With increasing translation of genetic testing to clinical practice, the cost-effectiveness of clinical applications of molecular genetic tests has become a “hot” question in health economics and genomics. Recent systematic reviews have examined “genetic testing technologies” [1], “genomic technologies” [2], “personalized medicine” [3], and “individualized medicine” [4]. Hatz et al. encourage researchers to seek to understand “how” or “when” rather than “whether” genetic testing is cost-effective [4]. That requires researchers to address the heterogeneity in cost-effectiveness estimates for the same genetic test in order to identify influential contextual factors, which is rarely done.

A precondition for the demonstration of cost-effectiveness is evidence of effectiveness, i.e., the ability to prevent mortality and morbidity. As a recent commentary on economic evaluation in genomic medicine put it, evidence of clinical utility (effectiveness) is needed before cost-utility or cost-effectiveness can be shown [5]. Public health genomics focuses on promoting the implementation of evidence-based genetic testing applications that have been shown to have health impact. In particular, the US Centers for Disease Control and Prevention (CDC) in 2012 began a process for identifying Tier 1 genomic applications that have a synthesized evidence base supporting implementation [6].

One of the most prominent Tier 1 public health genomic applications is testing for Lynch Syndrome (LS) in tumor specimens from patients newly diagnosed with colorectal cancer (CRC) using preliminary testing using either immunohistochemistry (IHC) or microsatellite instability (MSI), followed by genetic sequencing and deletion testing to identify a mutation on an MMR gene [7–9]. Lynch Syndrome (LS) is an autosomal dominant cancer syndrome associated with a very high risk of both CRC and endometrial cancer (EC) [10]. Testing for LS in adults with CRC permits cascade testing of relatives of probands. Relatives who are found to have LS and agree to undergo intensive surveillance for CRC through colonoscopy every 1–2 years can substantially reduce the risks of: developing CRC, an advanced stage tumor if cancer does occur, and the likelihood of death from CRC. In addition, women who are found to have LS may take action to reduce the risk of developing EC. Most of the benefit from identification of LS consists of the gains in life expectancy among relatives identified through cascade testing.

Testing of patients with CRC for LS can be either selective or universal. Traditionally, patients have been selected for testing based on either Amsterdam II criteria [11] or Revised Bethesda Guidelines (RBG) criteria [12]. The RBG criteria combine information on age of cancer diagnosis, tumor type, and stringent family history criteria suggestive of LS. Such criteria in a research setting are highly sensitive in identifying individuals with LS [13]. In some countries, RBG criteria are considered standard care [14]. Complex statistical models which integrate age at onset, tumor characteristics and detailed family history of LS-related cancers, can also be used [15]. The limitation to selective testing is that collecting, analyzing, interpreting, and appropriately using the information to guide testing is challenging. In practice few LS patients may be identified through the use of RBG criteria [16,17].

Universal testing for LS in adults with newly diagnosed CRC was first recommended in 2009 by the CDC sponsored Evaluation of Genomics in Practice and Prevention (EGAPP) working group [18–20]. Universal testing was subsequently endorsed by other US groups [17,21,22]. Testing can also be targeted to CRC patients based on age cutoffs such as 60 [23] or 70 years [24] or combination of universal testing below an age cutoff of 70 and selective testing in patients over age 70 who meet RBG criteria [25,26].

We explore under what assumptions universal or near-universal tumor testing for LS in patients with CRC followed by cascade testing of relatives of those found to have LS is likely to be cost-effective in terms of promoting survival or quality-adjusted survival. Specifically, we assessed estimates from “full” CEAs that calculate health outcomes to address the following questions: (1) how does universal testing for LS compare with selective or targeted testing including use of RBG criteria or statistical models; (2) how do different age cutoffs for targeted testing affect the incremental cost-effectiveness of universal testing relative to targeted testing; and (3) to what extent do differences in epidemiologic assumptions account for differences in estimates between studies of the incremental cost-effectiveness ratio (ICER) of testing for LS? Previous discussions of differences in ICER estimates of testing for LS have either touched lightly on epidemiological parameters or been limited to comparisons of pairs of studies [14,27].

Is testing for LS in adults with CRC along with cascade testing of relatives cost-effective? That in large part is a function of projected effectiveness. The effectiveness of testing for LS, whether measured in discounted LYs or QALYs gained, is primarily a product of the reduction in CRC-associated mortality. No two articles reviewed calculated mortality reductions in quite the same way, which makes direct comparisons difficult. Cost-effectiveness is also a function of the comparator. Universal testing may appear cost-effective relative to no testing but not necessarily in comparison with selective or age-targeted testing strategies.

Whether testing is considered cost-effective may also depend on the meaning of “cost-effective.” Do decision-makers classify an intervention with an ICER above a single threshold value such as $ 50,000 as not cost-effective? Or, do they use a range of values, such as $ 50,000 to $ 100,000? How does the definition of cost-effective health intervention vary across countries? The World Health Organization suggests that interventions be considered “highly cost-effective” if the cost per disability-adjusted life-year or DALY is less than per capita GDP, and “cost effective” if the ICER is less than three times the per capita GDP [48]. For the United States, that approach implies a range of $ 53,000 to $ 160,000 in 2013 values.

A related question is whether the outcome metric is LYs or health-adjusted LYs. CEAs of cancer prevention strategies often use LYs because mortality benefits dominate quality of life impacts [32,49]. Because preventing deaths among adults results in fewer gains in QALYs than LYs, ICERs are higher when QALYs are used. However, QALYs are challenging to implement in LS due to lack of consensus on the disutilities associated with intermediate outcomes [29,33,43]. Two CEA studies assumed dramatically different magnitudes of negative psychosocial effects on mutation carriers of knowledge of LS, of genetic testing, and on other family members [29,33]. These differences had major implications for the calculation of QALYs and the estimated cost per QALY, which was much higher in the UCSF study which assumed large negative effects on carriers and family members [33]. The more conservative assumptions about psychosocial effects made in the PenTAG model [29] appear consistent with previously published literature as well as expert opinion.

Whether routine testing for LS in adults with newly diagnosed CRC is considered cost-effective depends on the assumed feasibility, cost, and sensitivity of selective testing using detailed family history data. One older modeling study concluded that universal testing would not be cost-effective in comparison with selective testing based on Bethesda criteria [50]. Two of the US studies reviewed here similarly suggest that universal testing does not appear cost-effective relative to selective testing, with an ICER > $ 100,000 per LY [30,31]. A German study reached qualitatively similar conclusions [14]. That presumes that such selective testing is feasible to implement in routine clinical practice, which is uncertain [51]. One US study reported that two-thirds of suspected cases of LS would not have been identified if the RBG criteria had been strictly followed [16]. No published study has demonstrated that testing based on complicated predictive models achieves close to 90% sensitivity in routine practice. One study of selective screening at the UCSF teaching hospital during 2007–2010 compared with universal testing conducted during 2010–2013 found that twice as many cases of LS were identified with universal testing [52]. CEA studies that include testing based on family history as a comparator should include the cost to obtain and interpret family history data [29–31].

Both universal tumor testing and testing just patients with CRC up to age 70, i.e., the Jerusalem criterion [24], have been implemented in large healthcare systems [53–55]. Ladabaum et al. calculated that the ICER for universal testing in that comparison was approximately $ 100,000 per LY [31]. Two CEA studies that excluded patients over age 70 both concluded that universal testing of patients up to age 70 would be highly cost-effective relative to no testing [14,34]. Other experts have also suggested that testing for LS using an age cutoff of 70 would be more cost-effective than universal testing. One commentary argues that testing patients under age 70 would reduce testing costs by 49% while detecting 91% of cases of LS [56]. Another analysis calculated that an age cutoff of 70 would miss 14% of cases and reduce costs by 35% [57]. More work is needed to assess the incremental cost-effectiveness of universal testing relative to testing with an age cutoff of 70.

The analytical perspective and the methods for assessing the costs of clinical services also influence cost-effectiveness calculations. CEAs of testing for LS have not followed a consistent approach. Studies which stated that they followed a societal or healthcare sector perspective, both of which presume estimates of resource costs, for the most part relied on reported expenditures [30,32]. Just one study attempted to estimate resource costs for laboratory tests [32], but that study relied on 1998 Medicare payment rates adjusted for inflation; the actual cost of colonoscopy is likely to have been lower. For analyses from the payer perspective [14,29,31,33], it is the stakeholder expenditure that is pertinent, not resource costs [58].

One important factor is the number of relatives tested per proband, which varies across studies from just over 1 to 8. Within a model, the ICER is roughly proportional to the number of relatives tested per proband. For example, when the number of relatives tested per proband in the revised CDC model [27] is reduced by one-half from the baseline value, the ICER almost doubled from roughly $ 32,400 to roughly $ 56,200 per LY. Other authors have also noted that the cost-effectiveness of testing for LS may be contingent on the number of relatives tested and found to carry mutations [59]. Evidence from one US implementation study suggests that the assumption of 2.1 at-risk relatives tested per proband [29,32] might be conservative; Marquez et al. report that 13 relatives were tested for four probands identified at one hospital and six were found to be mutation carriers [55]. Larger studies are needed with data from diverse health institutions to produce reliable estimates.

This review is not exhaustive. It does not address the efficiency and costs of different tumor testing strategies, such as IHC first vs. MSI first. The cost per case detected has been modeled in other studies in addition to the CEAs included in this review [19,60–64]. The implications of use of preventive strategies for endometrial and ovarian cancer in females with LS require further investigation. Future studies of the cost-effectiveness of LS testing might model the effectiveness of colonoscopy and aspirin preventive strategies on the basis of observational data among LS mutation carriers regarding the use of just one or both preventive strategies. An important question that still needs to be addressed is whether long-term prophylactic use of aspirin by mutation carriers might make LS testing more cost-effective.

Finally, this paper does not address the controversial proposal to offer direct gene sequencing testing for LS in primary care patients, as proposed in a commercially-produced CEA modeling study [65]. That analysis assumed that detailed, fully accurate family history data would be available cost-free to primary care providers, which is implausible [51,66]. A recent modeling study concluded that such testing is not cost-effective under plausible assumptions [30].

The diversity in assumptions and estimates in this review preclude drawing conclusions about whether universal testing for LS is or is not cost-effective in absolute terms. Testing either all CRC patients for LS or patients up to age 70 years vs. no testing is likely to be considered cost-effective in the United States. Conclusions in other countries may vary depending on ICER thresholds and reimbursement rates for genetic testing and counseling. Cost-effectiveness of testing is contingent on systems of care in which relatives are counseled and offered genetic testing that is reimbursed by payers. Further, demonstration of actual cost-effectiveness requires documentation from routine clinical practice (not research protocols) of numbers of diagnoses of LS among relatives and uptake of intensive surveillance for CRC. A more challenging question is to identify the optimal testing strategy. We need real-world data on the sensitivity and costs of different testing strategies, including universal testing, age-targeted testing using different age cutoffs, and use of predictive models, including detailed family history data.

This case study has implications for the economic evaluation of genomic testing applications in general. First, context matters. Testing costs and uptake of genomic testing vary across populations and healthcare systems. Strategies to improve the uptake of counseling and testing may be crucial to improving the effectiveness and cost-effectiveness of genomic testing. Second, the cost-effectiveness of a testing strategy depends on the alternative to which is it compared. Universal testing for a condition for which highly effective prevention strategies are available is likely to prove cost-effective in many settings compared with no testing but may not be cost-effective in comparison with selective testing based on predictive models or criteria. The accuracy and cost of selective testing strategies should be documented in real-world settings before drawing conclusions about cost-effectiveness.