Overall, eleven papers assessed the cost-effectiveness of the corresponding drugs (14–25), while seven papers evaluated the cost-effectiveness of companion biomarkers per se (4, 5, 26–30). The most frequently used modeling type was a Markov model (eleven papers), followed by partitioned survival model (two papers) and semi-Markov model (two papers). All economic evaluations were performed from a third-party payer perspective except for one study which took a societal perspective. All studies were performed for high income countries except for four studies of China.
Study characteristics of included literature are detailed in Table 2. Figure 2 provides the synthesized overview of whether the key methodological areas were addressed or not in the evaluations. The most frequently ignored model inputs related to companion biomarkers were preference-based outcomes, clinical utility, resource use, and the timing of the test. The detailed analysis of key methodological areas per publication is provided in additional file 3.
Table 2
Detailed characteristics of the included studies
Study
|
Focu-s
|
Objective
|
Biomarker test
|
Corresponding therapy compared
|
Strategies compared
|
Biomarker related model inputs considered
|
Country
|
Perspecti-ve
|
Model type
|
Time horizon
|
Outcome measure
|
Funding
|
Aguiar 2017
|
Rx
|
To assess cost-effectiveness of immune checkpoint inhibitor with and without the use of PD-L1 testing for patient selection.
|
PD-L1 expression.
|
Immunotherapy (Nivolumab, Pembrolizumab,
Atezolizumab)
|
3 strategies compared:
Treat-all with docetaxel.
Treat-all with immunotherapy.
Test-treat (if PD-L1 expressed with 1% or more, patients were treated with immunotherapy; if not, treated with docetaxel.)
|
PD-L1 testing cost.
PD-L1 expression cut-off points (PD-L1 > 1% used in base-case analysis, while 5%, 10%and 50% tested in sensitivity analysis.)
|
USA
|
Third-party payer.
|
Decision- analytic model. (No further details given.)
|
5-year horizon
|
QALY
|
No funding declared.
|
Chouaid 2017
|
Rx
|
To assess the cost-effectiveness of afatinib versus gefitinib for EGFR mutation-positive NSCLCs.
|
EGFR mutation.
|
Afatinib, Gefitinib.
|
2 strategies compared on pre-specified patients:
Treated with afatinib.
Treated with gefitinib.
|
EGFR testing cost.
|
France
|
Third-party payer.
|
Partitioned survival model.
|
10-year horizon.
|
QALY
|
Commercial funding.
|
Curl 2014
|
Rx
|
To compare three strategies (dacarbazine, vemurafenib, vemurafenib plus ipilimumab) for patients with BRAF positive metastatic melanoma.
|
BRAF mutation.
|
Dacarbazine, Vemurafenib, Vemurafenib plus Ipilimumab
|
3 strategies compared on pre-specified patients:
Treated with dacarbazine.
Treated with vemurafenib.
Treated with vemurafenib plus ipilimumab.
|
BRAF testing cost (Cobas®)
|
USA
|
Third-party payer.
|
Decision tree model.
|
Lifetime
|
QALY
|
No funding.
|
Ewara 2014
|
Rx
|
To assess the cost-effectiveness of three strategies (bevacizumab plus FOLFIRI, cetuximab plus FOLFIRI, panitumumab plus FOLFIRI) for mCRC patients with KRAS WT.
|
KRAS mutation.
|
Bevacizumab, Cetuximab, Panitumumab.
|
3 strategies compared on pre-specified patients:
Treated with bevacizumab plus FOLFIRI.
Treated with cetuximab plus FOLFIRI.
Treated with panitumumab plus FOLFIRI.
|
KRAS testing cost.
|
Canada
|
Third-party payer.
|
Markov model
|
100-month horizon.
|
QALY
|
No funding.
|
Graham 2014
|
Rx
|
To assess the cost-effectiveness of panitumumab plus mFOLFOX6 compared with bevacizumab plus mFOLFOX6.
|
RAS mutation.
|
Panitumumab, Bevacizumab.
|
2 strategies compared on pre-specified patients:
Treated with panitumumab plus mFOLFOX6.
Treated with bevacizumab pus mFOLFOX6.
|
KRAS and RAS testing cost.
RAS frequency.
|
USA
|
Third-party payer.
|
Semi-Markov model.
|
Lifetime
|
QALY
|
Commercial funding.
|
Graham 2016
|
Rx
|
To assess the cost-effectiveness of subsequent-line treatment with cetuximab or panitumumab in patients with WT KRAS mCRC.
|
KRAS mutation.
|
Cetuximab, Panitumumab.
|
2 strategies compared on pre-specified patients:
Treated with cetuximab.
Treated with panitumumab.
|
KRAS testing cost.
|
USA
|
Third-party payer.
|
Semi-Markov model.
|
Lifetime
|
QALY
|
Commercial funding.
|
Harty 2018
|
Rx
|
To investigate the clinical effectiveness and cost-effectiveness of panitumumab plus chemotherapy and cetuximab plus chemotherapy for rat scarcoma (RAS) wild-type (WT) patients for the first-line treatment of mCRC.
|
KRAS/RAS mutation.
|
Cetuximab.
|
2 strategies compared:
Treated with FOLFIRI alone.
Treated with cetuximab plus FOLFIRI.
|
EGFR testing cost.
RAS testing cost.
|
UK
|
Third-party payer.
|
Markov model.
|
10-year horizon.
|
QALY
|
Commercial funding.
|
Huxley 2017
|
Rx
|
To investigate the clinical effectiveness and cost-effectiveness of panitumumab plus chemotherapy and cetuximab plus chemotherapy for rat scarcoma (RAS) wild-type (WT) patients for the first-line treatment of mCRC.
|
RAS mutation.
|
Cetuximab, Panitumumab.
|
5 strategies compared on pre-specified patients:
Treated with FOLFOX/FOLFIRI.
Treated with cetuximab plus FOLFOX/FOLFIRI.
Treated with panitumumab plus FOLFOX.
|
RAS testing cost.
RAS prevalence (50% of patients assumed to be RAS wild-type).
|
UK
|
Third-party payer.
|
Markov model.
|
30-year horizon.
|
QALY
|
Governmental funding.
|
Janmaat 2016
|
Rx
|
To determine the ICER of adding cetuximab to first-line chemotherapeutic treatment of patients with advanced esophageal squamous cell carcinoma (ESCC), based on RCT II trial.
|
EGFR expression.
|
Cetuximab.
|
2 strategies compared on pre-specified patients:
Treated with cetuximab plus cisplatin-5-fluorouracil.
Treated with cisplatin-5-fluorouracil.
|
EGFR testing cost.
EGFR prevalence (60% patients assumed to be EGFR positive).
|
Netherlands
|
Third-party payer.
|
Monte Carlo simulation using individual patient data.
|
0.9 years.
|
QALY
|
No funding.
|
Lim 2016
|
Dx
|
To evaluate the cost-effectiveness of treating patients guided by EGFR testing compared to no-testing (which is current practice in South Korea).
|
EGFR expression.
|
Erlotinib.
|
2 strategies compared:
Test-treat (if EGFR positive, treated with erlotinib; if EGFR wild-type, treated with conventional chemotherapy; if unknown, re-biopsy required).
No-testing (Treat all with conventional chemotherapy).
|
EGFR testing cost (Therascreen®, Cobas®).
Testing accuracy (sensitivity/specificity).
|
South Korea.
|
Third-party payer.
|
Markov model.
|
5-year horizon.
|
QALY
|
Governmental funding.
|
Lu 2018
|
Dx
|
To examine the economic outcome of three techniques for testing ALK gene rearrangement combining with crizotinib (first-line), compared with traditional regimen.
|
ALK gene rearrangement
|
Crizotinib.
|
3 ALK rearrangement testing techniques prior to crizotinib were compared (4 strategies compared):
No gene screening - all treated with standard chemotherapy.
Ventana IHC - if ALK rearrangement positive, treated with crizotinib; if ALK rearrangement negative, treated with standard chemotherapy.
qRT-PCR - if ALK rearrangement positive, treated with crizotinib; if ALK rearrangement negative, treated with standard chemotherapy
Conventional IHC - if IHC ALK rearrangement negative, treated with standard chemotherapy; if IHC ALK rearrangement positive, FISH testing (to confirm) to be performed and then, if FISH ALK rearrangement negative, treated with standard chemotherapy, if FISH ALK rearrangement positive, treated with crizotinib.
|
Cost of ALK rearrangement testing (Ventana IHC; IHC; qRT-PCR; FISH)
Sensitivity and specificity respectively for Ventana IHC; IHC; qRT-PCR).
ALK prevalence
|
China
|
Third-party payer.
|
Markov model.
|
10-year horizon.
|
QALY
|
Commercial funding.
|
Morgan 2017
|
Rx
|
To assess the cost-effectiveness of crizotinib in untreated anaplastic lymphoma kinase-positive (ALK-positive) non-small-cell-lung cancer (NSCLC).
|
ALK expression.
|
Crizotinib
|
2 strategies compared on pre-specified patients:
Treat all with crizotinib.
Treat all with pemetrexed chemotherapy in combination with cisplatin or carboplatin.
|
ALK testing cost
ImmunoHistoChemistry (IHC) testing cost
Fluorescence in situ hybridisation (FISH) testing cost
|
UK
|
Third-party payer.
|
‘area-under-the curve’ Markov model.
|
15-year horizon
|
QALY
|
Governmental funding.
|
Wen 2015
|
Dx
|
To explore the costs and effectiveness of RAS screening before monoclonal antibodies in mCRC based on FIRE-3 study.
|
RAS mutation.
|
Cetuximab, Bevacizumab.
|
Four strategies compared:
KRAS tested - treated with cetuximab and FOLFIRI.
RAS tested - treated with cetuximab and FOLFIRI.
KRAS tested - treated with bevacizumab and FOLFIRI.
RAS tested - treated with bevacizumab and FOLFIRI.
|
KRAS/RAS testing cost.
|
China
|
Third-party payer.
|
Markov model.
|
10-year horizon.
|
QALY
|
No funding.
|
Westwood 2014
|
Dx
|
To compare the performance and cost-effectiveness of KRAS mutation tests in differentiating adults with mCRC who may benefit from first-line treatment of cetuximab in combination with standard chemotherapy from those who should receive standard chemotherapy alone.
|
KRAS mutation.
|
Cetuximab.
|
10 different tests for KRAS mutation status. No comparator approach taken.
Cobas KRAS Mutation Test Kit (Roche Molecular Systems).
Therascreen KRAS RGQ PCR Kit (QIAGEN).
Therascreen KRAS Pyro Kit (QIAGEN).
KRAS LightMix Kit (TIB MOLBIOL).
KRAS StripAssay (ViennaLab).
HRM analysis.
Pyrosequencing.
MALDI-TOF mass spectrometry.
Next-generation sequencing.
Sanger sequencing.
|
KRAS testing cost.
KRAS testing accuracy (sensitivity/specificity)
KRAS prevalence (KRAS mutant, KRAS wild-type, KRAS unknown test result).
Timing of the test – justifications given.
|
UK
|
Third-party payer.
|
Markov model
|
Lifetime (23 years)
|
QALY
|
Governmental funding.
|
Wu 2017
|
Rx
|
To evaluate the economic outcome of adding cetuximab to the standard chemotherapy.
|
RAS mutation.
|
Cetuximab.
|
2 strategies compared:
No testing – treat all with FLOFIRI.
Test-treat (if RAS wild-type, treated with cetuximab plus FOLFIRI, if RAS mutant, treated with FOLFIRI).
|
RAS testing cost.
RAS prevalence.
|
China
|
Third-party payer.
|
Markov model.
|
Lifetime
|
QALY
|
No funding.
|
Zhou 2016
|
Dx
|
To evaluate the cost-effectiveness of predictive testing for extended RAS WT status in the context of targeting the use of cetuximab/bevacizumab.
|
RAS mutation.
|
Cetuximab, Bevacizumab.
|
4 strategies compared:
KRAS WT tested-treated with cetuximab plus chemotherapy.
KRAS WT tested-treated with bevacizumab plus chemotherapy.
RAS WT tested-treated with cetuximab plus chemotherapy.
RAS WT tested-treated with bevacizumab plus chemotherapy.
|
KRAS/RAS testing cost.
|
China
|
Societal perspective.
|
Markov model.
|
Lifetime
|
QALY
|
No funding.
|
Saito 2017
|
Dx
|
To determine the cost-effectiveness of comprehensive molecular profiling before initiating anti-EGFR therapies in mCRC.
|
RAS mutation.
Comprehensive profiling that includes PTEN + ERBB2, PTEN + SRC, and BRAF + RNF43 mutations (CancerPlex®).
|
Bevacizumab, Panitumumab.
|
3 strategies compared:
No testing
RAS screening
Comprehensive screening
|
Biomarker testing cost.
Proportion of molecular subgroups (proportion of patients per biomarker status).
|
Japan
|
Third-party payer.
|
Markov model
|
5-year horizon.
|
QALY
|
Unclear (Not reported)
|
Butzke 2015
|
Dx
|
To evaluate the cost-effectiveness of UGT1A1 genotyping in patients with mCRC undergoing irinotecan-based chemotherapy compared to no-testing.
|
UGT1A1 genotyping
|
Irinotecan
|
3 strategies compared:
No testing-treat all with standard dose of irinotecan.
Test-treat (if tested wild-type, standard dose of irinotecan treated; if hetero-and homozygotes, treated with a dose reduction of irinotecan by 25%).
Test-treat (all patients receive standard dose, and hetero-and homozygotes additionally received the growth factor 'pegfilgrastim').
|
Sensitivity/specificity.
|
Germany
|
Third-party payer.
|
Markov model
|
Lifetime
|
QALY
|
No funding.
|
Rx; Drugs, Dx; Companion biomark |
Target population
The patient population targeted in EEs of biomarker-guided therapies was varied but it can be broadly classified into two categories; one is a subgroup of patients with a specific biomarker status confirmed and the other is a group of patients with disease conditions regardless of biomarker status. Eight studies were performed on a pre-defined group of patients with a particular biomarker status (15–19, 21–23) however, they considered at least one characteristic of companion biomarkers in their evaluations. Many EEs were conducted using a pre-specified patient group with a particular confirmed biomarker status, and authors used this to justify excluding some of the key characteristics of companion biomarkers from their evaluations. In addition, two studies were conducted on all patients regardless of biomarker status, while additional analyses were done for a subgroup of patients with a specific biomarker status (4, 20).
Analysis viewpoint
The analysis viewpoint defines the scope of costs and health benefits to be assessed in EEs; often referred to as study perspective. All included studies clearly reported the perspective of EEs conducted. A majority of studies showed that EEs were performed applying the third-party payer perspective. Only two studies stated that they employed a societal perspective (16, 30); one from China and the other from the US. However, the US study (16) was found to be more appropriately described as a third-party payer perspective (e.g. Medicare).
Given the nature of multiple purposes of biomarker testing application or use, and the indirect impact of companion biomarker diagnostics on patient health benefits, taking a perspective of third party payers might not be sufficient to capture all costs and benefits relevant to companion biomarkers in the clinical context of selecting patients suitable for the corresponding therapy. However, only one study considered indirect costs such as travel fees and absenteeism costs together with the cost of adverse events (30). However, this study did not consider any biomarker-related indirect costs either. For example, Schnell-Inderst and colleagues conducted a targeted review and highlighted measuring the potential effect modifiers such as the dependency of treatment effects on contextual factors and learning curve (31).
Choice of treatment alternatives (comparators)
It is widely accepted that the alternative strategy to be compared in EEs should be based on the current practice with respect to the target population (32, 33). Several different types of comparator strategies were employed in the EEs of companion biomarkers for targeted therapies. These different strategies can be categorized in five forms as below. Some papers used more than one comparator strategy arm (14, 17, 27).
First, all patients were tested prior to the administration of the corresponding biomarker-guided therapy and treated depending on the test result. For example, if the patients tested positive for a particular biomarker, they received the guided therapy; however, they were treated with the non-guided therapy if they tested negative. This ‘test-treat strategy’ strategy was often employed as an intervention strategy rather than as a comparator in EEs of companion biomarker therapies. Five studies employed this strategy type as a comparator (4, 27–30) however, these studies focused on comparing the analysis among different biomarker types or testing kits rather than comparing biomarker-guided against non-guided strategy.
Second, patients were not tested but were treated with the biomarker-guided therapy; so-called ‘no-testing-treat-all with the guided therapy’. Only one study fell into this category (14). This study aimed to assess the cost-effectiveness of a new guided-therapy with and without the use of biomarker testing.
Third, no patients were tested but all patients were treated with the non-guided therapy; so-called ‘no-testing treat-all with the non-guided therapy’. Six studies used this strategy as their comparator (5, 14, 20, 24, 26, 27), and mostly a standard chemotherapy was chosen as the non-guided therapy.
Fourth, all patients modelled in EEs were already pre-specified like biomarker positive or negative, and all treated with the guided therapy; called ‘biomarker-specified group treating all with the guided therapy’. This type of comparator strategy is also commonly observed in EEs of biomarker-guided therapies in addition to the test-treat strategy. Two studies used this as their comparator strategy (15, 17). Both studies focused on assessing different guided therapies for the group of patients confirmed with a particular biomarker status. Only a handful of model parameters of companion biomarker tests were considered in their EEs and thus, they often failed to provide a full spectrum of decision-making information relevant to the use of companion biomarker medicines.
Fifth, all patients were biomarker positive or negative and treated with the non-guided therapy; called ‘biomarker-specified group treating all with the non-guided therapy’. Seven studies employed this as their comparator strategy (16–19, 21–23). This strategy is the most frequently employed comparator arm in EEs of companion biomarker medicines in cancer.
Structure of strategy comparisons
We found a wide range of inconsistencies in structuring the strategies to be compared in EEs of companion biomarker therapies. Structuring the comparative strategy arms can be determined by various factors such as eligible patient populations, decision-making bodies’ EE guidelines, and local clinical settings. For example, an EE study aiming to compare a guided therapy against a standard of care applied the structure of comparing the test-treat therapy against treat-all with the guided therapy or with the non-guided therapy. Or, a similar study aiming to assess the cost-effectiveness of a new therapy with or without biomarker testing could employ the comparative structure of a testing strategy against a no-testing strategy on a particular group of patients with known biomarker status. The structure of comparing strategies in comparative analysis can be classified into five types as described in Fig. 3.
The comparative structure of applying strategy arms in EEs of companion biomarkers was so varied, it would likely lead to a different or even conflicting conclusion in terms of cost-effectiveness of companion biomarker therapies depending on the comparator strategy chosen.
Measuring the clinical value of companion biomarkers
No consensus currently exists on data requirements when incorporating the clinical value of biomarkers into the modeling of EEs of biomarker-guided therapies. For example, the Diagnostic Assessment Program requires testing accuracy in appraisal of diagnostic tests (34), although it is not always feasible in practice especially when assessors are faced with no data on test accuracy at all. On the other hand, NICE methods guide of technology appraisal does not necessarily require the testing accuracy but requires the incorporation of the associated costs of biomarker testing (32). Furthermore, none of the EEs reviewed examined the accuracy of a companion biomarker diagnostic test separately, for example by testing different cut-off thresholds including false positive and false negative results as part of uncertainty analysis. The cut-off threshold is the cut-off point defining the presence of the biomarker, determining biomarker-positive and biomarker-negative patients for the administration of corresponding co-dependent therapeutic agents (35–37). Varying levels of accuracy may lead to different patient subgroups being eligible for the corresponding drugs. According to previous studies (9, 11), the clinical value of biomarker tests could be assessed in three ways; analytic validity, clinical validity, and clinical utility. Analytic validity is about how well a test detects the presence or absence of a particular marker (33). Clinical validity refers to the performance of a test (diagnostic accuracy) in detecting the presence of a specific disorder; so-called sensitivity and specificity (11). Clinical utility is defined in the ACCE (analytical validity, clinical validity, clinical utility, and ethical/legal/social implications) model project as “how likely the test is to significantly improve patient outcomes”, which goes beyond sensitivity and specificity and then which may change treatment options for the patient (38). In other words, clinical utility (effectiveness) of companion testing technology is based on the ability to improve patient health outcomes by altering treatment decisions (39, 40).
Relatively few EEs considered the diagnostic accuracy of biomarker testing using data on sensitivity and specificity (26, 27, 29). Many EEs did not consider the performance of biomarker testing or often did not mention this at all (4, 5, 14–19, 22, 30). Otherwise, some studies provided some assumptions or justifications why they did not consider the clinical value of a companion diagnostic test (20, 21, 23, 24, 28). It is often assumed that the technical accuracy of patient stratification by biomarker testing is perfect and thus, the sensitivity and specificity were either not considered or assumed to be 100%. However, no studies explicitly considered or assumed the clinical utility of companion biomarkers in their EEs. For example, no studies stated that the clinical value of companion biomarker testing was supposedly incorporated into the clinical effectiveness of the corresponding drug based on the clinical trial of the sub-population delineated by the diagnostic.
Meanwhile, a handful of studies considered the frequency or prevalence of a particular biomarker status among their target patient populations (4, 18, 22, 26, 27, 29). Among them, only one study considered the probability of unknown test result in the analysis (29).
Measurement and valuation of preference-based outcomes
The quality-adjusted life-year (QALY) is a preference-based health outcomes widely used in EEs of therapeutic products(41, 42). It is widely accepted because it allows comparisons of health benefits and costs across different disease areas and therapeutic interventions. However, challenges emerge with the economic assessment of companion biomarkers given the nature of targeted therapies guided by companion biomarker testing and indirect impact of companion biomarker testing on patient outcomes. The current metrics for measuring preference-based outcomes using population-based preferences cannot fully capture patient preferences for biomarker tests (43). There seems to be more aspects of individual patient preference when valuing biomarker tests compared to the valuation of conventional drugs. For example, patients could be informed in advance of the likelihood of therapeutic response or unresponsiveness prior to the provision of treatment.
Or, patients can have an improved sense of controlling their own choices of therapeutic options informed by their biomarker status. Shared decision making (SDM) and communication between patients and clinicians will put patients at the centre of treatment decisions guided by companion biomarker test results. Patients may feel empowered to make informed decisions about their own treatment and care (44–46). Although the provision of biomarker-guided therapy is dictated by the patient’s biomarker status, being informed of the biomarker status can support the SDM of both clinicians and patients to explore more fully the potential benefits and risks. It can then potentially improve patient satisfaction with health services.
Or, companion diagnostics for cancer patients usually require collecting a bio-sample for analysis, and this gives rise to the existence of process utility (including reassurance or information) (47–49). Brennan and Dixon’s study (50) supported the existence of process utility and found that different approaches were being used to detect and measure process utility such as gamble techniques, time trade-off, conjoint analysis. Some biomarker tests involve relatively invasive methods to collect the bio-sample, such as tissue biopsy, needle biopsy, skin biopsy in diagnosing cancer (51, 52), that can be measured and incorporated into QALY estimates. Yet, how to measure and incorporate process utility into cost-utility analyses needs to be further researched with more empirical studies in HTA. Or, if companion biomarker tests were already integrated into the clinical study of measuring patient reported outcomes (PROs) for co-dependent therapeutic agents, it can be assumed that the disutility or utility value of companion biomarker testing is already embedded or indirectly expressed in PROs of the corresponding therapy. Yet, this aspect should be transparently reported in health economic models of companion biomarkers or biomarker-guided therapies. Nevertheless, none of the EEs included in this systematic review discussed these aspects of companion biomarker testing or indicated how preference-based outcomes of companion biomarker devices were measured and valued. For example, no studies explicitly included utility or disutility values for biomarker testing. Where biomarker testing uses tissues collected in a previous biopsy, it can be argued that patient preferences do not need to be considered in the economic modeling. However, none of the EEs mentioned this aspect or attempted to justify the omission of preference-based outcomes of biomarker testing. As an example, patients might need to undergo another biopsy for the purpose of biomarker testing after the cancer has progressed to metastasis. Or, a second biopsy might be needed to confirm the biomarker status when the testing accuracy was unsatisfactory. Or, turnaround time of biomarker testing may lead to additional waiting time for patients to access the treatment. Or, patients might experience anxiety or hopelessness when they are informed that the test predicts non-response to the targeted therapy and no alternative therapy options are available.
Estimating resource use and costs
All included EE studies considered the costs of biomarker testing however, some details were ignored. Some papers did not report the cost of biomarker testing devices (14) and often a total lump sum cost was modelled without providing details on how the total cost calculated (15–17, 22, 28, 30). Several studies reported at least some details regarding data source or the names/types of biomarker testing kits (4, 5, 18–21, 23, 24, 26, 27), but many EEs did not consider or report the resource use parameters relevant to the testing of companion biomarkers. None of the studies considered the capital cost related to the initial purchase of a biomarker test kit or diagnostic equipment as well as other costs such as training staff, relevant consumables, or lab reporting tools. Even in the situation where laboratories can re-purpose existing testing platforms to deliver the new test, relevant costs of consumables and staff with appropriate skills need to be considered. As an example, the NICE committee was aware that ALK testing would be not carried out in this specific clinical setting if crizotinib was not available (53), and therefore it is highly likely that the hospitals will need to purchase the testing equipment (i.e. capital costing items) however, it was not considered in their EE.
Timing of the test use
Details of where in the clinical pathway testing was undertaken were often not reported. Only two studies (4, 29) provided some explanation on this aspect, however, it was not clear how the timing of the test use was considered in the analysis of the Westwood study (54). Whereas, Saito and colleagues (4) provided and justified their assumptions. Given the nature of companion biomarkers, the health benefit to the patient arises from the corresponding therapy guided by the testing result, which is best understood as it being part of the clinical pathway in relation to its indirect impact on patient outcomes. Therefore, the value of companion biomarkers is best assessed while considering the timing of the test use; for example, whether the testing was done at diagnosis or following progression to metastasis. Westwood and colleagues (29) noted that the timing of KRAS testing may vary; some clinicians might undertake routine testing for all patients at diagnosis or some might wait until metastases have been detected. Yet, they did not specify how their evaluation was done in this respect.
Uncertainty analysis
Six studies (14, 20–22, 26, 27) explored the impact of cost-effectiveness of varying at least one component of the characteristics of companion biomarker tests being evaluated such as unit cost, total testing cost, test accuracy, cut-off thresholds, and biomarker prevalence. However, many studies did not examine the characteristics of a test separately from that of the corresponding therapy. According to HTA guideline, “if a diagnostic test to establish the presence or absence of the biomarker is carried out solely to support the treatment decision… a sensitivity analysis should be provided without the cost of the diagnostic test” (32). However, out of three UK studies, two studies performed a sensitivity analysis on biomarker testing cost (20, 21).
Data sources for biomarker-related data inputs
All papers except for three studies (14, 17, 22) provided data sources used for the characteristics of biomarker tests. However, several studies did not provide a specific name of companion biomarker testing kits, although some of them reported a general biomarker testing type (e.g. RAS testing) and therefore, several studies were not transparent and reproducible. The most frequently used data sources were previous published literature. However, testing cost inputs were mostly sourced from reimbursement schedules (16, 19, 20, 22, 24), manufactures or laboratories (18, 29, 30), and if such information was unavailable, expert opinions were sought (21).