Pay for performance in health care: strategic issues for Australian experiments

In recent years, pay-for-performance (P4P) strategies have attracted considerable interest in the United States,1-5 the United Kingdom,6 Australia,7 Canada8 and other Western countries (Box 1). Their key attribute is a defined change in reimbursement to a clinical provider (individual clinician, clinician group or hospital) in direct response to a change in one or more performance measures, as a result of one or more practice innovations (Box 2).

Currently, across the US, over 170 P4P programs (or quality incentive payment systems) are in various stages of implementation in hospitals and group clinics, in both public9 and private sectors,10 covering 50 million beneficiaries. In 2004, the UK National Health Service (NHS) launched the General Medical Services Contract – Quality and Outcomes Framework, which gives family practitioners up to a 25% increase in income if various quality indicators are met.6 In Australia, Medicare’s Practice Incentives Program targets quality in general practice,7 and, in Queensland, a P4P program targeting public hospitals is being piloted from July 2007.11

P4P is popular because of the propositions that:

• Traditional approaches to optimising care, including education and certification, do not appear to guarantee minimum standards;

• Current quality improvement efforts are slow at reforming systems of care; and

• Few financial incentives exist for clinicians and managers to modify the status quo and reward high performance.12

However, despite its rising popularity, how effective is P4P, and what are the key determinants of its success? Are its benefits sustainable over the longer term? Are there unintended adverse effects?13 What forms of P4P might be applicable to Australian settings? In this article, I attempt to provide some answers based on empirical evidence obtained preferentially from controlled evaluations,14 and propose a framework under which P4P might evolve in Australia.

In a review of 17 studies (including 12 controlled trials),15 of which 13 examined process-of-care measures, five of the six studies of doctor-level financial incentives and seven of the nine studies of provider group-level incentives showed positive (improvement in all measures) or partial (improvement in some measures) effects on measures of quality, while four studies suggested unintended effects. Quality gains for most studies were modest (eg, a 3.6% increase in rates of cervical cancer screening, a 6.8%–7.4% increase in immunisation rates, and a 7.9% increase in rates of smoking status determination).

The US Centers for Medicare and Medicaid Services (CMS)/Premier Hospital Quality Incentive Demonstration (HQID) program recently reported 2-year results, comparing changes in process-of-care measures for acute myocardial infarction, heart failure and pneumonia between 207 P4P hospitals and 406 matched control institutions.1 After adjusting for differences between groups in baseline performance, condition-specific volumes and other hospital characteristics, P4P was associated with composite process improvements of 2.6% for acute myocardial infarction, 4.1% for heart failure and 3.4% for pneumonia.

Other uncontrolled, before–after studies suggest improvements in selected quality metrics for other large-scale programs.2,5,16 While initial reports on the NHS experiment appear encouraging (median of 83% achievement across six practice domains),6 no baseline analysis was undertaken, with anecdotal reports suggesting a pre-P4P mean performance improvement of 60%–80% that had been rising over previous years.17

Cost-effectiveness studies are few in number. An incentives package for improving nursing home care and access saved an estimated US$3000 per stay.18 Another study of P4P for cardiac care across eight hospitals estimated that 24 418 patients received improved care, translating to between 733 and 1701 additional quality-adjusted life-years (QALYs), with cost per QALY ranging from US$12 967 to US$30 081.19 A study of physician-level payments for diabetes care within a large health maintenance organisation reported savings of US$4.7 million due to better quality care over 2 years compared with a program cost of US$2.2 million.20 However, on the negative side, program expenditure within the first year of the NHS experiment vastly exceeded (by US$700 million) the US$3.2 billion additional funding allocated over 3 years, suggesting that quality improvements may not be commensurate with the rapid rise in costs.17

Most controlled trials have assessed preventive care and care of chronic diseases within primary care settings, with fewer targeting acute hospital care.15 High-volume, high-cost conditions with demonstrated practice variation and a strong evidence base to guide practice improvement are preferred as initial targets. However, others argue that minority populations with special needs and demonstrably suboptimal care should not be neglected.21

Should incentives be given to individual clinicians, clinician groups (multi-clinician associations, from single clinic or hospital department to multi-site collaborations or networks), hospitals, or all three? Currently, 14% of US physician P4P programs target individual clinicians alone, 25% target both individuals and groups, and 61% target groups alone.16 Results of the NHS trial,6 and of studies reporting the largest absolute increases in quality measures,15 suggest that paying individuals or small clinician groups, rather than large groups or hospitals, may be the better strategy. However, deciding which of several clinicians caring for a patient with multiple illnesses should be assigned primary responsibility for care is problematic,22 and detracts from using P4P to entice large groups to share risk and invest in population-level systems of care improvement.

Non-financial strategies for improving care quality (such as clinical guidelines, audits and feedback) yield median improvements of 6%–13% in process-of-care measures.23 Do they provide an equal or better return on investment compared with financial incentives when used alone, or do both strategies in combination yield synergistic effects? In one randomised trial involving 60 physicians, bonus payments combined with feedback led to a 25% increase in documentation of immunisation status compared with no change with feedback alone.24 In addition, non-cash incentives such as professional pride, recognition, opportunities to do more challenging work and the removal of administrative burdens may, in some circumstances, be equally motivating and should not be underrated.

The “dose–response” relationship for incentives in terms of size, frequency and duration remains uncertain. Some P4P programs pay as little as $2 per patient and exert impact, while others offer bonuses of up to $US10 000 per practice and have no effect.16 The 25% income increase possible for NHS family doctors is postulated as the reason for the observed high levels of performance.6 In contrast, data from US health plans, which on average pay 9% of base income as bonus payments, show no consistent relationship between incentive size and response.16 One private-sector survey found incentives had to average 5%–20% of income for individual physicians and 1%–4% for hospitals.16 Payments in the CMS/Premier HQID, for example, can be up to a 2% increase for high performance versus up to a 2% loss for poor performance.1

It is possible that changes to the base funding formula, rather than add-on incentives, could facilitate higher-quality care,12 especially within capitated or salaried environments. Under such circumstances, care improvements, which may incur additional “downstream” costs but no increased revenue, could financially negate incentives (which tend to be smaller than those under fee-for-service).25 Another debate centres on whether incentives should involve reallocation of fixed funds from low performers to high performers or elicit additional (“new”) funding. Clearly, payments must compensate for the incremental net costs of undertaking all actions required of P4P programs.

Quality measures should ideally be clinically relevant, reasonably stable over periods of 2–3 years to allow serial comparisons (yet responsive to important new changes in practice), feasibly ascertainable, accurate, and actionable with high impact.

Eligibility for incentive payment may rest on one or more performance thresholds:

• absolute — achievement of a predefined absolute threshold (eg, 75% of eligible patients receiving specific interventions);

• relative — improvement over baseline performance by a specified margin (eg, a 30% increase), or ranking measures (often as percentiles) relative to some external benchmark;

• all cases — payment for each instance of high-quality care regardless of overall performance (as in the NHS experiment); or

• some combination of these.

If absolute thresholds are to be used, should they be set at 75%, 85%, or 95%?

Also, will thresholds pertain to:

• individual instances of care — percentage of eligible patients who receive specific interventions;

• composite care — percentage of total instances in which eligible patients receive all of several interventions; or

• all-or-nothing care — percentage of patients who each receive all the interventions they are eligible to receive?30

Most programs to date use a combination of individual and composite care measures, and those that use both relative and absolute thresholds31 appear to show more consistently positive results and exhibit the largest improvements in individuals or groups with the lowest baseline performance. The danger of ceiling effects with use of absolute thresholds was seen in the small gains in cervical cancer screening, mammography, and glycated haemoglobin testing within one program involving 172 physician groups — 75% of the bonus payments went to those already at or above the target 75th percentile.32

The clinical dictum “first, do no harm” underlies the need for comprehensive evaluation frameworks to accompany P4P programs. Unless clinicians are convinced that performance measures are adequately risk-adjusted and take account of patient preferences, they may avoid very sick or challenging patients, or engage in other “gaming” strategies, such as reclassifying patient conditions or “ticking the box” even when care has not been provided or has been incompletely administered.13 In the NHS experiment, the strongest predictor of increases in reported achievement was a higher rate at which patients were excluded from the program (0.31% increase for every 1% increase in exception reporting).6 Clinicians or institutions serving disadvantaged populations, in whom performance thresholds may be difficult to achieve, might see falls in revenue. A predominant focus on process-of-care measures may promote inappropriate over-treatment in patients with multiple diseases.33 Financial incentives may further undermine morale and professional altruism and erode holistic patient care.

In one study of 22 clinician groups and nine hospitals, views on incentives varied widely depending on institutional culture, structure and stability, community context, quality measurement issues, nature and size of the incentives, and sustainability of interventions.34 Among 35 health plans which had initiated P4P, providers expressed concerns about administrative burdens of customised programs, absence of standardised performance measures, and potential for conflicting financial incentives.35 Physicians in California, while supporting P4P in general, wanted accurate and timely performance data, more patient time and staff support, and greater access to colleagues.36

In light of the above discussion, I propose the following framework under which P4P programs might evolve in this country, as opposed to the employer-subsidised managed care and competitive market environment of the US.

• A phased approach is suggested, which starts with “pay-for-participation” schemes in which participants focus on the development and testing of robust, standardised and preferably nationally consistent performance measures and systems of measurement that are integral to P4P.11,16 Only then should there be moves towards “pay-for-performance” which, initially, might be a bonus to base funding for care improvement (which may include one-off investments in information technology upgrades), but which eventually becomes embedded into funding formulas (such as casemix funding for hospitals and fee-for-service items in private care), with financial penalties in cases of consistent and clearly evident poor performance.

• Pilot demonstration programs adopting this phased approach could be funded, by state or federal government, or private health funds (as appropriate), over a 3–4-year period, in different geographical (metropolitan, urban, rural/remote) and clinical (primary/community care, hospital care) settings in both public and private sectors. Funding could be focused on several well defined, common clinical conditions to provide proof of concept and assess the relative impacts of different design elements.37

• Program design should: (i) target hospitals and clinician groups (arranged as clinical service teams, networks or collaborations), not individual clinicians; (ii) reward all high-quality care, not just that reaching predefined thresholds; (iii) include predominantly process measures with a select few unambiguous outcome measures (such as mortality) that, where feasible, not only relate to discrete episodes of care of individuals, but to cycles of care over time for whole populations (utilising linked national databases such as hospital discharge abstracts, the Pharmaceutical Benefits Scheme and the Medicare Benefits Schedule); and (iv) provide flexible payments for both capital purchases, particularly clinical information technology, and provider incentives linked to future performance guarantees (and associated penalties if these are not met).

• All P4P programs should have an appropriate governance structure comprising clinicians, health managers, quality improvement methodologists, and data managers/analysts, with advice being sought from health economists and epidemiologists.

While P4P programs can potentially improve care, they are not without problems if poorly designed or implemented in the absence of clinician involvement, or not subjected to ongoing evaluation and modification as circumstances demand. Key design elements discussed here deserve consideration if P4P programs are to be supported by clinicians, successful in meeting their objectives, and capable of providing useful answers to research questions.