, 55–64, 75+ years). The data source was 1- and 5-year survival for CRCs diagnosed between 2013 and 2017 and followed up to 2018 as reported by the Office for National Statistics. 163 The fitting was performed using maximum likelihood estimation, using survival estimates and their 95% confidence limits to construct a likelihood function using normal distribution probability density functions. The variance–covariance matrix from the maximum likelihood estimation was used to represent the uncertainty and correlation in these parameters.

Determining the baseline state

Even though we consider a population that has not been diagnosed with EC, at the start of the model it is not appropriate to assume the entire population is free from endometrial lesions (unless the population is being modelled from birth). We must model the state of the endometrium conditional on the observation that the individual has not been diagnosed with EC.

While the EC model cannot be described as a finite state cohort model since there is no upper limit on the number of endometrial lesions, it can be approximated by a cohort model in which no more than three endometrial lesions can arise. With this assumption, the size of the state space was reduced to 286. For each genotype, we performed cohort simulation and recorded the probability distribution over the states at each integer value of age.

Management of atypical endometrial hyperplasia

If an AEH is identified, the management may be conservative (hormone therapy via insertion of a LNG-IUS, with or without oral progestin) or may be surgical (hysterectomy with or without BSO). Per clinical expert opinion, this decision is strongly influenced by the age of the patient, since younger patients may wish to subsequently attempt conception and may be concerned about the prospect of surgical menopause (if surgery includes BSO) or premature menopause.

We therefore modelled the probability of surgical management of AEH as:

This model allows for some people to choose surgery regardless of age (θ1), some people to choose conservative management regardless of age (1−θ2−θ1), and for the remainder to be influenced by age. We assumed that the age effect would be centred around θ3~N(40,2) and would have a maximum slope determined by:

We assumed that (θ1,θ2,1−θ1−θ2)~Dir ​​[​​ (1,8,1) ​​]​​ , so that 10% would always choose conservative management and 10% would always choose surgical management, but these figures are subject to uncertainty.

We assumed that surgical management would be 100% effective, that is no residual risk of EC from an AEH following hysterectomy. For conservative management we assumed that within 1 year, treatment would result in regression (disappearance of the AEH) with probability. 168–170

Clinical expert opinion is that the probability of successful medical management may be lower for individuals with LS. During conservative management patients would undergo endometrial sampling or hysteroscopy every 3 months to check for disease progression. We assumed that if conservative management was not successful after 1 year, the patient would undergo surgical management.

It is possible that an AEH may be diagnosed when there is fact also an EC. If the AEH is managed surgically we assume that the EC is detected during pathological examination and from that point on it is treated as a diagnosed EC (either due to surveillance or symptoms depending on how the AEH was diagnosed). If the AEH is managed conservatively and the EC is stage I, we assumed there was a probability of the EC regressing within 3 months ofp~Beta(11,5). 171 If the cancer did not regress, we assumed that it would be detected at the first check for disease progression.

Survival

It has been observed that survival for LS-associated EC is better than for sporadic EC. 172 There are multiple possible explanations for this, including different distributions of stage, histology, grade, age at diagnosis and risk factors (e.g. obesity).

As we model age and cancer stage we ideally require an estimate of survival for LS-associated EC according to age and stage, and similar for sporadic EC.

We considered three sources when estimating the survival of LS-associated EC.

First, we considered survival from the Prospective Lynch Syndrome Database (PLSD). 2 Many of the women in this database will have been undergoing gynaecological surveillance, so it is possible that survival in this database is biased upwards. The authors estimated crude overall survival from EC in those diagnosed aged under 65 years to be 0.89 (95% CI 0.82 to 0.94) at 5 years (and also at 10 years). This source has the advantage that all included participants have confirmed LS and it includes data from multiple centres across the world, but it has the disadvantages of possible bias due to surveillance, and it does not present according to stage at diagnosis, nor does it report stage at diagnosis.

Second, we considered a study published by Carr et al. (2020). 173 They reported the results of MMR deficiency testing and MLH1 promoter hypermethylation testing in 1018 ECs treated by hysterectomy at a single centre. Patients were therefore classified as MMR-I (MMR intact), MMR-DM (MMR deficiency explained by MLH1 promoter hypermethylation) and MMR-DU (MMR deficiency without MLH1 promoter hypermethylation). Of the MMR-DU subgroup, approximately half of those undergoing genetic testing for LS had a pathogenic variant of an MMR gene identified (i.e. confirmed LS). The authors presented progression-free and overall survival at 3 years, stratified according to stage at diagnosis: stage I/II, progression-free survival 0.912 (95% CI 0.792 to 1.00), overall survival 1.00 (95% CI 1.00 to 1.00); stage III/IV, progression-free survival 0.656 (95% CI 0.247 to 1.00), overall survival 0.750 (95% CI 0.450 to 1.00).

Finally, we considered the study published by Post et al. (2021),174 which included an evaluation of recurrence-free survival and overall survival in women recruited to the PORTEC-1, -2 and -3 studies who were found to have LS. At 5 years, recurrence-free survival was 0.917 (95% CI 0.831 to 1.00) and overall survival was 0.885 (95% CI 0.785 to 0.998).

We modelled survival from LS-associated EC as follows. For cancers diagnosed at stage III/IV, we assumed an exponential distribution for the time to cancer-specific mortality with rate

on the basis that two of eight participants in the study by Carr et al. (2020) with stage III/IV MMR-DU EC died within 3 years.

For cancers diagnosed at stage I/II, we assumed an exponential distribution for time to recurrence, followed by an exponential distribution for time from recurrence to cancer-specific mortality. We modelled the rate of recurrence as