Cost Effectiveness Analysis and the Design of Cost-Sharing in Insurance: Solving a Puzzle

By Mark Pauly
National Bureau of Economic Research, October 2012


The conventional model for the use of cost effectiveness analysis for health programs involves determining whether the cost per unit of effectiveness of the program is better than some socially determined maximum acceptable cost per unit of effectiveness. If a program is better, the policy implication is that it should be implemented by full coverage of its cost by insurance; if not, no coverage should be provided and the program should not be implemented.  This paper examines the unanswered question of how cost effectiveness analysis should be performed and interpreted when insurance coverage can involve non-negligible cost sharing.  It explores both the question of how cost effectiveness is affected by the presence of cost sharing, and the more fundamental question of cost effectiveness when cost sharing is itself set at the cost effective level.  Both a benchmark model where only “societal” preferences (embodied in a threshold value of dollars per unit of health) matter and a model where individual willingness to pay can be combined with societal values are considered. A common view that cost sharing should vary inversely with program cost effectiveness is shown to be incorrect. A key issue in correct analysis is whether there is heterogeneity either in marginal effectiveness of care or marginal values of care that cannot be perceived by the social planner but is known by the demander. The cost effectiveness of a program is shown to depend upon the level of cost sharing; it is possible that some programs that would fail the social test at both zero coverage and full coverage will be acceptable with positive cost sharing.  Combining individual and social preferences affects both the choice of programs and the extent of cost sharing.

From the Introduction

Of course, the answer to the question of the relationship between cost effectiveness values and cost sharing depends both on the perspective taken and the empirical facts.  So I first outline the simple and correct application of a decision rule to treatment choices based on cost effectiveness in the “binary coverage” setting, when insurance either covers 100 percent of the cost of a given type of care or leaves it entirely uncovered,  and the extra welfarist approach is taken.  I show that this approach usually is based on two assumptions: a single value for expected improvement in health outcomes is to be applied to all patients, and a single monetary value for those expected marginal benefits prevails.  This is the approach, avowedly “extra-welfarist,” much favored at present in the United Kingdom by the NICE advisory body.

However, I then show that opening the door to consideration of cost sharing means that many things, including this perspective, might appropriately be modified.  Modifications are needed if there is heterogeneity in either effectiveness of the treatment across patients or in the values citizens place on health outcomes, and that heterogeneity is determined to be relevant to policy.  I show that the ideal level of “interior” cost sharing depends on whether consumer values are assumed to be relevant, on how much consumer values really vary, and most especially on whether the extent of variation in expected effectiveness across patients is perceived by patients but cannot be known by the insurer.  I briefly consider as well the possibility that social values are variable or uncertain.


These are somewhat discouraging conclusions.  They definitely imply that there is no simple but correct way to move from findings of a typical cost effectiveness study to saying what the coinsurance rate should be for a non-poor population.  The most one could hope for would be a binary decision of whether or not a particular treatment should or should not be covered by insurance with a particular predetermined cost sharing rate (usually but not necessarily zero).  They also imply that the cost effectiveness of a treatment cannot be properly determined unless coinsurance is set at the optimal level.  So, unless there is perfect information to identify heterogeneity of benefits, considerations of consumer demand (in the classic economic sense of the shape of the demand curve) need to be added, regardless of the normative model.

The fundamental problem is the assumption of a uniform benefit of uniform value which is central to the societal cost effectiveness model.  This assumption is presumably made for administrative and expository reasons, not because anyone believes that marginal health benefits are uniform, or that the marginal value of a health benefit (to a consumer or society) is independent of the current level of health or allocation of resources.  It was hard enough to get policymakers to accept the need for considering costs and the need to establish a money value for health outcomes, however arbitrary, and in the United States neither of these concepts is as yet effective.  But paradoxically it might be more feasible to get political acceptance if more attention to reasonable variation in effectiveness and value were explicit rather than suppressed in the analysis.  As always, there is a case to prefer approximating the perfect rather than precisely hitting the imperfect as a method of policy analysis.

Mark Pauly has contributed extensively to the policy literature on the moral hazard of health insurance (the hazard that individuals will obtain care that they do not need if they do not have to pay for it), a concept which has been used to support cost sharing (deductibles, co-payments and coinsurance) as a means to create consumer price sensitivity. Pauly now expands on the concept by applying it to cost effectiveness analysis.

With our very high health care costs, it is inevitable that more attention will be paid to cost effectiveness. Although most cost effectiveness analyses are aimed at whether or not a particular health care service or product provides adequate societal value (measured benefit per unit cost, especially costs that society pays through public programs or private insurance), Pauly now suggests that the value to the individual, as expressed by the level of cost sharing tolerated, should also be introduced in determining cost effectiveness.

First, a few words about cost effectiveness. Most health care professionals do attempt to provide cost effective care. If a very expensive procedure likely would provide no health care value, the practitioner would advise against its use. Likewise, low cost but high volume interventions that are eventually shown to be ineffective also would be abandoned by the practitioner.

Sometimes we simply don’t know whether the benefit is worth the cost. This is where cost effectiveness analysis can be helpful. If a cancer drug regimen that costs $300,000 results in maybe three months of poorer quality life due to side effects, but prolongs life by only three days, most reasonable individuals would decide that this is not cost effective and should not be paid for through our collective funds, whether government taxes or private insurance funds. Three months of hospice is better than three months and three days of therapeutic misery.

An example of a low cost, high volume intervention might be the use of an anti-hypertensive drug that in long term studies showed absolutely no benefit in reducing morbidity or mortality. Obviously that would not be cost effective, and its use would be abandoned.

Okay. So we determine that most medical interventions fall somewhere between 100 percent cost effective and not cost effective at all. Then we are supposed to determine a level of cost sharing for each intervention that would motivate the patient to make a correct decision on whether or not to accept the care based on consumer-directed cost effectiveness decisions that still provide adequate societal value? Come on!

Forget moral hazard. Through cost effectiveness analyses we can determine whether or not medical interventions should be available to patients, and we don’t need to have patients making spending decisions to determine health care value. Even if the benefit may not be uniform between patients, the decisions should be made at the clinical level based strictly on medical benefit and not on cost. Cost decisions should be made at the societal level.

The purpose of cost sharing is to reduce health care spending. Our message earlier this week demonstrated that controlling moral hazard through cost sharing introduces behavioral hazard; that is, decisions that patents make when exposed to out-of-pocket costs as a consequence of accessing health care can be detrimental. Policies that promote detrimental medical decisions are bad policies.

Keep in mind that we spend almost twice as much per capita on health care as the United Kingdom, yet we have ever-increasing cost sharing whereas their system pays 100 percent of the cost of covered services. There are many features of their national health services that result in lower costs, but one that applies to today’s discussion is their application of cost effectiveness and evidence-based analyses through their National Institute for Health and Clinical Effectiveness (NICE). Eliminating ineffective and detrimental care reduces costs, with a net gain in quality.

Just think of what we could have if we established a single payer financing system. We could reduce administrative and clinical waste while eliminating financial barriers to care so that everyone would have access to cost effective and evidence-based, high quality care – care that is really NICE.