Escaping the Hemoglobin A1c-Centric World in Evaluating Diabetes Mellitus Interventions

Abstract

Any intervention in patients with diabetes must consider its effect on both the incidence of hypoglycemia and hemoglobin A1c. Yet, there is no single metric that expresses these key factors simultaneously. Such a composite metric would permit clinicians, regulators, manufacturers, payers, and researchers to more easily evaluate the merits of an intervention as well as enable the comparison of qualitatively different interventions. This article proposes a composite metric, the hypoglycemia-A1c score (HAS), as the basis for a more comprehensive approach for the stakeholders in diabetes treatment to better understand how an intervention affects diabetes management. The article also demonstrates how additional parameters such as effects on weight, quality of life, and costs could be included in such a scoring system.

The traditional outcome metric for determining the efficacy for all studies since the Diabetes Control and Complications Trial study has been the change in hemoglobin A1c (A1c).¹ Since that time several new classes of medications (rapid and long-acting analog insulins, inhaled insulin, biguanides, GLP1 receptor agonists, DPP-4 inhibitors, and SGLT2 inhibitors) and better technologies (continuous glucose monitors, insulin pumps, sensor-augmented pumping with or without low-glucose threshold suspend features) have emerged. Many of these newer interventions can improve hemoglobin A1c with either a decreased or neutral effect on hypoglycemia rates or produce a decreased hypoglycemia rate without an increase in A1c. Less invasive approaches such as psychoeducational interventions have also proven effective in improving overall glycemic control.²

Most researchers and clinicians recognize that A1c insufficiently captures the quality of glycemic control because it gives no information about the prevalence and severity of hyper- and hypoglycemia or glycemic variability. Indeed, Cryer has stated that hypoglycemia is the most important limitation in achieving better glycemic control in patients with both type 1 and type 2 diabetes.³ The importance of hypoglycemia in understanding glucose data has been further emphasized by Rodbard, who suggests that glucose data should be semilogarithmically transformed so that a “penalty” for hypoglycemia is imposed.⁴ Bergenstal and colleagues (in the context of continuous glucose monitoring) have suggested a “dashboard” display of glucose data that provides information about glucose exposure, time in range, glycemic variability, and hypoglycemic episodes at various thresholds.⁵

What has been missing in these and other attempts to capture the salient features of glycemic control is a composite metric that reflects the key elements of glycemic control, A1c and severe hypoglycemia (SH). According to the American Diabetes Association (ADA), this means that for “healthy adults with diabetes, a reasonable glycemic goal might be the lowest A1c that does not cause severe hypoglycemia, preserves awareness of hypoglycemia, and results in an acceptable number of documented episodes of symptomatic hypoglycemia.”⁶ The ADA and Endocrine Society definition of SH is “an event requiring the assistance of another person.”⁷ Similarly, the American Association of Clinical Endocrinologists recommends that “glycemic control aimed at normal (or near-normal) glycemia may be considered . . . if it can be achieved without substantial hypoglycemia or other unacceptable adverse consequences” in patients with type 2 diabetes.⁸ Thus, at a minimum we should express both the rate/severity of hypoglycemia and A1c, the hypoglycemia-A1c score (HAS), as the basis for a combined outcome metric of effectiveness of any intervention in patients with diabetes. Such a composite metric should not be limited to just these 2 factors but should include other “patient-centered” factors including but not limited to weight change, cost, and/or quality of life (see below). Ideally, it would include confidence intervals so that it provides a clearer understanding of the significance of a specific score.

This approach is similar to a number of other ways to evaluate outcomes in medicine including FRAX scores which use a multivariate algorithm to assign fracture risk⁹ and error grids which have been used to assess clinical accuracy of glucose monitoring devices.^9-11 Similar approaches are used in the sports world where since 1973 the National Football League has employed quarterback ratings (using touchdowns, interceptions, etc), which correlate with wins.¹² The HAS would be expected to help diabetes care providers, regulators, and payers better understand what is best for patients by allowing a comparison of efficacy across the different types of interventions. For example, the prescription of hypoglycemia awareness training or a device may be more beneficial overall than the addition of another pharmacologic agent. In addition, the HAS will allow investigators to assess the effect of such a metric on long-term complications. I propose such a metric with the understanding that (1) the categorical limits set forth are somewhat arbitrary although based on current practice, (2) further development is required to refine this approach, and (3) ultimately, the HAS may be developed using continuous variables.

The highest HAS intervention would be one resulting in lowering of A1C with an improvement in the rate of SH; a lower HAS would be given to one in which the A1c is improved but there is no change in the SH rate; and the lowest HAS would be assigned to an intervention that resulted in no change in A1c with an increased SH rate or one that worsened both the A1c and SH rate The intervention could be a new drug, a new technology, or a new educational approach. The base case HAS, shown in the middle section (B) of Table 1, uses an improvement in A1c of >0.4% as recommended by the Food and Drug Administration (FDA) to show a significant effect on hyperglycemia¹³ and either a reduction of the SH rate (≥30% based on the recommendations of the 2005 ADA Working Group¹⁴) or an absolute SH rate per year. A good to excellent intervention would have an HAS ≥75 (green). If more outcome variables were to be used to refine the score (eg, effect on weight, HAS-W; the effect on blood pressure, HAS-BP; the effect of lipids, HAS-L; the cost of the intervention, HAS-C; or the effect of quality of life, HAS-QOL), bonus or penalty points would be assigned accordingly. An example of how weight changes would affect the HAS is shown in Table 1; the effect of a weight decrease is shown in section A and an increase in section C. In all cases, a penalty is assessed for increasing the hypoglycemia rate using the principles discussed by Rodbard.⁴ Applying the results of diverse interventions using this approach is instructive. For example, comparing 2 pharmaceutical interventions added to metformin, canagliflozin has a higher score than glimepiride (95 vs 50) despite substantial and equivalent A1c reductions based on the weight reduction versus weight gain and lower severe hypoglycemia rate.¹⁴ Since baseline hypoglycemia rates determined, these scores may not accurately estimate the overall glycemic effect. In the Juvenile Diabetes Research Foundation Continuous Glucose Monitoring (CGM) study, the scores differ depending on the baseline A1c.¹⁵ CGM results in a score of 70 for those with a baseline A1c of ≥7% and 80 for those <7% because of the reduction in the severe hypoglycemia rate in the latter group (15). These scores assume a neutral effect on weight, an outcome that was not reported. The low glucose threshold suspend pump scores an 80 based on a 31% decrease in severe hypoglycemia, no change in A1c and no change in body weight.¹⁶ A similar score of 80 is obtained in a psychoeducational intervention for those with hypoglycemia unawareness (assuming no weight change) because of a large reduction in severe hypoglycemia and no change in A1c.²

Table 1.

Composite Score Using 3 Variables: Change in A1c, Absolute Severe Hypoglycemia Rate (Events per 100 Person-Years), and Percentage Change in Severe Hypoglycemia Rate, and Weight.

		0 events/100-person years	1-10 events/100-person years	10-20 events/100-person years	21-30 events/100-person years	>30 events/100-person years
OR
A		Hypo rate better by ≥30%	Hypo rate better by 15%-30%	Hypo rate unchanged	Hypo rate worse by 15%- 30%	Hypo rate worse ≥30%	Weight
	A1c better by ≥1%	100	95	85	70	50	DECREASE
	A1c better 0.5%-0.9%	95	90	80	60	40
	A1c unchanged to better by 0.4%	90	85	75	40	25
	A1c unchanged to worse by 0.4%	70	70	65	25	15
	A1c worse by 0.5%-0.9%	65	60	55	20	10
	A1c worse by ≥1%	50	40	25	10	0
B	A1c better by ≥1%	90	85	75	60	40	NEUTRAL
	A1c better 0.5%-0.9%	85	80	70	50	30
	A1c unchanged to better by 0.4%	80	75	65	30	15
	A1c unchanged to worse by 0.4%	60	60	55	15	5
	A1c worse by 0.5%-0.9%	55	50	45	10	0
	A1c worse by ≥1%	40	30	15	0	0
C	A1c better by ≥1%	80	75	65	50	30	INCREASE
	A1c better 0.5%-0.9%	75	70	60	40	20
	A1c unchanged to better by 0.4%	70	65	55	20	15
	A1c unchanged to worse by 0.4%	50	45	40	5	0
	A1c worse by 0.5%-0.9%	45	50	35	0	0
	A1c worse by ≥1%	30	20	5	0	0

This framework is admittedly arbitrary at this point, it uses SH as a “worst case” metric, and the specific scores are both subjective and lacking confidence intervals. However, this proposal is intended to start a conversation about how to best assess, prescribe, and ultimately pay for current and future interventions in diabetes management. Not surprisingly, the devil is in the details. With respect to hypoglycemia, for example, the improvement of ≥30% in hypoglycemia was given a broad definition by the ADA’s Working Group which said that it could be a decrease in the proportion of patients with hypoglycemia, the hypoglycemia event rates, or both.¹⁷ Indeed, various metrics have been used in reporting hypoglycemia in drug and technology trials. For example, recently reported SH rates in patients with type 1 diabetes (variably defined) include episodes per patient-year or patient-month, episodes per 100 patient-years, percentage of patients with severe episodes/year, percentage with glucose ≤60 mg/dl or ≤70 mg/dl, symptomatic events per year, and area on the curve for nocturnal hypoglycemic events in mg/dl/min. Given the discordant time frames of A1c (3 months) and the various hypoglycemia metrics, a hypoglycemia rate per patient-months and/or an area under the curve might work best for shorter duration studies. Rates in patients with type 2 diabetes are generally lower and would require different rates for comparison. Thus, for a combined metric to be practically applied hypoglycemia must be strictly defined and standardized in new clinical trials as previously suggested.⁵ A consensus conference could be convened by professional societies, pharmaceutical and device manufacturers, and the FDA to work out the details. Finally, it will be important to develop a tool that is interactive and flexible so that different stakeholders can use it to understand how an intervention impacts their concerns.

In summary, I believe that the current system of A1c-centricity falls substantially short of real-world diabetes management where clinicians integrate multiple factors into their daily therapeutic decision making. While this kind of clinical judgment will be continue in the near term, it behooves us to think ahead to the day when this “mental modeling” can be more precisely expressed in a multivariable metric. I propose a new metric, the HAS, as the basis for expressing overall glycemic control using both hypoglycemia and A1c in addition to other variables. This sophisticated and patient-centric approach will hopefully reframe the discussion so that new interventions can be assessed and approved by regulatory agencies not only on the basis of a change (or lack thereof) in A1c. This may also help clinicians and third-party payers better understand the comparative effectiveness of the wide variety of options available for treatment of patients with both type 1 and type 2 diabetes.