Principal-Agent Problem

Principal-Agent Problem, Hidden Action and Incentives in Health Care

In health care the principal-agent problem is the gap between what a patient, insurer or government wants and what the doctor, hospital or purchaser acting for it actually does, a gap that persists because the principal cannot fully see the agent's effort or knowledge. It matters because almost every resource decision in a health system is delegated. Patients delegate treatment choices, insurers delegate the certification of need, and taxpayers delegate spending through a chain of agencies. The design of the contracts themselves is covered under contract theory and the payment methods under provider payment.

Who is the principal and who is the agent in health care

Jensen and Meckling define an agency relationship as a contract under which a principal engages an agent to act on its behalf, delegating some decision-making authority. If both parties maximise their own utility, there is good reason to expect that the agent will not always act in the principal's best interests. Health care contains several such relationships at once.

In 1963 Arrow described the clinical case. Because the patient cannot completely enforce standards of care, the patient must delegate much freedom of choice to the physician and rely on a relationship of trust. He also noted that, by certifying the need for treatment, the physician acts as a controlling agent for insurers, which makes the doctor an agent of two principals, a situation Blomqvist analysed as the doctor as double agent. Smith and colleagues extended the idea to whole systems. Every transfer of funds, from the population to financing agencies, to hospitals and within hospitals, involves a principal-agent problem, and at each stage society's objectives for the health system can be lost.

The perfect agent and the ways agency fails

The benchmark is the perfect agent. McGuire's review of physician agency cites Pauly's definition, under which the physician helps the patient demand exactly the care the patient would have chosen with the physician's information and knowledge. Laffont and Martimort point out that differing objectives alone are not enough for a problem; if the agent held no private information, the principal could write a contract that controlled the agent completely. The problem needs both diverging interests and information the principal lacks.

McGuire identifies two forms of physician diligence that patients cannot verify, which the agency literature calls hidden information and hidden action. These are revealing what the doctor knows about the patient's condition, and taking the steps needed to find it out. A third gap is the doctor's fixed ability, such as diagnostic acumen. A doctor is an experience good and outcomes also depend on luck, so patients learn about quality slowly; a simple learning model McGuire describes implies that poor doctors are valued too highly and good ones too low. Hidden action is the supply-side form of moral hazard, and private information held before a contract is agreed is adverse selection; both are forms of information asymmetry.

Doctors are not usually modelled as pure income maximisers either. McGuire reports that most health economics models put the patient's benefit into the physician's utility function, so concern for the patient is traded off against income, and that the empirical evidence supports non-profit-maximising behaviour.

What the problem costs the principal

Jensen and Meckling define agency costs as the sum of the principal's monitoring expenditures, the agent's bonding expenditures and the residual loss, which is the money value of the principal's welfare lost when the agent's decisions still diverge from the principal's interest. They use monitoring broadly, to include budget restrictions, compensation policies and operating rules. In a health system, utilisation review, clinical audit and payment rules fall on the monitoring side, and the residual loss is the value lost when care still differs from what a perfect agent would have provided.

Arrow saw one theoretical way out. Under ideal insurance the physician would be paid according to the patient's benefit, and the patient would have no concern with the informational inequality because payment would be by results. McGuire notes that Dranove and White, and later Mooney and Ryan, judged this infeasible because health status generally cannot be written into a contract.

Responses through incentives, monitoring, reputation and professional norms

No single instrument closes the agency gap, because each one acts on a different part of it. Health systems therefore combine several responses.

• Incentive design. The payment rule changes which choices pay. McGuire notes that if good outcomes can be paid on they can motivate doctors, the basis of performance contracting; fee-for-service, capitation and blended rules are compared on the provider payment page.
• Monitoring and control. Smith and colleagues concluded that adequate control depends on how patient entry to hospital is controlled, how hospitals are paid for additional patients and how hospital management controls physicians.
• Reputation. Where an outcome cannot be paid on but the patient can see it, McGuire notes that a doctor who does better is rewarded through patients returning and recommending the doctor.
• Professional ethics and trust. Arrow argued that the physician's social obligation for best practice is part of the commodity sold, even though the buyer cannot fully inspect it. McGuire cites Hillman's view that most physicians act in the patient's interest when the medical decision is clear-cut, and that financial incentives may matter more where the correct decision is not.

Worked example on the price of an imperfect quality indicator

The numbers are illustrative. A purchaser wants a general practice to carry out a thorough annual review for each patient with diabetes, which costs the practice £30 of extra staff time per patient. The purchaser cannot see the review, only whether a recorded indicator is met. The indicator is met for 70% of patients if the review is thorough and 45% if it is not. The practice is risk-neutral and cannot be paid less than nothing, so the purchaser offers a bonus $b$ for each patient meeting the indicator. The practice chooses the thorough review only if it pays at least as well, which gives the condition below.

$$p_H\, b - c \;\ge\; p_L\, b \quad\Rightarrow\quad b^{*} = \frac{c}{p_H - p_L} = \frac{30}{0.70 - 0.45} = 120$$

where $p_H$ is the probability that the indicator is met after a thorough review, $p_L$ the probability without one, $c$ the cost of the review per patient and $b^{*}$ the smallest bonus, in pounds, that makes the thorough review the practice's best choice.

1. The expected payment. The purchaser expects to pay 0.70 × £120 = £84 per patient for a review that costs £30. The difference of £54 is a rent to the practice, calculated as follows.

$$R = p_H\, b^{*} - c = \frac{c\, p_L}{p_H - p_L} = \frac{30 \times 0.45}{0.25} = 54$$

where $R$ is the expected payment above cost per patient. The rent equals what the practice would earn without any review, 0.45 × £120 = £54, because the indicator is often met by chance.

2. A list of 200 patients. The purchaser expects to pay 200 × £84 = £16,800 for reviews costing 200 × £30 = £6,000, so £10,800 is the cost of paying a practice that cannot be charged a fee on a noisy indicator rather than on the review itself.

3. A sharper indicator. Suppose an audited measure is met for 90% of patients after a thorough review and 20% without. The bonus falls to 30 ÷ (0.90 − 0.20), or £42.86, the expected payment to 0.90 × £42.86 = £38.57 and the rent to 30 × 0.20 ÷ 0.70 = £8.57 per patient, or £1,714 for the list.

4. Interpretation. The incentive works in both cases, but the weak indicator costs the purchaser about six times as much in rent. With pay only when the indicator is met, the rent equals $c/(p_H/p_L - 1)$, so it falls as the indicator better separates thorough from cursory reviews (the ratio rises from 1.56 to 4.5). Holmström's informativeness principle, as summarised in the Nobel committee's background paper, makes a related general point, that pay should depend on signals that tell the principal more about the agent's action. The rent exists because the practice cannot be paid less than nothing. If the purchaser could also charge a fixed fee of £54 per patient, the same bonus would still buy the review and the rent would disappear. The committee's paper notes that a risk-neutral agent with sufficient financial resources can be given full incentives through a franchise contract, paying a fixed fee and keeping the realised benefit, but that wealth constraints may be binding.

Measurement problems of multitasking and gaming

The example assumes one task. Holmström and Milgrom showed that when an agent has several tasks, incentive pay also directs attention between them, so paying for a measured task draws effort away from unmeasured ones. The desirability of incentives for any one activity falls as other competing activities become harder to measure, and a fixed salary can be the optimal contract.

The UK Quality and Outcomes Framework showed both sides in its first years. Gravelle, Sutton and Ma, studying 916 Scottish practices, describe a scheme with 65 clinical indicators at the time, measured as ratios, where practices may exception report patients for whom an indicator was not intended. They estimate that in practices below the upper threshold in 2004/05, 0.87% of patients might have been inappropriately exception reported in 2005/06, and that reported prevalence also responded to the incentives. The same study found that practices could have treated 11.8% fewer patients without losing income, which the authors read as partial altruism.

Using the agency lens in economic evaluation and HTA

Agency shapes the evidence that economic evaluations use. Observed utilisation reflects the doctor's choices as well as patient demand, so it cannot be read as revealed patient preference without care; supplier-induced demand is the contested case. An evaluation of a payment reform or a pay-for-performance scheme needs to model behavioural responses, including coding and exception reporting, as well as the intended change in care. Recorded indicator data need checking before use as model inputs, since Gravelle and colleagues concluded that Quality and Outcomes Framework prevalence reports may not be a reliable epidemiological resource. The agency chain also reaches the decision-maker, since a purchaser spends, and a health technology assessment body advises on spending, funds held on behalf of a population.

Boundaries and common misreadings

The principal-agent problem is the relationship-level problem; information asymmetry is the condition that makes it possible. Contract theory studies the agreements written in response, while market failure is the wider welfare diagnosis within which Arrow placed the informational inequality between doctor and patient.

Three misreadings recur. The problem does not assume dishonest doctors; it arises whenever interests differ and information is private, and the evidence points to partial altruism. Influence over a patient is not proof of inducement, and McGuire stresses that moving a patient towards what the patient would choose if informed is useful agency. Stronger or more numerous incentives are not always better, because multitasking and gaming can make weak incentives optimal. Mooney and Ryan add that basic agency theory raises more questions than it answers for the doctor-patient relationship, and that more research into both parties' utility functions is needed before optimal incentive structures can be designed.

Sources

• Arrow KJ. Uncertainty and the welfare economics of medical care. American Economic Review. 1963;53(5):941-973.
• Blomqvist Å. The doctor as double agent: information asymmetry, health insurance, and medical care. Journal of Health Economics. 1991;10(4):411-432.
• Gravelle H, Sutton M, Ma A. Doctor behaviour under a pay for performance contract: further evidence from the Quality and Outcomes Framework. CHE Research Paper 34. York: Centre for Health Economics, University of York; 2008.
• Holmström B, Milgrom P. Multitask principal-agent analyses: incentive contracts, asset ownership, and job design. Journal of Law, Economics, and Organization. 1991;7(Special Issue):24-52.
• Jensen MC, Meckling WH. Theory of the firm: managerial behavior, agency costs and ownership structure. Journal of Financial Economics. 1976;3(4):305-360.
• Laffont JJ, Martimort D. The Theory of Incentives: The Principal-Agent Model. Princeton: Princeton University Press; 2002. Introduction.
• McGuire TG. Physician agency. In: Culyer AJ, Newhouse JP, eds. Handbook of Health Economics, Volume 1A. Amsterdam: Elsevier; 2000:461-536. https://doi.org/10.1016/S1574-0064(00)80168-7
• Mooney G, Ryan M. Agency in health care: getting beyond first principles. Journal of Health Economics. 1993;12(2):125-135.
• Royal Swedish Academy of Sciences, Committee for the Prize in Economic Sciences in Memory of Alfred Nobel. Oliver Hart and Bengt Holmström: contract theory. Scientific background. 2016.
• Smith PC, Stepan A, Valdmanis V, Verheyen P. Principal-agent problems in health care systems: an international perspective. Health Policy. 1997;41(1):37-60.