Genomic parameters

The ACMG identified 56 genes associated with 24 inherited conditions that they recommended be analyzed and returned to individuals receiving NGS for any clinical purpose. The conditions were selected because of their high penetrance and clinical actionability, and the variants were limited to those known or expected to be pathogenic. The probabilities of receiving specific IFs after genome sequencing were based on population-level estimates of prevalence from Gene Reviews, a collection of expert-authored, peer-reviewed articles on the diagnosis and management of specific genetic diseases (Table 1).¹⁸

Policy model structure

In the decision model, individuals undergoing genomic sequencing as part of clinical care can either receive or not receive the ACMG-recommended IFs (Figure 1). If a patient has a pathogenic variant in one of the dominant genes, or in two for a recessive disorder, then disclosure could lead to additional diagnostic tests, increased surveillance, and/or prophylactic treatment. These early interventions generally reduce the increased risk of developing adverse outcomes such as cancer or sudden cardiac death. If patients have a pathogenic variant but do not receive IFs as part of genomic sequencing, then they are assumed to remain at increased risk for development of the condition for their lifetime.

Open in a separate window

Figure 1

Decision tree schematic comparing a policy whereby individuals who receive genomic sequencing may receive or not receive the American College of Medical Genetics and Genomics (ACMG)-recommended incidental findings (IFs).

Given the complexity of independently modeling 56 IFs, we devised an approach to capture the major impacts of returning these findings while providing a tractable and useful policy model. First, 7 of the 24 conditions represented by the ACMG list would encompass ~95% of the expected IFs. We therefore focused our analysis on these seven conditions and evaluated the incremental costs and quality-adjusted life years (QALYs) saved by the return of results for each individually. For five conditions, we derived estimates of costs and effectiveness primarily from previously published CEAs; we created novel decision models for the remaining two conditions. Second, given the very small population-level impact of the rare genetic conditions, and given the paucity of data regarding natural history progression and treatment effectiveness, we modeled these conditions collectively. Complete details regarding the modeling of each condition are described in the Supplementary Materials and Methods online. Third, we multiplied the individual-level estimates for costs and QALYs associated with returning an IF by the expected prevalence of identifying and returning those results to estimate the implications of returning IFs at the population level. We assumed that cardiac-related variants from the ACMG list would not be considered incidental for cardiomyopathy patients and that Lynch syndrome and BRCA variants would not be incidental for patients with colorectal cancer. Finally, we conducted sensitivity analyses in which we explored the impact of a wide range of estimates and assumptions for the remaining 17 rare conditions.

Overview of IF-specific modeling

We searched PubMed for previously published CEAs of genetic testing, particularly cascade or population-based testing strategies, for each of the seven most prevalent conditions using as search terms both the phenotype and pathogenic variants. We identified two CEAs for familial hypercholesterolemia^19–21; however, given the high rate of lipid screening in the United States,²² we assumed in our base case that all individuals receiving an FH IF would already be identified from lipid screening. We found one CEA of cascade genetic testing for hypertrophic or dilated cardiomyopathy¹⁶ and long QT syndrome²³; however, these CEAs did not publish their results in sufficient detail to be included in our model directly; therefore, we reconstructed these disease-specific decision models using Markov modeling techniques and, in so doing, confirmed the key model inputs that were used in the original publications with several clinical genetics experts (G.P.J., C.J.G., and W.B.). We identified two CEAs for hereditary breast and ovarian cancer^24,25 that reported sufficient data regarding incremental costs and health effects to be included in our model. Finally, we identified four CEAs for Lynch syndrome testing.^17,26–28 Upon request, the authors of one CEA provided us with their original model of genetic testing for Lynch syndrome¹⁷; the rationale and implications of choosing this CEA are discussed in the Supplementary Materials and Methods online. See Table 1 , Supplementary Table S1 online, and the Supplementary Materials and Methods online for complete details about how we modeled each IF.

Cost parameters

In our base case analysis, we did not include the cost of sequencing because we assumed it was ordered as part of the diagnostic workup; however, we included the cost of sequencing in a scenario analysis for generally healthy individuals as a preliminary exploration of primary screening with NGS. For all analyses, we assumed that each individual who received IFs would require additional pre-test genetic counseling to discuss the potential of returning IFs (15 minutes) and each IF would require review to determine pathogenicity (23 minutes).²⁹ Finally, we assumed that each individual with a positive IF would receive an additional hour of posttest counseling in addition to condition-specific diagnostic testing, surveillance, and/or treatment.

Analyses

We calculated QALYs, lifetime health-care costs, and the incremental differences between strategies of disclosing or not disclosing the ACMG-recommended list of IFs. We calculated the incremental cost-effectiveness ratio (ICER) as the difference in costs divided by the difference in QALYs. We performed oneway sensitivity analyses for individual variables over their plausible ranges and propagated uncertainty in our model inputs using probabilistic sensitivity analysis with 10,000 Monte Carlo simulations. We used a lifetime horizon and a health-care payer perspective in 2013 US dollars; both costs and outcomes were discounted at 3% annually.³⁰ Additional details regarding these analyses are included in the Supplementary Materials and Methods online.

Implications of our findings

The ACMG Working Group invested a considerable amount of time and effort into developing the initial list of IFs it felt should be returned to all patients receiving genomic sequencing for an underlying clinical condition. Such careful selection of specific genes and variants is a critical first step in developing a policy for the return of results. Our results indicate that by restricting the list of variants to those that are unequivocally pathogenic, lead to disease with very high probability, and for which some evidence supports the benefits of early intervention, returning results could improve health outcomes at a reasonable cost in certain patient populations. Although the return of each IF that we individually modeled provided some clinical benefit to patients, these benefits came at widely varying costs. To improve transparency and consistency, we recommend that future iterations of the IFs guidelines provide an explicit estimate of the benefits and costs for recommending disclosure of specific IFs.

Comparison with other studies

Two previous studies have evaluated the expected prevalence of IFs at the population level.^29,31 Ding et al.³¹ used an approach similar to ours to estimate the rate of IFs and reported results that were, in general, consistent with our projections; however, our estimated rates for MUYTH-associated mutations are several orders of magnitude lower because we restricted IFs to individuals with biallelic mutations, whereas Ding et al. also included heterozygous mutations. Our decision to only include biallelic mutations follows the ACMG recommendations.³ Furthermore, including heterozygous mutations in our model would not have importantly changed our results because only individuals with biallelic mutations are recommended to receive increased surveillance.³² Dorschner et al.²⁹ also evaluated the expected prevalence of IFs at the population level using data from 1,000 participants in the Exome Sequencing Project and found pathogenic or likely pathogenic variants from the ACMG list in 20 participants. Based on this early empirical evaluation of population-based prevalence, our projected rate of IFs may represent a lower boundary; however, a lower prevalence of pathogenic variants is conservative with respect to cost-effectiveness.

Limitations

This work has a number of limitations that warrant discussion. First, we derived estimates of incremental costs and effectiveness from previously published CEAs for several of the clinical conditions in our model and, as such, did not have as much control of or flexibility in the models, particularly in conducting sensitivity analyses. Another limitation was the lack of CEAs of genetic testing for FH that incorporate potentially suboptimal adherence to statins; however, in scenario analyses we found that even under conservative assumptions regarding treatment adherence, including FH IFs would slightly improve the overall cost-effectiveness of returning IFs. We also needed to make several assumptions regarding the costs and benefits of treatment for the rare conditions on the ACMG list given the varying quality of evidence and considerable uncertainty associated with these conditions. Our approach was pragmatic and efficient, and it was intended to provide a first approximation within a reasonable time frame to inform the ongoing debate and policy development for the return of IFs. We assessed the potential impact of uncertainty in our results by performing scenario and multivariate probabilistic sensitivity analyses, which showed fairly robust findings.

Second, the clinical conditions that could prompt exome or genome sequencing are heterogeneous and evolving. We therefore did not consider the clinical and economic implications of returning IFs to any patient who might receive NGS, but rather presented the results for three archetypal patient populations that may receive NGS and that illustrate the range of downstream clinical and economic consequences associated with returning IFs across a wide range of settings. It is also important to note that, except for the exploratory scenario analysis of primary screening with NGS, we have not assessed the cost-effectiveness of NGS, but rather that of the policy of returning IFs after NGS. The costs of adopting NGS could be much higher than those associated with returning IFs if this technology is more expensive than alternative genetic testing strategies. Whether these increased costs are warranted will depend not only on the return of IFs but also on the improved diagnosis of the primary genetic conditions, which will need to be evaluated in each clinical scenario.

Third, current estimates of prevalence and penetrance are derived primarily from selected high-risk populations and may not appropriately reflect estimates in unselected populations.^32,33 Begg et al.³² found substantial between-family residual variation in the risk for development of breast cancer among those with BRCA1 or BRCA2 pathogenic variants that was attributable to unobserved risk factors. They suggested that the reduced risk seen between families of probands with contralateral versus unilateral breast cancer—reduced up to one-third by age 80 in probands with BRCA2 pathogenic variants— might mirror the reduced risk seen between carriers identified because of family history and those identified from unselected populations. We found that reducing the penetrance had a moderate impact on our results; however, the ICER remained below $80,000 per QALY gained for healthy individuals and cardiomyopathy patients across a range of plausible estimates for reduced penetrance.

Finally, it is important to note that our policy model, like the ACMG guidelines, is only a first iteration. We anticipate frequent revisions and updates as new evidence emerges or as policies are developed regarding the use of NGS in clinical care.³⁴ In particular, only the CEA of genetic testing for BRCA1/2 explicitly incorporated the disutility of a positive result. Furthermore, patient preferences for Lynch syndrome outcomes suggest that although patients value genetic information on average, there is substantial heterogeneity in preferences.³⁵ Thus, it will be important to incorporate additional data regarding utility for genetic information once such data are available and to explore whether and how differences in personal utility provide a rationale for more personalized return of results policies.³⁶ The ACMG Working Group recently updated its recommendations to allow patients to opt out of the return of IFs; however, it is not clear whether such all-or-nothing disclosure policies will adequately address the heterogeneity in patient preferences for genomic information.³⁷ Our policy model provides a timely and relevant evaluation of the long-term health and economic implications of the current ACMG recommendations and, importantly, also provides a framework for evaluating future iterations of these guidelines.

This work was supported by grant U01 HG0006507-01 from the National Human Genome Research Institute, by an R36 Award (1R36HS023340-01) from the Agency for Healthcare Research and Quality, a doctoral dissertation grant from the National Science Foundation (award 1424250), and by a grant from the University of Washington Northwest Institute of Genetic Medicine awarded by the Washington State Life Sciences Discovery Funds (grant 265508). The funding sources had no role in the design, conduct, or interpretation of the study or the preparation, review, or approval of the manuscript.