Diabetes Control and Lower Weight Is Associated With Statistically Significant Savings

A complication of diabetes

The study is reported here in the American Journal of Managed Care.

Researchers at the western Massachusetts Reliant Medical Group were interested in knowing whether there was any association between control of blood sugar among persons with diabetes and their health insurance claims expense. Using combined data from their electronic health record (EHR) plus insurance claims, they identified all (continuously enrolled) adults with at least one physician encounter between January 2007 through December 2011 for a diagnosis of diabetes, who also were being treated with metformin and a sulfonylurea drug.  This yielded a population of 2044 patients.

This study had some useful benchmark data for other providers with caring for a similar diabetic population. 27% had an A1C goal of less than 7%, 64% had an A1C goal of less than 8%, 33% had a blood pressure less than 140/90 mm Hg, 68 % had a LDL of less than 100 mg/d, and 34% had a body mass index (BMI) less than 30.

After controlling for age, gender, enrollment date, race, payer type and comorbidities, the researchers found that having a target A1c less than 7% (indicating good diabetes control) was associated with $992 per member per year (PMPY) in savings. There were $1445 PMPY in savings for an A1c less than 8%, and $1218 for a BMI less than 30 - all versus persons who did not achieve those goals.  Just controlling blood pressure, or cholesterol levels did not result in statistically significant savings  

Combining a low A1c, blood pressure control less than 140/90, low LDL cholesterol and BMI in various combinations seemed to result in savings in excess of $2000 per member PMPY.  Most of the savings related to diabetes control appeared in the outpatient category, while most of the savings related to BMI appeared in the inpatient category.

The DMCB's take:

1. Not all persons with diabetes may have a "diabetes" diagnosis in the EHR or a bill submitted to an insurance company that uses that particular code. The DMCB likes this study because it's more likely that a person with diabetes will eventually show up in a 5 year period. Plus, the use of metformin and a sulfonylurea medication makes it easier to accurately capture persons with diabetes.

2. This is another in a series of observational studies that infers that persons with target control of blood glucose or who are not overweight experience lower claims expense compared to persons with poor control or who are overweight.

3. What this study does not prove is that blood glucose control or weight control causes low insurance claims expense.  Association does not equal causality any more than "white hair" causes heart attacks.  In order to prove causality, persons with diabetes and similar baseline claims would need to be randomly allocated to good vs. bad control of their blood glucoses (or good vs. bad weight) with prospective and simultaneous comparison of the future expenses over time.  That's called a randomized controlled trial.

4. Remember the ACCORD study?  Persons were randomly allocated to tight vs. very tight control of their blood glucoses and the death rate unexpectedly causally went up for persons with very tight control. The DMCB brings that up because, in the AJMC study described above, less tight control of diabetes (A1c less than 8% vs. the target of 7%) was associated with even greater reductions in claims expense.  Is this further evidence that tight control of diabetes leads to problems?

5.  Unexpectedly, blood pressure or LDL control was not independently associated with lower claims expense.  That's important because Accountable Care Organizations that assume blood pressure and cholesterol lowering pill compliance will result in shared savings may need to reconsider.

Ultimately, while this study doesn't prove that successful population health management for diabetes would save money, it's one more piece of evidence pointing in that direction.  Let the studies continue.

Diabetes, ADVANCE, ACCORD and the Timely Perspective of Dr. Gauss

If you are a regular reader of the Disease Management Care Blog, you already know about ACCORD. This randomized prospective study of tight versus not so tight blood glucose control among persons with diabetes showed an unexpectedly increased death rate associated with tight control. While the preliminary results initially appeared online, the New England Journal of Medicine has now published a full report. The mean A1c was 6.4% in the tight control group and was 7.5% in the not-so-tight control group. While there was a non-statistically significant decline in heart attack, stroke and cardiovascular death in the group with the lower A1c, the death rate was statistically higher: these patients were less likely to have a stroke or heart attack, but it if it did happen, it was more likely to be lethal.

At about the same time ACCORD was announced, there was another study called ADVANCE that was also randomized and prospective and also designed to test the merits of tight versus not so tight blood glucose control. It differed, however, in that patients were initially started on a drug called ‘glicazide.’ In ACCORD, docs were able to choose the drugs. In ADVANCE, there was no difference in death rates between the two groups but the tight control group had fewer kidney complications. The rate of heart attack and stroke between the two groups was not different.

If you are confused by the contrasting results, welcome to the club. As Dr. Cefalu points out in his accompanying editorial, there were differences in the rate at which patients achieved their targeted A1c levels, the medications used,the duration of follow-up and the degree of weight gain. Whether one some or all account for the confusion will be answered in future studies.

The DMCB preferred to focus on the what-does-this-mean-for-diabetes-guidelines commentary by Drs. Krumholz and Lee. Wondering if the differing medications had much to do with the disparate ACCORD and ADVANCE outcomes, they have two recommendations:

1. Treatment targets (for example, the American Diabetes Association recommends an A1c of 7%) need to be paired with recommendations on how to get there. Some drugs may be better than others.

2. Treatment targets need to be tailored to an individual patient’s risk. For example, ADVANCE tells us that an increased pre-existing burden of cardiovascular disease may warrant avoiding tight blood glucose control.

The DMCB cannot disagree with these eminent editorialists, but humbly offers an additional population-based point of view that contrasts with the ‘guideline-ish’ one-patient-at-a-time perspective of Drs. Krumholz and Lee. By the way, the pic above is of Herr Professor Gauss, of population-distribution fame. He'd probably appreciate what follows.

Assuming that an A1c target of 6.4% results in a distribution of A1cs around a mean, it is possible that persons to the left will have an even higher rate of harm. Imagine, then, that docs serving a population in a network are successful in achieving an average state-of-the-art A1c 7%. Assuming there is a distribution around THAT mean, the result is that there will be a fraction of patients that are being exposed to a low A1c, some as low as 6.4%, which ACCORD tells us results in a greater rate of lethal heart attack and stroke.

This potential for an increased risk of harm for some persons in a population-based A1c that otherwise 'looks' good at 7% can be addressed one of two ways:

1) move the curve to the right (say to a mean A1c of 7.5%). Before you blanch at this apostasy, consider that there are other diabetes treatment recommendations that have a more flexible approach to the A1c.

2) narrow the distribution around that mean, so there are fewer persons with low blood glucoses.

Either way, fewer patients are exposed to harm. While the first strategy means there will be patients to the right of the mean with higher A1c, the ‘harm ratio’ of the left versus the right of an A1c strategy of 7.5% is unclear. Of the two approaches, a narrower distribution is probably better.

Either way, the DMCB suspects savvy population-based approaches to care, backed up by the right kind of registries, patient education, disease management and physician incentives (among other things) can result in a better ‘distribution’ of A1cs with fewer deaths than the one-on-one approach advocated by the editorialists.

ACCORD & the Implications for Diabetes Disease Management

The Disease Management Blog doesn’t presume that its readers rely on this corner of the blogsphere for news and information, so you’ve probably already heard that the ACCORD (the catchy acronym stands for the Action to Control Cardiovascular Risk in Diabetes) was suspended. This my blog though, so that doesn’t mean I can’t weigh in with what I’ve learned and offer up some additional speculation through the lens of disease management. Read on if you are so inclined…….

Interesting stuff. Over 10,000 persons were randomly allocated to either tight blood glucose control (target A1c less than 6%: that is VERY aggressive) versus moderate control (an A1c between 7% and 8%, which isn’t bad but doesn’t meet guidelines of the American Diabetes Association or HEDIS). As an aside, there was an additional “2x2 factorial design” that also tests the benefits of 1) aggressive blood pressure control and 2) treatment to increase the “good” or HDL cholesterol. Research subjects began to be recruited in January 2001 in 77 outpatient clinics across the United States and Canada.

What was found? Over an average of 4 years, the researchers noted an increase in the death rate among the approximately 5000 subjects assigned to the A1c less than 6% group versus the 5000 in the other group. This was quite counterintuitive: 257 died in the tight control group, versus 203 in the group assigned to an A1c between 7% and 8%. This was statistically significant and apparently not a function of the types of diabetes drugs used. The portion of the trial on tight blood sugar control has been halted; the other research on blood pressure and HDL is continuing.

To put the calculated excess death rate of 3 per 1000 into perspective, the numbers suggest that a doctor (ensconced in a population-based program of course) would need to aggressively target 333 persons with diabetes down to an A1c of 6% or less to provoke one extra death (for more on number needed to treat go here). Deaths were evenly split between cardiovascular categories and “other” (for example, cancer). Persons assigned to the low A1c group had a lower rate of heart attacks, but the irony is that they were more likely to die if that happened. Participants are being notified of the trial result and the persons in the low A1c group are being reassigned to an A1c between 7% and 8%.

Note this is an “intention to treat” analysis. In other words, the data hasn’t been sorted by the actual A1c. A technically correct interpretation is that trying to get a person with diabetes to an A1c less than 6% is associated with excess mortality. That is slightly different than the conclusion getting a person to an A1c less than 6% is associated with excess mortality. Not everyone in the aggressive control group actually got to an A1c of less than 6%.

Why is this interesting? Older readers with a background in patient care may recall the debate about the “J curve” in essential hypertension back in the 1990s. The moniker “J curve” was used because the plot of BP control on the horizontal axis versus complications on the vertical axis looked like a “J.” Some studies had suggested that lowering the blood pressure “too much” (for example, less than 70 diastolic) among persons with hypertension seemed to be associated with increased mortality. Folks speculated that a lower “head of pressure” in arteries partially blocked with atherosclerosis led to premature clotting/thrombosis and death. The HOT (another catchy acronym – Hypertension Optimal Treatment) Trial found out that was not true and that aggressive lowering of blood pressure does no harm. What’s more, for persons with diabetes, HOT showed aggressive lowering of blood pressure is beneficial.

The preliminary review of the data from ACCORD makes me wonder if the previously obsolete concept of a J curve can be resurrected for diabetes mellitus. If we forego “intention to treat” for a moment and speculate persons in the ACCORD Trial who maintained an A1c between 6% and 7% did “better” than those with an A1c greater than 7%, the question is how did the persons in the less than 6% group do compared to those with an A1c between 6% and 7%? If they did worse, J curve! If they did the same, hockey stick.

Stay tuned. It will take some time for the health services researchers to pull all that apart and present all this in a peer review, transparent forum.

So what do we know?

This is another great example of why we need to uncouple short term process (the incidence of A1c testing) or clinical measures (the A1c results themselves) from the outcome measures that people really care about. And people care about death.

Patients who are engaged in their diabetes care now have even more information to better gauge how they should be treated. They can be counseled that targeting an A1c lower than 6% isn't necessarily in their best interest. However, they should factor in the intention to treat dimensions.

The mortality was observed in an “intention to treat” context. Just because a patient has an A1c less than 6% isn’t necessarily bad, it’s having that be the target that’s apparently bad. For example, a patient with an A1c between 7% and 8% who is being aggressively treated to achieve an A1c less than 6% is also in potential trouble.

The difference in mortality rates could have happened as a result of random variation and may have nothing to do with the A1c target. While possible, it’s unlikely because the researchers probably used the "chance of variation being 5% or less" (otherwise known as p<.05) threshold. In other words, the likelihood of random variation being responsible for this is less than 5% or 1 in 20.

It took four years for the difference in mortality to become apparent. Even if we forgo the intention to treat context, a brief dip of an A1c to less than 6% in an individual patient is not cause for alarm.

Note that the participants in this study are being re-targeted to an A1c between 7% and 8%, not the ADA recommended level of less than 7%. In looking over the protocol (see page 12) the researchers argue the 7%-8% treatment range what was obtained in previous studies on the benefits of diabetes control, particularly with the drug metformin. There are some interesting data out there that says a level between 7% and 8% may be good enough. This is despite what the ADA says and what the NCQA’s HEDIS optimum measure is. Once again, enrollees in disease management program have a basis to assess the ADA/HEDIS recommendations in the context of their own preferences and values. And deciding to let the A1c creep up over 7% may not be an egregious foolhardy sin.

Keep an eye out for the J curve.