This study synthesized published evidence on Lynch syndrome screening and expanded that evidence to match the decision needs of internal decision makers.
Published Online: August 08, 2011
James M. Gudgeon, MS, MBA; Janet L. Williams, MS, CGC; Randall W. Burt, MD; Wade S. Samowitz, MD; Gregory L. Snow, PhD; and Marc S. Williams, MD
Our study summarizes the evidence on the cost-effectiveness of Lynch syndrome (LS) screening in colorectal cancer (CRC) populations and, through computer simulation models, extended the evidence base to address the specific information needs of decision makers in our integrated healthcare system.
Published evidence suggests that LS screening in unselected CRC population is cost effective.
Our simulation models have identified the most efficient screening protocol as well as other decision metrics.
While LS screening offers promise in reducing morbidity and mortality from CRC, important questions remain about implementation of such programs.
Colorectal cancer (CRC) is one of the most common cancers worldwide, with more than a million cases detected and half a million who die from the disease annually.1,2 Approximately 5% to 10% of CRC is familial, reflecting shared genetic and environmental predisposition.3 However, some CRC is due to mutations in highly penetrant genes such as APC, MYH, and PTEN, as well as the DNA mismatch repair (MMR) genes MLH1, MSH2, MSH6, and PMS2. These are referred to as hereditary CRC to distinguish them as single gene disorders.4 Between 2% and 5% of all CRC is due to genetic mutations in MMR genes, collectively known as Lynch syndrome (LS), previously referred to as hereditary non-polyposis colorectal cancer. 5 Individuals with LS have a lifetime risk of CRC as high as 70% to 80% as well as higher risk of endometrial, ovarian, pancreatic, and urologic cancers.6
Importantly, risk for CRC can be dramatically lowered by intensive surveillance of LS mutation carriers, principally colonoscopy beginning at an early age and at increased frequency.7,8 This can be achieved by identifying MMR mutations in index cases and if a mutation is identified, performing family-specific mutation analysis on at-risk relatives. Relatives who test negative revert to routine population screening for CRC.
The optimal way to identify index cases remains undetermined. Family history was initially the recommended screening approach, but recent evidence has suggested substantial deficiencies in this approach. Emerging evidence suggested that tumor-based testing might be a clinically effective and cost-effective alternative to family history. However, multiple approaches could be used to identify individuals at risk, about which evidence was immature.
The goals of this study were 3-fold. The first goal was to thoroughly understand the current evidence base on the effectiveness and “value” of the different LS screening approaches. Based on the results of the first goal, the second goal was to make decisions about the details required for simulation models and construct appropriate models that would address screening efficiency and other questions from the perspective of our healthcare system. The third goal was to communicate the results to our clinical and administrative leaders to inform decisions on implementation of a screening program.
To address our first goal, we performed a series of narrative or qualitative reviews, which shaped our strategies for addressing the layers of questions important to remaining goals. The qualitative reviews were followed by more focused reviews, including review of unpublished data when available to provide data and summary statistics for the simulation models. Queries of our electronic data warehouse were performed to determine system-specific statistics for the models.
Narrative reviews were chosen for their flexible nature, both to extract the background information needed to understand the subject area and allow for incorporation of new data in a rapidly evolving field.9 The evidence base searched was PubMed, augmented by hand searches, the expertise of our team, contact with Hampel and colleagues of the Ohio State University LS research group, and information garnered from professional meetings.
As the nature and scope of the important questions to be addressed became clear from the narrative reviews, more focused reviews were used to address the questions. The first was the clinical effectiveness of available methods for LS case finding among CRC patients. The second, given our commitment to bringing value to our integrated system, was the costeffectiveness of these different approaches. Finally, the task of identifying numerical parameters to populate the analytic models involved both focused reviews and expert opinion.
The primary study period was mid-2007 to fall 2009, with final updates made up to fall 2010 in order to rerun the model with the most recent data available. All searches and analyses were performed by the first author (JMG); input was provided by all co-authors as well as other sources.
The computer simulation models were theoretic models, thus exempt from institutional review board approval. Our theoretic cohort consisted of new CRC patients expected to be diagnosed and treated within our delivery system. The models were constructed in TreeAge Pro software (Williamstown, Massachusetts) and Excel software (Microsoft; Redmond, Washington) using @Risk software as a decision analysis “addon” (Palisade, Ithaca, New York). Final model constructs were constructed in both platforms for validation purposes.
Model structures were constructed to estimate a combination of budget impact and cost-consequences analyses.10,11 The primary outcomes included the absolute and relative effectiveness and costs of the screening protocols, including total costs of testing, number of cases detected, cost per case detected, and the incremental differences between the different protocols. Secondary outcomes were defined by decision makers as we proceeded through decision making, which we describe in the Results section.
Sensitivity analyses (SAs) were performed to test the models as well as examine their output. The influence of model variables on outcomes was explored via multiple 1-way SAs. Calculation of point estimates and plausible ranges of outcomes was performed by Monte Carlo analysis.
Data Sources for Models
Test performance values, except those of methylation testing, were taken largely from the recent Hampel et al and Palomaki et al studies.5,8,12 Methylation testing performance values, including conditional dependence (CD) with the BRAF test, were extracted from recent studies.13,14 When gaps in published data were identified, we contacted outside investigators to obtain unpublished data. Sensitivity and specificity values were based on clinical validity (ie, the tests’ abilities to detect the presence or absence of LS). See Table 1 for primary model input.
Costs of tests were obtained from our reference laboratory, based on prices available to our healthcare system (current as of fall 2010), plus a $30 charge by our pathology department when applicable. At the time of completion of this study, there were no other labs in the United States that offered all the tests (with all 4 MMR proteins and genes) considered in the models.
The preponderance of evidence identified by our review suggested that screening unselected CRC patients for LS index cases with systematically applied tumor-based testing protocols could lead to a substantial reduction in CRC mortality in our patient population.7,8 The early modeling studies provided preliminary evidence suggesting that tumor-based LS screening protocols can provide good economic value at the societal level on the order of $5000 to $10,000 per life-year (LY) saved.13,15-22 A recent modeling study by Mvundura et al, using conservative baseline estimates and modeling more contemporary screening protocols, provided further evidence of the value of this screening; it demonstrated a cost-effectiveness ratio of about $22,500 per life-year saved (using an immunohistochemistry-first [IHC-first] approach) compared with no LS screening.23
The review also established that family history approaches have poor sensitivity or specificity for LS case finding under ideal circumstances.24,25 Performance in real-world practice would be far less than ideal because of barriers in the collection and interpretation of family history data.26
Limiting screening of CRC cases to persons under 50 years of age, while decreasing the cost of the screening program, would miss approximately 50% of LS cases; thus this approach was judged not sufficiently sensitive for our purposes.23,27 There is early evidence that using a prediction algorithm combining pathology and other features might efficiently predict a positive microsatellite instability (MSI) test as a marker for the presence of LS.28 This screening method requires validation, as its performance is likely operator dependent.
The tests determined to be candidates for initial screening of tumors and thus represented in our simulation models included IHC staining for the 4 MMR proteins, MSI, BRAF mutation, and MMR gene sequencing/rearrangement (ie, deletion/ duplication) analyses (Seq-Rearr). Reflex testing after a positive screening test included various combinations of these tests, always finishing with confirmatory Seq-Rearr. Seq-Rearr tests are bundled by our reference lab and thus are treated as a single test in our models. In the final months of the study, evidence regarding the utility of an additional test emerged: methylation of the MLH1 promoter (methyl), which was added to the models.
Based on published evidence, several options were eliminated early. Screening by Seq-Rearr was not viable due to poor cost-effectiveness.16,23 Initial screening by BRAF analysis was ruled out given its low prevalence in the CRC population. MSI-first screening was removed from the final models as evidence emerged that it would be substantially less cost-effective than IHC-first screening because of its inability to target subsequent sequencing with equivalent sensitivity.5,7,12,23
PDF is available on the last page.