Overview of Chapters

This chapter has introduced some central issues for health economic analysis to enable addressing constrained optimisation of societal decision making objectives informed by community values in a principled and robust way, whether in technology, program, policy or practice comparisons and in health promotion, preventative, curative, rehabilitative or palliative settings.

Chapter 2 further cements coverage and comparability principles as the robust foundation underlying unbiased decision making and health economic analysis and starts to consider robust approaches and methods to inform unbiased cost effectiveness analysis and adoption decisions. Satisfying coverage and comparability principles to avoid biases in undertaking cost effectiveness analysis is shown to require jointly considering adequate scope and duration of downstream cost and health effect impacts across strategies compared and relative treatment effect(s).

The advantages that the net benefit metric has over incremental cost effectiveness ratios (ICERs) - in summarising cost effectiveness evidence for such decisions - are illustrated and shown as particularly important when allowing for decision uncertainty. Useful presentation and summary measures for comparison under uncertainty of costs and effects of two strategies, the incremental cost effectiveness plane and net benefit and cost effectiveness acceptability curves, are introduced and illustrated for trial-based analysis. These presentation and summary measures are shown to be simply constructed in appropriately allowing for joint cost and effect distributions non-parametrically with bootstrapping and parametrically with Fieller’s method. The need for joint consideration of costs and effects in avoiding bias and inferential fallacies when informing decisions under uncertainty is highlighted with consideration of seminal papers including ‘The death of cost minimisation’ (Briggs and O’Brien 2001) and ‘Thinking outside the box’ (Briggs et al. 2002). These papers also begin to point to more general problems of bias with reductionist approaches, a theme which is expanded on in:

• (i) Chapter 3 for modelled cost effectiveness analysis;
• (ii) Chapters 5, 6 and 7 for value of information (VOI) analysis;
• (iii) Chapters 4, 8 and 10 for multiple strategy and outcome comparisons;
• (iv) Chapter 9 in efficiency measurement across providers in practice consistent with maximising net benefit; and
• (v) Chapters 11 and 12 in appropriately considering alternative actions for identifying the opportunity costs of investing in, and pricing of, new technology.

Chapter 3 highlights some further common problems and dangers in inherently inconsistent and biased methods for modelled cost effectiveness analysis where coverage and comparability principles are violated with choice of methods and metrics in synthesising, translating and extrapolating evidence. These are illustrated with inferential fallacies and inconsistencies arising with use of relative risk in indirect comparison (Eckermann et al. 2009) and translation of evidence (Eckermann et al. 2011). They are also illustrated with parametric methods in extrapolation of costs, effects and cost effectiveness inconsistent with indication or associated factors over time such as compliance, resistance and side effects, as well as in inconsistent extrapolation across cost and effects. More importantly, methods which solve these problems are identified in each case. Odds ratio methods are shown to enable unbiased consistent estimates with alternative framing of outcomes in indirect comparison and translation (Eckermann et al. 2009, 2011). Decision analytic modelling approaches with extrapolated treatment effects conditional on indication, continuation rules and compliance and side effect profiles in surviving populations in practice are indicated to allow unbiased and consistent extrapolation of costs, effects and cost effectiveness. Solutions to these problems also serve to illustrate the need for complementary approaches to health economic evaluation with trial-based and model- based evaluation to allow evidence relevant to decision making in a jurisdiction of interest such as that of the Pharmaceutical Benefit Advisory Committee (PBAC) in Australia (Commonwealth of Australia 2016), where the seminal paper ‘Frankenstein’s Monster or the Vampire of Trials’ (O’Brien 1996) takes centre stage.

In general, marrying coverage and comparability principles are required to avoid biases in divining how cost effectiveness presentation and summary measures under uncertainty can be robustly applied. These principles are illustrated with two-strategy comparison for modelled analysis in Chap. 3, while robust presentation and summary measures for more than two-strategy comparison are identified and illustrated in Chap. 8 for multiple strategies and additionally with multiple outcomes in Chap. 10. Problems of partialisation and failure to reflect community values are also pointed to as particularly important considerations in prevention and health promotion strategies in complex community settings, issues which Chaps. 4 and 12 expand on.

Chapter 4 explores some of the challenges faced when undertaking health economic analysis in comparing prevention and health promotion strategies in complex community settings such as schools and palliative care settings with multiple domain comparisons, and some principled approaches and methods developed to address these challenges. Evaluating community-based primary prevention programs makes clear that the principles and evaluation approach to health system decision making need to consider community population impacts over time. Conventional within-study cost effectiveness and extrapolated modelling methods are shown to struggle within typical short-term evaluation time frames to appropriately assess or tractably capture or model community acceptance or the diffusion of impacts over time in populations across community networks conditional on health promotion strategy acceptance. Hence, the need is shown for alternative evaluation methods in navigating coverage (scope and duration) and comparability of the acceptance, diffusion and incremental impact of prevention and health promotion strategies. The research of Shiell and Hawe, pointing to the value of assessing network multiplier impacts from investment on community activity, is illustrated as a more robust and appropriate approach to informing decision makers of the longterm acceptance and success of community-based health promotion and prevention interventions. In modelling terms, such multipliers and their trajectory over time represent the key prognostic factors, or surrogates, for long-term acceptance and success of community-based health promotion and prevention programs and their network impacts over time. Multiplier methods for assessing complex interventions are illustrated in evaluating the Stephanie Alexander Kitchen Garden National Program (SAKGNP), a health promotion and primary prevention program undertaken in primary schools (Eckermann et al. 2014; Yeatman et al. 2014).

The research of McCaffrey et al. (2010, 2013, 2015) is also highlighted in Chap. 4 as enabling robust comparison of multiple outcome domains under uncertainty and illustrated in greater detail with associated methods in Chap. 10. Multiple outcome domain comparisons are shown to be valuable in many settings to consider diffuse outcomes beyond single health metrics that inform wider community utility functions but also alternative values and domain aspects of utility within health. This is particularly the case in areas such as palliative care where domains such as finalising affairs and process of death are not amenable to being integrated with survival time and hence unable to be incorporated into quality-adjusted life years. Further, even within a quality-adjusted life year (QALY) framework, significant value to decision makers in many circumstances arises from being able to explicitly present multiple events or effects underlying QALY estimates and robustly consider their joint uncertainty. Such analysis allows the potential for baseline risk of effect and/or utility weights for states or domains of effect to differ across populations and jurisdictions, as well as over time.

In Chap. 5, optimal decision making in relation to evidence-based reimbursement of technologies based on their incremental cost effectiveness (net benefit) under uncertainty, is shown to be inextricably linked to research decisions. Frequentist approaches to trial design such as the use of type I error, type II error and minimum significant difference to power hypothesis tests don’t consider or reflect the expected value or expected cost of information and hence are unable to efficiently design trials or optimally inform such decisions. Bayesian methods are shown to enable joint optimisation of research and reimbursement decisions with robust estimation of expected value and cost of further research to decision makers’ conditional on critical decision contexts given prior uncertainty in incremental net benefit and as a function of trial size and designs.

Nevertheless, to estimate the distribution of INB and undertake meaningful value of information (VOI) analysis in any jurisdiction of interest, unbiased estimates of incremental costs and effects (following Chaps. 2, 3 and 4) and a meaningful threshold value for effects are required. Hence, Chaps. 2, 3 and 4 should be considered alongside Chap. 11 in deriving a robust estimate of where the INB distribution lies given local decision contexts before undertaking VOI analysis such as that in Chaps. 5, 6 and 7. That is, an unbiased estimate of the expected value of INB is primary to informing societal decision making under the Arrow-Lind theorem (Arrow and Lind 1970) before consideration of uncertainty. While estimating expected INB is the key information decision makers require to assess reimbursement decisions, it also informs the location of tail distribution and associated estimation of expected value of sample information (EVSI) and any opportunity cost of delaying a decision to adopt while research is undertaken.

Value of information (VoI) principles and methods enabling optimisation of expected net gain (expected value less costs) of local trial design and decision making are identified and illustrated in Chap. 5. Importantly, central limit theorem (CLT)-based VOI methods presented are shown to be both:

• (i) Simply applied in estimating expected value of actual trial designs (expected value of sample information) given estimates of mean cost and effects of their variance and covariance; and
• (ii) Allow for relevant decision contexts that jurisdictions face in estimating expected value and cost to make these decisions locally, including rate of recruitment, follow-up and analysis time, opportunity cost and option value of delay and imperfect implementation.

Hence these CLT methods are shown to satisfy Occam’s Razor in realtion to VOI methods (Eckermann et al. 2010), enabling simple optimisation of ENG under relevant decision contexts, providing the necessary and sufficient conditions to locally inform decisions including:

• (i) Is further research for a specific HTA potentially worthwhile?
• (ii) Is a given research design worthwhile?
• (iii) What is the optimal research design?
• (iv) How can funding best be prioritised across alternative research proposals?

The ability to optimise ENG, while allowing for key decision contexts in addressing these questions, is particularly suggested to better inform research grant allocation bodies who have mission statements emphasising ‘value for research dollar’, ‘efficiency in research design’ and ‘research making a difference to practice’. From a researcher perspective, research designed to address decision making (DM) uncertainty and relevant DM contexts in maximising value relative to cost or ENG from limited budgets connect with decision making and funding bodies underlying objectives. Hence, VOI principles and methods enabling optimising ENG return on research should also increase research chances of success given the centrality of these factors to research funder aims, mission and objective statements, as well as budget-constrained expected impact on policy and implementation.

In Chap. 6, the methods for optimally and efficiently informing joint research and reimbursement decisions locally identified in Chap. 5 are extended to allow for optimal global trial design and local decision making across jurisdictions (Eckermann and Willan 2009). The ability to adopt and trial in jurisdictions as part of a global trial is shown to be particularly advantageous in moving from the local to global setting and avoiding opportunity costs of delay while obtaining best evidence globally. Optimally designed global trials allow promising technologies to be adopted early in jurisdiction to avoid opportunity costs of delay for societal decision makers and manufacturers alike. Such advantages of jurisdictions adopting and trialling with promising therapies arise provided evidence translates from other jurisdictions who undertake research, creating appropriate requirements and incentives for evidence coverage, which are made explicit in optimal global trial design and early adoption assessment.

In Chap. 7, the methods for optimal societal decision maker trial design and joint research and reimbursement decision making in Chaps. 5 and 6 are extended to allow for pricing under uncertainty locally (Willan and Eckermann 2012) and risk sharing in the case of jurisdictions who adopt as part of an optimal global manufacturer trial design (Eckermann and Willan 2013). Optimally designed global trials with explicit consideration of evidence translation in trial design are shown to allow earlier adoption of promising programs or technologies while this evidence is collected, with the ability to feasibly adopt and trial with translatable evidence. Further, the greater strength of evidence from larger trials expected a priori is also expected to result in improved implementation (Willan and Eckermann 2010). Such optimal global trials also provide the ability to feasibly and meaningfully risk share for jurisdictions who adopt and trial, with price changes able to be informed by prospective randomised controlled trial (RCT) evidence from the global trial and local evidence of performance of the new technology in practice (Eckermann and Willan 2013).

The ability to feasibly adopt and trial and risk share in such optimally designed global trials better aligns societal decision maker and manufacturer research interests for best research design globally and evidence translation across jurisdictions. Further, such optimal global trial designs also provide an option for feasible collection of RCT evidence for existing strategies which have already been adopted, key to informing opportunity cost and health shadow price assessment, as highlighted in Chap. 11 and policy options in Chap. 12.

Chapter 8 moves beyond two-strategy cost effectiveness analysis to allow for multiple strategy cost effectiveness comparison, presentation and summary measures. Such multiple strategy comparisons are increasingly important with multiple treatment modalities, diagnostic and treatment options and combinations of customised strategies such as genetic testing and initiatives towards individualised care. When comparing multiple strategies, presentation on the cost disutility plane and use of expected net loss (ENL) curves and frontiers are shown to overcome limitations of methods for two-strategy comparison on the C-E plane with CEA and NB curves (Eckermann and Willan 2011; Eckermann et al. 2008).

When comparing multiple strategies, the optimal strategy for comparison is not fixed, as in two-strategy comparisons, but rather changes across replicates and/or threshold values. Flexible axes on the cost disutility plane explicitly addresses this, overcoming problems of fixed axes on the cost effectiveness plane and associated confounding of graphical inference (Eckermann and Willan 2011). Similarly, the expected net loss (ENL) statistic and associated summary measures enable flexible while consistent comparison of differences in expected net benefit with the optimal strategy in any given replicate at any threshold value. Hence, ENL curves and frontiers are shown to overcome problems of CEA curves and frontiers not presenting differences in expected net benefit and the fixed nature of the comparator with incremental net benefit statistics.

Consequently, for multiple strategy comparisons, ENL curves and frontiers are illustrated to fully inform asymptotically risk neutral societal decision making under the Arrow-Lind theorem. If societal decision makers are somewhat risk averse, ENL curves and the ENL frontier provide primary evidence of expected values for making decisions which can be supplemented by appropriate uncertainty evidence. Such evidence is highlighted as needing to be derived from bilateral CEA curves between potentially optimal strategies of interest to prevent confounding of probabilities from other strategies. Further, the ENL frontier identifies both the strategy minimising ENL (equivalently maximising expected net benefit (ENB)) across strategies at any given threshold value and the per-patient potential value of future research. That is, the ENL curve also represents the expected opportunity loss that could be avoided with perfect information and hence the expected per-patient value of perfect information. Thus, the ENL frontier makes explicit the link between optimal reimbursement and research, further supporting the joint nature of research and reimbursement decisions locally and globally, as highlighted in Chaps. 5, 6 and 7.

Chapter 9 shows how the advantages of comparing multiple strategies consistent with maximising net benefit on the cost-disutility plane in Chap. 8 naturally extend to efficiency measures across providers in practice consistent with maximising net benefit. The net benefit correspondence theorem (NBCT) providing the robust theoretical framework underlying methods in Chaps. 8, 9 and 10 is derived. In efficiency comparisons in practice, the NBCT is shown to uniquely provide explicit and joint consideration of the value and costs of quality of care in efficiency measures consistent with maximising net benefit. This overcomes problems of conventional efficiency measures in practice, such as cost per case-mix-adjusted admission in hospitals, implicitly including cost of quality while ignoring the value of quality of care and hence creating incentives for cost minimising quality of care.

More generally, the one-to-one correspondence of the NBCT underlying efficiency comparison with radial properties in cost-disutility space is shown to provide distinct advantages over alternative specifications (Eckermann 2004; Eckermann and Coelli 2013) in enabling:

• (i) Identification of net benefit maximising peers over threshold value for effects where they maximise NB;
• (ii) Practice and policy relevant net benefit (economic) efficiency of providers and decomposition into technical, allocative and scale efficiency consistent with maximising net benefit; and
• (iii) Shadow price for service quality across industry behaviour without requiring prices for admissions.

Importantly, coverage and comparability conditions of the NBCT are also shown to provide an accountable framework to prevent cost-shifting and cream-skimming incentives in practice (Eckermann 2004; Eckermann and Coelli 2013). These explicit coverage and comparability conditions continuously support evidence- based approaches to joint accountability for cost and quality including risk adjustment and standardisation methods (Eckermann et al. 2009, 2011) and data linkage and/or modelling of expected effects beyond service to a common meaningful time point (e.g. 30 days or 1 year beyond admission in hospital).

The NBCT as a generalised method can more generally be applied to efficiency measure with these advantages in any health, care, service or industry setting where maximising net benefit is the appropriate economic objective. Further, radial properties on the cost-disutility (C-DU) plane enable robust comparison, presentation and summary measures for as many domains of effect as appropriate, as highlighted in Chap. 10 following the research of McCaffrey et al. (2013, 2014, 2015).

Chapter 10 shows how the framework presented in Chaps. 8 and 9, for optimal comparison across multiple strategies or providers’ costs and effects on the incremental cost-disutility plane and expected net loss curves and frontiers, naturally extends to multiple effect comparisons. Radial properties on the cost disutility plane enable robust comparison of multiple outcomes under uncertainty, providing distinct advantages over cost consequences analysis. That is, allowing for uncertainty in joint consideration of multiple effects and cost effectiveness analysis summary measures to avoid inferential problems of partial analysis with single effect comparison and summary measures. Summary measures of ENL planes and surfaces and cost effectiveness analysis (CEA) planes developed by McCaffrey (McCaffrey et al. 2010, 2013, 2015) are shown to have further distinct advantages over conventional methods in presenting cost effectiveness across multiple effects and potential threshold values for multiple domains of effect.

Chapter 11 addresses the requirement of net benefit assessment across joint research, reimbursement and regulatory decisions (Chaps. 2, 3, 4, 5, 6, 7, 8, 9 and 10), for threshold values for effects that reflect the opportunity costs (best alternative actions) for relevant decision contexts in the jurisdiction of interest to enable budget-constrained optimisation. The research of Pekarsky (2012, 2015) is highlighted in deriving the health shadow price of reimbursement (adoption and financing) actions for investments with net incremental cost (NW quadrant on the C-E plane) under characteristic health system allocative and displacement inefficiency conditions. The best alternative action to adopting a new technology financed with displacement of programs (ICER = d) is adopting the most cost effective expansion of existing programs (ICER = n) financed by contraction of the least cost effective existing program (ICER = m) leading to a health shadow price for effects of

where the subscript c refers to the prevailing economic context in the jurisdiction of interest.

Importantly, this health shadow price reflects conditions of allocative inefficiency (n < m) and displacement inefficiency (d < m) characteristic of current health systems that can be improved with optimal decision making. The implications of the health shadow price in providing a pathway to allocative efficiency and addressing market failure in provision of evidence for n, d and m are discussed following Eckermann and Pekarsky (2014).

Shadow prices are also considered for the less usual case, on the south-west (SW) quadrant, where new investment is expected to lead to health system cost savings while being potentially less effective. The opportunity cost of decisions to invest in new technology on the SW quadrant, expected to generate net funding relative to current practice over time while some potential health loss, is shown to differ qualitatively as well as quantitatively (Eckermann 2015). If the budget is free to contract, strategies on the SW quadrant should be compared with the least health-reducing way of generating funds for the health budget, and hence the health shadow price reflects an ICER of m. However, where budgets are fixed from going down as well as up, then funds raised by such cost-saving technologies are required to be spent on adoption of other programs. In that case, the health shadow price on the SW quadrant in generating funds (fi_f) is shown to be derived equating returns of funding generated with the cost-saving technology and adopted with ICER a and that of the best alternative fund generating and adoption actions, leading to:

Where adoption is efficient (a = n), then p_f = m, while if adoption is inefficient (a > n) as with threshold based on displaced services, then p_f is greater than m, or can even be required to be dominant to be optimal, as illustrated for the UK with current adoption thresholds. In general, a kink in the economically meaningful threshold value is shown to arise under characteristic health system conditions of allocative and displacement inefficiency, where the threshold value is higher in the SW relative to NE quadrant (Eckermann 2015). The extent of this kink reflects the degree to which there is allocative and displacement inefficiency.

Chapter 12 highlights application of health economic principles and methods to address the challenge of budget-constrained successful ageing of baby boomers with publicly provided universal access health systems (Eckermann 2014a, b; Eckermann et al. 2016; Eckermann and Sheridan 2016) in Australia and internationally. This points to the need for reform that addresses historical inefficiencies across the spectrum from prevention to palliative care including:

• (i) Community age and dementia-friendly policies to successfully age while minimising the need for aged care and nursing home facilities in line with community health promotion and prevention considerations from Chap. 4;
• (ii) Dementia-friendly aged care and nursing home design and care, illustrating the key need for better use of factor priced environmental design approaches to better care for and meet community needs and preferences;
• (iii) Effective factor-priced promising palliative care options to address palliative care primary preferences for key palliative domains - finalising affairs in community of choice while minimising family and carer distress - identified in Chaps. 4 and 10 - in particular policies for optimising net benefit of medicinal cannabis cultivation and program provision and a promising reformulation of 5-FU; and
• (iv) Extending NBCT efficiency measures from Chap. 9 to funding mechanisms in providing active incentives for budget-constrained health and aged care system net benefit optimising quality of care rather than for minimum cost per service quality of care, cost shifting and cream skimming with current case-mix funding methods.

Implications are drawn in each case for optimal policy direction and options and methods that should be adopted to support better joint research, reimbursement and regulatory societal decisions made locally and internationally.

Chapter 13 concludes, showing how a principled approach to health economic evaluation and research can optimise community objectives under resource and budget constraints, but only where key bigger picture structural issues are jointly addressed for research, reimbursement and regulation (pricing, performance monitoring and funding). Optimisation and robust analysis with health economic-related decision making requires satisfying coverage and comparability principles in addressing research, reimbursement and regulatory decisions in HTA and practice. The need to systematically address critical weaknesses of the current political economy in research, reimbursement and regulation biasing towards new technology and away from better use of existing technology is identified. The failure of community preferences to be reflected in resource allocation and policy making in key areas such as palliative and end of life care are also highlighted.

Optimal global trials with coverage of evidence translation and the ability to adopt and trial with use of the NBCT to monitor performance in practice while providing evidence to enable robust risk share are suggested as a first best solution that overcomes many otherwise intractable joint decision making and political economy issues. In addition to optimising joint research, reimbursement and regulation decisions across jurisdictions globally, such designs in avoiding opportunity costs of delay while providing best evidence for decision making also better align societal decision maker and provider (non-patented programs, strategies or technologies) or manufacturer (patented products) interests. Critically such global trials would also enable an optimal pathway to providing evidence for existing or new technologies required to inform health shadow prices in any jurisdiction and optimise budget constrained reimbursement decisions and pricing of new technology given best alternative actions. Importantly, this provides a pathway towards allocative and displacement efficiency with appropriate research, reimbursement and regulatory incentives.