Living in a material world: tumor-treating fields at the top of the charts

The potential role of low-intensity, intermediate-frequency alternating electrical fields (tumor-treating fields, or TTFields [Optune, Novocure]) in treating glioblastoma has generated excitement, hope, and controversy. Initial cell line and animal model studies^1,2 were quickly translated to the EF-11 Phase III trial in recurrent glioblastoma. EF-11 confirmed device safety and suggested that TTFields had antitumor activity comparable to second- and later-line chemotherapy,³ a result leading to FDA approval but uncompelling to some neuro-oncologists given the poor results with chemotherapy for recurrent glioblastoma. Stronger results emerged from EF-14, a Phase III trial in newly diagnosed glioblastoma that randomized patients postradiation to TTFields plus temozolomide versus temozolomide alone. Interim analysis of EF-14 demonstrated a 3.0-month overall survival advantage in the intent-to-treat analysis (4.9 months in the per-protocol subset) and a 3.1-month improvement in progression-free survival.⁴ These positive results were based upon the first 315 of 695 enrolled patients; analysis of the entire cohort did not substantially change the results. Consequently, in October 2015 the FDA approved the use of TTFields in newly diagnosed glioblastoma.

In this issue of Neuro-Oncology, Bernard-Amoux and colleagues report on the results of a cost-effectiveness analysis evaluating the incremental benefit of TTFields for glioblastoma in the context of the French health care system. Their principal finding is that the use of TTFields in conjunction with standard of care chemoradiation is not within the realm of what is considered “cost-effective” by most regulatory bodies that have established explicit thresholds. While France does not use explicit thresholds, Commonwealth countries with national health systems, such as the UK, Australia, New Zealand, and Canada, generally consider treatment interventions that have ratios of $50 000–$100 000 per life-year gained to be cost-effective.⁵ In the US, regulatory authorities are prohibited from considering cost in making decisions about coverage, but informally, similar thresholds are used to label treatments as cost-effective.⁶

Cost-effectiveness analyses such as Bernard-Amoux et al's are determined on the basis of comparing 2 alternative strategies. In this case, the comparison is between chemoradiation plus TTFields versus standard chemoradiation. The incremental cost-effectiveness ratio (ICER) computes the cost of chemoradiation/TTFields versus that of chemoradiation while also assessing the benefits (expressed in terms of survival time) for these treatments. The ICER is the ratio of these differences.

Several methodological points are worth noting about the reported analysis. First, this was a post-hoc cost-effectiveness analysis that relied entirely on the published report of the trial. Why does this matter? The modeling approach does not have data for each trial participant and therefore does not calculate the precise amount of time that each individual spent receiving protocol-based treatment, no treatment without progression, second-line treatment, or palliative end-of-life care. The analysis relies on the study-reported average differences to model these parameters. The underlying assumption is that the key cost driver in this comparison stems from the costs of treatment itself. The huge cost of TTFields in the commercial setting makes this assumption reasonable, albeit uncertain. A major source of health care expenditures (in all health systems) is hospitalization. More hospitalization days in one group should correspond to higher costs. If TTFields prolong survival and keep glioblastoma patients out of the hospital, some of the excess costs of the therapy might be offset by reduced hospitalizations. However, if the intervention merely shifts hospitalizations to later time points without reducing them, then no such cost savings would be realized. Post-hoc models that do not collect detailed information about cost drivers such as hospitalizations, stays in rehabilitation centers, and the use of other expensive interventions are necessarily somewhat crude estimates.

A superior strategy is to embed a cost-effectiveness analysis in the up-front design of the clinical trial. This is often not performed when the drug or device manufacturer is the clinical trial sponsor. However, additional collection of a small amount of information can greatly facilitate the capacity for conducting a post-hoc cost-effectiveness analysis. For example, collection of data characterizing the total number of hospital days, ICU days, ER visits, and nursing-home days for subjects in each arm would increase the validity of the cost comparison. Information about the amounts of other chemotherapy treatments received would also be helpful. Simple case report forms can be embedded into clinical trial design to make this information available, and preplanned analyses can state thresholds considered cost-effective. In the absence of such prospective data collection and analysis plans, the results of cost-effectiveness models should be interpreted with caution.

The second methodological issue relates to incorporation of quality of life assessments. The accepted standard for cost-effectiveness is to incorporate measures of well-being in different health states. These measures are known as utilities. Perfect health corresponds to a utility of 1 and death to a utility of 0. For example, time spent with symptoms of progressive glioblastoma has lower utility than time spent without such symptoms. Ideally, patients' health states are measured directly through the use of simple tools such as the 5-item EQ5D scale to evaluate patients' valuations of particular treatment and disease states.⁷ When this is not possible, utilities can be obtained from libraries that have evaluated typical health states for patients with the index condition (glioblastoma). In the case of TTFields, it would be extremely helpful to know to what extent patients found the device cumbersome and inconvenient and whether it impeded their utility to a meaningful extent (eg, skin reactions and mild anxiety were more common in the TTFields arm). The goal is not simply to compare costs and survival but to understand how costs relate to quality-adjusted survival time. This ratio is known as a cost-utility ratio and is obtained by discounting survival time to adjust for the patient-perceived quality of survival. Bernard-Amoux and colleagues lacked information on the utility of wearing the TTFields device and therefore did not perform a cost-utility analysis. That said, it is highly unlikely that incorporating utilities would have tipped the scales and made TTFields treatment flip below typical cost-effectiveness thresholds. However, information about utilities and well-being of glioblastoma patients treated with and without TTFields is critical information to facilitate informed decision making.

Using their model, Bernard-Amoux and colleagues found that across a broad range of plausible thresholds, TTFields treatment is not likely to be cost-effective. This will pose a challenge for health care systems that explicitly consider cost in deciding which health interventions to cover. When a therapy is considerably outside the usual bounds, it is more likely that a health care system will deny coverage or require extensive review prior to authorization for reimbursement. At some threshold this should pressure manufacturers to lower their prices.

In the US, there has historically been immense concern that explicit consideration of costs in regulatory treatment decisions will deny patients access to valuable treatments. Two powerful lobbies, one representing retired and older Americans and another representing the pharmaceutical industry, dedicate vast resources to ensuring that cost considerations are not allowed to influence regulatory decisions; rather, they support decision making based purely on benefits such as survival and patient well-being. This approach has led to a backlash as prices for drugs and devices skyrocket and become increasingly unattainable for people earning average incomes. Outside the US, the use of explicit cost-effectiveness thresholds is better accepted. When items meet these thresholds, they are considered “good value” and are covered by the national health care system. When they are just above the thresholds, they may or may not be covered after careful review. When they greatly exceed thresholds, treatments may be available but require individuals rather than national health insurance to cover.

Although comparison of cost-effectiveness analyses conducted in different health care systems and with varied methodologies is rightly discouraged, it is nonetheless of interest to compare the ICER for TTFields with those reported for other approved therapies of newly diagnosed glioblastoma. A recent study of brand-name temozolomide with the Stupp regimen found an ICER of $102 364 per quality-adjusted life-year (QALY),⁸ while a UK analysis reported a figure of £35 861 ($56 000) per QALY.⁹ The addition of carmustine wafers to radiation therapy for grades III and IV astrocytoma was associated with £54 500 per QALY.¹⁰ Since patient quality of life is always less than perfect, the ICER per life-year gained irrespective of quality with temozolomide and carmustine wafers would be less than these estimates. The ICER per life-year gained with TTFields in the Bernard-Amoux study, at €550 000, exceeds these figures by a factor of 5–10. As such, the onus is on the device manufacturer either to provide data justifying the price of TTFields or to bring the cost in line with other approved therapies providing comparable clinical benefit.