Measuring Progress Toward Achieving Hemoglobin A_1c Goals in Diabetes Care: Title and subTitle BreakPass/Fail or Partial Credit

Despite increasing recognition of the effect of diabetes on the health of the US population, there has been only modest improvement in glycemic control among most demographic groups from the 1990s to the early 2000s regardless of health care setting.¹ Performance measurement should assess quality of care, which is defined by the Institute of Medicine as the degree to which health services for individuals and populations increase the likelihood of desired health outcomes and are consistent with current professional knowledge.² Consequently, hemoglobin A_1c (HbA_1c) control is an important intermediate outcome for measuring the quality of care provided to the 21 million Americans with diabetes. However, considerable debate exists concerning which HbA_1c levels are appropriate for public reporting when quality assessments are compared across multiple health care systems.

It is important to distinguish between practice guidelines and performance measures. Although efficacy trials are sufficient for guideline recommendations and inform accountability measure development, effectiveness studies, technical considerations (bias, variability in practice, and definition of population at risk), and policy perspectives are also pertinent. However, their relative importance may differ among different stakeholders. For example, persons with diabetes and their advocates might hope that health care systems will provide all the resources prescribed while minimizing out-of-pocket cost; managers of health plans might seek to minimize their costs, especially when there is high turnover of health plan members so the benefits of the more expensive optimal treatment will accrue to a different health plan; and policymakers need to balance mandates and patient-physician autonomy.

Differences of opinion on how to weight these factors have led to the selection of different HbA_1c thresholds by various coalitions and organizations. For example, the National Quality Forum³ measures are consistent with those recommended by the National Diabetes Quality Improvement Alliance⁴ : a public accountability measure for poor control (HbA_1c >9% or not measured) and an internal (ie, to the health plan) quality improvement measure to assess excellent/optimal control (HbA_1c <7%). The Ambulatory Care Quality Alliance⁵ has chosen to include only the poor-control HbA_1c measure for pay for performance, but the National Committee for Quality Assurance (NCQA) has recently adopted the less than 7% measure for public accountability in addition to the existing poor glycemic control measure for all persons aged 18 to 75 years without exclusion criteria.⁶ Although there is agreement that a greater than 9% HbA_1c level reflects poor glycemic control in all individuals even without risk adjustment,^{3 - 6} there is disagreement about how to define an optimal measure that can compare quality of care provided by physicians in different populations across plans. In this Commentary, we propose 3 questions that organizations and coalitions contemplating adoption or endorsement of a less than 7% threshold measure for public reporting or pay for performance should consider:

Is the generalization of an HbA_1c threshold measure of less than 7% to all persons with diabetes justified by currently available evidence and concordant with nationally recognized guidelines?

Epidemiological evidence from the Diabetes Control and Complications Trial (DCCT)⁷ and the United Kingdom Prospective Diabetes Study (UKPDS),⁸ the major clinical trials demonstrating efficacy of glycemic control in type 1 and type 2 diabetes, indicates that while the relative risk reduction of adverse outcomes is linear over a wide range, the absolute risk reduction in adverse events is log-linear; that is, the number of adverse outcomes prevented by improving HbA_1c decreases as baseline HbA_1c is reduced. Thus, health benefit (ie, absolute risk reduction) is greater by decreasing HbA_1c from 9% to 8% than from 8% to 7%. Also, a reduction in composite diabetes-related complications accrues after years of control; the mean duration of treatment was 10 years in the UKPDS, although benefits occurred earlier in the DCCT for asymptomatic microvascular outcomes. Furthermore, the macrovascular benefits of tight glycemic control for individuals with type 2 diabetes, who comprise 90% to 95% of all persons with diabetes, remain to be defined by the ongoing National Heart, Lung, and Blood Institute–funded Action to Control Cardiovascular Risk in Diabetes (ACCORD) study and the Veterans Affairs Diabetes Trial. In addition, disease models based on UKPDS data suggest that the lifetime benefit of decreasing HbA_1c from 7.9% to 7.0% is about 2-fold greater for individuals aged 45 to 54 years than for those aged 55 to 64 years and about 5-fold greater than for individuals aged 65 to 74 years at time of diagnosis.⁹ Moreover, the majority of even these highly selected clinical trial participants failed to achieve and maintain HbA_1c levels less than 7%, thus precluding broad inferences about generalizability to individuals with coexisting medical illnesses and psychiatric conditions that would have excluded them from the trials.

Consequently, nationally developed diabetes guidelines emphasize the need to individualize HbA_1c targets based on age, life expectancy, comorbid conditions, patient preferences, and medication adverse effects.^{10 - 12} Thus, a dichotomous performance measure of less than 7% for all adults with diabetes is not concordant with existing practice guidelines.

Is achieving and maintaining an HbA_1c threshold of less than 7% for most persons with diabetes sufficiently actionable by health care systems and by practicing physicians to justify its use for public accountability?

Effectiveness studies suggest that the ability to achieve an HbA_1c level of less than 7% in a population of patients is at least partially under the control of health plans and physicians. Considerable evidence indicates that clinical inertia, defined as failure of clinicians to alter therapy in the face of clear indications for change, hinders treatment of glycemic control above an HbA_1c threshold level of 8%.^{13 - 14} However, a recent meta-analysis suggested that quality improvement strategies have at best a modest effect and largely benefit persons with the highest initial HbA_1c levels, not those with levels less than 8%.¹⁵ Greater adherence to a less than 7% measure may be easiest to achieve by focusing resources on diagnosing incident diabetes, aggressive treatment of individuals whose HbA_1c is marginally greater than 7%, or both, rather than the more resource-intensive but more beneficial effect of addressing HbA_1c levels between 8% and 9%.

Achieving and maintaining HbA_1c levels of less than 7% have proved difficult in patients with a longer duration of diabetes, even with the addition of a third oral agent.¹⁶ Hypoglycemia remains a major limiting factor in achieving tight glycemic control with insulin. A meta-analysis indicated that only about one third of persons receiving either glargine or isophane insulin could achieve HbA_1c levels of less than 7%.¹⁷

Furthermore, a 7% HbA_1c value obtained from a routine clinical laboratory could represent an actual value range of 6.5% to 7.5% because of total error (laboratory imprecision as well as bias from the National Glycosylated Hemoglobin Standardization Program Target).¹⁸ Although this imprecision would not affect mean population HbA_1c levels, greater reliance on the most recent HbA_1c measurement for intensification of insulin treatment, rather than multiple HbA_1c levels and monitoring results, could result in an increased risk of severe hypoglycemia, especially in patients with other underlying conditions and polypharmacy. Without a surveillance system in place, it is not possible to prospectively track possible individual and societal harms.

Factors outside of health plan control are also critical. For example, differences in prescription coverage among individual patients because of employer or patient choice may affect adherence to therapy,¹⁹ and socioeconomic status influences glycemic control even in managed care plans.²⁰ These factors might not be distributed equally among health plans or individual physicians. The benefit and marginal cost-effectiveness of using newer agents, especially for HbA_1c values between 7% and 7.9%, is unknown. However, based on the use of agents available in 2001, the cost per quality-adjusted life-year (QALY) for decreasing HbA_1c levels from 7.9% to 7% was $71 000 for individuals aged 55 to 64 years and $144 000 for individuals aged 65 to 74 years.⁹ The cost-effectiveness of intensive glycemic treatment would be even less favorable if costlier medications were initiated at HbA_1c levels closer to 7%.

Thus, the extent to which optimal target levels can be achieved in selective clinical trials and with structured quality improvement efforts should give pause to the face validity of using “optimal” values for the purpose of public reporting. Unintended consequences could include targeting of individuals with measurements marginally above target values, selection biases, patient safety, and less regard for patient preferences.²¹

Are there other approaches to measuring glycemic control that can accurately assess the benefit to population health, compare plans for public reporting, and inform quality improvement?

The American Diabetes Association–NCQA Bridges to Excellence Program uses a weighting system based on predefined percentages of patients meeting the measure. However, the differential weighting is arbitrary: achieving HbA_1c levels less than 7% counts half as much as achieving levels less than 9%.²² In addition, physicians who believe they have more complex patient populations can request risk adjustment. Clinician recognition and payment can be achieved even without achieving the target proportion of patients with HbA_1c levels less than 7%. Moreover, clinician adherence to the program performance measures is not publicly reported.

Alternate approaches need to be considered. For instance, risk adjustment methods could be applied, although there are no currently validated and published methods for HbA_1c. Moreover, patients could be risk-stratified, such as by duration of disease or complications, or goal-stratified with different thresholds. Alternatively, a threshold measure and absolute improvement could be combined, although this approach would require collection of longitudinal data. Assessing clinical inertia is difficult for individuals receiving insulin because dosage changes may not be easily identified.¹⁴

Another potential approach is to use a continuous and weighted measure that assesses progress toward achieving an optimal HbA_1c level rather than any specific threshold and that could be initially implemented using the last HbA_1c value. This approach is more concordant with the clinical epidemiology of glycemic control, permits greater discretion in balancing benefits, harm, and preferences in clinical practice, and, thus, may be viewed as better simultaneously reflecting both physician/plan performance and achieved population health. For example, statistical models predict that for a 55-year-old patient with diabetes, decreasing the HbA_1c level from 10% to 9% will reduce lifetime risk of blindness by 1.3%, whereas decreasing the HbA_1c level from 8% to 7% would reduce the risk by only 0.4%, a 3-fold difference.²³

However, there remains the issue of simultaneous accounting of the effect of the different diabetes complications—blindness, dialysis, amputation, stroke, myocardial infarction, or premature death—over the remainder of the patient's expected life span. Health-adjusted life-years (HALYs), of which QALYs are a subset, are widely accepted by researchers and policymakers as a composite measure by which such disparate outcomes can be summarized both within a study or population and in long-term modeling. This approach has been widely used in cost-effectiveness analyses of diabetes outcomes.⁹ A comparison of rankings of 141 health care facilities within the Veterans Health Administration based on population QALYs achieved from controlling HbA_1c from 7.9% to 7.0% vs rankings based on population adherence to a less than 7% threshold showed that 1 in 5 facilities in the best and worse deciles markedly changed ranking despite high overall correlation.²⁴

There is ongoing scientific debate regarding the use of QALYs to model diabetes outcomes for policy decisions.²⁵ How should models account for serious comorbid conditions that affect the life expectancy of actual patients within a plan? How should duration of diabetes be incorporated? How can longitudinal cohorts be used to avoid biases introduced by case finding of new diabetes and out-migration of patients?

Despite the challenges of developing a continuous and weighted measure, assessing improvement toward optimal HbA_1c, rather than a threshold optimal target, is a more precise assessment of individual clinical benefit and achieved improvement in population health. This approach would reward health plans and clinicians for focusing improvement strategies on risk factor control at higher HbA_1c levels than levels marginally above the “optimal” target. While research continues, a public accountability level for HbA_1c of greater than 9%, as currently endorsed by the National Quality Forum, along with a threshold of less than 7% for internal quality improvement, on balance, seem reasonable.

The desire to set and measure an “optimal goal” for HbA_1c for all patients is understandable. However, while accountability measures assess population health and plan/physician performance without the opportunity to defend “exceptions,” clinicians must rely on guidelines, their judgment, and individual context to treat persons with diabetes one at a time. The potential risks and benefits of performance measures must be balanced with respect to overtreatment or undertreatment of individual patients to maximize intended population health benefits while minimizing individual patient-level harms and avoiding unintended consequences.