Background
For many tumor types, choosing the optimal treatment for patients with advanced or metastatic disease depends on the outcomes of biomarker testing. Biomarker testing helps select the optimal treatment and avoid overtreatment with ineffective treatments. The role of biomarkers for treatment selection is especially substantial in lung cancer [14]. Therefore, lung cancer is one of the first tumor types for which WGS will potentially be implemented.
However, it is not clear whether the potential value of WGS outweighs the incremental costs that WGS incurs. Its clinical utility is currently limited to those genes for which a targeted treatment is available. Critics assume that current standard of care (SoC) testing that entails the use of TGP and other tests that test one or a few genes, provide enough information for a clinical diagnosis in most cases. However, proponents hypothesize that WGS adds value in cases where SoC would not have identified a biomarker. Recently, a study concluded that the actionable genome shows limited evolution while under therapeutic pressure, meaning that conducting WGS once is sufficient for most patients [15].
The clinical utility of WGS must be weighed against the incremental costs. WGS requires a significant upfront investment due to the required lab facilities and infrastructure for data storage amongst others. Additionally, WGS has a higher cost per patient. Changes in the organization of care, such as adapting diagnostic workflows to accommodate WGS and putting the required infrastructure in place, will help to realize the potential value of WGS. The need to transform health services underlines the importance of assessing the full, system-wide requirements posed by WGS to support its implementation in routine clinical practice.
Problem conceptualization
Current workflows for biomarker testing in the Netherlands
Figure 1 depicts a schematic representation of the healthcare system considered to implement WGS. The system elements that are shown in Fig. 1 interact with other system elements. For example, patients visit hospitals to be diagnosed and treated, while hospitals use WGS services and molecular tumor boards (MTB) to provide that care. Currently, WGS for cancer patients is primarily used in the clinical research setting as the clinical and/or economic value of WGS has not been demonstrated. One central facility in the Netherlands conducts WGS for cancer patients in hospitals participating in the Centre for Personalized Cancer Treatment study [16]. However, this centralized organization may shift to a regional organization in the near future if hospitals invest in building up their own WGS capacity. Interpretation of the complex genetic information that WGS provides is preferably performed by a group of multidisciplinary experts in an MTB [17]. Currently, the development of MTBs is still in an early phase. Nonetheless, Dutch academic hospitals each organize an MTB who meet regularly.
Biomarker testing for treatment selection is used by all hospitals (n = 78) that treat lung cancer patients. Patients can receive treatment with chemotherapy and targeted therapy in most hospitals [18]. Conversely, immunotherapy prescription is concentrated in a subset of hospitals that meet specific quality requirements [19]. Most hospitals that meet these requirements are academic or teaching hospitals or general hospitals with a high patient volume. Enrollment into clinical trials is also initiated via these hospitals. In most cases, patients are referred to one of these hospitals for treatment, or the patient's tumor material is sent to a hospital with a more elaborate testing capability.
The potential value of WGS from a systems perspective
Diagnostic workflows
A major challenge related to the diagnostic workflows is adapting the current workflows to accommodate WGS. Currently WGS is only used in the clinical research setting in oncology. Therefore, an important step is to determine the patient subgroups that will receive WGS, as the current price level of WGS [1] makes it prohibitive to provide WGS to all patients with lung cancer. This is linked to determining which tests will be substituted by WGS and how tests will be planned. Careful planning is essential, as there is a risk that the time-to-treatment will increase beyond the recommended maxima [18].
Additionally, it has been widely recognized that MTBs should interpret the genetic information that WGS provides [20]. For an adequate interpretation, an MTB should at least consist of clinicians, pathologists, clinical biologists, geneticists, and bioinformaticians [17]. However, there is still a large variety in the composition of MTBs in the Netherlands [21], and MTBs need to be able to cope with a potential increase in the number of patients who receive WGS. This can be achieved, for instance, by automating workflows and setting up clinical decision support systems [22].
The process of conducting WGS differs substantially from the current SoC biomarker tests. When conducting WGS, the patient’s tumor material is sent to a WGS facility. Once sequencing is completed, a report containing the results is sent to the MTB. The MTB discusses the results from WGS and reports an evidence-based treatment recommendation back to the hospital. Ultimately, the treating physicians can, together with the patient, use this recommendation to make a treatment decision. Hence, using WGS involves more and different steps than SoC, which is usually conducted in-house and typically does not use the services of an MTB.
Impact of policy decisions
At present, WGS is offered from one location in the Netherlands. While the evidence is still lacking on the effects of centralization [23], focusing all sequencing in one facility can potentially lead to improved efficiency and economies of scale as the throughput increases [24, 25]. However, it is possible or perhaps even desirable that, over time, a regional organization emerges, such that several hospitals can conduct WGS independently. The required capacity to conduct WGS should be carefully predicted, as a decentralized organization potentially leads to overcapacity, similar to what happened with proton therapy in the Netherlands [26]. Overcapacity may be utilized to conduct WGS for new patient indications, for whom a clinical benefit is perhaps not demonstrated yet. This may lead to an increase in the overall budget impact of WGS.
Additionally, the reimbursement status of WGS plays a role in how affordable and accessible WGS is. Presently, WGS is not reimbursed through health insurance in the Netherlands. Especially at the current price level of WGS, the lack of reimbursement presents a substantial barrier to wide-scale use [27]. If the reimbursement status of WGS does not change, only a few hospitals will likely implement WGS into their clinical practice, and then only for narrowly defined patient subgroups. Hence, the reimbursement decision will influence the required sequencing capacity and the likelihood of decentralization.
Technical considerations
Technical considerations that separate WGS from other biomarker tests are primarily related to the tissue used for WGS. While WGS is increasingly able to handle formalin-fixed, paraffin-embedded tissue [28], WGS using fresh frozen tissue remains more accurate. Fresh frozen biopsies are not routinely taken, which means that an additional biopsy needs to be taken for WGS. Moreover, biopsies for WGS need to comprise at least 20% of tumor cells for successful sequencing, meaning that approximately 28% of biopsies are not suitable for WGS [29]. These biopsy requirements pose substantial hurdles for successfully conducting WGS, as tumor material is often limited and difficult to access.
Model implementation
The conceptual model has been implemented as a hybrid dynamic simulation model using both DES and ABM. The SIMULATE checklist [30] was used to describe the systems model and can be found in supplementary file 2. We have opted for a hybrid model as it allows us to benefit from the comparative advantage of each modeling paradigm. Furthermore, both DES and ABM are individual-level modeling paradigms. Individual-level models can make optimal use of available patient-level data to make future events or trajectories dependent on each individual's history and characteristics, which is very informative in the context of precision medicine. For instance, when modeling care pathways, a treatment decision can be based on the outcome of a diagnostic test and patient characteristics.
The model has been developed in AnyLogic 8.3.3 (The AnyLogic Company). AnyLogic is one of several software packages in which multiple DSM model types can be combined in a single, hybrid model, thus providing high flexibility to model developers.
Model structure
Figure 2 provides a high-level representation of the model structure. Defining a model boundary is a necessary but subjective decision. The focus of this study is on the required changes in the organization of care. Therefore, system elements that have the largest potential influence on how care is organized or system elements most affected by changes in the organization of care are included in the model.
A hypothetical stage IV NSCLC patient who requires biomarker testing for the initial diagnosis is generated. Upon entering the nearest hospital's workflow, it is determined whether that specific hospital has implemented WGS and whether the patient matches the indication. If the patient receives SoC, all care processes are conducted within that hospital. If the patient should receive WGS, it is assessed whether the pathologist has adopted WGS and whether the patient prefers WGS over SoC. Subsequently, the patient's biopsy is sent to the WGS facility (n = 1), and once sequencing is completed, a report is sent to the MTB (n = 7). Finally, once either SoC or WGS has been concluded, a guideline-based treatment recommendation is given. Thus, the model's starting point is the diagnosis of stage IV NSCLC, and the endpoint is either death during the diagnostic pathway or the conclusion of the diagnostic pathway.
All hospitals that provide biomarker testing for lung cancer patients are reflected in the model. Hospitals are stratified according to type: general (n = 43), teaching (n = 21), and academic hospitals (n = 8). They differ in the testing strategy they employ. General hospitals have a relatively simple testing strategy; they test ALK rearrangement status using IHC and test the EGFR and KRAS oncogenes' mutation status with Sanger Sequencing. Teaching and academic hospitals test PD-L1 expression and ALK with immunohistochemistry (IHC) and use the same TGP to test for EGFR, ROS1, BRAF, and KRAS. It is assumed that these tests are conducted in parallel.
If SoC testing in a general hospital did not identify a biomarker, that patient is referred to a teaching hospital. If the biomarker testing strategy in a teaching hospital also did not identify a biomarker, that patient is referred to an academic hospital. Academic hospitals have implemented WGS for referred patients and patients for whom SoC testing in that academic hospital did not identify a biomarker. If biomarker testing in the academic hospital also did not identify a biomarker, that patient is not referred further. Hence, WGS is implemented as a last-resort diagnostic test. A technical model description, describing the different agent types and parametrization is available in Additional file 1.
Model transparency and validation
We aimed to create model transparency by providing a clear description of the model and its software implementation. Furthermore, the model has been uploaded to AnyLogic Cloud [31]. Systems models are typically relatively complex and, therefore, difficult to extensively validate. In this case, validating the outcomes of a scenario in which WGS is not used against real-world data was not possible as those data were not available. Achieving face validity is often seen as an important first step [32]. Face validity was achieved through several discussions with stakeholders during and after model development to discuss modeling choices, assumptions, and outcomes. During model development, interactive discussions were held with the Technology Assessment of Next Generation Sequencing in Personalized Oncology (TANGO) consortium [33], which investigates the added value of WGS for clinical diagnostics in the Netherlands. This group consisted of experts on oncology, pathology, genetics, bioinformatics, ethics, and health economics. Once model development was concluded, an interactive discussion with patient representatives, stakeholders from the current genomic services provider, and the TANGO consortium was organized to evaluate whether the model’s face validity was sufficient.
Model-based analysis
Sensitivity analyses were conducted for model verification and to illustrate several relationships within the model. The following parameters were varied: the cost of WGS, the percentage of patients who need to be referred to another hospital that are referred, and the capacity to conduct WGS. For each parameter setting, the model was run 500 times to quantify the stochastic uncertainty in the outcomes [34]. To achieve stable outcomes, each simulation ran for 2000 days. With an annual expected patient population of 5313 [18], each run approximately simulated 29,000 patients.
Results
The cost of WGS
Figure 3 shows the impact of changes in the cost of WGS on the mean cost per patient. Figure 3 includes all patients; patients who received only SoC and patients who received both SoC and WGS. The changes in the cost of WGS have no impact on the mean cost of patients who did not receive WGS and only received SoC. Additionally, not every patient received WGS, and therefore, the mean cost per patient does not increase on a one-to-one basis with the cost level of WGS.
Referral patterns for biomarker testing
All hospitals are placed in networks with other hospitals to facilitate referrals of patients among hospitals. To benefit from more extensive biomarker testing, general hospitals refer patients to the nearest teaching hospital, and teaching hospitals refer patients to the nearest academic hospital. Patients who received WGS will not be referred, as there is no additional biomarker test available. Figure 4 shows these hospital networks, as well as the size of the referral flows and patient volume per hospital. For example, general hospital 5 (GH[5]) has a patient volume of below 1500 patients and referred between 367 and 671 patients to teaching hospital 2 (TH[2]). TH[2] has a patient volume of between 1501 and 3000 patients. While TH[2] also received referred patients from general hospitals 4 and 7 but refers only to academic hospital 0 (AH[0]), with a referral volume exceeding 642 patients. AH[0] has a patient volume of between 4501 and 6000 patients. AH[0] does not refer patients, but did receive referred patients from teaching hospitals 0, 1, 2, and 4. Note that Fig. 4 is a visualization based on data from one simulation run. In each simulation run, the distribution of hospitals across networks can vary, but how patients are referred is constant across runs. From Fig. 4, we can observe that hospitals vary in patient volume, patient referrals (both sending and receiving), and the degree of the relative importance of hospitals in the network.
Patients are referred to other hospitals if no actionable target has been found and more elaborate biomarker testing is available elsewhere. Figure 5 shows that a higher percentage of referrals lead to, on average, a longer diagnostic pathway. The diagnostic pathway's mean duration increases when more patients are referred due to a model mechanism that extends the diagnostic pathway for several days when a patient is referred, reflecting that referrals cause a delay [18]. Moreover, the uncertainty in the mean diagnostic pathway duration increases once more patients are referred.
Capacity constraints for WGS
Figure 6 illustrates how constraining the capacity to conduct WGS and the MTB capacity to give a clinical interpretation of the WGS report impacts the percentage of patients who died before receiving a treatment recommendation. Figure 6 is stratified by MTB meeting frequency; weekly or every two weeks. Once the sequencing capacity is below 1600 biopsies annually, which is enough capacity to prevent long queues in this scenario, the diagnostic pathway's mean duration increases. At a capacity of 1450 biopsies annually, the effects are noticeable but not as extreme compared with a capacity of 1300 biopsies annually. This extreme undercapacity leads to a significantly increased mean duration of the diagnostic pathway and increased uncertainty surrounding that mean estimate. The MTB meeting frequency is also a form of capacity constraint, as it affects the waiting time for the clinical interpretation of WGS results. If MTBs meet once every 14 days, the duration of the diagnostic pathway increases slightly, approximately equal to seven days.