NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Bonis PA, Trikalinos TA, Chung M, et al. Hereditary Nonpolyposis Colorectal Cancer: Diagnostic Strategies and Their Implications. Rockville (MD): Agency for Healthcare Research and Quality (US); 2007 May. (Evidence Reports/Technology Assessments, No. 150.)

  • This publication is provided for historical reference only and the information may be out of date.

This publication is provided for historical reference only and the information may be out of date.

Cover of Hereditary Nonpolyposis Colorectal Cancer

Hereditary Nonpolyposis Colorectal Cancer: Diagnostic Strategies and Their Implications.

Show details

3Results

Overview

The MEDLINE® literature search produced 3,018 abstracts, which were screened for their relevance. The full text article was retrieved for all studies that appeared to be relevant to the Key Questions. With the addition of titles from technical experts and reviews of bibliographies, 523 publications were retrieved for review. One hundred fifteen papers fulfilled eligibility criteria but 11 were duplicate reports and were therefore excluded (or used to provide supplementary information as described in Chapter 2). Thus, 104 unique studies were included for this review (40 for clinical validity, 3 for analytic validity and 61 for benefits and harms) while 410 did not meet the inclusion criteria and were rejected (see Figure 2; see also Appendix D *).

Figure 1. Analytic framework that served as the basis of the key questions proposed by the CDC Office of Genomics and Disease Prevention 2005.

Figure

Figure 1. Analytic framework that served as the basis of the key questions proposed by the CDC Office of Genomics and Disease Prevention 2005.

Figure 2. Literature search results.

Figure

Figure 2. Literature search results. * Analytic framework that served as the basis of the key questions

In the following sections, we present results for Key Questions as described in Chapter 2. The Key Questions pertaining to each domain (i.e., clinical validity, analytic validity, and benefits and harms) are presented within each section.

Analytic Validity

Key Question 4: What is Known About the Analytic (Sensitivity, Specificity, Reproducibility, Reliability) and Clinical Validity of Tests That Identify HNPCC Mutations?

Image er-hnpccfu2.jpg

Definition

The analytic validity of a laboratory test refers to its accuracy and reliability for identifying a finding of interest (such as a genotype). There are four general elements of analytic validity (see http://www.cdc.gov/genomics/gtesting/ACCE.htm for further details):

1)

Sensitivity (i.e., the ability to detect a finding when it is present)

2)

Specificity (i.e., the ability to exclude a finding when it is absent)

3)

Laboratory quality control (i.e., the procedures used in a testing laboratory to ensure that results fall within specified limits)

4)

Assay robustness (i.e., how resistant the assay is to changes in pre-analytic and analytic variables).

Laboratory Tests Used in HNPCC

The major laboratory tests used in the evaluation of patients suspected of having HNPCC include testing of tumor tissue using IHC, MSI testing, or germline (generally from peripheral blood mononuclear cells) testing for mismatch repair defects. Family members generally undergo only germline genetic testing (unless they have also developed a relevant cancer), ideally based upon the genotype of the proband. Detection of a pathogenic mutation (i.e., one known to be associated with HNPCC) in a proband permits testing of at-risk family members for the genotype. Family members with the same genotype have HNPCC while HNPCC can be excluded in those who do not. The situation is more complex when the proband does not have a detectable DNA alteration associated with HNPCC or when an alteration with unclear clinical significance is detected. In such cases, it may not be possible to exclude HNPCC in family members; as a result, they may be empirically offered enhanced cancer surveillance.

The laboratory tests used in HNPCC have been described in several reviews, including a comprehensive summary published in 2006 to which we refer readers seeking a detailed review.7 The specific tests and order of testing used in the studies included in this review are summarized in Table 11.

Table 11. Genetic testing strategies used by studies included in quantitative analyses.

Table 11

Genetic testing strategies used by studies included in quantitative analyses.

Immunohistochemistry. Pathogenic mutations in mismatch repair proteins usually lead to the absence of a detectable gene product providing the rationale for IHC techniques used to detect underexpression. Tumors from patients suspected of having HNPCC (based upon clinical or pathologic findings) can be stained for mismatch repair proteins. The surrounding normal colonic tissue can be used as a positive control.

As noted in the table below, testing is available commercially for MLH1, MSH2, MSH6 and PMS2, although the extent to which various laboratories stain for all of these proteins is unclear. Furthermore, such staining is relatively easy to perform and available in kits. As a result, local pathologists, who may select to stain for some or all of these proteins, can perform it.

Knowledge of how the mismatch repair proteins interact during DNA repair can help interpret the results of such testing and be useful for guiding germline genetic testing. For example, MSH2 forms a heterodimer with MHS6; MLH1 complexes with PMS2 and binds to the MSH2–MSH6 heterodimer. When MSH2 is not expressed in a tumor, MSH6 is also not expressed. Because MSH2 and MLH1 are the most common mismatch repair proteins implicated in HNPCC, a patient who has a tumor that does not stain for MSH2 (and whose normal surrounding colonic mucosal demonstrates preserved staining) is most likely to have a mutation in MSH2 (but could also possibly have a mutation in MHS6). Similarly, MSH6 may be the likely gene involved in a patient with a tumor that expresses MLH1 and not MSH6 (again with normal staining patterns in surrounding colonic mucosal). The situation is more complex with lack of expression of MLH1; promoter hypermethylation of MLH1 is common with sporadic colorectal cancer and may lead to its underexpression. Some laboratories offer methylation analysis to help determine whether lack of expression is due to promoter hypermethylation. Such an approach has been suggested in various reviews of HNPCC.

Immunohistochemistry has an advantage over other techniques (particularly MSI testing) since it is much easier to perform and is less expensive. However, the technique is vulnerable to the quality of tissue preparation, staining, and interpretation. This concern is not merely, hypothetical; our literature search revealed that sensitivity to detect loss of MSH2 expression ranged from 84 to 100 percent in a study of 18 participating centers performing such testing.56 Variability in specificity was even greater (see below).

Microsatellite Instability. Microsatellite instability (MSI) refers to a variety of patterns of microsatellite repeats observed when DNA is amplified from a tumor with defective mismatch repair compared with DNA amplified from surrounding normal colonic tissue. Repetitive mono- or dinucleotide DNA regions are particularly vulnerable to defective mismatch repair. For example, the mononucleotide sequence “AAAAAAAAAAAAAAAAAAAAAAAAAA” is located on chromosome 2P, near the MSH2 gene locus. This sequence is referred to as the “Big A Tract-26” (BAT26). As a result, a tumor suspected to result from mismatch repair defects can be tested for the presence of these repeats.

MSI testing involves amplification of a standardized panel of DNA markers; five markers were agreed upon by a consensus panel convened by the National Institutes of Health in 1997 (BAT25, BAT26, D2S123, D5S345, and D17S250) as described in Chapter 1. Three categories of MSI have been recognized based upon these panels: MSI-high (instability of two or more markers), MSI-low (instability of one marker), and MS-stable (no instability). More recently, some laboratories have begun using ten or more markers. In such cases MSI is defined as “stable” when fewer than 10% of markers are unstable, “low” when 10 to 30% of markers are unstable, and “high” when greater than 30% of markers are unstable (some laboratories use 40%). We recorded the markers used in each study included in this review (see section on clinical validity and accompanying tables).

There are several pitfalls of MSI testing. First, it is labor intensive, relatively costly (compared with IHC), and requires expert pathologic services. In addition, tissue to be amplified should ideally be microdissected to avoid amplifying DNA from normal colonic mucosa. We systematically recorded whether microdissection was performed for all studies related to clinical validity. As a practical consideration, tissue may not always be available since the diagnosis of HNPCC may not be suspected when the cancer was first diagnosed.

Genetic Testing. Multiple methods have been used for genetic testing in HNPCC. The methods used should ideally be able to detect the many potential genotypes associated with HNPCC (e.g., nonsense, missense, and frameshift mutations, genomic deletions, duplications, and rearrangements). We recorded the specific methods and order of testing used in all studies included in this report and, in the case of clinical validity, attempted to discern whether the specific testing methods were associated with the accuracy of HNPCC testing (see sections and corresponding tables on clinical validity).

A detailed review of these methods is beyond the scope of this report. However, the following summarizes major categories of testing that are used currently, and that were reported in the studies included in this report.

High Output Screening Techniques. High output screening techniques include single stranded conformation polymorphisms (SSCP), conformation sensitive gel electrophoresis (CSGE), denaturing gradient gel electrophoresis (DGGE) and denaturing high-pressure liquid chromatography (DHPLC). These methods all take advantage of the observation that alteration of DNA (due to a polymorphisms or mutation) confers chemical properties that allow it to be differentiated from normal DNA. These approaches can be performed relatively rapidly and allow more detailed studies (such as DNA sequencing) to be targeted to specific regions of DNA.

DNA Sequencing. DNA sequencing can be used following a high output screening technique or as a primary approach (particularly when IHC patterns allow for targeting of a specific mismatch repair gene). It is considered the method of choice for detecting most mismatch repair gene mutation. However, it does not reliably allow for detection of deletions or rearrangements, which are important in HNPCC. DNA sequencing has become automated in recent years, greatly reducing the required time, costs, and expertise.

Conversion analysis. Conversion analysis involves converting diploid cells to haploid cells so that only a single allele is analyzed at a time. The rationale is based upon the observation that a wild-type allele can mask the presence of a mutant allele when performing DNA sequencing (thereby obscuring the presence of a mutation). Conversion testing can increase the yield of genetic testing in HNPCC but is technically complicated, expensive, and, as a result, not widely available. We recorded whether conversion analysis was performed in all studies included in this report.

Methods To Detect Large Structural DNA Abnormalities. Large structural DNA abnormalities (such as large genomic deletions, rearrangements) are potentially important in HNPCC but are not detected by the high output screening techniques or DNA sequencing. There are several methods for detecting these defects. We recorded the specific methods used for all studies included in this report.

Southern blotting involves digestion of genomic DNA (which breaks it into pieces), separation of fragments using electrophoresis, transfer of the fragments to a membrane, and hybridization using probes to recognize deletions, duplications or rearrangements. Southern blotting has not yet been automated to the extent of DNA sequencing and is time-consuming.

Multiplex ligation-dependent probe amplification (MLPA) is a newer technique. It involves measurement of the relative copy number of DNA sequences. MLPA has evolved to become a standard approach for analyzing mismatch repair genes for deletions.

Family History. The family history (and related risk assessment tools such as the Amsterdam criteria) can also be considered as a type of laboratory testing, which can be applied to both the probands and their family members. The accuracy of the family history in predicting the presence of germline mismatch repair mutations is described in the sections on clinical validity. By contrast, the analytic validity of the family history can be considered the accuracy with which individuals are able to report their family history.

We did not identify any studies that assessed the analytic validity of the family history in patients or families with HNPCC. However, we confined our literature search specifically to HNPCC, while there are several studies that have assessed the validity of the family history for a variety of tumor types, including those associated with HNPCC such as colorectal or endometrial cancer.5760 A systematic review of these data found that (in individuals without a personal history of cancer) the positive and negative likelihood ratios of a family history for colon cancer in a 1st degree relative were 23.0 (95% CI 6.4–81.0) and 0.25 (95% CI 0.1–0.63), respectively.59 These values were 14.0 (95% CI 2.2–83.4) and 0.68 (95% CI 0.31–1.52), respectively for endometrial cancer. In another report, the degree of relationship to the probands, type of cancer, age at diagnosis of the probands, and source of ascertainment of probands were all statistically significant predictors of the accuracy of reporting.60 The extent to which these data can be generalized to patients and families with HNPCC is unclear.

Limitations of a Literature-Based Approach to Analytic Validity

We encountered only three studies that were eligible based upon our inclusion criteria, and these are summarized below. This was disappointing but expected, since a literature-based approach to understanding analytic validity has severe limitations as described in Chapter 1:

A)

Many of the laboratory techniques (e.g., DNA sequencing) are used not only in testing for HNPCC but also in testing for a variety of genetic disorders. As a result, a comprehensive search of issues related to analytic validity should consider the entire literature evaluating such testing, an undertaking that is neither practical nor feasible. A search in MEDLINE® on DNA sequencing, for example, will yield more than 416,000 citations.

As noted in the methods section, we confined our search to studies that used biological materials from patients or family members suspected of having HNPCC to make the results as applicable to the population of interest as possible. The accuracy of these methods used for other disorders may not be generalizable to HNPCC. The extent to which such sources of indirect evidence may be applicable to HNPCC was beyond the scope of this review.

B)

The methods used in laboratory testing evolve rapidly. The specific methods used are commercialized by companies or individuals who developed them. As a result, the published literature may not (and often does not) reflect the actual testing performed in clinical laboratories. It is customary for commercial laboratories to assess at least some components of analytic validity for these contemporary tests, but laboratories are under no obligation to publish the results.

A partial list of laboratories offering testing services for HNPCC is available through “Genetests” an organization funded through the National Institutes of Health (see http://www​.genetests​.org/servlet/access?id​=8888891&key​=AmmF6UdFjuxHB&fcn​=y&fw​=vDlR&filename=/). A summary of the tests listed on this Website is presented in Table 12 below. We attempted to contact one of the laboratories (chosen at random) to determine whether they would share information related to analytic validity. Our attempt had limited success. Such a review of “gray” literature may be helpful for understanding the analytic validity of these tests but would require appropriate resources, including incentives, for participating laboratories.

Some commercial laboratories participate in proficiency testing, but the results are generally not published. Proficiency testing involves distributing samples of tissue that are known to be positive or negative for findings of interest. Laboratories participating in such programs can obtain a benchmark for how they are performing in detecting (or excluding) such findings (e.g., genotypes of interest). Proficiency testing also helps understand issues related to reliability.

A major limitation of proficiency testing is that it does not consider the pre-analytic components of testing; for example, the accuracy of a test may be greatly affected by the methods used to acquire tissue. In addition, proficiency testing does not allow for a detailed understanding of the spectrum of abnormalities (e.g., various genotypes) that may be clinically relevant.

Proficiency testing for MSI was introduced in 2006 and is being conducted by the College of American Pathologists. Further information is available at their Website (see http://www​.cap.org/apps/cap.portal) but the initial results of their program have not been made public.

C)

Studies evaluating analytic validity often suffer from serious methodologic limitations that potentially invalidate the results. Particularly problematic was the inappropriate use of reference standards or lack of clear definitions of references standards altogether. For example, a common pitfall was the inclusion of the test under evaluation as part of the reference standard.

We chose inclusion and exclusion criteria carefully to select for high quality studies while excluding studies whose results are misleading or clinically irrelevant. There were remarkably few high-quality studies directly addressing analytic validity using biological materials from probands or family members proven or suspected to have HNPCC.

Table 12. North American laboratories offering clinical testing for Hereditary Nonpolyposis Colon Cancer (Lynch Syndrome)*.

Table 12

North American laboratories offering clinical testing for Hereditary Nonpolyposis Colon Cancer (Lynch Syndrome)*.

Eligible Studies. We identified only three studies that fulfilled eligibility criteria for analytic validity.

  • One study compared conversion analysis (in which alleleles are separated into hybrids prior to mutation screening) with DNA sequencing alone to detect heterogeneous germline mutations in MLH1, MSH2 and MSH6 in patients with colorectal cancer.61 The authors estimated that conversion analysis provided an increased yield of 56% (35/63 cases) compared with DNA sequencing alone. The study was rated a methodologic quality B because the investigators were not blinded to the method used.
  • Another study included 20 patients with CRC with known mutations in MLH1 or MSH2.56 A set of two unstained slides from each case were sent to participating medical centers with capability of performing immunoperoxidase assays for MLH1 and MSH2.

Of 18 participating centers 2 were excluded: one because slides were damaged in transit and the other because of insufficient staining. Sensitivity for detecting loss of MSH2 expression ranged from 84 to 100%; 10 centers identified all six. Five out of six false positive results were in the same case suggesting that staining or interpretation were not random. Fourteen out of 16 laboratories showed 100% specificity (one laboratory had 93% specificity due to staining failure on one slide and one lab demonstrated 45% specificity due to weak or absent staining in most cases.)

Re-review of returned MSH2 slides showed lack of internal positive control staining in at least 2 of the 6 MSH2-negative cases from 8 of 16 centers. The other 8 centers had 100% sensitivity and 93–100% specificity on re-review. The slides that lacked internal positive control staining were largely accounted for by two cases.

The study was rated a methodologic quality B because of its small sample size, and because it did not describe quality control methods or whether microdissection had been performed when preparing the specimens.

  • A third study evaluated the sensitivity and specificity of DHPLC analysis compared with DNA sequencing in 46 patients with colorectal cancer from families with HNPCC.62 DHPLC analysis identified 19 changes previously identified by DNA sequencing and 16 new alterations not previously described. DHPLC was considered to be highly sensitive, detecting a mutation in all patients with no false negative results. The study was rated methodologic quality C because of several deficiencies in reporting.

Summary. In summary, the analytic validity of the tests used to evaluate HNPCC is substantially uncertain. However, there is heterogeneity in the type of testing offered by commercial laboratories and the available data suggest that there may be variability across testing facilities. Additional information that could shed light upon analytic validity is available but would require evaluation of non-published data sources. Committees of experts could also review the strengths and limitations of specific testing techniques based upon clinical experience, with particular reference given to experience with these methods in other genetically-based disorders.

Table 12 presents information from North American laboratories offering genetic testing for HNPCC. Most of the data was retrieved through the website, GeneTests.org. When available, links to the clinical laboratory websites offering genetic testing service were accessed to gather additional detail on type of tests offered. The information presented is not comprehensive as reporting to GeneTests.org is voluntary.

Clinical Validity

Key Question 2 and Pertinent Subquestions

Image er-hnpccfu3.jpg

HNPCC has been defined clinically and genetically as described in Chapter 1. The following sections present analyses using either the clinical or the genetic definition of the condition, depending on which Key Question was being addressed.

Key Questions 2a to 2c seek to evaluate the prevalence of MMR mutations and suggestive MSI and IHC among patients with the Lynch Syndrome. Thus, they assume a clinical definition of the condition.

Key Question 2 pertains to the ability of clinical and laboratory predictors to identify the presence of MMR mutations. Thus, it assumes a genetic definition of the condition.

We first present analyses based on a clinical definition of HNPCC. In these analyses we estimated the frequency of mismatch repair gene mutations and tumors with MSI and IHC among CRC patients fulfilling the Amsterdam I and II criteria.

We present subsequent analyses based on the genetic definition of the Lynch syndrome. These help define test characteristics of various strategies (such as the combination of a clinical history with laboratory testing of tumor tissue) for predicting the presence of mismatch repair mutations.

Finally, we used a decision tree model to calculate the expected number of patients with MMR mutations among unselected patients presenting with CRC using various predictive strategies. The parameters used in the calculations are based on our best estimates derived from this systematic review.

Key Question 2a: Assuming a Clinical Definition of the Lynch Syndrome, What Proportion of Patients has a Mismatch Repair Mutation?

Among CRC fulfilling Amsterdam I criteria, the random effects summary prevalence of MLH1 and MSH2 gene mutations was 44% (95% CI: 35, 52; n=19 studies, 464 CRC patients), with evidence for substantial between-study heterogeneity (p<0.01; I2=52%). The six studies that performed genetic testing among all Amsterdam I patients had a summary prevalence of 51% (95% CI: 35, 66%; 84 CRC patients).

For patients fulfilling the Amsterdam II criteria, the corresponding prevalence values were 39% (95% CI: 30, 49%; 10 studies, 279 CRC patients) and 40% (95% CI: 30, 52%) based upon 2 studies that performed sequencing on all 87 Amsterdam II patients.

Only three studies examined other MMR genes (MSH6, PMS1 and PMS2) in Amsterdam I patients, without identifying any additional mutations. Two additional MSH6 mutations were found among 20 Amsterdam II patients in a single study.

Frequency of MMR Gene Mutations Among Patients Fulfilling Amsterdam Criteria I

Description of studies. Nineteen studies (described in 21 papers 6383) provided data for the prevalence of MMR mutations among CRC probands who fulfilled the Amsterdam I criteria. The median number of patients with relevant data in each study was only 13 (interquartile range: 10, 22), and the actual number ranged between 3 and 154 patients. However, study sizes (all included patients) ranged between 19 and 509. Only five studies included at least 20 probands fulfilling the Amsterdam I criteria.67, 72, 76, 7981 Nine studies were rated grade B for their overall methodologic quality63, 6567, 70, 71, 74, 77, 7981 while the rest were rated grade C (Appendix F-1 *).

Six studies performed bidirectional sequencing on all patients who fulfilled the Amsterdam I criteria.69, 75, 77, 79, 81, 83 (Colombino 200583 performed full sequencing only among familial CRC cases, which included Amsterdam I patients). Ten studies performed full sequencing only on patients who were selected by gene screening methods such as DHPLC or SSCP.63, 6668, 70, 71, 73, 74, 76, 82 The remaining three studies did not describe the presence of any mutation with bidirectional sequencing in any patient64, 72 or did not provide any details on the genetic testing strategy they used.78 None of these 19 studies used conversion analysis to detect mutations.

Assessment of mutations in MMR genes other than MLH1 and MSH2. All reports assessed mutations in the MLH1 and MSH2 genes. Additional MMR genes were assessed in three studies.76, 78, 81 Park 199976 assessed the PMS1 and PMS2 genes in a minority of their Amsterdam I patients (24 and 27 patients respectively, less than 20% of their sample), without identifying any additional mutations(Appendix F-1 *). Wang 199981 also tested for MSH6, PMS1 and PMS2, but found no additional mutations among the subgroup of patients fulfilling the Amsterdam I criteria. Stormorken 200178 also tested for MSH6 mutations, but found none among Amsterdam I patients.

Analyses. Overall, the random effects summary estimate from the 19 studies was 44% (95% CI: 35, 52%), with evidence for substantial between-study heterogeneity (p<0.01; I2=52%) (Table 13, Figure 3). The six studies that performed full sequencing in all available samples identified mutations in 51% (95% CI: 35, 66%) of 84 patients fulfilling the Amsterdam I criteria (Table 13). The corresponding prevalence estimate was 44% when sequencing was performed after suggestive SSCP, DHPLC or other gene screening methods, and only 33% among studies that used screening methods without further confirmation by sequencing (Table 13). The prevalence of MMR mutations did not differ beyond chance with respect to overall study quality score, whether the study was small (i.e., analyzed fewer than 20 patients), or whether the authors described how they classified mutations as being pathogenic.

Table 13. Summary estimates of the prevalence of MMR mutations among CRC fulfilling Amsterdam I criteria.

Table 13

Summary estimates of the prevalence of MMR mutations among CRC fulfilling Amsterdam I criteria.

Figure 3. Prevalence of mismatch repair gene mutations among colorectal cancer patients fulfilling the Amsterdam I criteria.

Figure

Figure 3. Prevalence of mismatch repair gene mutations among colorectal cancer patients fulfilling the Amsterdam I criteria.

Finally, three studies (n=28 eligible patients) that sampled unselected, nonreferral patients with CRC found a higher prevalence of MMR mutations (66%) among patients who fulfilled the Amsterdam I criteria, compared to studies that sampled among referral or otherwise selected cases (prevalence 41%; Table 13). The precision of these estimates is uncertain because of the very small number of eligible patients.

Frequency of MMR Gene Mutations Among Patients Fulfilling Amsterdam II Criteria

Description of studies. Ten studies assessed the prevalence of MMR mutations among 271 CRC probands who fulfilled the Amsterdam II criteria (Appendix F-2 *). They were described in 12 papers.65, 67, 71, 72, 7780, 8487

The total number of patients in the various study populations ranged from 45 to 535 patients with CRC. However, the median number of patients fulfilling Amsterdam II criteria was 19.5 (interquartile range: 5, 35). Only five studies assessed more than 20 probands from families fulfilling the Amsterdam II criteria.67, 72, 79, 80, 86, 87 Eight out of ten studies received grade B in the overall quality rating while two72, 78 received grade C.

Four studies performed bidirectional sequencing in all patients fulfilling Amsterdam II criteria.77, 79, 84, 86 Four studies performed full sequencing only to patients who were selected by gene screening methods.65, 67, 71, 85, 87 The remaining two studies did not describe the presence of any mutation with bidirectional sequencing in any patient72 or did not provide any details on the genetic testing strategy they used.78 None of these 10 studies used conversion analysis.

Assessment of mutations in MMR genes other than MLH1 and MSH2. All studies tested for MLH1 and MSH2 mutations. Only Stormorken 200178 also tested for MSH6 mutations. In this study five out of 20 Amsterdam II patients had deleterious MLH1 and MSH2 mutations (25% [95% CI: 9, 49%]) and another two had deleterious MSH6 mutations (10% [95% CI: 1, 32%]).

Analyses. As expected, the overall prevalence of MMR gene mutations was lower compared with patients who fulfilled the stricter Amsterdam I criteria (39% [95% CI: 30, 49%]; Table 14, Figure 4). Between-study heterogeneity was statistically significant.

Table 14. Summary estimates of the prevalence of MMR mutations among CRC fulfilling Amsterdam II criteria.

Table 14

Summary estimates of the prevalence of MMR mutations among CRC fulfilling Amsterdam II criteria.

Figure 4. Prevalence of mismatch repair gene mutations among colorectal cancer patients fulfilling the Amsterdam II criteria.

Figure

Figure 4. Prevalence of mismatch repair gene mutations among colorectal cancer patients fulfilling the Amsterdam II criteria.

Studies that performed sequencing in all Amsterdam II patients estimated similar prevalence of MMR mutations compared to studies that performed sequencing only after suggestive gene screening with SSCP, DHPLC, or other methods (40% versus 45%, respectively; Table 14). Studies that were rated C for their overall quality estimated significantly lower prevalence rates (27%; Table 14) compared to studies that received B for their overall quality. In the latter, the summary estimate was 43% (95% CI: 34, 54%) with no evidence for heterogeneity.

Studies that mentioned how pathogenic mutations were characterized as such, contributed fewer than 20 patients, or sampled unselected, non-referral patients with CRC, tended to provide higher estimates for the prevalence for MMR mutations. However, all of these studies were small and thus it is possible that the observed variability could have been due to chance (Table 14).

Interpretation of the MMR Prevalence Eestimates. The summary prevalence was 44% among all studies on Amsterdam I patients, and 51% for studies that sequenced all patients. For Amsterdam II the corresponding numbers were 39% and 40%. These estimates pertain mainly to MLH1 and MSH2 mutations, because most studies tested only these two genes. The true prevalence of MMR mutations among Amsterdam I (or Amsterdam II) patients may be different (possibly higher rather than lower) than these estimates for several reasons:

  • Genes other than MLH1 and MSH2: Only a minority of studies tested for mutations in other MMR genes (especially MSH6, the gene most commonly implicated in patients without mutations in MLH1 or MSH288).
    • Stormorken 2001 and Wang 1999 did not find any deleterious MSH6 mutations among 34 Amsterdam I patients in total. The upper boundary of the exact binomial 95% CI for the frequency of MSH6 mutations in these two studies is approximately 10%.
    • Stormorken 2001 found two deleterious MSH6 mutations in addition to five MLH1 and MSH2 mutations out of 20 Amsterdam II patients. Thus, MSH6 mutations resulted in an increment of 10% (95% CI: 1, 32%).
    Presumably, if other studies had assessed MSH6 routinely, the total prevalence of MMR mutations would be higher. However, there were insufficient data to provide robust estimates.
  • Comprehensiveness of genetic testing - sequencing of all samples and testing for large deletions/rearrangements: Studies that performed sequencing in all samples estimated higher frequency of MMR mutations among Amsterdam I patients compared to other studies. This was not true among studies that focused on Amsterdam II patients. Thus, more comprehensive testing resulted in more identified MMR mutations (at least in Amsterdam I studies), as expected.
  • Role of founder mutations: The degree to which a search for mutations other than MLH1 and MSH2 will identify mutations in other MMR genes depends upon the population being studied. The spectrum of MMR mutations is variable and some are specific to certain populations. An example of the latter is the presence of founder mutations in the MLH1 genes that are common among patients of Finnish origin, but are not prevalent in patients of different descent. The Finnish study63 included only 4 patients fulfilling the Amsterdam criteria I, all of whom were found to have MMR mutations. An American founder mutation in the MSH2 gene has been described in one kindred, probably originating from German immigrants, but it is unlikely to have a large impact in the general population.90, 91 The Finnish founder mutations are probably not applicable to the US population. Nevertheless, the summary estimates are practically identical if the Finnish study63 is excluded.
  • Identifying deleterious mutations: As mentioned in the methods, only mutations that were characterized by the primary studies as deleterious or pathogenic were analyzed. Misclassifications may have occurred in both directions:
    • Some deleterious missense mutations may have been erroneously misclassified as non-pathogenic.
    • The opposite is also likely, especially when pathogenicity is inferred on the basis that the mutation was absent from cancer-free controls. This was done in five studies.63, 66, 67, 82, 83 However, most of the studies included only small numbers of cancer-free controls (only around 100 patients were used as controls in the five studies). Absence of a mutation in such a small population of cancer-free controls does not necessarily prove that the mutation is not pathogenic.
    There were insufficient data to provide precise estimates of how many missense mutations were found among the subgroup of Amsterdam I (or Amsterdam II) patients in the studies that performed comprehensive genetic testing.
  • Sampling of tested Amsterdam I and Amsterdam II cases: The strategy used for selecting patients with CRC can have a substantial impact on the prevalence estimates (see comments on additional studies that were suggested for consideration by the TEP, below). Only three studies on Amsterdam I patients studied unselected patients with CRC, but they identified a total of only 28 patients. By contrast, studies that identified cases that had been included in cancer databases (e.g., the ICG-HNPCC database that was used in the Park 1999 study) represent a highly selected population,88 or may pertain almost entirely to individuals with a high probability of having mutations (e.g., because of suggestive MSI testing prior to their registration in the database).61 Biases resulting from sampling are not only a theoretical concern (see below comments on additional studies that were suggested for consideration by the TEP).

Additional studies suggested for consideration by the TEP. Several additional studies were considered by the TEP during the peer review process as important and potentially relevant that had not been included in the draft report. Although none of these studies was eligible for these analyses according to the eligibility criteria, they provide insight and complementary data, directly supporting the analyses and their interpretation. Most importantly, the findings from these studies are in accordance with the notion that sampling of Amsterdam I or II patients from different sources may result in very different prevalence estimates, even when the genetic testing strategies are comprehensive.

  • Balmana 200692 reported the prevalence of MLH1 and MSH2 gene mutations among 1914 unrelated people who had been tested for MMR mutations in the Myriad Genetic Laboratories. Genetic analysis was performed using a combination of sequencing and southern blotting. The cohort included 534 unrelated people fulfilling the Amsterdam II criteria (although it was unclear if all of them had CRC). Amsterdam I patients were not reported separately. Only half of the Amsterdam II patients (274/534=51%) were tested for large deletions or rearrangements.
    In total, 180 Amsterdam II participants were found to be carriers of MLH1 or MSH2 mutations (34% [9%% CI: 30, 38%]). This estimate includes large deletions or rearrangements and pathogenic and non-pathogenic mutations. About one fourth (27% [95% CI: 20, 35%] of the mutations were due to large deletions/rearrangements in a subgroup of the 1016 participants who underwent such testing. Extrapolating the proportion of large deletions to all Amsterdam II patients would result in a total prevalence of approximately 38%, very close to the summary estimate provided above. This study was excluded from the analyses because it was unclear whether all Amsterdam II participants had CRC.
  • Wagner 200393 evaluated 49 individuals from the Henry Lynch Cohort fulfilling the Amsterdam criteria (presumably Amsterdam I criteria), at least 9 of whom were not CRC patients. They used DGGE to screen for MMR mutations in the MLH1, MSH2 and MSH6 genes, and performed sequencing on exons with altered migration patterns. They tested for large deletions, and performed mono-allelic mutation analysis (MAMA) in very few samples, mainly to demonstrate the feasibility of the technique.
    Overall, 25 out of 49 had deleterious MLH1 or MSH2 mutations, 1 had deleterious MSH6 mutations, and 14 had genomic rearrangements. Thus, the overall proportion was 82% (95% CI: 68, 91%). In addition, 5 missense mutations were found; were they to be considered pathogenic, the overall proportion would be 92% (95% CI: 80, 98%). The detection of large deletions/rearrangements resulted in 29% increase in the estimated prevalence of MMR mutations. The degree to which the Henry Lynch Cohort is representative of the whole Amsterdam I population is unclear.
  • Liu 199694 assessed the MLH1, MSH2 and PMS2 genes 48 families of Amsterdam I patients who had suggestive MSI (note that more than one person per family was tested for mutations -i.e, reporting was at the family not the individual level). They found a prevalence of “drastic” (deleterious) mutations of 54%. Counting all identified mutations, drastic or not, 71% (95% CI: 56, 83%) of the families carried a mutation. Fifteen families had large genomic deletions/rearrangements (31%), but many families had more than one type of mutations.
    It is unclear how many unrelated people had only large deletions or rearrangements in this study. There might be some overlap with the Wanger 200393 families for patients from North America. The authors note that their sample was not representative of the general Amsterdam I population.

Caveat on true overall prevalence of MMR mutation carriers. Limited data suggested that approximately one-fourth to one-third of genotypes associated with HNPCC are related to deletions or rearrangements that would be missed through sequencing alone. As a result, the prevalence of mutations in Amsterdam I (or II patients) assessed from studies that performed sequencing alone are likely to be underestimates.

Accounting for this effect, one may derive a prevalence of MMR mutations of approximately 63% to 67% in Amsterdam I patients. For Amsterdam II patients the corresponding prevalence values would be 50 to 53%.

Furthermore, limited data suggest that approximately 10% of MMR genotypes involve MMR genes other than MLH1 and MSH2. Thus, assuming an additional 10% increase in mutations by assessing more genes would result in an overall prevalence of up to 70% to 75% for Amsterdam I patients, and 55 to 59% for Amsterdam II patients.

Key Question 2b: Assuming a Clinical Definition of the Lynch Syndrome, What Proportion of Patients has MSI?

Among patients fulfilling the Amsterdam I criteria (data from 11 studies, n=159 patients fulfilling the criteria), 71% (95% CI: 63, 78%) of tumors were found to be MSI-H (p=0.51 for heterogeneity; I2=0%). Among patients who fulfilled the Amsterdam II criteria (data from four studies, n=102 patients fulfilling the criteria), the corresponding summary prevalence was 68% (95% CI: 58, 76%) without evidence for between-study heterogeneity (p=0.91, I2=0%).

Patients Fulfilling Amsterdam I Criteria. Eleven studies with 159 patients fulfilling the Amsterdam I were criteria were included in the analysis (Figure 5). These were described in 13 papers.6366, 6872, 75, 79, 80, 95 Although the total number of patients with CRC included in the primary studies ranged from 11 to 509, the median number of CRC patients eligible for this analysis was 10 (interquartile range: 9, 18), and only one study assessed more than 20 CRC patients. (Lamberti 1999,72 n=57). Five studies 63, 65, 66, 70, 71, 79, 80 received a grade B in the overall quality rating while the others were rated a grade C.

Figure 5. Prevalence of MSI in colorectal cancer tumors from patients fulfilling the Amsterdam I criteria.

Figure

Figure 5. Prevalence of MSI in colorectal cancer tumors from patients fulfilling the Amsterdam I criteria.

We considered specific study features as described in Chapter 2 (i.e., whether the study reported use of microdissection and used the microsatellite marker set recommended by the National Cancer Institute). Only six studies reported that they used microdissection (which helps assure that the tissue sample that was tested for MSI consisted mainly of malignant cells).6466, 68, 71, 79, 80, 95 Only two studies (reported in three papers64, 79, 80) used the marker set recommended by the National Cancer Institute (Appendix F-3 *).

Overall, 71% (95% CI: 63, 78%) of tumors were MSI-H (p=0.51 for heterogeneity; I2=0%). The estimate was consistent among the various subgroups that we assessed (Table 15). The prevalence rates were generally higher among studies that used the marker sets recommended by the National Cancer Institute (Table 15).

Table 15. Summary estimates of the prevalence of MSI-H in colorectal cancer tumors from patients fulfilling Amsterdam I criteria.

Table 15

Summary estimates of the prevalence of MSI-H in colorectal cancer tumors from patients fulfilling Amsterdam I criteria.

The prevalence of combined MSI-H and MSI-low was similar to that of MSI-H (Appendix F-3 *).

Patients Fulfilling Amsterdam II Criteria. Only four studies including 102 patients fulfilling the Amsterdam II criteria were identified (Appendix F-4 *). These were described in five papers.65, 71, 72, 85, 86 The largest study (Lamberti 199972) analyzed 59 available eligible tumors, and the smallest study (Salovaara 200085) had only 4 eligible tumors. The total number of patients with CRC included in these studies ranged from 45 to 535. All studies except for the one by Lamberti 199972 received grade B for overall quality. None of the studies used the National Cancer Institute recommended marker sets, and only two 79, 80, 86 used microdissection (Appendix F-4). Overall, the summary prevalence was 68% (95% CI: 58, 76%) without any evidence for between-study heterogeneity (p=0.91, I2=0%) (Figure 6).

Figure 6. Prevalence of MSI-H in colorectal cancer tumors from patients fulfilling the Amsterdam II criteria.

Figure

Figure 6. Prevalence of MSI-H in colorectal cancer tumors from patients fulfilling the Amsterdam II criteria.

A pooled estimated for the three fair quality (B overall quality) studies was 66% (95% CI: 51, 79%) with no evidence for heterogeneity. Estimates from the two studies with more than 20 patients that fulfilled the Amsterdam II were similar to the overall estimate. The results were also similar for studies that used the markers sets recommended by the National Cancer Institute.

Finally, results were consistent when the MSI-L and MSI-H tumors were combined in two studies (described in three papers 65, 71, 72, Appendix F-4 *). The overall prevalence was 71% (95%CI: 60, 81%) with no statistically significant heterogeneity.

Interpretation of the MSI Prevalence Estimates. Some of the biases that may affect the calculated prevalence for MMR mutations are also applicable in this section, specifically biases that pertain to the selection/sampling of Amsterdam I and Amsterdam II patients who were studied. In addition, even among the selected Amsterdam I or Amsterdam II patients in each study, MSI testing often depended upon practical and logistical considerations (e.g., patient availability and consent and availability of tumor tissue). Thus, not all patients had all tests and it was unclear whether additional bias may have been introduced in selecting patients for testing.

Key Question 2c: Assuming a Clinical Definition of the Lynch Syndrome, What Proportion of Patients has Abnormal Protein Expression by Immunohistochemistry?

Among patients fulfilling the Amsterdam I criteria, the overall prevalence of tumors with loss of protein expression was 40% (95% CI: 28, 53%; n= 6 studies, 63 patients) with no evidence for between-study heterogeneity (p=0.75, I2=0%).

Only one eligible study provided relevant data for 20 patients fulfilling the Amsterdam II criteria. Eight out of 20 tumors had suggestive IHC for the MLH1, MSH2 or MSH6 genes (40% [95% CI: 9, 64%]).

Patients Fulfilling Amsterdam I Criteria. Only six studies that included a total of 63 eligible patients (median 12 tumors in each study) provided relevant data; no study contributed more than 15 tumors in this analysis. The total number of patients included in the pertinent studies ranged from 30 to 168. These were described in eight publications.65, 66, 6971, 7880.

Five out of six studies were characterized as grade B quality while two69, 78 were rated grade C. Only Stormorken 200178 assessed the expression of the MSH6 genes (in addition to MLH1 and MSH2) (Appendix F-5 *). However, none of the samples was positive for only MSH6.

The overall prevalence of tumors with loss of protein expression was 43% (95% CI: 30, 57%) with no evidence for between-study heterogeneity (p=0.85, I2=0%) (Figure 7). The summary estimate was similar (42% [95% CI: 29, 56%] without any heterogeneity) after excluding the report by Debniak 2000,69 which was the only one graded C for overall quality.

Figure 7. Prevalence of suggestive IHC in colorectal cancer tumors from patients fulfilling the Amsterdam I criteria.

Figure

Figure 7. Prevalence of suggestive IHC in colorectal cancer tumors from patients fulfilling the Amsterdam I criteria. Studies are ordered by decreasing number of patients.

Patients Fulfilling Amsterdam II Criteria. Only Stormorken 200178 assessed the prevalence of suggestive IHC among 20 patients fulfilling Amsterdam criteria II. This study was rated C for its overall methodologic quality. The expression of three MMR genes was sought, namely MLH1, MSH2 and MSH6. Eight out of 20 Amsterdam II patients had suggestive IHC (40% [95% CI: 19, 64%]).

Tumors that have suggestive IHC for MSH2 may also have suggestive IHC for MSH6, because in the absence of functional MSH2 protein the MSH2/MSH6 heterodimer is not formed correctly. Two out of 20 Amsterdam II patients had deleterious mutations in the MSH6 gene in Stormorken 2001.78 None had suggestive IHC only for the MSH6 antibody.

Interpretation of the IHC Prevalence Estimates. In contrast to MSI testing that detects replication errors, and is generally not MMR-gene specific, IHC uses antibodies that target specific MMR genes. Only one of the eligible studies used antibodies against MSH6.78 The MSH6 gene is the third most common gene with pathogenic mutations in HNPCC patients (approximately 6% of the mutations in the ICG-HNPCC database are in MSH688). Thus, the prevalence of suggestive IHC would probably be higher if anti-MSH6 or additional antibodies were used.

Some of the biases that may affect the calculated prevalence for MMR mutations are also applicable in this section, specifically biases that pertain to the selection/sampling of Amsterdam I and Amsterdam II patients who were studied. In addition, even among the selected Amsterdam I or Amsterdam II patients in each study, IHC testing often depended upon practical and logistical considerations (e.g., patient availability and consent and availability of tumor tissue). Thus, not all patients had all tests and it was unclear whether additional bias may have been introduced in selecting patients for testing.

Key Question 2: How Accurate Are Various Predictors Assuming a Genetic Definition of the Lynch Syndrome?

We examined a variety of clinical and laboratory predictors as described in Chapters 1 and 2. A central question of this report relates to the accuracy of various predictors in unselected patients with CRC. Because testing for MMR mutations is expensive, most studies performed testing only in patients who had been pre-selected based on clinical or laboratory features. Thus, it is critically important to describe the specific CRC population when attempting to make comparisons among studies.

The various clinical predictors defined populations at different risk for HNPCC. Figure 8 illustrates the relationship among patient populations defined by Amsterdam I criteria, Amsterdam criteria II, Bethesda guidelines, and unselected CRC patients.

Figure 8. Relationship between the CRC populations defined by different clinical predictors.

Figure

Figure 8. Relationship between the CRC populations defined by different clinical predictors.

Table 16 presents the different populations that underwent MMR testing in all studies included in this analysis. Moving to the right, columns represent increasingly selected patient populations (by analogy, moving to the top of the pyramid in Figure 8). Table 16 provides a roadmap for the analyses presented in the following sections.

Table 16. Overview of available evidence on the sensitivity and specificity of various predictors for MMR mutation rates.

Table 16

Overview of available evidence on the sensitivity and specificity of various predictors for MMR mutation rates.

Table 17 summarizes the overall sensitivity and specificity of the various predictors among several CRC populations. These are described in detail in the sections below.

Table 17. Sensitivity and specificity of various predictors for detecting MMR gene mutations.

Table 17

Sensitivity and specificity of various predictors for detecting MMR gene mutations.

We first present an overview of the clinical predictors among unselected CRC patients. We then present each one of the clinical predictors separately.

Overview of Clinical Predictors Among Unselected Patients With CRC

In total, five studies provided information on eight clinical predictors among unselected CRC probands (Salovaara 2000,85 Aaltonen 1998,63 Colombino 2005,83, Samowitz 2001,10 and Pinol 200512) (Appendices F-7, F-11, F-16, F-17, F-19, F-21, F-22, F-23, F-25 *). All studies received a grade B in the overall quality rating. All were limited by verification bias. In the two studies from Finland63, 85 founder mutations that are unique in the Finnish population were also assessed. These specific founder mutations have not been found in the US population.

The five studies differed in their testing strategies. Aaltonen 199863 and Salovaara 200085 performed comprehensive genetic testing only among patients whose tumors exhibited MSI. The remaining patients were tested only for common founder mutations in the Finnish population (large deletions in the MLH1 gene). Samowitz 200110 and Pinol 200512 performed genetic testing only among patients with tumors that had suggestive MSI (or suggestive IHC testing for Pinol 200512). Finally, Colombino 200583 performed comprehensive genetic testing only among patients with cancer history in 1st or 2nd degree family, and tested the remaining cases only for the mutations that were identified in the former.

As noted in the methods, we assumed that all patients who were not selected for genetic testing based on the MSI/IHC results were mutation negative.

Figure 9 illustrates the sensitivity and specificity of the clinical criteria among unselected CRC patients.

Figure 9. Sensitivity and specificity of clinical criteria for detecting mismatch repair mutations among unselected patients with CRC.

Figure

Figure 9. Sensitivity and specificity of clinical criteria for detecting mismatch repair mutations among unselected patients with CRC.

As shown in Figure 9 and Table 17, the revised Bethesda guidelines had the best sensitivity (91%, [95% CI: 59, 100%]). However, clinical predictors that are relatively easier to assess such as the family history of colorectal or endometrial cancer, or the fulfillment of at least one of three simple clinical criteria (proband age less than 50 years at diagnosis, history of CRC or endometrial cancer in 1st degree family, or multiple synchronous or metachronous CRC or endometrial cancer in the proband) appeared to have better test performance than the Amsterdam I and II criteria, or even the Bethesda guidelines. Specific summary estimates are presented in Table 17 and are also mentioned in the following sections.

Amsterdam I Criteria

Nineteen studies (described in 21 publications29, 61, 6368, 7177, 7983, 96) provided information needed to calculate sensitivity and specificity for this analysis (Appendix F-6 *). Fifteen assessed the clinical predictors among all patients, and three (Casey 2005,61 Raedle 2001,29 and Nakahara 1997,96 Appendix F-6 *) among patients with suggestive MSI.

Ten studies were rated grade B for their overall methodologic quality29, 61, 63, 6567, 71, 74, 77, 7981 and nine received grade C.

Genetic Testing Irrespective of MSI or IHC Status. Sixteen studies performed some form of genetic testing in all people, irrespective of suggestive MSI or IHC (Appendix F-6 *). Only four of these studies performed sequencing on all available samples,75, 77, 79, 81 and only Wang 199981 also tested for large deletions/rearrangements. Most studies tested only for mutations in MLH1 and MSH2. Wang 199981 also assessed the MSH6 gene, the PMS2 and PMS1 genes (no deleterious mutations were identified in non-MLH1 and non-MSH2 genes). Park 199976 sought mutations in additional MMR genes (PMS1 and PMS2) and in only a minority of patients (without finding any additional mutations). Eight out of the sixteen 2×2 tables had at least 40 patients in total.

There was variability in the CRC populations that were examined (Appendix F-6 *). The sample selection process was considered to be transparent (based on clearly stated selection criteria applied to all available patients) in only eight studies.63, 65, 67, 71, 74, 77, 7981, 83 Of these, three sampled from unselected, non-referral CRC populations.63, 65, 71, 83 The remaining five sampled from referral or otherwise selected CRC patients67, 74, 77, 7981 (Appendix F-6 *).

Figure 10 graphically depicts the sensitivity and specificity of the 16 studies that performed some form of genetic testing in all people irrespective of suggestive MSI/IHC. A study suggesting that the Amsterdam I criteria had perfect test characteristics (i.e., sensitivity and specificity of 100%) would appear in the upper left corner. The graph shows that the Amsterdam I criteria are not highly sensitive for detecting MMR gene mutations. In two studies that enrolled unselected patients with CRC, the summary estimate of sensitivity was only 45% (95% CI: 29, 63%), and the summary specificity was 99% (95% CI: 74, 100%).63, 83 However, both studies were limited by verification bias, as described below.

Figure 10. Sensitivity and specificity of the Amsterdam I criteria for detecting mismatch repair mutations in patients with CRC.

Figure

Figure 10. Sensitivity and specificity of the Amsterdam I criteria for detecting mismatch repair mutations in patients with CRC.

Overall, the distribution of sensitivity and specificity of the various studies did not follow a discernable pattern. The differences may be related to variation in genetic strategies (gene screening and deletion analysis), the definition of pathogenic mutations, sample selection processes (clearly described versus not), overall study methodologic quality (B versus C), or within-study heterogeneity of “baseline risk” for HNPCC (study sample included patients at very different risks for HNPCC). However, as the graphs in Figure 11 show, none of these factors alone appears to account for the variability. The variability is most likely attributable to the very different CRC populations assessed in these studies.

FIgure 11. Sensitivity and specificity of the Amsterdam I criteria for detecting mismatch repair mutations: study subgroups according to various factors.

Figure

FIgure 11. Sensitivity and specificity of the Amsterdam I criteria for detecting mismatch repair mutations: study subgroups according to various factors.

Amsterdam I Criteria Among Different Populations. As discussed in the Methods section, clinical predictors are expected to have substantially different sensitivity and specificity to detect mismatch repair gene mutations in CRC populations selected using different clinical criteria. We assessed the sensitivity and specificity of Amsterdam I criteria among CRC subpopulations that were defined consistently across studies (Appendices F-7, F-8 and F-9 *).

Figure 12 depicts the change in the sensitivity and specificity of the Amsterdam I criteria among CRC populations that had an increasing likelihood of having HNPCC: unselected CRC, CRC fulfilling the Bethesda criteria, and CRC fulfilling Amsterdam II criteria. As the likelihood for HNPCC increases, the studies shift toward the upper right (i.e., better sensitivity and worse specificity). Thus, there is evidence for spectrum effects in the diagnostic ability of the Amsterdam I set of criteria.

Figure 12. Sensitivity and specificity of the Amsterdam I criteria for detecting mismatch repair gene mutations among populations with an increasing likelihood of having HNPCC.

Figure

Figure 12. Sensitivity and specificity of the Amsterdam I criteria for detecting mismatch repair gene mutations among populations with an increasing likelihood of having HNPCC.

Amsterdam I Criteria Among Unselected Patients With CRC. Two studies (Aaltonen 1998,63 and Colombino 200583) assessed the performance of the Amsterdam I criteria among unselected CRC patients (Appendix F-7 *). The first63 assessed all new incident CRC cases from nine Finnish hospitals, while the second83 was retrospective and assessed all CRC patients who were registered in a tertiary hospital during the 3 years before the study (the authors commented that their hospital registered practically all CRC cases diagnosed in Sardinia).

Both studies were limited by verification bias. Aaltonen 199863 performed comprehensive genetic testing only among 63 patients who had tumors with MSI, and tested the remaining 446 for a common founder mutation in the MLH1 gene. Colombino 200583 performed detailed genetic testing only among patients with family history and assessed patients with apparently sporadic CRC only for mutations found in the familial cases. Thus, both studies clearly overestimated the sensitivity and the specificity of Amsterdam I criteria to detect mutations. Their summary sensitivity was 45% (95% CI: 29, 63%) and their summary specificity was 99% (95% CI: 74, 100%). The wide confidence intervals of the sensitivity values reflect that few patients fulfilled Amsterdam I criteria (n=4 in Aaltonen 1998,63 and n=21 in Colombino 200583).

Genetic Testing Only Among People Who Have MSI Positive (or IHC Negative) Tumors. In the three studies that were pertinent to this category29, 61, 96 the sensitivities ranged between 55% and 78% and the specificities between 40% and 100% (Appendix F-6 *).

Amsterdam II Criteria

Twelve studies described in 14 publications12, 29, 65, 67, 71, 77, 79, 80, 8487, 97, 98 provided data for determining the sensitivity and specificity of the Amsterdam II criteria for predicting the presence of mismatch repair mutations (Appendix F-10 *). Eight assessed sensitivity and specificity of the Amsterdam II criteria among all available patients,65, 67, 71, 77, 79, 80, 8487 three among patients who had tumors with MSI instability,29, 97, 98 and one among patients whose tumors were selected after MSI or IHC testing.12

Genetic Testing Irrespective of MSI or IHC Status. Eight studies provided data on genetic testing irrespective of MSI or IHC status.65, 67, 71, 77, 79, 80, 8487 The populations evaluated were heterogeneous (Appendix F-10 *). One study84 received grade C for overall quality rating while the rest were rated grade B.65, 67, 71, 77, 79, 80, 8587 All eight studies tested only for mutations in MLH1 and MSH2 genes. Seven studies had more than 40 patients in total in the pertinent 2×2 tables (all except for Rossi 200277). The sampling process was described in all but one study.87

Figure 13 provides an overview of the eight studies while Figure 14 shows the various studies according to the presence or absence of various characteristics that may affect the sensitivity and specificity of the Amsterdam II criteria. Overall, sensitivities varied between 22% and 80% and specificities varied between 61% and 100%. The plot suggests that Amsterdam II criteria are not a good screening test for HNPCC. The seven studies did not follow any particular pattern with respect to the factors addressed in Figure 14. The variability is mostly likely due to the very different CRC populations assessed in these studies.

Figure 13. Sensitivity and specificity of the Amsterdam II criteria for detecting mismatch repair gene mutations.

Figure

Figure 13. Sensitivity and specificity of the Amsterdam II criteria for detecting mismatch repair gene mutations.

Figure 14. Sensitivity and specificity of the Amsterdam II criteria for detecting mismatch repair mutations: study subgroups according to various factors.

Figure

Figure 14. Sensitivity and specificity of the Amsterdam II criteria for detecting mismatch repair mutations: study subgroups according to various factors.

Genetic Testing Only Among People Who Have MSI Positive (or IHC Negative) Tumors. Three studies performed genetic testing only among patients who fulfilled the Amsterdam II criteria and had MSI high tumors,29, 97, 98 and one12 among patients selected based on the results of MSI and IHC testing (Figure 15).

Figure 15. Sensitivity and specificity of the Amsterdam II criteria for detecting mismatch repair mutations among patients selected after MSI or IHC testing.

Figure

Figure 15. Sensitivity and specificity of the Amsterdam II criteria for detecting mismatch repair mutations among patients selected after MSI or IHC testing.

Pinol 200512 and Raedle 200129 assessed all available consecutive CRC patients. Terdiman 200198 selected patients who fulfilled the Bethesda guidelines and Pistorius 2000,97 patients in whom there was increased clinical suspicion for HNPCC (at least two 1st degree relatives with CRC or young age at diagnosis or multiple tumors in the same patient). Sensitivities ranged from 36% to 80% and specificities ranged from 64% to 99% (Appendix F-10 *).

Amsterdam II Criteria Among Unselected CRC Cases. As noted above, a question of central importance in this report is the accuracy of predictors in unselected patients with CRC. Only two studies assessed the accuracy of the Amsterdam II criteria in such patients (Pinol 2005,12 and Salovaara 200085) (Appendix F-11 *). The latter study85 performed comprehensive genetic testing in a subset of 66 patients who had tumors with MSI but tested the remaining 469 only for founder mutations in the MLH1 gene common in the Finnish population. The study by Pinol 200512 also performed comprehensive testing in only a subset of patients; however we included it by making an assumption that patients who were not selected for genetic testing based on MSI/IHC results would test negative for mutations.

Two studies that focused on unselected CRC probands (Pinol 2005,12 and Salovaara 200085) provided a summary estimate of 28% (95% CI: 15, 47) for sensitivity and 99% (95% CI: 97, 100) for specificity. However, both studies were limited by verification bias.

Amsterdam II Criteria Among Different Populations. Figure 16 depicts sensitivity and specificity of the Amsterdam II criteria for identifying MMR mutations, among populations at variable risk for HNPCC. The studies are described in Appendices F-11, F-12 and F-13 *. Studies focusing on populations at increasing risk appear to be associated with better sensitivity and worse specificity, similar to the findings described above for the Amsterdam I criteria.

Figure 16. Sensitivity and specificity of the Amsterdam II criteria for detecting mismatch repair mutations among populations with an increasing likelihood for HNPCC.

Figure

Figure 16. Sensitivity and specificity of the Amsterdam II criteria for detecting mismatch repair mutations among populations with an increasing likelihood for HNPCC.

Modified Amsterdam Criteria

The modified Amsterdam criteria are fulfilled when CRC has appeared in more than one generation, there are at least two CRC in 1st degree family, and at least one CRC diagnosed at age younger than 55 years; or when there are at least two 1st degree relatives affected by CRC plus another relative with an unusually early onset neoplasm or endometrial cancer. Two studies (described in three papers79, 80, 86) provided data on the modified Amsterdam criteria (Appendix F-14 *). Both studies received grade B for their overall quality. Neither assessed mutations in genes other than MLH1 or MSH2. The studies differed in their inclusion criteria and neither included unselected patients. Despite these differences, the estimates of sensitivity and specificity were similar. Overall, the summary sensitivity was 78% (95% CI: 61, 89%) and the summary specificity was 46% (95% CI: 37, 55%). It is likely that the sensitivity of the modified Amsterdam criteria in an unselected CRC population would be lower.

Bethesda Guidelines

Five studies described in six papers assessed the sensitivity and specificity of the Bethesda guidelines to detect MMR mutations: Syngal 2000,79, 80 Wolf 2005,86 de Abajo 2005,67 Raedle 2001,29 and Pinol 2005.12 Three studies performed genetic testing irrespective of tumor MSI status67, 79, 80, 86 while the other two performed genetic testing only among patients who were selected based on MSI testing29 or MSI and IHC testing.12 All four studies were rated grade B for the overall quality. All studies examined only the MLH1 and MSH2 genes. Only Syngal 2000 and Wolf 2005 performed sequencing to all available samples.

Only the study by Pinol 200512 assessed incident, newly diagnosed CRC cases. We included it by making an assumption that patients who were not selected for genetic testing based on MSI/IHC results would test negative for mutations. Figure 17 depicts the sensitivity and specificity of the five studies.

Figure 17. Sensitivity and specificity of the Bethesda guidelines for detecting mismatch repair gene mutations: all available studies.

Figure

Figure 17. Sensitivity and specificity of the Bethesda guidelines for detecting mismatch repair gene mutations: all available studies.

Based on the study by Pinol 2005,12 the Bethesda guidelines had a sensitivity of 73% (95% CI: 39, 94%) and a specificity of 82% (95% CI: 80, 84%) for detecting MMR mutations among unselected patients with CRC (Appendix F-15 *). However, these are probably overestimates because of verification bias.

Among a referral population (or otherwise selected CRC populations) the Bethesda guidelines appeared to have high sensitivity (over 90%) but low specificity (below 50%) in all three studies (including the study by Raedle 200129 that focused on patients whose tumors exhibited MSI-H) (Appendix F-15 *).

Revised Bethesda Guidelines

Only one study (Pinol 200512) provided data on the sensitivity and specificity of the revised Bethesda guidelines in a unselected population of patients with CRC (Appendix F-17 *). The sensitivity and specificity of the clinical criteria were calculated to be 91% (95% CI: 59, 100%) and 77% (95% CI: 75, 79%), respectively. Thus, the revised Bethesda guidelines were more sensitive and as specific as the original Bethesda guidelines (Appendices F-16 and F-17 *). However, these may be overestimates because of the possibility of verification bias as discussed above.

Young Age of Onset

Appendix F-18 summarizes five studies that assessed young age of onset as a clinical predictor to identify mismatch repair gene mutations (Salovaara 2000,85 Aaltonen 1998,63 Colombino 2005,83 Samowitz 2001,10 and Pinol 200512) (Appendix F-18 *). Young age of onset was defined as less 50 years in all studies, with the exception of Samowitz 2001,10 where the definition was less than 55 years. Colombino 200583 provided data only in patients with familial CRC (Appendix F-18 *). All studies suffered from verification bias because genetic testing was not thorough63, 85 or was not performed at all10, 12 among patients without MSI (MSI or IHC for Pinol 200512).

Among the three studies of unselected patients with CRC (Appendix F-19) that defined young age of onset as less than 50 years old, a summary estimate of sensitivity was 31% (95% CI: 18, 47%) and specificity was 95% (95% CI: 94, 96). Only the specificity estimate was heterogeneous (p<0.01, I2=81%) (Appendix F-18 and F-19 *). These estimates might be inflated because of verification bias. Sensitivity and specificity were similar in the studies by Colombino 200583 and Samowitz 200110 (Appendix F-19 *).

Familial Cancer History

Nine studies provided data eligible for this analysis10, 12, 63, 75, 81, 84, 85, 87, 99 (Appendix F-20 *). Six studies were rated with grade B in the overall quality scale,10, 12, 63, 81, 85, 87 and the remaining three studies were rated with grade C. Only one study had fewer than 40 patients in total in the 2×2 tables.75

There was variability in the populations that were examined and the definition of familial disease. A familial history of cancer was defined as presence of colorectal cancer or endometrial cancer in a first-degree family member of the proband in three prospective studies that assessed consecutive unselected CRC cases.12, 63, 85 Another study on incident CRC (Samowitz 200110) defined familial cancer as CRC in a first degree family. In these studies, all other CRC cases were used as a comparator.

In the remaining studies, familial cases were highly selected, based on sets of clinical criteria such as the Amsterdam I or II criteria, the Korean HNPCC criteria, the aggregation of HNPCC related tumors in a family, or combinations thereof. Similarly, in these studies the comparator (“sporadic CRC cases”) was defined variably (Appendix F-20 *), and was most often a group of “apparently sporadic” CRC that were selected based on their young age of onset or a diagnosis of multiple tumors.

Figure 18 depicts all studies according to whether they assessed unselected CRC cases.

Figure 18. Sensitivity and specificity of familial cancer history for detecting mismatch repair gene mutations: all available studies.

Figure

Figure 18. Sensitivity and specificity of familial cancer history for detecting mismatch repair gene mutations: all available studies.

Four studies reported estimates in unselected, consecutive patients with CRC.10, 12, 63, 83 Three used identical definitions (CRC or endometrial cancer in first degree family),12, 63, 83 and one defined familial cancer as CRC in a first degree family member10 (Appendix F-21 *).

A summary estimate of sensitivity based on the four studies that used similar definitions was 76% (95% CI: 50, 91%) with borderline between-study heterogeneity (p=0.13, I2=50%); the summary estimate of specificity was 87% (95% CI: 86, 89%) (without between-study heterogeneity: p=0.51, I2=0%). The estimates were similar when Samowitz 200110 was included in the calculations, despite the different definition of familial cases. The other studies assessed substantially different populations to allow for a meaningful synthesis (Appendix F-20 *). However, they uniformly suggested high sensitivities (above 75%) and variable, but generally low, specificities (ranging from 24% to 65%).

Multiple CRC or Endometrial Cancer in the Same Patient

Appendix F-22 * summarizes three studies that included unselected CRC and assessed the sensitivity and specificity of multiple, synchronous or metachronous CRC or endometrial cancer in the same proband to predict MMR mutations (Salovaara 2000,85 Aaltonen 1998,63 and Pinol 200512). Their characteristics and limitations have already been discussed. The respective summary estimates of sensitivity and specificity were 38% (95% CI: 25, 54%) with no evidence for between-study heterogeneity (p=0.98, I2=0%) and 97% (95% CI: 91, 99%) with significant between-study heterogeneity (p<0.01, I2=94%).

Familial History of Cancer, Age of Onset Less Than 50 Years, or Multiple Cancers in the Same Proband

Appendix F-23 * summarizes three studies that enrolled unselected CRC (Salovaara 2000,85 Aaltonen 1998,63 and Pinol 200512). The sensitivity and specificity of the presence of either of the following three simple clinical criteria was assessed: history of colorectal or endometrial cancer in first degree family, CRC onset at less than 50 years of age, or multiple tumors in the same proband versus all other cases. Their characteristics and limitations have already been discussed. Summary estimates of sensitivity and specificity were 88% (95% CI: 60, 97%) with little evidence for between-study heterogeneity (p=0.19, I2=39%) and 77% (95% CI: 74, 81%) with significant between-study heterogeneity (p=0.03, I2=73%), respectively.

MSI

The presence of MSI is a marker for replication errors and thus a predictor of MMR mutations. We assessed the sensitivity and specificity of MSI-H versus MSS. We also analyzed combined MSI-H and MSI-L versus MSS

Studies were considered eligible if they performed both MSI testing and MMR genetic testing in CRC patients and each test was applied irrespective of the results of the other test.

In these analyses we assumed that any genetic testing was the reference standard.

Overview of MSI Testing. Overall, MSI testing appeared to have modest to good discriminating ability to identify carriers of MLH1 and MSH2 mutations. The sensitivity of MSI-H versus MSS ranged between 56% and 100% and the specificity between 17% and 93%.

MSI-H Versus MSS. Sixteen studies described in 18 publications11, 6366, 6971, 75, 79, 80, 8486, 95, 100102 were eligible. One received grade A in the overall quality rating,11 nine received grade B,63, 65, 66, 70, 71, 79, 80, 85, 86, 101, 102 and six received grade C (Appendix F-24 *).

Eight studies performed sequencing on all available samples11, 69, 75, 80, 84, 86, 95, 101 and one of them (Barnetson 200611) also used specific methods to detect large genomic deletions or rearrangements. All studies tested for MLH1 and MSH2 gene mutations. One study also assessed the presence of MSH6 mutations (Barnetson 200611) and another tested for MSH6 and PMS2 mutations (Southey 2005102). It was not possible to examine the changes in the sensitivity and specificity of MSI with and without the non-MLH1/non-MSH2 mutations in these studies. Seven out of 16 had information in the 2×2 tables on at least 40 patients (Figure 19, Appendix F-24 *).

Figure 19. Sensitivity and specificity of MSI-H to identify mismatch repair gene mutations.

Figure

Figure 19. Sensitivity and specificity of MSI-H to identify mismatch repair gene mutations.

Two studies analyzed all available incident CRC cases (Aaltonen 199863 and Salovaara 200085), but had substantial verification bias since individuals who were negative in the MSI testing were assessed only for the presence of founder mutations in the MLH1 gene. Two other prospective studies analyzed incident CRC patients (Barnetson 2006,11 Southey 2005102) who were diagnosed at young age (<55 and <45 years, respectively).

The sample selection process was based on clearly stated selection criteria that were applied to all available patients in ten studies; in five of them, cases were selected among incident CRC patients (Aaltonen 199863, Salovaara 200085, Katballe 200265, 71, Southey 2005102 and Barnetson 200611). Twelve studies used microdissection to help assure that the analyzed tumor tissue contained a high proportion of malignant cells,11, 6466, 69, 71, 79, 80, 84, 86, 95, 100102 and eight used the marker sets proposed by the NCI.11, 64, 69, 79, 80, 84, 100102

Overall, the sensitivity of MSI ranged between 56% and 100% and the specificity between 17% and 93%. Two small studies found a specificity of only 25% (Durno 2005100 among only four patients without a mutation) and 17% (Curia 199966 among 18 patients without a mutation). Notably, the study by Curia 199966 found few pathogenic mismatch repair gene mutations among CRC cases with HNPCC-related cancers (3/30=10%) and thus its applicability is unclear.

We could not explain the heterogeneity among estimates based on the overall study quality, the comprehensiveness of the genetic testing, the presence of microdissection, the use of NCI-recommended marker sets, or whether the study used a transparent sample selection process (Figure 20). Larger studies (with at least 40 patients in the 2×2 tables) tended to cluster more around the upper left corner of the plot (Figure 19) than studies that used deletion analysis to test for MMR mutations (all four of which had more than 40 patients in the 2×2 tables) (Figure 20).

Figure 20. Sensitivity and specificity of MSI-H to identify mismatch repair gene mutations based upon study characteristic.

Figure

Figure 20. Sensitivity and specificity of MSI-H to identify mismatch repair gene mutations based upon study characteristic.

Among studies with more than 40 patients in the 2×2 tables, a summary estimate of sensitivity was 83% (95% CI: 65, 92%), with little evidence for between study heterogeneity (p=0.15, I2=37%). A summary estimate of specificity was 87% (95% CI: 80, 91%) with high between-study heterogeneity (p<0.01, I2=83%). The estimates were similar after the exclusion of the two Finnish studies (Salovaara 200085 and Aaltonen 199863) in which genetic testing was less rigorous for patients without MSI-H tumors. The latter studies also included the Finnish founder mutations in their analyses; these mutations are not found in populations of non-Finnish origin.

A summary ROC analysis is shown in Figure 21.

Figure 21. Summary ROC curve for the diagnostic ability of MSI-H and MSI-L to identify mismatch repair gene mutations.

Figure

Figure 21. Summary ROC curve for the diagnostic ability of MSI-H and MSI-L to identify mismatch repair gene mutations.

Combined MSI-H and MSI-L Versus MSS. There were seven studies in which allowed the comparison of combined MSI-H and MSI-L versus MSS. These were described in eight publications.11, 63, 65, 70, 71, 84, 85, 102

The summary estimate for sensitivity was 94% (95% CI: 86, 97) (no heterogeneity: p=0.85, I2=0%) and for specificity was 83% (95% CI: 77 to 88%) (with substantial heterogeneity: p<0.01, I2=80%). When analyses were limited to the four studies that had at least 40 patients in the 2×2 tables,11, 63, 85, 102 the corresponding results were 87% (95% CI: 82, 91%) and 95% (95% CI: 86, 98%); both estimates had substantial between-study heterogeneity.

What is the role of MSI-L in the detection of MMR mutations? We evaluated the sensitivity and specificity in the seven studies excluding the MSI-L tumors. 11, 63, 65, 70, 71, 84, 85, 102 The summary sensitivity became smaller (80% [95% CI: 63, 90%]) and the summary specificity increased (88% [95% CI: 83, 91%]). The summary sensitivity estimate was not heterogeneous (p=0.21, I2=28%) but the summary specificity estimate was heterogeneous (p<0.01, I2=71%).

MSI-H and Combined MSI-H and MSI-L Versus MSS Among Unselected CRC Probands. There were only two relevant studies, both from Finland (Salovaara 200085 and Aaltonen 199863 (Appendix F-25 *). MSI was assessed based upon the BAT26 marker set in the former, and on at least 30% of at least seven markers being unstable in the latter. Patients without MSI tumors were screened only for founder mutations in the MLH1 gene, which are common in Finland, whereas patients with MSI tumors received more comprehensive genetic testing (for both MLH1 and MSH2 genes). Thus, their estimates are likely to be biased. Summary estimates of sensitivity and specificity were 100% (95% CI: 88, 100%) and 90% (95% CI: 88, 92%), respectively.

MSI-H Versus MSS Among Patients Fulfilling the Revised Bethesda Guidelines. Only the study by Wolf 200586 provided relevant data (Table 17, Appendix F-26 *). Sensitivity was 100% (95% CI: 75, 100%) and specificity was 79% (95% CI: 63, 90%).

MSI-H Versus MSS Among Patients Fulfilling Amsterdam I Criteria. There were very few patients for these analyses and thus the precision of the estimates is poor (Table 17, Appendix F-27 *). Sensitivity is of MSI-H (and combined MSI-H and MSI-low) versus MSI-stable was universally excellent (100% [95% CI: 62, 100%]) in the three studies that provided the relevant information,70, 75, 95 although none contributed more than 10 patients who fulfilled the Amsterdam I criteria. Overall specificity was 69% (95% CI: 20, 95%).

IHC

Nine studies (described in 11 publications11, 65, 66, 6971, 79, 80, 98, 100, 102) provided information needed to calculate the sensitivity and specificity of IHC for predicting MMR mutations (Appendix F-28 *). One of these (Terdiman 200198) assessed the ability of IHC only in individuals with MSI-H positive tumors.

Studies That Assessed IHC Irrespective of MSI Testing. Eight studies assessed IHC in patients with available tumor tissue; one received grade A in overall quality,11 five grade B65, 66, 70, 71, 79, 80, 102 and two grade C.69, 100

Three studies performed sequencing in all available patients,11, 69, 79 and one of them (Barnetson 200611)also used specific methods to detect large genomic deletions or rearrangements.

All studies tested for MLH1 and MSH2 gene mutations. One study also tested for the presence of MSH6 mutations (Barnetson 200611) and another tested for MSH6 and PMS2 mutations (Southey 2005102). IHC testing was performed for all assessed MMR genes (MLH1, MSH2 and the additional MMR genes) in the latter two studies.11, 102 It was not possible to examine the changes in the sensitivity and specificity of IHC with and without the non-MLH1/non-MSH2 mutations in these studies.

Three studies had information in the 2×2 tables on at least 40 patients.11, 69, 102 Two papers described prospective studies that analyzed incident CRC patient (Barnetson 200611 and Southey 2005102) diagnosed at young age (<55 and <45 years, respectively).

Figure 22 shows the distribution of the studies according to the presence or absence of several study characteristics. Overall, the sensitivity ranged from 27% to 100% and the specificity from 43% to 100%. The six good or fair quality studies (grade A or B respectively) had a summary sensitivity of 74% (95% CI: 54, 87%) and specificity of 77% (95% CI: 61, 88%).

Figure 22. Sensitivity and specificity of IHC to identify mismatch repair gene mutations based on study characteristics.

Figure

Figure 22. Sensitivity and specificity of IHC to identify mismatch repair gene mutations based on study characteristics.

Figure 23 shows the summary ROC curve for the pertinent studies.

Figure 23. Summary ROC curve for the diagnostic ability of IHC to identify mismatch repair gene mutations.

Figure

Figure 23. Summary ROC curve for the diagnostic ability of IHC to identify mismatch repair gene mutations.

IHC among people fulfilling Amsterdam I criteria. Only the studies described by Katballe 2000 (another paper by Christensen 2002 reported the same patients65, 71) and the study by Dieumegard 200070 provided relevant information, and they are described in Appendix F-29 *. Both contributed nine or fewer patients and thus estimates of sensitivity and specificity lack precision (Table 17).

Studies That Assessed IHC After Suggestive MSI Testing. A single study by Terdiman 200198 assessed IHC among individuals who had tumors with MSI-H. The sensitivity of IHC was 94% (95% CI: 71, 100%) and the specificity was only 13% (95% CI: 4, 30%).

Expected Outcomes With Different Testing Strategies

Outline of the Problem

The prevalence of MMR mutations among people who are newly diagnosed with CRC is low.7, 10, 12, 63, 85 Because of practical, economic, and logistical considerations, genetic testing would ideally be performed only in patients with high probability of HNPCC. Such patients may be selected based on heightened suspicion from the clinical history, suggestive laboratory testing of tumor tissue (in particular MSI or IHC), or complex combinations of all of the above. The various strategies differ in the number of tests (MMR, MSI or IHC) that need to be performed and the accuracy of the diagnosis.

We performed analyses using decision trees to model the expected outcomes with different testing strategies from the payers'/third party perspective. The outcomes were the number of incident CRC with positive diagnosis for HNPCC, and the number of tests (MMR, MSI, or IHC) needed to detect them. We also assessed how many patients found to be mutation carriers with each strategy would be truly positive.

This analysis pertains to a cross-section in time. We used a hypothetical population of 100,000 incident cases of CRC. This number is in the order of incident cases expected annually in the US (approximately 150,000, given an annual incidence of 50/100,000 and a population of 300 million) and is a number convenient for calculations. The interested reader could multiply the reported numbers by 1.5 to extrapolate to the whole population in the United States.

We assessed nine different strategies, which were most commonly represented in the studies summarized in this review. The strategies used clinical criteria, MSI, IHC, or a combination of clinical criteria with MSI or IHC to select patients for MMR mutation testing.

Among the various clinical criteria that have been proposed, we opted to model the revised Bethesda guidelines and a set of three clinical criteria that are more easily ascertained (fulfillment of at least one of the following: age less than 50 year old at diagnosis, history of CRC or endometrial cancer in 1st degree family, or presence of multiple -synchronous or metachronous CRC or endometrial cancer in the proband). The revised Bethesda guidelines were selected because they appear to have the best sensitivity and specificity among unselected incident patients with CRC (data from Pinol 200512).

The set of three clinical criteria is a simple alternative that appears to have comparably high sensitivity and specificity, but is much simpler to assess. Interestingly, in the predictive models that have been developed in the literature, the aforementioned three simple clinical criteria have been the most influential predictors of HNPCC status (based on the coefficients from the logistic regression models from Barnetson 2006,11 Wijnen 1998103 and Balmana 200692).

The selected strategies are not exhaustive. However, these strategies have been discussed in the literature as possible options, and some of them have been evaluated in previous decision analyses.19 It would be impractical to provide analyses on a more extensive list of strategies, especially given the paucity of relevant data in the literature.

Brief Description of Strategies and Values Used in the Decision Tree

Nine strategies were modeled as described briefly below.

  • MMR-All: Perform MMR testing on all patients with CRC.
  • BethR-All: Perform MMR testing only among those fulfilling the revised Bethesda guidelines.
  • Clinical-All: Perform MMR testing only among those fulfilling at least one of the three simple clinical criteria (age <50y at diagnosis; 1st degree family history of CRC or endometrial cancer; or multiple, synchronous or metachronous, CRC or endometrial cancer in the same patient).
  • MSI-All: Perform MSI testing on all patients; followed by MMR testing only among those with suggestive MSI test.
  • IHC-All: Perform IHC testing on all patients; followed by MMR testing only among those with suggestive IHC test.
  • BethR-MSI: Perform MSI testing on patients fulfilling the revised Bethesda guidelines; perform MMR only among those with suggestive MSI test.
  • 3Clinical-MSI: Perform MSI testing on patients fulfilling at least one of the three simple clinical criteria (age <50y at diagnosis; 1st degree family history of CRC or endometrial cancer; or multiple, synchronous or metachronous, CRC or endometrial cancer in the same patient); perform MMR only among those with suggestive MSI test.
  • BethR-IHC: Perform IHC testing on patients fulfilling the revised Bethesda guidelines; perform MMR only among those with suggestive IHC test.
  • 3Clinical-IHC: Perform IHC testing on patients fulfilling at least one of the three simple clinical criteria (age <50y at diagnosis; 1st degree family history of CRC or endometrial cancer; or multiple, synchronous or metachronous, CRC or endometrial cancer in the same patient); perform MMR only among those with suggestive IHC test.

Figure 24 illustrates the general concepts in the architecture of these decision trees.

Figure 24. Decision tree model used to calculate the impact of different testing strategies.

Figure

Figure 24. Decision tree model used to calculate the impact of different testing strategies.

The strategies are conceptually organized in similar groups depending upon the pattern they follow in selecting who would be sequenced for MMR mutations.

  • Group 1: Test everyone (MMR-All)
  • Group 2: Screen with a set of clinical criteria (BethR-All, 3Clinical-All)
  • Group 3: Screen with a laboratory test (MSI-All, IHC-All)
  • Group 4: Screen using two serial tests: a set of clinical criteria first and then a laboratory test (BethR-MSI, BethR-IHC, 3Clinical-MSI, 3Clinical-IHC).

Table 18 shows the selected baseline values for the decision tree parameters, as well as the ranges of the sensitivity analyses. It should be noted that only some of the conditional probabilities that were presented in the previous sections are applicable to the following modeling approach; furthermore some conditional probabilities are imputed based on realistic assumptions because they were not extractable from the reviewed literature. All probabilities, their derivation and the rationale for any assumptions are described in more detail in the sections that follow.

Table 18. Probabilities and parameters used in the decision trees for different strategies to detect MMR mutations.

Table 18

Probabilities and parameters used in the decision trees for different strategies to detect MMR mutations.

Prevalence of Mutations Among Incident Unselected CRC Patients ( Prev ). The prevalence of mutations among incident unselected CRC is the most influential variable in the analyses (see Appendix G *). Based on the following considerations we present here two sets of results; one using a low prevalence estimate (0.9%) and one using a higher prevalence estimate (2.7%). The rationale for these values is given below. The prevalence value ranged from 0.6% to 5.1% in the sensitivity analyses.

Relevant studies. Among the 40 studies that were assessed for Key Question 2, four (3,332 analyzed patients in total) evaluated incident unselected CRC patients (Aaltonen 1998,63 Salovaara 2000,85 Pinol 2005,12 and Samowitz 200110). All were limited by verification bias, in that comprehensive genetic testing was performed only in patients with suggestive MSI results,10, 63, 85 or MSI and IHC results.12

We also considered the study by Hampel 2005,8 although it was not eligible (did not provide needed data) for the quantitative analyses described in the previous sections. Patients with suggestive MSI were sequenced for MLH1, MSH2, MSH6 and PMS2 mutations; analyses for large genomic deletions/rearrangements were also performed.

Low estimate for mutation prevalence (0.9%). The two Finnish studies63, 85 reported founder mutations (large genomic deletions) in the MLH1 gene of several CRC patients; however the Finnish founder mutations have not been identified in patients of non-Finnish origin.12 Thus, they are not applicable to the US population. Excluding the founder mutations from the analyses, the proportion of mutation carriers among unselected patients with CRC was estimated at 0.9% (95% CI: 0.6 to 1.3%).

An American Founder Mutation (AFM) in the MSH2 gene has been described in one kindred probably originating from German immigrants, but it is unlikely to have a large impact in the general population.90, 91 AFM is expected to account for very few HNPCC incident cases (between 51 and 290 CRC in the US for AMF) annually.90, 91 Thus, we did not model the AFM.

Higher estimate for mutation prevalence (2.7%). In the Hampel 2005 study twenty-three out of 1066 CRC were identified as carriers (prevalence 2.2% [95% CI: 1.4, 3.2%]), after suggestive MSI. Conservative accounting for carriers who did not have suggestive MSI (using the lower 95% CI boundary for the sensitivity of MSI among unselected CRC) raises the prevalence value to 2.7%.

Range of sensitivity analyses. We used a wide range of prevalence values in the sensitivity analyses from 0.6% to 5.1%. 0.6% is the lower 95% CI boundary for the low prevalence estimate. 5.1% is 20% more than the upper 95% CI of the higher prevalence estimate and was deemed to be high enough to encompass practically all realistic prevalence estimates that have been suggested by experts.7

Sensitivity and Specificity of Genetic Testing ( S_MMR, C_MMR ). We assumed:

  • That complete gene sequencing for MMR testing would be done.
  • Near perfect sensitivity and specificity for genetic testing.
  • The proportion of inconclusive genetic test results (that is mutations of unknown clinical significance) was not incorporated in the estimates of sensitivity and specificity of genetic testing; it is rather modeled separately (see below).

Because the available estimates for the diagnostic performance of all clinical predictors were based on MLH1 and MSH2 gene mutation testing in the primary studies, we “penalized” the maximum sensitivity of genetic testing to account for HNPCC cases with mutations other than MLH1 and MSH2. There were limited data on this proportion in the eligible studies. We therefore assumed that 5% of patients with HNPCC would have mutation in MMR genes other than MLH1 and MSH2 (thus we set sensitivity to 95% in the base case). In further analyses, we examined sensitivity as low as 72% based on data from Southey 2005,102 where 28% of the detected mutations among CRC aged less than 45 years of age (a selected population) were due to MSH6 and PMS2 mutations. Thus, the modeling accounts for the incremental value of testing for additional MMR genes other than MLH1 and MSH2.

Proportion of Inconclusive Results Among HNPCC With Non-Positive Test Results ( Pinconcl1 ), and Proportion of Inconclusive Results Among Patients Without HNPCC With Non-Positive Test Results ( Pinconcl2). These estimates are associated with considerable uncertainty. Syngal 199955 found that 2/18 patients with pathogenic mutations also had inconclusive mismatch mutations. Thus we assumed that 11% of HNPCC carriers also have inconclusive mutations.

The proportion of patients with CRC without HNPCC who carry mismatch mutations is more difficult to estimate. We based the estimate on the only study on incident, unselected CRC that provided the relevant information.12 This study was published recently and thus the authors presumably had access to the latest knowledge on which mutations were pathogenic; the corresponding estimate was 0.25%.

Both the aforementioned proportions were varied from 0% to their upper 95% CI values in sensitivity analyses.

Sensitivity and Specificity of MSI ( S_MSI, C_MSI ) and IHC ( S_IHC, C_IHC ) Among Unselected Incident CRC. These values were based on estimates from population-based studies that performed some form of genetic testing in all available patients. Because such IHC estimates were not available in unselected CRC, we assumed that two studies on patients with young age of onset for CRC (Barnetson 200611 and Southey 2005102) provided reliable approximations. This is based on the assumption that the diagnostic ability of laboratory predictors is relatively independent of clinical criteria that were used to select a population.

Similarly, for MSI we used a synthesis from the estimates obtained from four studies on incident CRC (Salovaara 200085, Aaltonen 199863, Barnetson 200611 and Southey 2005102). The values for MSI refer to combined MSI high and MSI low. The range of the sensitivity analyses includes the estimates for MSI high only.

Sensitivity and Specificity of the Revised Bethesda Guidelines Among Unselected CRC ( S_BethR, C_BethR ). These were obtained from the study by Pinol 2005.12

Sensitivity and Specificity of the Three Clinical Criteria (Young Age at Diagnosis, CRC in Family or Multiple Tumors in Proband) Among Unselected Patients With CRC ( S_3Clinical, C_3Clinical ). These were obtained from a meta-analysis of pertinent studies (Aaltonen 1998,63 Salovaara 2000,85 and Pinol 200512). The estimates for the two Finnish studies63, 85 were calculated accounting for the presence of founder mutations.

Sensitivity and Specificity of MSI ( S_MSI_3Clinical, C_MSI_3Clinical ) and IHC ( S_IHC_3Clinical, C_IHC_3Clinical ) Among People Fulfilling the Three Clinical Criteria (Young Age at Diagnosis, CRC in Family or Multiple Tumors in Proband). The values for MSI were based on the studies by Aaltonen 1998 and Salovaara 2000.63, 85 Values include both MSI-H and MSI-low, as discussed above.

Similar values for IHC were not available. We assumed that these could be substituted by the corresponding estimates for the unselected CRC (which were based on data from Barnetson 2006 and Southey 2005, two studies11, 102 that assessed people at young age of CRC diagnosis).

Sensitivity and Specificity of MSI ( S_MSI_BethR, C_MSI_BethR ) and IHC ( S_IHC_BethR, C_IHC_BethR ) Among People Fulfilling the Revised Bethesda Guidelines. A small study on 81 patients by Wolf 200586 provided the relevant information for the accuracy of MSI. However, the study reported perfect sensitivity (100%) for MSI, which was considered improbable. We therefore used the same estimates as for patients fulfilling at least one of the three simple clinical criteria (<50 years of age at diagnosis, CRC or endometrial cancer in family or multiple colorectal or endometrial cancers in proband).

There were no available data for the diagnostic accuracy of IHC among people fulfilling the revised Bethesda guidelines. Therefore, we assumed that Southey 2005102 (a study that assessed incident CRC aged less that 45 at diagnosis) provided suitable corresponding estimates.

Results With the Different Strategies

Testing for MSI (or even IHC) among CRC patients with high clinical suspicion for HNPCC may be a reasonable compromise between the number of laboratory tests that would be required and an accurate diagnosis in the majority of people with HNPCC (i.e., strategies in group 4).

More specifically, group 4 strategies selected relatively fewer patients for genetic testing (6% or less) and missed at most 27% of patients with HNPCC. In contrast, for strategies in groups 1 to 3, more than 13% and up to 100% of newly diagnosed CRC patients would be genetically tested, and 5% to 16% of patients with HNPCC would be missed. These descriptives are true for both the low (0.9%) and the higher estimate for the prevalence of mutation carriers among incident unselected CRC (Table 19).

Table 19. Expected number of MMR, MSI or IHC tests and expected MMR testing results with the nine strategies, assuming a population of 100,000 incident cases of CRC.

Table 19

Expected number of MMR, MSI or IHC tests and expected MMR testing results with the nine strategies, assuming a population of 100,000 incident cases of CRC.

Because the revised Bethesda guidelines are more difficult to ascertain compared to the fulfillment of at least one of the three simple clinical criteria, it might be easier to use the combination of the three clinical criteria. Although it seemed that MSI was preferable over IHC as a screening tool among people with high clinical suspicion, estimates were based on several assumptions due to missing information, compromising the validity of this observation.

We decided not to highlight the relative ranking of the strategies within each group because of the many assumptions in the used probabilities. We therefore focus more on the differences across the four groups of strategies. The relative ranking of the 4 strategy groups with respect to each outcome was robust in all one-way sensitivity analyses.

Baseline Analyses.

Number of patients identified as positive - low estimate for prevalence of mutation carriers. In the hypothetical population of 100,000 CRC, 900 individuals had MMR mutations. The number of positive diagnoses with each strategy ranged from 1,351 (MMR-All) to 659 (Beth-IHC) (Table 19). The number of HNPCC patients who were truly identified was highest in the MMR-All strategy (n=855 out of 900) and lowest in group 4 strategies (ranging between 654 and 739, depending on the strategy). Reciprocally, the number of HNPCC patients who were missed with each strategy increased for strategies that performed fewer MMR tests. While only 45 patients would be missed if all CRC were to be genetically tested, up to one out of five (between 161 and 246 out of 900) would be missed if MSI or IHC were performed only on people with a high clinical suspicion for HNPCC.

As shown in Table 19, the number of inconclusive MMR tests increases with the number of total MMR tests. In group 4 strategies it ranged between 6 and 12, and in the MMR-All strategy it was approximately 250.

Number of patients identified as positive - higher estimate for prevalence of mutation carriers. In the hypothetical population of 100,000 CRC, 2,750 individuals had MMR mutations. The number of positive diagnoses with each strategy ranged from 3,098 (MMR-All) to 2,002 (Beth-IHC) (Table 19). The number of HNPCC patients who were truly identified was highest in the MMR-All strategy (n=2,612 out of 2,750) and lowest in group 4 strategies (ranging between 1,997 and 2,258, depending on the strategy). The number of HNPCC patients who were missed with each strategy increased for strategies that performed fewer MMR tests. While only 138 patients would be missed if all CRC were to be genetically tested, up to one out of four (between 492 and 753 out of 2,750) would be missed if MSI or IHC were performed only on people with high clinical suspicion for HNPCC.

As shown in Table 19, the number of inconclusive MMR tests increases with the number of total MMR tests. In group 4 strategies it ranged between 14 and 20, and in the MMR-All strategy it was a bit over 260.

Number of tests performed for assessments - low estimate for prevalence of mutation carriers. The expected percentage of the CRC cohort receiving MMR tests ranged from 100% (MMR-All) to 1.83% (BethR-IHC) (Table 19).

Generally, strategies that used clinical criteria to select patients to be tested with MSI or IHC (BethR-MSI, BethR-IHC, 3Clinical-MSI, 3Clinical-IHC) required fewer MMR genetic tests. Fewer than 4% of newly diagnosed patients with CRC would be expected to need MMR genetic testing based on this approach. Furthermore, for the strategies in group 4, approximately one out of four patients would receive MSI or IHC testing, based upon suggestive clinical criteria.

The use of clinical criteria alone to screen cases (BethR-All, 3 Clinical-All) would require MMR genetic testing in almost a quarter of the incident cases of CRC. Using laboratory tests as a first screening would result in approximately one out of seven patients requiring genetic testing (Table 19). By definition, all patients in the strategies that first screen with MSI or IHC would receive MSI or IHC testing (MSI-All or IHC-All, respectively), but fewer than one out of four patients would need to be screened with MSI or IHC based upon suggestive clinical criteria (Table 19).

Number of tests performed for assessments - higher estimate for prevalence of mutation carriers. The expected percentage of the CRC cohort receiving MMR tests ranged from 100% (MMR-All) to 3.22% (BethR-IHC) (Table 19). The overall percentages of people receiving genetic testing, MSI testing or IHC testing with the various strategies was similar to what was calculated for the low prevalence estimates (Table 19).

One may consider the nine strategies as nine (composite) diagnostic tests, and thus calculate their overall (strategy-level) sensitivity and specificity (Table 20). In these calculations, inconclusive tests were assumed to be false negative (for the calculation of overall sensitivity for each strategy) or false positive (for the calculation of overall specificity for each strategy). The overall specificity was high in all strategies. Group 4 strategies had the lowest overall sensitivity (ranging between 73% and 82%). The overall sensitivity and specificity values were similar for both the low and the higher estimates for the prevalence of MMR mutation carriers.

Table 20. Overall sensitivity and specificity for each of the nine strategies.

Table 20

Overall sensitivity and specificity for each of the nine strategies.

Benefits and Harms

Key Question 1: Does Risk Assessment and HNPCC Mutation Testing in Patients With Newly Diagnosed CRC Lead to Improved Outcomes for the Patient or Family Members, or is it Useful in Medical, Personal, or Public Health Decision Making? (Overarching Question)

Image er-hnpccfu4.jpg

Key Question 1 is a general question that includes all other Key Questions. An ideal study addressing Key Question 1 would enroll patients with CRC patients and/or their family members and randomize them to risk assessment and mutation testing or a control intervention, and follow them up prospectively. The study would compare a spectrum of health outcomes among those who received screening for HNPCC to those who did not, while considering subsequent treatments or interventions in each group.

Summary of Findings. No study directly addressed Key Question 1 based on the ideal framework described above.

Gaps in the Literature. While our literature review did not identify a comprehensive study that directly answered Key Question 1, it is unlikely that such an “ideal” study is feasible. Such a study would likely be prohibitively complex, require multiple years of follow-up and randomization of patients or family members to care that is no longer considered to reflect contemporary practices. However, studies addressing specific components of this general question (as described in the remaining Key Questions) provide some of necessary pieces to better understand this overarching question.

Key Question 3: What Are the Harms Associated With Screening High-Risk Individuals for HNPCC?

Image er-hnpccfu5.jpg

Studies were considered eligible for Key Question 3 if they reported harms of a risk assessment process (e.g., Amsterdam, Bethesda and/or MSI, IHC) used to identify CRC patients at increased risk for HNPCC.

Summary of Findings. No study described any harms of the risk assessment process in CRC patients at increased risk for HNPCC.

Gaps in the Literature. The goal of the risk assessment process is to identify individuals suspected of having HNPCC who might benefit from subsequent genetic testing. It is possible that such a process could lead to harm, such as stigmatization or worry, but no study directly assessed such associations.

The risk assessment tools also have the potential to misclassify patients. The test characteristics (i.e., sensitivity, specificity and predictive values) of these methods are summarized in the “Clinical Validity” section. We did not consider misclassified patients (i.e., false positive or negatives) as having been harmed. However, it is possible that subsequent genetic testing and/or management options could harm patients who had been misclassified. The harms associated with genetic testing and/or management options are described later in this report.

Future research is needed to identify the best strategy of the risk assessment process that increases the accuracy, feasibility and applicability of diagnostic methodologies to determine HNPCC so the potential harms can be minimized.

Key Question 5: What Are the Harms Associated With Screening for High-Risk Individuals?

Image er-hnpccfu6.jpg

Studies were considered eligible for Key Question 5 if they reported the harms associated with testing CRC patients for MMR mutations. Common harms that are thought to be associated with genetic testing are labeling, discrimination in health coverage, and emotional distress.

Three quantitative, comparative studies of quality A and B, and one qualitative study of B quality reported harms associated with MMR mutation testing in CRC patients104106 (Table 21).

Table 21. Key Question 5. What are the harms associated with genetic testing for HNPCC mutations?

Table 21

Key Question 5. What are the harms associated with genetic testing for HNPCC mutations?

Summary of Findings. One 1-year prospective study (Grade A) compared the psychological impact of MMR mutation testing between mutation carriers and non-carriers.104 Results for non-carriers are summarized below and also in Table 21. Subjects in this study were CRC probands or relatives from HNPCC families with a prior diagnosis of any cancer (excluding non-melanoma skin cancer). Anxiety, depression, and quality of life measures did not change over time, and there were no differences in these measures between mutation carriers and non-carriers. Distress levels were significantly decreased 2 weeks and 6 months after revealing the genetic testing results, but were not significantly different from the baseline at 1-year follow-up. There was no difference in the distress levels between mutation carriers and non-carriers.

Another 1-month prospective study (Grade B) found that three of the 27 probands (11%) had minor depression at 1 month after revealing the genetic testing results, but the prevalence of minor depression was not significantly different compared to the prevalence at baseline or between mutation carriers and non-carriers.105 Of the six probands who received a positive result, two (33%) felt severe guilt regarding their children.

One prospective study (Grade B) reported changes in the psychological outcomes of CRC patients from self-completed questionnaires pre- and 4–6 weeks post-genetic counseling.107 There was no genetic testing performed in this study. There was a trend toward greater anticipated ability to cope with a positive gene test after counseling, as reflected by a decreased in anxiety and cancer-specific distress.

The qualitative study of 111 newly diagnosed CRC patients reported a high acceptance and understanding about information on HNPCC.106 Nineteen percent of participants rated their current level of worry caused by the genetics information at or above the midpoint of 4 on a 1 (not at all) to 7 (all the time) scale.

Gaps in the Literature. Most research on psychosocial aspects of genetic counseling and testing for cancer risk has focused on hereditary breast and ovarian cancer. Genetic testing for hereditary colorectal cancer has become available only in the past decade. As a result, only few studies with small number of CRC patients have evaluated the acceptability and psychosocial sequelae of genetic counseling and testing. Some of the findings from studies of counseling and testing in other forms of cancer may be applicable to the CRC setting.

Key Question 6a: What Are the Management Options for CRC Patients Who Are HNPCC Positive? b: Does the Identification of HNPCC Mutations Lead to Improved Patient Outcomes in Terms of Early Detection, Mortality/Morbidity, or Management Decisions (e.g., Counseling, Surveillance, Treatment, Other Decision Making) by Patients and Providers?

Image er-hnpccfu7.jpg

Key Question 6 examines the outcomes of using MMR mutation status to make management decisions compared to conventional management without MMR mutation status. There are three aspects to Key Question 6: 1) Are management options for patients with CRC with a MMR mutation different from those without a MMR mutation? 2) Does the knowledge of MMR mutation status change management decisions by patients and providers? 3) Does changing management options for MMR positive patients with CRC improve outcomes (e.g., prognosis and survival) compared to standard approaches for patients with CRC?

We encountered a variety of surgical and medical management options in patients with CRC who were MMR positive. These included various forms of colonic resection108110 and chemotherapy for patients with stage III tumors.111 However, these studies did not directly address the components of Key Question 6 described above.

Because of the limited data, we broadened the scope of this question to include studies evaluating all forms of cancer related to HNPCC, since patients with CRC from HNPCC families are potentially at increased risk for all of these cancers. We also included studies using clinical criteria and laboratory testing (such as MSI and IHC) as a surrogate for identification of HNPCC mutations.

Despite having broadened the inclusion criteria, we did not identify any prospective, comparative study (direct evidence) addressing any aspect of Key Question 6. However, two retrospective cohort studies indirectly addressed the prognosis of CRC and the survival outcome of patients with endometrial cancer from HNPCC families, in relation to MMR mutation testing.112, 113 Both were of C quality, due to potential selection bias and unclear effects of treatments or interventions on the outcomes.

In addition, five retrospective cohort studies indirectly addressed mortality or morbidity in relation to risk assessments (clinical criteria and laboratory testing) for HNPCC mutations108, 109, 111, 114, 115 (Table 22). Of these, three studies of B quality described clinical outcomes in patients with CRC who were screened for HNPCC using clinical criteria.108, 109, 115 Another two described clinical outcomes in patients with CRC who were screened for HNPCC with laboratory testing.111, 114 Both were of C quality and limited by the potential selection bias, and/or unclear effects of treatments or interventions on the outcomes.

Table 22. Key Question 6b. Does the identification of HNPCC mutations lead to improved patient outcomes in terms of early detection, mortality/morbidity or management decisions (e.g., counseling, surveillance, treatment, other decision making) by patients and providers?

Table 22

Key Question 6b. Does the identification of HNPCC mutations lead to improved patient outcomes in terms of early detection, mortality/morbidity or management decisions (e.g., counseling, surveillance, treatment, other decision making) by patients and providers? (more...)

Summary of Findings. Indirect evidence from one study suggested that identification of HNPCC mutations was associated with better prognosis of CRC. However, there was no data on whether management options for CRC differed based on MMR mutation status.

Indirect evidence from one study showed no difference in survival of patients with endometrial cancer, comparing those who were mutation positive to those who were mutation negative.

In five studies with indirect evidence, there was no evidence in favor or against differences in survival, when comparing CRC patients who fulfilled different clinical criteria for HNPCC or screened positive for HNPCC by laboratory testing with those who did not (Overview Table 1; Table 22).

Gaps in the Literature. Some studies described the management of patients with HNPCC but surprisingly, there were no data directly exploring how MMR mutation status influenced the choice among these management options or their outcomes. As a result, the available data do not directly answer whether knowledge of MMR mutation status changes management decisions by patients and providers or whether changes in management for MMR positive patients improves their outcomes compared to standard CRC management. Issues related to MMR mutation testing in influencing decisions to undergo cancer screening procedures is discussed below with Key Question 10.

Key Question 7: What Are the Harms Associated With Subsequent Management Options After Identification of HNPCC Mutations in CRC Patients?

Image er-hnpccfu8.jpg

Due to limited data on harms associated with subsequent management options or interventions, we broadened the scope of this question to include studies evaluating all forms of cancer related to HNPCC, since patients with CRC from HNPCC families are potentially at risk for all forms of HNPCC-related cancers. We also included studies that reported any outcome relating to subsequent management options or interventions in these patients.

Two retrospective studies of B and one of C quality reported outcomes related to subsequent actions or interventions in patients with CRC and gastric cancers, respectively (Table 23).110, 116 The validity of the study of C quality (gastric cancers) is questionable because of unclear participation rates and an unclear description of how survival/death was ascertained.

Table 23. Key Question 7. What are the harms associated with subsequent management options after identification of HNPCC mutations in CRC patients?

Table 23

Key Question 7. What are the harms associated with subsequent management options after identification of HNPCC mutations in CRC patients?

Summary of Findings. No study described harms associated with subsequent management options after identification of HNPCC mutations in patients with CRC or other forms of HNPCC-related cancers.

One study (involving two centers) described the types of colorectal surgery performed on CRC patients who were part of an Amsterdam criteria-positive family, and compared rates of metachronous cancers that followed each type of index operation.110 The overall rate of second surgeries for metachronous cancer were 23% in patients who underwent right colectomy, 17% in patients who underwent left/sigmoid colectomy or proctosigmoidectomy, 0% in patients who underwent total/subtotal colectomy, and 44% in patients who underwent segmental colectomy. The two centers had significantly different second resection rates for metachronous cancer.

One study described the survival rate of 45 patients with gastric cancer from HNPCC families with MMR mutations.116 Many of these patients had already had treatments for other HNPCC-related cancers, including CRC. The 5-year survival was higher in patients in whom radical surgery was performed (48%) than in patients in whom radical palliative surgery or explorative laparotomy alone was performed (15%).

Gaps in the Literature. There are no data on harms associated with surgical procedures or chemotherapy in CRC patients with HNPCC, although it is likely that such patients encounter the same spectrum of harms associated with chemotherapy and surgery as patients with CRC (or other forms of HNPCC-related cancer) without HNPCC. One cannot assume “no harm” when a study did not report data on harms.

There were also no data on whether cancer patients with MMR mutations derive a greater benefit from specific management options compared with patients without MMR mutations. However, several studies have suggested a relatively favorable prognosis of colorectal cancer in patients with HNPCC or those who have clinical features associated with HNPCC such as tumors that are MSI-H.117120 These data have also suggested that patients with colorectal cancer that demonstrate MSI may not derive a benefit from adjuvant chemotherapy with 5-fluorouracil (5-FU), possibly because of a higher cure rate with surgery alone and/or intrinsic resistance to 5-FU.120

However, a 2004 consensus statement issued by the American Society of Clinical Oncology121 found insufficient evidence to recommend modification of overall treatment recommendations in patients with stage II colorectal cancer with a known mismatch repair mutation or tumors that demonstrate MSI, noting the retrospective and inconclusive nature of the existing evidence. The report cites an ongoing European trial (FOLFIRI), which is comparing irinotecan and 5-FU and is stratifying patients based on tumor MSI status, and a United States Intergroup trial that is also incorporating MSI status as a prognostic factor. The final results of these trials have not been published.

Our review of the literature did not provide additional strong evidence in favor or against the use of adjuvant chemotherapy in patients with HNPCC of any stage. However, our literature search strategy excluded studies with patients who had tumors that demonstrated MSI unless the study performed additional testing for HNPCC (either by clinical criteria, IHC, or genetic testing). Thus, the literature search was not comprehensive for establishing the relationship between tumor MSI status and the outcome of adjuvant chemotherapy. The ongoing trials described above will help clarify this issue (as well as the role of other biologic markers proposed for guiding adjuvant chemotherapy in patients with colorectal cancer).

Key Questions 8 to 11 summarize what is known about the processes and benefits of bringing family members into the testing process. Knowing that a CRC patient is MMR mutation positive has implications for family members. As a general rule, family members receive counseling and, subsequently, genetic testing for MMR mutations ideally based upon the results of MMR testing in the proband.

Key Question 8b: What is the Accuracy of HNPCC Testing in Family Members in Predicting the Risk of CRC?

Key Question 8c: Do Other Factors, Such as Race/Ethnicity, Age, Gender, or Co-Morbidities Affect the Accuracy of the Testing?

Image er-hnpccfu9.jpg

The analytic framework proposed above greatly restricted the pool of potentially eligible studies since it required that CRC probands were proven to have a MMR mutation. As a result, we broadened the scope of our search to include studies using clinical criteria, laboratory testing or mutation testing in predicting the risk of CRC or extracolonic cancers in family members of CRC probands with HNPCC (based on clinical or genetic criteria) and/or kindreds with HNPCC (based on clinical or genetic criteria). We considered the findings in such studies to be applicable to the Key Question. Any measure of relative risk (e.g., the standardized incidence ratio) was included as indirect evidence for the accuracy of the testing.

We summarize the findings for family members of CRC probands with HNPCC (in Table 24) and for kindreds with HNPCC (in Table 25) separately, because the risk of CRC and/or extracolonic cancers is different between the two groups.

Table 24. Key Question 8b. What is the accuracy of HNPCC testing in family members in predicting the risk of CRC? Key Question 8c. Do other factors, such as race/ethnicity, age, gender, or co-morbidities affect the accuracy of the testing?

Table 24

Key Question 8b. What is the accuracy of HNPCC testing in family members in predicting the risk of CRC? Key Question 8c. Do other factors, such as race/ethnicity, age, gender, or co-morbidities affect the accuracy of the testing?

Table 25. Key Questions 8b and 8c in kindreds with HNPCC based on clinical and/or genetic criteria.

Table 25

Key Questions 8b and 8c in kindreds with HNPCC based on clinical and/or genetic criteria.

Six studies reported the accuracy of HNPCC testing in family members of CRC probands (based on clinical or genetic criteria) in predicting the risk of CRC20, 122126 (Table 24). Of these, only two (both of quality B) reported the accuracy of MMR mutation testing in family members of CRC probands with a positive MMR mutation.20, 126The other four reported the accuracy of HNPCC testing in family members of CRC probands with HNPCC based on clinical criteria122125 Half studies were of B quality and half were of C quality. These studies were limited by verification bias, selection bias, unclear follow-up duration, and/or bias from post-hoc analyses.

Six studies indirectly reported the accuracy of the HNPCC testing in kindreds with HNPCC (based on clinical or genetic criteria)127132 (Table 25). Of these, two were of B quality and four were of C quality. These studies were limited by selection and verification bias. Most studies did not adjust for familial clustering of cancers.

Summary of Findings. Only two studies of quality B reported the risk of CRC in family members of probands with positive MMR mutations (the proposed framework). The lifetime risk of CRC was 68.7% for men and 52.2% for women with MMR mutations in one study,20 and it was 74% and 30% respectively in the other study.126 Men had a higher lifetime risk of CRC in both studies.

In a study that reported the risk of CRC in family members of probands with HNPCC based on clinical criteria, the cumulative risk of CRC by age 75 years old was 57% and 41% in families that fulfilled the Amsterdam I and II criteria, respectively.122 The cumulative risk of CRC by age 75 years old was 42% and 23% for men and women, respectively, from families that fulfilled Bethesda criteria. In another study123 family members of CRC probands who were younger than 50 years old at cancer diagnosis had a higher risk of CRC, compared to family members of CRC probands who were 50 years old and older. The risk of CRC was increased three-fold, comparing first-degree relatives of CRC probands who developed a second primary in the HNPCC spectrum with the single primary group in a third report124 (Overview Table 2; Table 24).

In a study that reported the risk of CRC in kindreds with HNPCC based on genetic criteria, the cumulative risk of CRC by age 70 yr was 82% in MLH1/MSH2 mutation carriers.132 The cumulative incidence of CRC was 100% in men and 54% in women. There was overall a higher risk of CRC in men but the sex difference was not consistent among HNPCC kindreds with different MMR mutations (Overview Table 2; Table 25).

Gaps in the Literature. We found only limited evidence on the accuracy of MMR mutation testing in predicting the risk of HNPCC-related cancers in family members of CRC probands. Only two studies based such predictions on proven MMR mutations in the proband.20, 126 The study by Hampel20 found that cancers in family members developed at a later age compared with previous reports that had used clinical criteria to establish the diagnosis of HNPCC, suggesting that additional studies are required to fully understand penetrance in family members of CRC probands who carry MMR mutations.

No study directly examined factors that influenced the accuracy of the HNPCC testing, although some found significant differences in the risks of CRC among HNPCC kindreds with different MMR mutations. More studies are needed to fully understand the relationships among specific MMR mutations and cancer risk in probands with HNPCC-related cancers and their family members.

Key Question 9: What Are the Harms Associated With Informing/ Counseling Family Members or With Subsequent Testing for HNPCC Mutations?

Key Question 8a: What is the Efficacy of Pre-Test Genetic Counseling for Informing Family Members of Potential Risks and Benefits of Testing?

Image er-hnpccfu10.jpg

Studies were considered eligible for Key Question 8a if they summarized the efficacy of pre-test genetic counseling immediately after counseling and before performing genetic testing. Studies addressing Key Question 9 were considered together with those addressing Key Question 8a because there was substantial overlap in the studies addressing these questions. The studies addressed the long-term efficacy of pre-test genetic counseling or harms associated with screening high-risk individuals (such as by using clinical criteria, MSI, or IHC), genetic testing, and informing/counseling family members or with subsequent testing for HNPCC mutations.

Four studies addressed the efficacy of pre-test genetic counseling immediately after counseling and before performing MMR genetic testing107, 133138 (Table 26). Of these, three were of B quality and one was of C quality.

Table 26. Key Question 8a. What is the efficacy of pre-test genetic counseling for informing family members of potential risks and benefits of testing?

Table 26

Key Question 8a. What is the efficacy of pre-test genetic counseling for informing family members of potential risks and benefits of testing?

Eight comparative and four qualitative studies addressed the harms associated with genetic testing for HNPCC mutation or with informing/counseling family members or with subsequent testing for HNPCC mutations4, 104106, 133144 (Table 27). Of these, one study was of A quality, six of B quality, and five of C quality.

Table 27. Key Question 9. What are the harms associated with informing/counseling family members or with subsequent testing for HNPCC mutations?

Table 27

Key Question 9. What are the harms associated with informing/counseling family members or with subsequent testing for HNPCC mutations?

Table 27a. Key Question 9. What are the harms associated with informing/counseling family members or with subsequent testing for HNPCC mutations?

Table 27a

Key Question 9. What are the harms associated with informing/counseling family members or with subsequent testing for HNPCC mutations?

Summary of Findings. Overall pre-test genetic counseling had good efficacy in improving knowledge about HNPCC, and resulted in high likelihood of proceeding with genetic testing, satisfaction in decision to undergo genetic testing, and decreasing in depression and distress levels among family members of HNPCC probands or among asymptomatic individuals from HNPCC families. Family members of HNPCC who received positive MMR mutation test results had higher psychological distress levels and anxiety compared to those who received negative test results, but this difference generally disappeared with time. However, all of these psychological measures were within the normal ranges for the general populations. Available evidence did not show differences in quality of life, comparing mutation carriers to non-carriers or the general population (Overview Table 3; Table 2627).

Gaps in the Literature. The results described above apply only to individuals who agreed to participate in the genetic studies and opted for genetic testing. Many individuals from HNPCC families chose not to participate in the study or pre-test genetic counseling. The reasons were often not described; however, they included fear of knowing the results, fear of insurance discrimination, survivor's guilt and disinterest.

The literature did not provide what key elements should be included in genetic counseling, and the effectiveness of different counseling strategies. Only a small RCT compared the efficacy of shortened to extended pre-test counseling and found no differences in psychological and decision making outcomes or knowledge about HNPCC between groups immediately after counseling and before revealing the results of genetic testing.133 At 3-month follow-up, mutation carriers in the shortened group experienced the least distress along with a high level of decision satisfaction, in contrast to carriers in the extended group who reported the most distress and least decision satisfaction. However, there were no details regarding baseline familiarity, education level and other potentially relevant characteristics across study groups to assure randomization was successful.

Some studies examined the factors that influenced decisions regarding genetic testing or the reasons for pursuing or for not pursuing genetic testing.134138, 145151 Although these studies (most of which were descriptive) were not specifically relevant to any of the Key Questions, they were clinically relevant and thus we attempted to summarize them in ancillary tables (Extra Table 1). People who were employed, who had higher education levels, who had CRC screening, and who had familial cancers were more likely to accept genetic testing.

Common reasons for not pursuing genetic testing include worry about losing health insurance, and concerns about coping with the emotional reactions if the test were positive. Women appeared to need more professional psychological support for coping with the results of genetic testing. These findings are important considerations for genetic counseling programs and support systems to minimize the harms associated with genetic testing for HNPCC mutations.

Findings from two surveys of genetic centers or insurance providers suggested the need for more empirical research on which components of pretest counseling are effective in promoting good psychological outcomes and the need to prevent insurance discriminations to the persons with genetic risk of CRC152, 153 (Extra Table 2).

Key Question 10a: What Are the Management Options for Family Members of CRC Patients Who Have a Positive HNPCC Test? b1: Does the Identification of HNPCC Mutations Lead to Improved Outcomes in Terms of Decision Making by Patients, Family Members and Providers, or Public Health Policy? b2: Does the Identification of HNPCC Mutations Lead to Improved Outcomes in Terms of Early Detection and Mortality/Morbidity of Patients, and Family Members?

Image er-hnpccfu11.jpg

Below we summarize management options for family members of CRC patients or asymptomatic individuals from HNPCC families (based on clinical or genetic definitions) that were reported in the literature. This is not necessarily a complete list of management options.

Many studies described management options for family members of CRC patients or asymptomatic individuals from HNPCC families. We encountered the following in the studies included in this report.

  • Annual colonoscopy beginning at age 20 to 25 or 5 years earlier than the youngest affected family members.140, 154159
  • Colonoscopy or double contrast barium enema and sigmoidoscopy at two- or three-year intervals.42, 155, 160162
  • Endometrial screening from age 30 to 35 generally with annual or biennial transvaginal ultrasound (TVU).44, 45, 154, 157, 158
  • Endometrial sampling in premenopausal women and annual TVU and CA 125 testing for postmenopausal women.45, 154
  • Prophylactic oophorectomy and hysterectomy (TAH) from ages 30 to 35 or when childbearing was complete.154
  • Screening for stomach, duodenum or urinary tract cancers.157
  • Prophylactic hysterectomy with or without bilateral salpingo-oophorectomy.163
  • 1.5 g calcium daily in a form of calcium carbonate tablet.164
  • Biennial breast mammography starting at age 35.158

We included studies evaluating all forms of cancer related to HNPCC since family members of CRC probands are potentially at risk for all such cancers.

Six studies examined the impact of mutation testing on the decision to undergo specific management recommendations among family members of CRC patients or asymptomatic individuals from HNPCC families139141, 154, 155, 165 (Table 28). Of these, four were of B quality and two were of C quality. No study examined the impact of HNPCC mutation testing on public health policy or decision making by insurance providers.

Table 28. Key Question 10b1. Does the identification of HNPCC mutations lead to improved outcomes in terms of decision making by patients, family members and providers, or public health policy?

Table 28

Key Question 10b1. Does the identification of HNPCC mutations lead to improved outcomes in terms of decision making by patients, family members and providers, or public health policy?

Two studies (in three publications) of B and C quality indirectly addressed outcomes of early detection and mortality/morbidity in relation to the identification of MMR mutations in family members of CRC probands or asymptomatic individuals from HNPCC families42, 112, 160 (Table 29). These studies were limited by potential selection bias, and/or unclear effects from treatments or subsequent interventions.

Table 29. Key Question 10b2. Does the identification of HNPCC mutations lead to improved outcomes in terms of early detection and mortality/morbidity, of patients, family members?

Table 29

Key Question 10b2. Does the identification of HNPCC mutations lead to improved outcomes in terms of early detection and mortality/morbidity, of patients, family members?

Summary of Findings. Identification of HNPCC mutations was associated with improved outcomes in terms of decision making to undergo screening for cancers in family members of HNPCC.

Survival was increased among asymptomatic HNPCC members who received colonoscopy screening regularly, regardless of their mutation status (Overview Table 4; Table 2829).

Gaps in the Literature. Identification of HNPCC mutation alone cannot lead to improved outcomes in terms of mortality or morbidity unless persons are willing to take subsequent actions or interventions. Thus, the Key Questions above have two main components: 1) whether the identification of HNPCC mutations led to improved outcomes in terms of decision making by patients, family members and providers, or public health policy, and 2) whether the decision making subsequently leads to improved outcomes in terms of mortality or morbidity. Studies that reported outcomes relating to subsequent management options or interventions in HNPCC family members are summarized later with Key Question 11.

No study examined the impact of HNPCC mutation testing on public health policy or decision making by insurance providers on an actual cohort of patients. However, a survey of 11 health insurers in Norway evaluated two hypothetical individuals' requests (one who had an HNPCC mutation; the other who had a BRCA1/BRCA2 mutation) for insurance153 (Extra Table 2). The insurers raised premiums on the one with genetic risk of CRC but not in the patient with a genetic risk of breast cancer.

HNPCC family members who were mutation carriers were more likely undergo screening and treatment for preventing cancers than non-carriers. Thus, the data presented above suggest that screening and testing for HNPCC mutations improves acceptance of screening procedures, at least in individuals who agree to undergo screening and testing.

Whether this benefit translates into improved morbidity or mortality is less clear from the available literature. A study that attempts to examine whether the identification of HNPCC mutations leads to improved mortality or morbidity should ideally control for the differences in the rates and intensity of screening or treatments for cancers. However, most studies did not provide such details. Furthermore, there may be complex interactions with age, gender, and time,since identification of HNPCC mutation, compliance with planed screening programs or treatments, and mutation status. No study considered all of these factors together. Large HNPCC registries with active data collection for the elements described above are needed.

Although not specifically asked in the Key Questions, a few studies reported factors that influenced compliance with screening (such as socioeconomic status and perceived barriers to screening).166, 167 We considered such observations clinically important and thus summarize them in ancillary tables. However, there were few such studies (Extra Table 3).

Key Question 11: What Are the Harms Associated With Subsequent Actions or Interventions for Family Members?

Image er-hnpccfu12.jpg

Due to limited data on harms associated with subsequent management options or interventions, we broadened the scope of this question to include studies that reported any outcome relating to subsequent management options or interventions in HNPCC family members.

Nine studies reported outcomes related to subsequent actions or interventions among family members. Of these, six were of B quality and three were of C quality. Four of these nine studies reported harms or adverse events associated with subsequent actions or interventions for family members. In addition, one study of C quality examined the psychological impacts associated with colonoscopies44, 45, 156159, 161164 (Table 30). Some of these studies did not have a control group of subjects who declined to undergo surveillance, and most results did not adjust for potential confounders such as age, personal history of cancer, and educational levels.

Table 30. Key Question 11. What are the harms associated with subsequent actions or interventions for family members?

Table 30

Key Question 11. What are the harms associated with subsequent actions or interventions for family members?

Summary of Findings. Fewer than 0.5 percent of family members experienced harms associated with screening or surveillance examination or surgical procedures. There was some negative psychological impact associated with colonoscopies.

HNPCC family members who took subsequent actions or interventions had a lower risk of developing HNPCC-related cancers and lower mortality rates, compared to those who did not take actions (Overview Table 5; Table 30).

Gaps in the Literature. There was little information on harms associated with screening and surveillance examinations or surgical procedures in CRC patients and their family members. However, the risks of these procedures are likely to be similar to the non-HNPCC setting. Only five studies reported data on harms or adverse events. One cannot assume “no harm” when a study did not report data on harms; therefore harms associated with the screening procedures may be underreported.

Footnotes

*

Appendixes cited in this report are provided electronically at http://www​.ahrq.gov/clinic/tp/hnpcctp​.htm

Views

  • PubReader
  • Print View
  • Cite this Page

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...