August 26, 2011


Changing The Rules: The incidence of conditions like hypertension and diabetes has skyrocketed in recent years. Some of that increase is real. But some of it is due to changes in the way diseases are defined. In this excerpt from a soon-to-be-published book, a member of the DMS faculty explains the downsides of that trend. (H. Gilbert Welch, M.D., M.P.H., and Lisa M. Schwartz, M.D., M.S., and Steven Woloshin, M.D., M.S., Winter 2010-11, Dartmouth Medicine)

Many modern diseases are defined by a numerical rule. If your blood pressure is above a certain number, for example, you have hypertension. If it isn't above that number, you don't. And hypertension isn't the only condition defined by a numerical rule. There are many diseases that you can be labeled with simply because you are on the wrong side of a number, not because you have any symptoms. Diabetes is defined by a number for blood sugar, hyperlipidemia by a number for cholesterol, and osteoporosis by a number for bone density (called a T score). By establishing these numerical targets, of course, we doctors are trying to get ahead of patients' symptoms--to make diagnoses early in order to prevent bad events such as leg amputation and blindness from diabetes, heart attacks and strokes from high cholesterol, and wrist and hip fractures from osteoporosis.

The conventional wisdom tells us this is good: finding problems early saves lives because we have the opportunity to fix small problems before they become big ones. What's more, we believe there are no downsides to looking for things to be wrong.

But the truth is that early diagnosis is a double-edged sword. While it has the potential to help some people, it also has a hidden danger: overdiagnosis--the detection of abnormalities that are not destined to ever bother us. Some people diagnosed with diabetes, high cholesterol, and osteoporosis, in other words, will never develop symptoms or die from those conditions. This is most likely the case for those in whom the condition is mild.

The numerical rules used to define conditions are really important. They typically involve a single number: if you fall on one side of the number you are defined as being well; if you're on the other, you are defined as being ill. These numbers--called cutoffs or thresholds--determine who has a condition and who doesn't. They determine who gets treatment and who doesn't. And they determine how much overdiagnosis occurs.

Cutoffs are set by expert panels of physicians. I wish I could say that their determinations result from purely scientific processes. But they are more haphazard than that: they involve value judgments and even financial interests. The experts who select the cutoffs have particular sets of beliefs about what is important. Because these doctors care greatly about the conditions they specialize in, I believe they sometimes lose a broader perspective. Their focus is to do everything they can to avoid the bad events associated with that condition; their main concern is not missing anyone who could possibly benefit from diagnosis and treatment. So they tend to set cutoffs that are expansive, leading many people to be labeled ill or abnormal. They tend to either ignore or downplay the major pitfall of this strategy: treating those who will not benefit. This is a problem because almost all treatments have the potential to do some harm.

Over the past few decades, many cutoffs have been changed in a way that dramatically increases the number of individuals who are labeled with these conditions and others. It means that the threshold to make a diagnosis has fallen. Even if this is done with the best of intentions--to avoid more bad events--it can lead to an undesirable consequence: more overdiagnosis and thus more treatment of people who won't benefit but can potentially be harmed. [...]

The problem of overdiagnosis was dramatically demonstrated in a recent randomized trial funded by the National Institutes of Health. The trial was designed to determine whether intensively lowering blood sugar reduced the risk of having or dying from a heart attack or stroke. The trial enrolled over 10,000 patients with diabetes who were at high risk for these events. About 5,000 were randomized to receive standard diabetes therapy--treatment to lower their average blood sugar to a more acceptable, though not normal, range. The other 5,000 were randomized to receive intensive drug therapy--treatment to bring their blood sugar down to a normal level. Half of the patients in the latter group achieved the goal: the average blood sugar level of those who got intensive therapy was below 140. Because the average included blood sugars measured right after eating (which tend to be high), it is safe to assume that their fasting blood sugars were considerably lower.

The trial started in 2003 and was supposed to continue to 2009. But on February 6, 2008, the National Heart, Blood, and Lung Institute issued a press release saying they were "changing" the intensive therapy regimen "due to safety concerns." Changing wasn't the most accurate word to describe what they were doing; stopping would have been a better choice. And the safety concern was that patients receiving the intensive therapy were dying more often than patients receiving the standard therapy. After three years, 5% of the patients receiving intensive therapy had died, compared with 4% of those receiving standard therapy. That is a 25% increase in the risk of death, and the researchers were confident that it was not a statistical fluke. There was little doubt: intensive treatment was worse than standard treatment.

You might wonder how making people's blood sugar normal could end up killing them. It's probably because we can't simply dial a patient's blood sugar to a specific number; our therapies aren't that precise. Instead, blood sugar bounces around, and if we try to have blood sugar bounce around normal, sometimes it will bounce too low. And having your blood sugar too low increases your risk of death. The investigators might argue that hypoglycemia (low blood sugar) was not the cause of the increased risk of death. But by their own admission, they were not sure what explained the increased mortality. In the official report, lead author Hertzel Gerstein wrote: "Despite detailed analyses, we have been unable to identify the precise cause of the increased risk of death in the intensive blood sugar strategy group. . . . Our analyses to date suggest that no specific medication or combination of medications is responsible. We believe that some unidentified combination of factors tied to the overall medical strategy is likely at play."

My view is that if the trial had shown a mortality benefit, the authors would have been quick to ascribe that benefit to intensive control of blood sugar (as I think would have been correct in that case). But since the trial showed a mortality harm, that must also be ascribed to intensive control of blood sugar. That's the point of a randomized trial.

What does this study tell us about where to set the threshold to diagnose diabetes? My take is this: if it's not good to make diabetics have nearly normal blood sugars, then it's not good to label those with nearly normal blood sugars as diabetics. Why? Because doctors will treat them. People with mild blood sugar elevations are the least likely to gain from treatment--and arguably the most likely to be harmed, as Mr. Roberts was.

Posted by at August 26, 2011 6:38 AM

blog comments powered by Disqus