New oncology drugs are starting to be licensed in conjunction with companion diagnostics, limiting prescribing only to those patients with maximal potential benefit from the drug. Consequently, cost-effectiveness analysis now has to address aspects of the screening as well as the traditional cost and benefit from the treatment ().
Using available information on the cost and performance of different ALK detection methods, we were able to assess the potential impact of these factors on the cost-effectiveness of biomarker screening in relation to ALK and crizotinib in NSCLC ( and ).
There are several assumptions and limitations associated with our work. First, we used estimated commercial charges for pathology testing derived from one institution for our analyses, limiting the direct transferability of specific values worldwide. However, the purpose of this paper is not to evaluate the cost-effectiveness of a particular test that will be applicable without additional analyses across all scenarios; instead, our purpose is to illustrate the issues that will determine the cost-effectiveness and the relative contribution of prevalence rates, screening cost, screening sensitivity and specificities, and the cost of the treatment. The precise screening cost in any given scenario will depend on a multitude of factors, including the particular provider of the test, and the health organisation and payer system involved. To generate data for modeling, we assumed that the PFS in single arm studies represented the actual health gain from the intervention and that the dramatic clinical responses seen when specific inhibitors are given to pre-selected oncogene-addicted tumours would produce a period of time with perfect quality of life. Inevitably, as either assumption weakens, the health gain from the intervention in our modeling would become correspondingly less. We assumed that the median age of lung cancer is 70-year-old, which in turn assumes that the molecular markers being screened for are uncorrelated with age. In fact, for several actionable molecular abnormalities, including ALK rearrangements, the median age of onset may be several years younger than for the general lung cancer population (Weickhardt and Camidge, 2011
). Consequently, in such scenarios there may be additional health benefits that could be addressed in the form of productivity gained. We also assumed that the proportion of never smokers is constant by histological subtype in the absence of additional data (Ramalingam et al, 2011
). However, if the never smoking rate were, in fact, >15% among adenocarcinomas, this would increase the denominator in the relevant enrichment step, increasing both the absolute costs and the cost per positive. Finally, we assume that PFS represents the dominant clinical benefit, as this is the only clinical data available in most cases; however, if overall survival is also longer then this would be an additional benefit and would reduce the cost per QALY gained.
Unlike most health economic analyses, we did not conduct a classical Markov model, comparing those who test positive and get the specific inhibitor with those who test negative and get standard chemotherapy. Instead, we focused our analyses on the pairing of the screening and the treatment, assuming that for those who test positive, after the treatment has run it's course, they simply proceed with the remaining standard therapies.
From our modeling, if all individuals with advanced NSCLC are screened for ALK positivity by FISH and the 1.6% who are found to positive are treated and gain 0.83 QALYs each, the total gain across the population will be ~0.013 QALYs (~5 extra days of perfect life) per individual screened. The health gain increases, on average, as each of the screening techniques is directed to populations with higher rates of ALK positivity (). The screening cost per QALY gained – omitting the treatment cost – is relatively high when screening all individuals with advanced NSCLC – $1
707 per QALY gained for the FISH test. However, as the population screened becomes more enriched and the mean health gain increases, the cost per QALY gained decreases. In addition, beyond increasing the screening ‘hit rate', sequential enrichment provides further absolute cost savings through smaller and smaller proportions of the total population being screened (column 3, ).
When comparing the cost-effectiveness of the different screening tests, both cost and performance of the test have to be considered. Specifically, an ‘imperfect' test may be more cost-effective than a ‘perfect' test, provided the proportion of cases missed remains less than the proportional reduction in the screening cost associated with using the ‘imperfect' assay. For example, although the RT–PCR test misses 30% of cases, our estimate is that the test is ~40% less costly ($875 vs $1400). Thus, the cost per QALY gained is uniformly 10% lower than for FISH testing.
When both drug and screening costs are included, in a more general oncology model that assumes perfect performance of all screening tests, shows that at very low biomarker frequencies, the price of the screening test is enormously important in the cost-effectiveness of the overall screening-treatment pairing. In the extreme, if the frequency of the marker is 1% and the price of the screening test is $1400, a cost per QALY gained of $100
000 is not achievable at any drug price. It just becomes achievable at a drug price of $1
000 per month if the screening cost drops to $600 per person tested. However, as the underlying frequency in the population increases above a relatively low threshold (
5% ), a cost per QALY gained of
000 is achievable with screening costs of $1400 per person tested when the drug is priced in the $1000–$5000 per month range.
As the frequency of the biomarker increases, the relative contribution of the screening costs to the overall cost per QALY gained, and, by extension, the effect of differences in the screening cost per person becomes correspondingly less ( and ). At low biomarker prevalence rates, there is a substantial difference in the drug price per month needed to achieve cost-effectiveness, which is heavily influenced by the price of the screening test per person. However, as biomarker prevalence rates increase, the drug price necessary to achieve cost-effectiveness for the differently priced screening tests converges. At a prevalence rate of 50%, the $800 difference in screening cost per person can be offset with only a $160 difference in drug price per month.
This can be seen visually in . If the cost of screening were zero dollars per person tested, the lines in would be flat with only the cost of the drug affecting the cost per QALY gained. However, when there is a cost associated with screening, the graph shows that at low biomarker prevalence rates, the cost of the screening test is the critical factor in the overall cost-effectiveness of the treatment. The cost of the screening test per person (shown for $1500 and $500 per person estimates at each of two different monthly drug prices) influences the point of inflexion where biomarker frequency becomes the dominant contributor to the cost per QALY gained. Lower screening costs per person left-shift, and higher screening costs right-shift the point of inflexion. However, once the biomarker prevalence rate exceeds ~3–5% (depending on the exact cost of the screening test per person) in , the price of the drug treatment alone dominates. Beyond (i) the cost per person of the screening assay, (ii) the frequency of biomarker positivity in the screened population and (iii) the price of the drug, the only other factor that will affect the cost per QALY gained will be the extent of benefit from the treatment in the biomarker-positive group. For a continuously dosed drug without screening costs, extent of benefit does not influence the cost-effectiveness, as changes in the denominator (QALYs gained) are balanced by equivalent changes in the numerator (required duration of therapy that is, total cost of drug). In contrast, when screening costs occur, as these are a one-time expenditure, they only appear in the numerator. With a greater benefit from the drug, the cost per QALY, taking into account screening costs, will therefore subtly decrease. In our modeling, we assumed a standardised gain of 0.83 QALYs. For other solid tumours and targeted therapies this figure could be different. However, we should recall that not only is this the least impactful of the four factors, it is also the only one that society and health care providers will not have any control over, whereas addressing the other three factors can significantly affect the cost-effectiveness of any screening-treatment pairings in the future.
Pulling all these data together, we can make certain observations about the cost-effectiveness of screening for predictive biomarkers in oncology. First, when considering the use of cheaper but ‘imperfect' assays, the cost of the assay and the drug, and the frequency of the biomarker have to be considered. We would suggest that, with drug and assay costs within the range we have modeled, it makes little sense to potentially miss marker-positive patients through the use of imperfect but cheaper assays, when the biomarker of interest is above a certain threshold (~5%). This is because at higher biomarker frequencies, the difference in assay cost makes minimal difference to the overall cost-effectiveness of the screening-treatment pairing. Second, for rare subtypes of a disease, in our model existing at <5% of the population, acceptable cost-effectiveness may only be achieved through either reducing the screening price or through adopting clinical enrichment strategies to narrow the focus to populations in which the biomarker exists at higher levels ().
The development of multiplexed assays may be one way to achieve the equivalent of lower screening costs per positive, without the risks of missing true positives through focused clinical enrichment policies. Although the frequency of each individual marker may still be low, tests such as the SNaPshot or sequenome assays used by the Lung Cancer Mutation Consortium can check for mutations in multiple different genes at the same time (Kris et al, 2011
). Using these assays, for the cost of one screening test, information may be generated on enough different mutations to produce a hit rate for ‘actionable biomarkers' of >40% in some NSCLCs, well above the level needed to avoid the screening costs being dominant in our modeled cost-effectiveness analysis. In our modeling, such multiplexing would effectively shift the cost per QALY calculation to the lower rows in and right-shift the curves in . With the advent of newer technologies, such as next generation sequencing, there may even be the potential to combine information on mutations, gene rearrangements and gene copy number, together within a single multiplexed platform in the future.
With regard to clinical enrichment, it is important to recognise that no enrichment step is perfect, and some true positives will be missed simply because they exist in subgroups that are not screened. For example, in columns 4 and 5 of , although a policy of only screening patients with adenocarcinoma who had never smoked will produce an ALK positivity rate of 13.7%, in our modeling this approach would leave half of the total ALK-positive cases undiagnosed.
Lowering drug costs will also improve the cost-effectiveness. However, a higher price for drugs with small market size is a precedent already partially established by some cost-effectiveness bodies (Rawlins and Culyer, 2004
), the logic being that the overall impact on society (cost × number of cases) will be low, whereas the gain for the individuals affected may be great. In addition, a higher permissible pricing may act as a partial incentive for the pharmaceutical industry to develop drugs for rare indications, as the incidence and prevalence of these newly discovered molecular subtypes of common cancers approaches that of traditional orphan diseases. Although given the protracted clinical benefit of targeted drugs in biomarker-positive patients compared with the treatment of unselected populations, the details of how and when to offer incentives may need to be updated in the era of molecular screening for drug sensitivity biomarkers in oncology.