It is an unfortunate reality that costs and expected benefits of any medical process, including diagnostics, need to be considered in determining the appropriateness of performing the process. The need for this consideration is equally relevant regardless of whether it is done in the context of a user-pay system (where the end user or patient needs to make a value judgement) or a centralized payer system, where difficult decisions must be made by administrators to attain maximal benefit for all covered people with a finite pool of funds to work from. Either case demands responsible, informed decision-making based on the best available data for the associated costs and benefits.
Frequently, reports of a novel MDx method will provide detailed information on the analytical performance of the test (sensitivity, specificity), turnaround times, throughput, and other “hard” metrics of performance. Costs, however, are often dealt with more vaguely; or, where firm per-test cost values are provided, close review suggests the reported costs may only be relevant in the narrow situation as applied at a specific site, with questionable generalizability. Detailed, in depth analysis not just of the direct costs of a test method but also of its indirect costs (or cost savings) in terms of such factors as reduced length of hospital stay, reduced risk of nosocomial infections, and the like are rare.
This common weakness is not a reflection on study authors, who are aware of and generally at least allude to these issues; rather, it reflects the complexities involved in addressing MDx cost-effectiveness in a truly rigorous and defensible fashion. In this month’s Primer, we’ll briefly consider what some of these complexities are and how they can possibly be addressed; then we’ll consider, for the narrower subset of oncology-related MDx, a representative handful of publications that have attempted to shed light on this issue. Understanding their assumptions—implicit and explicit—can help in determining goodness of fit of their conclusions to other situations.
Factors to consider
With that context in mind, let’s consider some of the inherent complexities in assigning costs and values to an MDx test in an oncology setting.
- What is the probability of the test returning an “informative” result? Consider, for example, an oncogene panel, of the sort discussed in last month’s Primer installment. (Brunstein J. Oncogene panels: a window into the individuality of cancers. MLO. 2015;48(4):18-20.) If a patient sample is tested by MDx panel and the results don’t add any actionable information (i.e., information that dictates a change in treatment strategy) over what was indicated by classical identification and classification means, then it could be argued that the test cost is wasted. If zero is the value assigned to this type of result, then a first step in assigning net value to the test is to know what fraction of samples will yield actionable information from the test. For example, if 20 percent of cases yield actionable results, then 80 percent do not and the immediate “value” of the test is only one-fifth of its possible maximum. Knowing this fraction with any certainty requires significant sample population sizes, but the knowledge can probably be gleaned from pooled multi-site data if uniform population inclusion criteria are applied.
- The term “informative” itself, as used in Point 1, is debatable and may not be applied equally in all settings. It may well be argued that even a non-actionable test result has value, in assuring the clinician that alternative treatment strategies are not warranted. If this line of reasoning is taken, then some partial value of a non-informative test result, as compared to an informative test result, must be applied.
- Individual cancers may change molecular characteristics over the course of disease progression, particularly if approaches such as chemotherapies apply selective pressures to the cancer cells. Detecting this change may be useful in modifying therapy, but the concept of detecting change implies one or more prior “baseline” measurements for comparison. If this sort of time course progression data is desirable, then, similarly to Point 2, all measurements, regardless of “actionability” of result, have some intrinsic value.
- Localized costs of assay performance can be highly variable. Particularly if a test is not a complete “sample to answer” system and relies on discrete steps or devices for nucleic acid preparation, molecular manipulation such as qPCR, and detection/analysis of results, then there exists a possibility that the laboratory may already have one or more of the required devices being used for other test functions. This, of course, amortizes the infrastructural cost over several test streams, making it cheaper per test. (This would also impact throughput considerations, our next point.)
- Test throughput and scale can dramatically alter the per-test performance cost. Higher throughputs (utilizing instrumentation near capacity) and batching of specimens to optimize use of laboratorian time both can act to significantly reduce direct per-test cost. Reference 1 is a specific study of this, and suggests in the study setting (Brazil) that a 30 percent utilization rate for various MDx tests resulted in cost-per-test increases of between 169 percent and 412 percent over at-capacity costs, depending on test methodology.
Insights from studies
Now that we can appreciate what some of the challenges are in assigning both cost and value to a given MDx test, and in taking cost/value data from one location to another, let’s consider a few representative published studies from the oncology field.
Our first example relates to thyroid cancers. Najafzadeh and coauthors2 examined adding MDx to fine needle aspirate biopsies (FNAB). The authors suggest that FNAB with classical methods alone may yield indeterminate results in up to 25 percent of cases, and prepare a model assuming 95 percent sensitivity and specificity for the MDx test. Results of this model suggested the addition of MDx methods gained 0.046 quality adjusted life years (QALY), with a per-patient cost savings of $1,087 (assuming the MDx test cost nothing to perform). While this might at first glance seem an odd way to present their findings, it’s actually perhaps the clearest way to allow another site to apply the results to their situation by directly inserting local cost-of-test estimates. The authors conclude that if the test costs are less than $1,087, then there is both a net cost savings and a gain in QALY. (Recall that one QALY equals one year at “perfect” health; a year at less than perfect health, then, equates to something under one QALY, by whatever fractional decrease is associated with disease states.) The authors provide highly detailed descriptions of their model assumptions as well as clear graphic representations of the impact of different levels of MDx sensitivity and specificity. Sensitivity more than specificity is shown to impact the value outcomes.
Our second example is by Hagen and coauthors3 and focuses on MDx in the context of HNPCC (hereditary non-polyposis colorectal cancer). This is an interesting analysis of four different testing models with and without MDx components. While the body text is in German, the English language abstract indicates cost per patient life-year gained, with an optimal tiered methodology (application of MDx only in suggestive family history contexts) showing roughly 3x better cost-effectiveness of the least effective approach of blind population MDx screening. While these are the rank order results one would expect a priori, the magnitude of the value is instructive. As the authors point out, however, decreases in assay cost will start to make blind screening more attractive. This highlights the importance of considering publication year of such studies, as the intervening eight years have made MDx methods significantly cheaper, with corresponding impacts on study conclusions.
Our third example, a study by Djalalov and coworkers4 in the context of NSCLC (non-small cell lung cancer), highlights additional complexities in making cost-benefit analyses for MDx when it is applied as a companion diagnostic. In this study, subsequent to a number of defined assumptions, the authors report that MDx (specifically, EML4-ALK fusion testing by FISH) did improve patient outcomes by an average of 0.011 QALY, with a relatively minimal cost differential of an additional $2,725 per patient—of which only a very modest $60 is directly attributable to the MDx assay component. The overall interpretation of cost-benefit in this situation, however, changes dramatically when the cost of the companion drug (crizotinib) is included. The authors’ final conclusion, in fact, is that testing in this case is “not likely to be considered cost-effective,” but, also tellingly, “the model was not sensitive to the costs of molecular testing.” The generalizable message from this example is that where companion diagnostic MDx is considered, these assays are by nature tightly coupled to the cost of the specific associated drug.
In conclusion, if your facility is considering introduction of an MDx method in an oncology setting, and wants to address cost-effectiveness issues in determining whether this is a good use of resources, the preceding provides some ideas on how to most logically go about doing so. While identification of relevant publications is a critical first step, assessing what assumptions to adjust in applying the conclusions to another location is far from trivial but not impossible. As more sites add MDx protocols to their oncology diagnostics workflows, we should appreciate those that publish or otherwise share their cost/benefit results; doing so is of great help to other sites in tackling the challenges of good economic stewardship.
As a final note, readers interested in a more in-depth discussion of how to formalize assessment of clinical utility for MDx assays may be interested in the final reference provided,5 which deals specifically with this issue.
- Schlatter RP, Matte U, Polanczyk CA, Koehler-Santos P, Ashton-Prolla P. Costs of genetic testing: supporting Brazilian public policies for the incorporating of molecular diagnostic technologies. Genetics and Molecular Biology. 2015;38(3):332-337.
- Najafzadeh M, Marra CA, Lynd LD, Wiseman SM. Cost-effectiveness of using a molecular diagnostic test to improve preoperative diagnosis of thyroid cancer. Value Health. 2012;15(8):1005-1013.
- Hagen A, Hessabi HK, Gorenoi V, Schönermark MP. Cost-effectiveness evaluation of predictive molecular diagnostics using the example of hereditary non-polyposis colorectal cancer (HNPCC). Gesundheitswesen. 2008;70(1):18-27.
- Djalalov S, Beca J, Hoch JS, et al. Cost-effectiveness of EML4-ALK fusion testing and first-line crizotinib treatment for patients with advanced ALK-positive non–small-cell lung cancer. Journal of Clinical Oncology. 2014;32(10):1012-1019.
- Parkinson DR, McCormack RT, Keating SM. Evidence of clinical utility: an unmet need in molecular diagnostics for patients with cancer. Clinical Cancer Research. 2014;20(6):1428-1444.
John Brunstein, PhD, is a member of the MLO Editorial Advisory Board. He serves as President and Chief Science Officer for British Columbia-based PathoID, Inc., which provides consulting for development and validation of molecular assays.