In a partial capitation contract, the remunerated party’s income comprises a fixed component f determined by the ex-ante characteristics of the population for whom care responsibility is assumed, and a variable component v for each unit of output q (for example, consultations) produced at a cost c (profit = f-q(c-v)). In a full capitation contract (v = 0, equivalent to budget funding of entities or salaried remuneration of individuals), the recipient’s income is invariant to the number of consultations provided. By decoupling remuneration from cost drivers, the higher is f and the lower is v, the greater the recipient’s reliance upon income from the prospective capitation payment than upon income determined by the number of consultations delivered. The recipient now faces incentives to reduce costs in order to increase profitability — for example, producing fewer consultations or finding cheaper ways of producing them. Behavioural change ensues because undesirable behaviour is no longer rewarded (Milgrom and Roberts, 1992). Health care capitation contracts, as a form of “supply-side cost sharing” (Ellis and McGuire, 1986), explicitly share financial risk otherwise borne by funders (for example, governments) and insurers with providers.
However, capitation contracts have complex effects on practitioner behaviour, as they share two specific types of financial risk between the purchaser and the provider: ‘controllable’ risk and ‘random’ risk. ‘Controllable’ risk relates to uncertainties of which the expected consequences can be anticipated in the aggregate, although are not predictable in respect of any specific occurrence. Cost consequences can be managed through the use of specific contract terms or institutional forms. Controllable risk is most efficiently managed when those capable of controlling its extent bear the financial costs of undertaking the undesirable actions or accrue the benefits of undertaking desirable actions. The specific controllable risk addressed by primary health care capitation is the provider propensity towards inefficient supplier-induced demand under subsidised fee-for-service remuneration (Zeckhauser, 1970). As providers bear at least some of the cost of their demand-raising choices, under capitation the number of unnecessary consultations reduces. Furthermore, providers are rewarded for engaging in consultation-reducing (and hence cost-reducing) preventative activities (Crampton, Sutton and Foley, 2001).
By contrast, ‘random’ risk relates to those factors for which the practitioner assumes either partial or complete financial responsibility via the risk-sharing contract, but is powerless to control. One such risk is an exogenous event causing an unpredictable and uncontrollable increase in demand for consultations (for example, a localised epidemic) — ‘exogenous’ risk. A second risk relates to differences between ex-ante anticipated demand for care within a population, and actual demand recorded ex post — ‘random demand variation’ risk. The aggregation of individuals’ demand risks into large insurance pools in order to reduce the cost of individual uncertainty by compensating from the pool in the event of the insured event materialising is an example of ‘controllable’ risk, more efficiently managed by an insurer with a diverse portfolio across which to spread the costs than by the risk-averse individual (Arrow, 1963). However, a discrepancy remains between anticipated demand, upon which premia or capitation payments are based, and actual demand which imposes actual costs.
When a large pool (for example, the population) is disaggregated into smaller pools (for example, patient lists), it is most unlikely that demand in each of the small pools will be the population average. At best, without explicit cream-skimming, half the pools will incur more demand than the population average, and half less. The smaller the size of the pool relative to the population, the greater will be the average absolute variation between the pool average upon which costs depend and the population average upon which remuneration is based — that is, the greater the extent of ‘random demand variation’ risk. Capitation contracts let by large risk-pool managers (insurers and government funders) to a large number of smaller pool managers (service providers) fragment the efficiency-raising aggregate risk pool. The capitation contract transfers not just the amount of ‘controllable’ risk desired to be shared with the practitioner in order to alter the practitioner’s behaviour, but also a share of responsibility for ‘random demand variation’ and ‘exogenous’ risk. The United States Health Care Financing Administration considers capitated primary health care physician groups to be at substantial financial risk from random effects if they have fewer than 25,000 registered patients (Hagen, 1999).
On the one hand, capitation increases efficiency by mitigating the effects of unnecessary supplier-induced costs. On the other hand, by transferring responsibility for managing uncontrollable risk to more risk-averse practitioners, with less scope for diversification and efficient risk management than the funder/insurer (Milgrom and Roberts, 1992), efficiency is reduced. Providers’ incomes now become subject to factors over which they have no control or capacity to anticipate. In effect, providers assume an insurance role which they do not carry under fee-for-service contracts. The overall efficiency of capitation relative to fee-for-service depends upon whether the gains from improved management of ‘controllable’ risk exceed the losses from less efficient ‘random’ risk-management practices.
The effects of ‘random risk sharing’ on provider incomes may be either positive (that is, fewer consultations than anticipated/remunerated are provided) or negative (more consultations than anticipated/remunerated demanded) but, importantly, is beyond the provider’s control. The amount of risk shared is crucial for contract efficiency. The stronger the capitation contract incentive (that is, the higher is f and the lower is v), the greater the proportion of ‘random’ risk that is shared, and the more the practitioner’s income comes to depend upon uncontrollable factors. If the ‘random’ effects are large compared to the ability to manage income via the ‘controllable’ factors, then irrespective of the amount of effort exerted pursuing the desired activities, the practitioner’s income becomes essentially a lottery. The incentive to pursue desirable behaviour is ‘crowded out’ by the random effects. Desired activities are not pursued. Rather, the practitioner will exert effort instead in activities that maintain or increase income given the amount of ‘random’ risk assumed (for example, ‘cream-skimming’ using information unknown to the funder to ensure that the patients for whom care management responsibility is assumed are more financially ‘desirable’) (Holmstrom and Milgrom, 1991).
United States evidence suggests that substantial changes in practitioner behaviour have been induced with only very weak capitation incentive contracts (Ma and Riordan, 2002), whereas only about 20 per cent to 25 per cent of the variation in individual demand for health care services can be predicted using individual characteristics such as age, gender, income, ethnicity and past consumption of care (Robinson, 2004; Newhouse, 1996). These data suggest that there are very real ‘random’ financial risks for practitioners associated with using strong capitation incentives to manage ‘controllable’ risks when risk-adjusted fixed capitation payments are based on only a small range of individual demographic characteristics.
If the uncontrollable and unpredictable events were truly random, then over time, losses incurred by a provider in ‘bad’ years will be cancelled out by ‘profits in ‘good’ years. This is true of ‘exogenous’ risks. However, Howell (2007) posits that primary health care capitation contracts are exposed to serial correlation of profitability between years as a consequence of repeated transacting between a single provider and the same patients with persisting, but unknown (and hence uncompensatable via the prospective payment) risk factors that have ongoing effects upon the demand for care — that is, ‘random demand variation’ risk. The consequence is the emergence of habitually profitable and habitually loss-making practices. The correlation problem is further exacerbated by primary care practices generally being composed of only a small pool of individuals, where the persistent, atypical demand patterns of a few individuals (for example, underpinned by unknown and unknowable genetic predisposition) can have a significant effect on long-term practice profitability.
Correlation factors are less problematic in secondary and tertiary care provision as practitioner interaction with specific individuals tends to be episodic rather than ongoing, and the catchment from which demand derives is generally very much larger than a typical primary practice pool (Scott, 2000; McGuire, 2000). These effects are mitigated only by the recreation of larger pools amongst which to share the risks. If an individual general practitioner serves a group of typically between 1200 and 2000 patients, then based upon the United States evidence above, even with low-strength capitation incentives, financially viable primary care patient pools require the aggregation of the lists of between 12 and 21 practitioners. The greater the capitation incentive strength, the greater the number of practitioner lists required to be merged to counter the effect of ‘random’ risk sharing.
Figure 1 illustrates. Assume it costs an average practice c to deliver an average primary care consultation (including all overheads and a fair return on the human capital and time invested by the practitioner) and that the practice delivers q consultations. Average revenue received per consultation is f/q+(v-c). Under a fee-for-service contract charged at cost (f=0; v=c), the practitioner makes no profits and no losses on all consultations delivered (that is, ‘breaks even’ at all values of q). The number of consultations delivered, q, is determined solely by consultation demand and the practitioner’s willingness to work. Under a capitation contract, however, the practitioner receives v<c for each consultation delivered. The maximum number of consultations that the practitioner will deliver is q=Q, where the practice ‘breaks even’ financially. If demand arising from the population for which capitation is received results in fewer than Q consultations being delivered, the practice makes a ‘windfall’ profit. If q>Q, the practice makes a financial loss.
A capitation contract incentivising desirable behaviours looks for providers delivering more than Q consultations through their own cost-causing choices (that is, ‘controllable risk’) to reduce the number to Q in order to remain financially viable, or for those with costs higher than c to reduce them to c. However, the same contract will financially penalise those practitioners with costs c and not over-producing, where random factors, rather than behavioural choices, lead to more than Q consultations being delivered. If these practices are to remain financially viable, they must reduce costs below c (for example, through shorter consultations), ration services (for example, institute waiting lists) or pass the extra costs on in some other way (for example, institute a patient payment y in addition to the capitation contract payment v, which is typically paid by the insurer or funding body).
The sharper the capitation contract incentive (that is, the lower is v and the higher is f), the steeper is the slope of the average revenue curve (Figure 1), the greater the profits and losses, and the greater the additional costs that must be borne by the patients of ‘unprofitable’ practices (q>Q). Ironically, ‘unprofitable’ practices meeting all ‘controllable’ risk expectations make losses in the first place because their patient base has higher demand (that is, is ‘sicker than average’). Capitation contracts result in sicker-than-average patients bearing more of the consequences of ‘random’ risk sharing, in either lower care quality (shorter consultations, waiting lists) or higher prices (y charged to them) than the ‘healthier-than-average’ patients of ‘profitable’ practices.
Figure 2 illustrates the effect of ‘unprofitable’ practices being able to charge patients y to recover ‘random’ risk costs. Assume that the ‘most unprofitable’ practitioner not over-producing at cost c must deliver Q1 consultations to meet demand at f and v. By charging patients y per consultation to break even, the practice will still produce Q1 consultations (technically, the practice’s average revenue curve moves upward). As primary care practitioners have some market power due to product differentiation arising from patient preferences for the attributes of individual practitioners and repeated transacting between the same individuals (Scott, 2000; Dranove and Satterthwaite, 2000), within bounds such price increases can be undertaken without invoking the loss of large numbers of patients to other practices. However, if one practice can charge a patient fee y without losing patients (or the failure to charge y has no substantial effect upon a given practitioner’s demand), all other ‘profitable’ and ‘less unprofitable’ practices will also be able to charge y. ‘Unprofitable’ practices delivering between Q and Q1 consultations now make profits instead of losses, and the ‘profitable’ practices producing fewer than Q consultations make higher profits than before. The total number of consultations produced is higher than anticipated by the capitation contract based upon remuneration from f and v alone.
If practitioners can charge patients y, there is no need to engage in cost- and service quality-reducing activities such as rationing or shortening consultations (these actions are commonplace where patient charging is prohibited; for example, in England’s NHS). However, all patients face an additional financial cost in lieu of the quality reduction that would otherwise be borne by patients of ‘unprofitable’ practices, regardless of whether their practice would have been required to engage in such activities in order to break even. All practices now make profits, with those facing least demand becoming substantially more profitable than if patient charging was prohibited.
Furthermore, the ability to charge patients eliminates the justification for using capitation to alter practitioner behaviour in the first place. If practices can levy charges to cover costs of ‘random’ risk allocation, they can also levy charges to cover the additional costs arising from their ‘controllable’ risk choices. There is no financial penalty from engaging in the delivery of over-many consultations. The number of consultations delivered returns to the level under fee-for-service remuneration, but the total cost of delivering those consultations increases as a consequence of practitioners reaping profits from random demand variation and patient charging that they were unable to appropriate under fee-for-service remuneration. Despite raising the costs of service delivery, the contracts are impotent in constraining practitioner supply. Undesirable practitioner behaviour must now be controlled using costly, overt means such as direct observation or regulation. Ironically, capitation is widely used precisely because it is more cost-effective than monitoring where the desired behaviour is either extremely costly or infeasible to directly observe (such as in third-party purchasing — Milgrom and Roberts, 1992).
Systems enabling patient charging (y) thus result in more costly consultations. They also invoke perverse distributional consequences. The higher costs of risk management are borne only by individuals consuming consultations — that is, the sick. The ‘sicker’ the patient (that is, the more consultations consumed) the greater the contribution towards the higher risk management costs. ‘Healthy’ patients (that is, those consuming no consultations) pay none of the inflated risk-management costs imposed by system design. Patient payment systems thus allocate the higher-than-expected costs in the form of a perfectly risk-rated ‘premium’ per consultation based upon patient health state (or equivalently, as a ‘consumption tax’ imposed on the sick). The ‘well’ are rewarded for their ‘good’ health state by not being required to pay any of the risk-management costs shared with practitioners via capitation contracts and subsequently ‘passed on’ to the sick. Such arrangements are particularly antithetic to the principles of socially motivated insurance schemes, where patient income (via taxation), rather than health state (via patient payments per consultation required), is the preferred metric via which the financial costs of the scheme are allocated.