CANOE and Diabetes as a Surrogate Endpoint

In my most recent post, I commented on the DREAM trial's problems with composite endpoints, but also touched on the issue of using diabetes as an endpoint that was assessed while patients were on rosiglitazone.

Today the Lancet (which published DREAM) published online CANOE, a trial of relatively low dose rosiglitazone and metformin for the "prevention" of diabetes. Given the timing right after my discussion of DREAM I feel like I should comment on CANOE as well.

Although I want to mainly focus on the notion of diabetes as an outcome, I'll first point out a major difference between DREAM and CANOE -- the number of patients. While DREAM enrolled 5269 patients at high risk for diabetes, CANOE enrolled only 207. Normally small numbers would not be a concern in a "positive" trial, but where DREAM found evidence of cardiovascular harm, CANOE was so small that it would be unlikely to detect any differences in clinical events -- likely an advantage from the point of view of the sponsor (who manufactures rosiglitazone and rosiglitazone/metformin).

Like DREAM, however, CANOE assessed development of diabetes while patients were still taking the active hypoglycemic medications. As such, the trial was virtually certain to yield a positive result. "Diabetes" was diagnosed by either two blood sugars above 126 mg/dL or an OGTT at 2 hours above 198 mg/dL.

Although we often don't notice it since we think of diabetes as "real", when diagnosed this way, diabetes is a surrogate outcome. The notion of a surrogate outcome can feel quite confusing given how entrenched these sorts of diagnoses (hyperlipidemia, hypertension) are in out psyches, but they should be contrasted with patient-important outcomes like stroke or clinically apparent MI (as opposed to a troponin leak, which is also a surrogate outcome). You can usually tell the difference by asking whether the outcome would matter to a patient in and of itself if you could prevent the potential outcomes that the surrogate is felt to be a risk factor for. For instance, LDL lowering is a surrogate outcome. No patient feels ill with and LDL-C of 200 mg/dL, and if you had a drug that raised LDL higher but lowered cardiovascular risk it would be a helpful intervention. LDL lowering IS NOT a patient-important outcome. Reduction of cardiovascular risk IS a patient important outcome.

Is diabetes a patient-important outcome?

The tricky aspect to diabetes is that, as originally conceived, it is a real disease of great patient importance. Type I diabetes with weight loss, polys, DKA, and rapid death is obviously not a surrogate outcome. Type II diabetes, diagnosed by polys, nonketotic coma, infections, and fatigue is also not a surrogate outcome.

But that is no longer how "diabetes" is diagnosed. Based on diabetes as a risk factor for various bad events, and following the recommendations of endocrinology societies and international organizations, we now routinely accept a diagnosis of diabetes following the rules used in CANOE (repeated fasting glucose above 126 mg/dL; OGTT above 200 mg/dL at two hours) and have added A1c to the mix. While this method of diagnosis may have some justification in thinking about when to initiate therapy (though evidence for this is limited), it clearly makes "diabetes" a surrogate outcome in a trial like DREAM or CANOE: patients do not feel ill with a fasting glucose of 130 mg/dL, and if we could prevent the various long-term outcomes associated with "diabetes" it would not matter at all to patients whether their measured blood sugar stayed at 130 mg/dL or was reduced to 75 mg/dL.

Because diabetes has been made into a surrogate outcome, the prevention of diabetes ceases to be a compelling result in a trial designed like DREAM or CANOE. DREAM and CANOE may show that taking hypoglycemic agents postpones the diagnosis of "diabetes", but the major clinical consequence of that diagnosis is that the patient would be told to start taking hypoglycemic agents. Thus these trials at best show that initiating hypoglycemic agents delays the time until a physician would have told the patient to initiate hypoglycemic agents -- hardly an exciting outcome. Wouldn't it be easier for the patient to just wait until they actually needed the drugs? Instead, the outcomes of the trials should be clinical endpoints like vascular events or renal failure. (As mentioned in my prior post, DREAM created a false clinical endpoint with the composite of "development of diabetes or death".)

At the least, it would have been nice if either DREAM or CANOE had assessed diabetes while the patients were off medication since there was never any doubt that while on medication the surrogate outcome of "diabetes" based on lab results would be less common. Although this issue is never mentioned in the abstract of CANOE, the final paragraph finally pays it some lip service: 

An important limitation of our study is that we cannot clearly differentiate whether the effect recorded represents true prevention of diabetes or early treatment of diabetes. This limitation applies to all pharmacological clinical trial interventions for diabetes prevention, and has been reviewed. Further clarification of this question will need a longer follow-up after drug washout, with careful assessment of changes in the underlying pathophysiological events that bring about type 2 diabetes.

The paragraph is an interesting one, since the authors appear to state in the second sentence that the problem of assessment is an inherent issue with "all" diabetes prevention trials, but then in the third sentence explain how a trial ought to be conducted to avoid the problem.

So, like DREAM before it, CANOE produced a completely predictable result that added little to our knowledge of whether diabetes prevention with hypoglycemic agents might actually be of clinical benefit. And unlike DREAM, CANOE was so small that it did not risk finding important harm from treatment with rosiglitzaone/metformin.

Comments

[Joseph Nicholas, MD, MPH]

Great points, and these surrogacy issues are really fundamental to the practice of American style medicine. Even in my acutely hospitalized patients, many treatment decisions are based on managing surrogate markers for which the impact on meaningful clinical outcomes is nebulous at best. LDL < 70, inpatient SBP < 130, glucose 70-120 all take on a clinical life of there own; the patient is diseased because we say so. Does the CAD patient with an LDL of 101 really have dyslipidemia (he's not at goal!) And once you start adding these "diseases" to your problem list- we've created a meta-reality that just might often be divorced from the real reality of disease, suffering, benefit and harm.

[sanjiva86]

What really annoyed me is some of the stories in the media saying that no increased cardiac risk was apparent with a lower dose of Avandia. Well, of course not! The trial was never big enough to determine that in the first place.

[Dan Hackam]

Response to Joseph Nicolas MD MPH.
I agree, all surrogate markers.
If we follow atherosclerosis as a biological disease without waiting for clinical events to occur or recur, but with some form of highly sensitive atheroma imaging, we don't have to rely on upstream surrogates like LDL, SBP, and glucose at all.