Life-cycle evaluation can be applied at any stage: from innovation to regulatory decisions, to reimbursement, re-appraisal, and disinvestment (Fig. 1)22,23. Each life-cycle phase can draw from complementary HTA methodologies. For example, in the innovation phase (discovery and early phase clinical trials), methods that combine business and economics, including early-stage economic evaluation (what is the range of a cost-effective drug price?) and real-options analysis (should we continue to invest in research?), inform the expected attractiveness of investing in phase II/III clinical trials. Prior to reimbursement, interventions with uncertain clinical trial evidence may be eligible for managed access, which could support conditional access to technologies whose effectiveness and cost-effectiveness are uncertain, providing time-limited access to patients, while generating evidence to decrease decision uncertainty24,25,26,27. In the post-reimbursement phase, continual surveillance and health technology management provide the opportunity for comparative evaluation using quasi-experimental comparisons of eligible versus ineligible patient cohorts, including through matched and pre/post study designs, such as propensity score or genetic matching28 or interrupted time series analysis29. The outcomes of these phases can inform disinvestment in expected low-value technologies, freeing resources for other areas of innovation or healthcare. Throughout each life-cycle phase, patients, families, and the public should be engaged as partners and participants in research. Their participation ensures that evidence produced and subsequent reimbursement decision-making meet their needs and preferences.
A precision oncology life-cycle health technology assessment framework
Generating life-cycle evidence to support timely and affordable access to promising technologies requires careful thought around an evidence and deliberation framework. We turn now to an advanced approach for life-cycle HTA, where there is initial full or conditional regulatory approval for the target technology, combined with uncertainty in comparative effectiveness and cost-effectiveness. We define life-cycle HTA as the standardization of data collected and methods needed to inform life-cycle appraisal, re-appraisal, and de-adoption of health technologies, all conducted within a living and adaptive LHS that periodically examines the value of continued research and evaluation in light of an evolving evidence base. Accordingly, life-cycle HTA is premised on an LHS that is iterative and ongoing, with data that aids serial decision-making. Life-cycle HTA differs from traditional HTA, which focuses on static estimates of effectiveness and expected present value30.
Our life-cycle HTA process framework is depicted in Fig. 2. The supporting criteria critical to each element of the framework are in Table 1. This framework is directed to generate and evaluate evidence for: (1) technologies with regulatory approval but that are not yet reimbursed, where RCT evidence is either based on small patient numbers or when RCT evidence will not be pursued because of disease rarity or lack of incentivization (e.g., patent expiry); or (2) reimbursed technologies, where there is uncertainty in comparative value, and where technology management is important for sustainability. Framework components are: (a) managed access that defines the time horizon and pricing conditions of real-world healthcare system trialing (including zero-cost or discounted cost drug provision); (b) collecting core data elements for RWD, including leveraging external data; (c) RWE generation to determine comparative effectiveness, cost-effectiveness (net-benefit), and the value of conducting additional research; and (d) interpretation of data and updating of decisions, including investment, continued evaluation, or disinvestment from managed access. Each life-cycle HTA step is briefly examined below in the context of CGP-directed precision oncology.
Managed access agreements allow patients to access new technologies while collecting additional data for RWE; for life-cycle HTA, the goal of managed access is to address key uncertainties and better inform reimbursement decisions. It is crucial to note that managed access agreements do not need to be promises of reimbursement but they do outline conditions under which reimbursement may be achieved. Managed access agreements can be informed by a combination of: decision-maker(s) conditions for allowing healthcare system access and industry sponsor conditions for funding health technology access; and the baseline value of initiating and continuing collection of RWD within the LHS using value of information analysis. While not extensively used in HTA, value of information analysis methods (e.g., expected value of sample information) can estimate the current expected value to society of collecting additional data from a sample of observations31. As such, the value of information analysis informs the time horizon of institutional support because it can quantify whether the value of collecting additional data is less (greater) than the cost of continuing to invest in research.
Key to supporting managed access and sustainable decision-making is set guidance regarding what constitutes comparative value. For an intervention to be considered cost-effective, decision-makers need to specify an endpoint, such as Net Monetary Benefit (NMB), which would need to be positive to demonstrate value for money. NMB represents the monetary value of an intervention conditional on both a particular willingness to pay for a health gain and the expected cost of technology implementation. This endpoint is calculated as the difference between incremental cost and the product of incremental effectiveness and payers’ willingness to pay for a gain in effectiveness, i.e., NMB = λ*ΔQALY − ΔC. Where ΔC is the cost difference between the treatment and comparator, termed incremental cost (Ct − Cc), ΔQALY is incremental effectiveness (QALYt − QALYc), and λ is payers’ willingness to pay for a QALY gain. QALYs are anchored on preference-based values between 0 (death) and 1 (perfect health). The willingness to pay threshold (λ) represents the opportunity cost of displacing QALYs from other interventions given budget constraints. For transparency, decision-makers will need to define λ. Decision makers will also deliberate on comparative clinical effectiveness. Determining the endpoint for clinical effectiveness will be disease-dependent, but patient-valued endpoints such as survival, health-related quality of life, or progression provide advantages over intermediate endpoints with uncertain patient utility (e.g., patient is matched to an experimental trial). When deliberating on clinical effectiveness, an endpoint with an established minimally important difference will facilitate deliberation. Understanding the impact of a defined clinical endpoint speaks to the importance of research with patients on their preferences and values.
Collecting and concatenating real world data
The specification of core data elements and how to achieve data collection is the second framework component, with RWD requirements informed by stakeholders and guided by the study design and analysis. Pollard et al. (2022) outline our consensus-based core data elements for precision oncology life-cycle assessment, emphasizing the importance of collecting data throughout the patient disease and care trajectory, including the time period prior to life-cycle HTA study initiation32. The core data elements necessary for life-cycle assessment are in Supplementary Table 1. Data spanning the pre-study time period enables RWE analyses that generate synthetic control cohorts. Synthetic control cohorts can include historical controls and controls from different jurisdictions, with data on the entire patient trajectory allowing for statistical analyses of patient outcomes that can adjust for time-varying confounding and other biases.
Our life-cycle HTA framework responds to situations with small benefiting patient populations and low event rates by considering access to external datasets and pooling of cross-jurisdictional data. Life-cycle HTA also needs to be responsive to administrative data shortcomings. Administrative databases were established to support routine patient tracking and financial planning. Their use in life-cycle HTA has the advantage of being often population-based and generalizable to the routine unselected population, unlike cohort or clinical trials databases. On the other hand, they may not fully capture the required data for analyses. Inconsistently collected data, missing variables, and incomplete data entry introduce additional challenges for comparative evaluations. Resultant effect estimates may be subject to both random and systematic error, unless life-cycle HTA considers methods for addressing and imputing missing data.
External data generated through cohort studies or clinical trials provides an opportunity for integration and linkage with health systems-generated RWD. Leveraging existing datasets (Fig. 2) will allow for initial analyses to be conducted based on current evidence and can facilitate components of the prospective data analysis, through informing intervention outcomes (e.g., variant identified and treatment given) or usual care outcomes (e.g., variant identified and treatment not given) based on individual-level data. As we will show next, careful thought has to be given to study design and analyses generating RWE from non-randomized cohorts.
Real world evidence generation
Life-cycle HTA requires an approach to study design adaptable to continuous data collection and iterative evaluation. Figure 3 presents an example non-RCT study design for evaluating effectiveness and cost-effectiveness of CGP-directed precision oncology targeting a rare biomarker. The study design is a retrospective cohort study analyzing linked routinely and prospectively collected data from multiple sources. The relevant study period spans the patient’s entire disease trajectory, from disease onset to date of intervention, to death or end of follow-up, with final outcomes analyses that allow for left- or right-censored data. With managed access, the primary cohort comprises consenting patients eligible to undergo CGP testing for a targeted therapy. The intervention group includes patients with a rare biomarker identified by CGP who receive targeted treatment. If the targeted therapy also requires the new reimbursement of a CGP technology, the intervention group may also include those who receive CGP but are biomarker negative. The comparison group are those with or without the rare biomarker who do not receive the targeted treatment. Given that patients with a rare biomarker who receive usual care are usually unobserved, quasi-experimental methods, pooling of available data from across cohorts or jurisdictions, and consideration of prognostic effects are needed to inform a synthetic counterfactual with adequate statistical power for effect detection.
Within-sample of the real-world evaluation, matching methods, such as propensity score matching, coarsened exact matching, or machine learning (e.g., genetic matching) can identify controls similar to the intervention group, who instead received usual care28. We caution that quasi-experimental approaches cannot adjust for all types of unmeasured confounders, emphasizing the importance of planning in advance to ensure that the LHS collects all necessary data elements to help avoid selection bias. Missing data, another source of bias, again speaks to the need of planning ahead when designing the LHS platform. Other potential biases for causal inference of CGP interventions are lead time bias or immortal time bias, in which treated patients must be outcome-free until their treatment date and thus have improved relative outcomes as well as time-dependent effects of subsequent lines of cancer therapy (systemic, radiation, or surgery)33. These can be minimally addressed through analytic methods and study design, built on an awareness of the clinical context of each case. Stakeholders should note that there is currently no optimal quasi-experimental method for adjusting for unmeasured confounders using observational study designs and a risk of bias will always be present. Externally, individual-level data can be leveraged to estimate and prognosticate the trajectories of those patients known to be biomarker positive or negative, but who did not receive the therapy. Further, clinical trial data from industry stakeholders can be leveraged to inform biomarker-positive patients who received the drug intervention.
Non-parametric and parametric methods for effectiveness and cost-effectiveness analyses on the entire cohort need to be agreed upon and applied periodically over time. Both Kaplan-Meier survival analysis and regression models can be considered to estimate incremental differences in endpoints, such as overall survival and quality-adjusted survival, within the time horizon of the available data34. Heterogeneity in clinical effectiveness across patient subgroups should be explored through stratified analysis or pooled analysis when sample sizes permit. For evaluating cost-effectiveness, regression-based methods based on NMB can be used. These methods have the advantages of being able to adjust for residual confounding and to characterize covariate impacts on marginal cost-effectiveness35,36. In the likely presence of censoring, net-benefit regression may be weighted by participants’ inverse probability of censoring weights to reduce estimation bias37. Additionally, decision modelling is a key tool that enables projection of future effectiveness and cost-effectiveness over a longer time horizon; it accounts for the patient trajectory beyond the observed intervention period. Conservative approaches for modelling should be used, including the assumption of no added health or survival benefit beyond the observed real-world trial period. Value of information analysis can identify which parameters are driving decision uncertainty and inform continued data collection and reimbursement decisions31,38.
Interpretation of the evidence and update decision
Decision-making is usually concerned with static estimates of efficacy, expected net present value of net benefit, and budget impact or feasibility, conditional on the current state of knowledge. Life-cycle HTA is a continuous decision-making process with the need for ongoing re-assessment informed by data emergence and horizon scanning. At pre-specified intervals, an interdisciplinary arms-length and independent prioritization committee should examine the updated evidence that is tailored for the deliberative processes. In Canada, prioritization committees exist both federally and within jurisdictions. Given life-cycle HTA is responsive to small sample sizes where high decision uncertainty is present, we suggest that an initial managed access recommendation is made at the national level, followed by coordination of data generation and prioritization at both the provincial and federal levels. The specifics around the managed access approach and the coordination of data sharing is a topic of ongoing debate and is a crucial area of additional research. Regarding data sharing between jurisdictions, we believe technology-enabled federated analysis is an important step for producing cross-jurisdictional evidence where the RWD does not have to leave the jurisdiction within which it was generated.
The decision to disinvest, to continue with evidence development, or to fund a technology (Fig. 2) should be based on previous knowledge and on the new evidence emerging from life-cycle evaluation. As discussed, the value of information analysis can inform whether to collect additional evidence. The value of collecting additional evidence will also be shaped by the committee’s tolerance for uncertainty of clinical effectiveness and the joint uncertainty in incremental costs and benefits.
Life-cycle HTA deliberations that are not supported by RCT evidence face the critical issue of potentially biased outcomes estimation. As a result, life-cycle HTA utilizing RWD absent of randomized study protocols should be subject to considerable scrutiny and oversight. We recommend that independent HTA outcomes units collaborate with decision-making committees to define the endpoints, study protocols, and supporting analyses needed for deliberation. Further, research infrastructure that allows for scrutinization of safety, effectiveness, and cost outcomes and that permits requests of additional analyses by experts will be an important component for ensuring trust in RWE. We recognize that HTA units will need to collaborate with for-profit entities, particularly when patient-level data from clinical trials to support RWE are key to addressing uncertainty and when financial resourcing may be required. Ideally, publicly and privately funded life-cycle HTA experts will work together to inform clinical trial protocols. This important collaboration is critical for enabling transparent life-cycle HTA analyses and subjecting the protocols and analytic approaches to scientific debate.