Sharing the risk of innovation

A major challenge facing healthcare systems is assessing value of new health technologies, which are generally more effective alternatives to current ones, but often fraught with financial uncertainty owing to the low sample sizes, relatively short study time frames of clinical trials, and the resulting long-term projections of costs and outcomes. Healthcare payers are also at risk that the costs and health benefits accrued during clinical trials will not translate into the same costs and outcome metrics under real-world conditions.

Since 1999, the National Institute for Health and Care Excellence (NICE) has been responsible for assessing the cost-effectiveness of new health technologies in England and making recommendations to the healthcare payer, the National Health Service (NHS), on the most efficient use of resources. The appraisal framework, which relies on the mean incremental cost-effectiveness ratio (ICER) as a main criteria for reimbursement, measures the additional cost of a new technology compared with an existing treatment or standard of care divided by the additional health benefits to patients.

The uncertainty around the ICER value is not currently explicitly defined in the decision-making, but NICE acknowledges that two technologies sharing the same mean ICER may not represent the same degree of risk to payers; hence the distribution of the ICER may start to receive more attention than the easy-to-understand mean.

Furthermore, NICE does not acknowledge an explicit ICER threshold for recommendation, but in practice it is estimated to lie between £20,000/QALY and £30,000/QALY or quality-adjusted life-year. QALY is the preferred metric to measure gains in outcomes, where each future year of life is weighted using a quality of life score from 0 to 1.

Our recent research paper explores some alternatives to the mean ICER, which can incorporate financial uncertainty more explicitly into the decision-making process, potentially giving preference to technologies with a higher mean ICER Markov model but lower variability. For instance, dictating that the median ICER, or a higher percentile, be below a specified threshold.
As the choice of percentile is inversely correlated to risk appetite, selecting a higher percentile, while reducing the potential for downside risk, may result in fewer highly uncertain technologies being recommended. Additionally, the paper presents ways to apportion the risk of recommending new health technologies among stakeholders.

 
Actuaries' contribution
Actuaries possess the right skill set to quantify financial uncertainty, yet their involvement in health technologies has been limited. The Society of Actuaries in the US recently funded a research project that encourages actuarial participation in this area, in particular to assist health plans to understand the financial implications of adding a new technology to their plan benefits.

Actuarial expertise can be particularly valuable in helping design risk-sharing agreements, which offer the potential to mitigate some of the financial risks linked to uncertainty. As the focus is on adverse experience and sometimes extreme, more volatile, scenarios, it can inform a more productive distribution of risk and allocation of costs among the various stakeholders, such as the NHS and manufacturers of health technologies.

Multiple approaches can be considered when designing risk-sharing agreements. These can also be implemented at a population or patient level, drawing on tools and techniques used for insurance purposes. For example, risk corridors are used when there are limitations on data, and provide a range of values for which no compensation to or from an organisation is required. In health economics terminology, this could translate into setting bounds around the model assumptions that are particularly sensitive and from which changes in the value of these assumptions could result in a significant budget impact. A symmetrical 95% confidence interval might be used initially, but other statistical methods may be appropriate as more data are collected.

A particular feature of risk corridors is the ability for healthcare payers to be compensated for adverse deviation. In one-way risk-sharing, the costs of the actual experience that fall outside the margin of tolerance would be borne partly or fully by the manufacturer according to pre-agreed terms. In two-way risk-sharing (see Figure 1), both parties share in parts of the risk, so any reduction in the variability of the ICER is observed across both sides of the mean. Potentially, the healthcare payer (such as the NHS) is subject to additional payments when actual experience is more favourable than expected.

Stop-loss insurance schemes can be designed at a patient level to limit the financial impact of providing care in extreme and more volatile circumstances. Using the 95th percentile of the modelled lifetime cost distribution as a threshold, excess costs would fall under the sole or shared responsibility of the manufacturer, subject to the risk-sharing terms. This threshold can be adjusted up or down in accordance with the tolerance level of the risk-bearing organisation.

As most economic models are presented using 30-year or even lifetime horizons, an alternative approach can be considered, whereby the timeframe is partitioned and costs analysed separately within each time segment. For curative treatments, this may better reflect the distribution of costs over time, as, traditionally, high upfront costs are observed in early years, and lower maintenance costs in later years.

Lastly, individual patient schemes may be incorporated under an aggregate stop-loss. A major benefit of this structure is to limit the overall budget to payers like the NHS for delivering care, as the entire expected treatment population is covered under the risk-sharing agreement.

A measure of variability, such as the coefficient of variation (standard deviation over the mean ICER), could be added to the current methodological framework to further limit potential downside risk to payers when compared with some predetermined risk tolerance. This standardised measure, which arguably could take many other forms, allows comparison of distributions with different means and has the potential to inform the magnitude of the risk sharing. For cost-effective technologies (under £20,000/QALY, for example) showing a measure of variability above the mandated limit, implementing risk-sharing agreements to bring the coefficient in line with the threshold would reduce the standard deviation of the ICER as some of the uncertainty is mitigated. This could be embedded in the reimbursement methodology.

Ultimately, every approach to mitigating risk presented above requires a cost or ICER distribution, likely to be derived from stochastic modelling. By leveraging big data and analytics, maybe one day 'real-world' data will be used by public health systems to parameterise these distributions.
This would represent a major improvement from the current approach of relying on clinical trial data and other, more limited cost-effectiveness studies to inform reimbursement decisions. The use of such evidence could also provide the rationale for revisiting past decisions through a feedback loop, enabling timely adjustments to reimbursement methodologies.

Where risk-sharing agreements require a comparison of actual with projected experience, retrospective reviews of historical medical services utilisation data could inform the amount of the recoverable. As many clinical processes and treatments are simulated using a state-dependent Markov model, the impact on the ICER from changes in sensitive assumptions such as transition probabilities (probability of progressing/regressing to another stage of the disease) can be quantified. For instance, values found in the drug manufacturer submission to NICE could serve as potential anchors for risk-sharing agreements.

An integrated approach
The recommendation of new technologies involves diverting resources away from existing services. It is therefore important to understand and be able to quantify the degree of uncertainty in each appraisal and to design, where appropriate, mechanisms to mitigate future risk. This is particularly key as health technologies, once recommended for routine use, are rarely decommissioned. Equally, it can assist care commissioners and healthcare payers in their budgeting, as the predictability of future medical expenses is enhanced.

The number of risk-sharing agreements between pharmaceutical manufacturers and healthcare payers is gradually increasing, particularly in countries with a single national health service subject to budgetary constraints. The majority of these agreements though remain financial-based, merely consisting of discounts or price/volume caps, as they are relatively simple to implement and maintain from an administrative perspective.

The move to more outcomes-based (or performance-based) schemes, which tie reimbursement to superior outcomes but require additional administration and data requirements, present significant opportunities for actuaries to play a key role in assessing the value of health technologies under real-life conditions.