Ending Medical Reversal

by Vinayak K Prasad

  What does this study tell us? A couple things. First, no one believes a placebo is acceptable if it is extremely costly (vertebroplasty, orthopedic surgery), delays proven therapies, or causes real harms. We can debate whether or not deception is permitted in select medical circumstances, and one might argue that if a placebo is cheap, proved to be better than the alternative (doing nothing), and patients are fully informed of what they are getting into (no deception), then, under those circumstances, placebo treatments can be considered. While vertebroplasty was unacceptable, what about sugar pills for patients with IBS?—this is less clear. Our bias is toward abandoning them, but we also believe there is no single right answer and much room for debate. There may be a role for promoting a low-cost placebo in a setting of total transparency.

  THE SUBJECTIVE NATURE OF ANGINAL CHEST PAIN

  Now, let’s apply these lessons to a final example using a truly classic medical experiment from the 1950s that tested a costly and invasive medical practice. Consider angina, which in chapter 1 plagued Anthony Baker. Anthony’s cardiologist placed a stent to open up a clogged artery—which everyone agreed was the source of the problem. Anthony felt better after the procedure and was happy to have a new lease on life. He left the office thinking that the stent would improve his survival (studies show that many patients believe this). Then, as we indicated, it was proved that stenting these lesions did not help patients live a single day longer. The study was named COURAGE.

  When Anthony heard this news, he was disappointed. Well, he reasoned, at least my pain is better. In fact, a follow-up publication from the COURAGE study shows just this. Placing stents in the coronary arteries of people with stable angina reduces chest pain and improves quality of life, somewhat, at 6, 12, and 18 months. By three years the benefit of the treatment wanes, and the study shows that pain is identical between the people who had gotten stents and those who were treated only with medicine. Fine, Anthony says, even if the benefit is small and short-lived, at least there is some benefit to the procedure. All in all, he is happy to have had it done. Many doctors agree with this reasoning. They fault the medical system for throwing the baby out with the bathwater. Even if stents do not make you live longer, you live better, they argue. That counts.

  Without a doubt, that counts. If placing stents in the coronary arteries reduces chest pain, people should have that option as a way of treating their pain (as long as they realize that is the only benefit). However, the problem with COURAGE data for chest pain is that, in studying this subjective end point, the control arm was inadequate. The COURAGE trial tested whether stents improved pain compared to medications. But getting a stent is a major intervention. A cardiologist talks you through the procedure and then you are sedated. Often, you are shown your angiograms afterward. You see an image of a clogged, disfigured artery transformed into an open pipe with beautiful flow. How could such a convincing improvement in the plumbing not improve symptoms? To truly see whether stenting improves pain, the control arm should be a sham coronary-artery intervention. Patients randomized to the control arm should have everything except the stent placement. In arthroscopic knee surgery and vertebroplasty, that was what it took to reveal the placebo effect. You might be reluctant to do this study. Angina is different from knee pain, you might argue. It has to do with blood flow. There cannot be a placebo response in angina, can there?

  Consider a study from 1959. In the years leading up to this seminal New England Journal of Medicine paper, doctors struggled with angina. They knew it was a problem with blood flow in the coronary arteries, but they did not have the tools or the medicines to really address the problem. One proposal was that if you tied off (ligated) the internal mammary artery, an artery in the chest that does not lead to the heart, there would be increased upstream flow to the coronaries. It was a bold idea that was not without a physiological basis.

  In 1955 an Italian team tried the procedure, internal mammary artery ligation, on 11 people with severe chest pain. The results were impressive; all 11 had improvement in their pain. By 1959 the same team of doctors had treated more than 300 people. Most were doing well.

  Then in 1959, Leonard Cobb published the results of his classic study. He had recruited patients with angina and brought them to the operating room to perform the procedure. At the point in the surgery when it was time to ligate the artery, the researchers randomized the patients to ligation or sham ligation. Those patients who had sham ligation were told the procedure had been done when it had not. Of the people who had the ligation procedure, 32 percent improved; 43 percent of the patients who received sham operations also improved. A similar study was later published confirming these results, and internal artery ligation was abandoned.

  In retrospect, the lesson of Cobb’s study was not that ligation of the artery did not work but that any subjective end point, knee pain in osteoarthritis, shortness of breath in asthma, and yes, even angina in people with coronary-artery disease, is remarkably susceptible to the placebo response. If the patient believes the treatment will work, often it does work.

  With this in mind, we can look at the results of the COURAGE trial with a bit more skepticism. The benefit in symptoms was small and completely lost by 36 months. The control arm received medicine alone, not a sham procedure, the closest thing to the procedure. Could stenting for people with angina from stable coronary-artery disease be like ligation of the internal mammary artery? Could the small, relatively short-lived benefit simply be a placebo effect? There is no answer to that question, not without a definitive study, but we can say the idea is plausible enough to consider. It should be tested. If stenting were found to be no better than sham stenting, then percutaneous coronary intervention for people with stable angina would be the greatest example of reversal, at least in terms of cost, in the past 20 years.

  CONCLUSION

  Reversals of treatments that made us feel better are disconcerting. We put tremendous faith in knowing how we feel, and it is confusing to find out that our own bodies can fool us (and our physicians). All the practices that we discussed really did make people feel better. However, in each example it was not really the treatment that helped; it was the idea of the treatment. This would not be a problem if our placebo therapies were cheap, carried no risks, did not involve deception, and had no potential of delaying receipt of therapies that would be more beneficial. Therapies that were thought to benefit objective outcomes can be overturned; likewise, therapies that affect subjective outcomes can be reversed.

  3 SURROGATE OUTCOMES

  THOMAS GALBRAITH WAS DISAPPOINTED. He had just gotten off the phone with his endocrinologist and his numbers were still too high. The doctor told him that a measure of his average blood sugar, the hemoglobin A1c (HbA1c), remained “above goal.”* It sounded ridiculous to Thomas. He had been doing everything right: eating well, watching his weight, and taking all his diabetes medications, which now totaled three pills and one shot of insulin each day. Thomas was a 64-year-old accountant who had, as we say in the business, multiple medical problems. He was overweight, had diabetes and high blood pressure, and had experienced a heart attack a few years earlier. Now, he also had an HbA1c of 7.5 percent—“above goal.” His doctor wanted the number below 7.0 percent. Thomas’s doctor told him that he would need to increase his insulin dose and emphasized that a lower blood-sugar level would improve his chances of avoiding another heart attack (as well as many other sordid complications that the doctor seemed to enjoy enumerating).

  That night over dinner—steamed green beans and baked tilapia— Thomas discussed the situation with his wife. Despite losing three pounds over the past three months, not missing a single pill (which now included pioglitazone, saxagliptin, and metformin), and taking his five units of insulin at bedtime, his HbA1c was a percentage point above where his doctor wanted it to be. He recounted what his doctor had said—for every 1 percent rise in HbA1c, his risk of stroke or heart attack went up nearly 20 percent and his risk of death went up 12 percent.
/>
  This whole scenario was playing out in May 2008. Thomas redoubled his efforts to lose weight. He grudgingly increased his insulin (which seemed to make him feel a bit foggy in the morning) and dedicated himself to an even more ascetic diet. After all this, it came as a shock when in June of that year Thomas read about a study. This study concluded that the lower HbA1c his doctor wanted would not reduce his risk of death; it would actually increase it. And his HbA1c of 7.5 percent was better for him in the long run than anything below 7.0 percent. “Are you kidding me?” Thomas asked, as he read the news story.

  THE TRUTH ABOUT SURROGACY

  In the first two chapters we presented reversals that involved objective end points, like heart attacks and death, and subjective ones, like knee pain and back pain. Now we are discussing another sort of end point—the surrogate end point. Surrogate end points are a whole different matter. Many of the reversals that we have discussed, and will discuss, occur because we have adopted therapies whose efficacy is supported by inadequate data. Surrogate end points are at the heart of a generous portion of that inadequate data.

  Strictly speaking, surrogate end points are objective ones. They are end points that we can easily and directly measure. Surrogate end points, however, unlike clinically important objective end points, are invisible to the patient. They are stand-ins. We use them because it is simpler to show that a treatment improves a surrogate end point than to show it improves a real clinical one. For this reason, the medical literature is full of surrogate end points. It is easier to show that a drug improves bone density than it is to show that it decreases the rate of fractures. It is easier, and far less expensive, to show that a drug lowers blood pressure than it is to show that it decreases the rate of strokes. It is easier to show that a drug lowers blood sugar than it is to show that it decreases the risk of blindness, kidney failure, heart attacks, and death. Studies with surrogate end points may require dozens of patients and take months to years; studies with clinical end points often require hundreds of patients and need to run for years or even decades.

  Thomas’s story illustrates what can happen when we rely on surrogate end points. What went wrong in this case is that HbA1c is not a perfect surrogate for what we are really interested in: all the harms that diabetes can do. Thomas is really not interested in lowering his HbA1c, and with an HbA1c of 7.5 percent he is not having any symptoms of diabetes. What he is interested in is lowering his risk of diabetic complications, such as going blind or dying of a heart attack. We have already presented a couple of reversals that were linked to the use of surrogate end points in chapter 1: antiarrhythmic drugs decrease the number of premature ventricular contractions but do not improve survival after a heart attack, and atenolol lowers blood pressure but does not improve survival.

  The surrogate end points that appear in the medical literature are varied: many are measurements made in the lab (cholesterol levels and HbA1c); others are measured by procedures: electrocardiograms (premature ventricular contractions), blood-pressure measurements, or CT scans (tumor size). In what follows, we provide a tour of surrogate end points to demonstrate just how common and how tricky they can be.

  Returning now to Thomas. After he read about this study in the news, he began to wonder why we adopted HbA1c as a surrogate marker in the first place? Turns out that doctors used the best observational studies of people in the real world and paired that information with their biological understanding. They first asked a simple question: Which patients with diabetes do best? When you ask that question, it turns out that people with HbA1c readings less than 7 percent live longer (and better) than those with higher levels. The researchers realized that this made perfect biologic sense; the closer the average blood sugar was to that of a normal person, the less damage the blood sugar would do to the organs. After this relationship was seen in countless studies, doctors began to recommend an HbA1c of less than 7 percent as the goal for most people. Some doctors strove for even lower numbers.

  Doctors were so convinced of the benefit of a low HbA1c that they codified it into law—or the closest thing to law in medical practice—clinical guidelines. Many medical practices used the HbA1c as a measure of how well doctors were managing their patients with diabetes. Hey Doc, bad news, only 28 percent of your diabetic patients have an HbA1c below 7.0 percent. Do you think you can do better?

  When we discuss the types of studies that we base decisions on in medicine (in chapter 9), we consider why observational studies, such as the ones that defined the association between low HbA1c and better outcomes, can be problematic. However, you may already begin to see why this is true. A person in the “real world” with a lower HbA1c might do better not because of the lower numbers, but because she is the sort of person who does better. In other words, if you have a strong social network, are wealthy, live in a good neighborhood, have a personal trainer, and generally are healthier, your HbA1c might be lower naturally. It might very well be everything else in your life (everything other than the low HbA1c) that makes you better off. Of course, these issues can get very confusing. If you have all these advantages, you might also have access to better health care, which would put you in touch with doctors who believe that a lower HbA1c is better. These doctors might be pushing you toward a lower HbA1c with medications. This might actually be harming you. But, because of all the other advantages in your life, you still come out ahead. As we said, it can get very confusing.

  HbA1c seemed to be an especially robust, well-validated surrogate marker because it was not just observational studies that predicted its utility. A few randomized trials, real experiments, demonstrated that bringing people with diabetes to HbA1c values of 7 to 7.5 percent was better than leaving their HbA1c’s at 9.0 percent or so. Of course, these were different populations of people (some even had type 1 diabetes, really a very different disease from Thomas’s type 2 diabetes), who were sometimes treated with different drugs.

  The study that got Thomas’s blood boiling was the ACCORD trial, published in 2008. ACCORD was a randomized study in which doctors enrolled more than 10,000 people with diabetes and divided them into two groups. Half had their sugar lowered to a goal of below 8.0 percent, basically the standard of care, and the other half aimed for an HbA1c below 7.0 percent. The standard-of-care group achieved an average HbA1c of 7.5 percent, while the intensive-treatment group achieved an HbA1c of 6.5 percent. As expected, the group with a lower sugar target, the “intensive” therapy group, used more drugs to achieve their goal. They often ended up on medical regimens similar to Thomas’s. At the end of the study, there was no difference between the groups in terms of the combined rate of death from a cardiovascular cause, heart attacks and strokes. There was one difference in outcomes between the groups: there was a clear difference between the two groups in terms of overall mortality. In one group, 4 percent of the patients died; in the other, 5 percent died. Unexpectedly, the group with a higher death rate was the intensive-treatment group.

  The findings of ACCORD were bolstered by a similar study called ADVANCE, which found that patients achieving a goal HbA1c of 6.5 percent did not live longer than those achieving an average HbA1c of 7.3 percent. Targeting a goal of less than 8.0 is no worse (and possibly better), when it comes to survival, than targeting a goal of less than 7.0. Thomas’s doctor, although with good intentions, was telling him to do the wrong thing.

  HOW TO TREAT CHOLESTEROL

  Studies in every corner of medicine rely on surrogate markers, and no discussion of the topic would be adequate without talking about hypercholesterolemia, or high cholesterol. This surrogate marker has essentially become accepted as a disease in the United States. You do not feel your cholesterol level. You may feel terrible after a fatty meal—we all do—but it is not cholesterol you are feeling. Cholesterol is an important molecule: it makes up part of your cell membranes and the backbone of key hormones. There are different cholesterol molecules. For decades we have known that people with low HDL cholesterol and high LDL cholesterol
do worse than patients with the opposite combination. When we say worse, we really mean worse. They have more heart attacks, more strokes, and a higher death rate. This relationship is so clear that we commonly refer to HDL as good cholesterol and LDL as bad cholesterol. Much of the information about cholesterol comes from the Framingham Heart Study, a famous study in which a large group of men and women (mostly white) from Framingham, Massachusetts, were studied for more than 40 years.

  Given the close relationship between people’s cholesterol profile and their health, it was not much of a leap to assume that lowering cholesterol would improve health. Many drugs have been developed to achieve this goal. Among the most commonly used drugs in the United States are the statins, a class of drug that lowers cholesterol and, most definitively in patients with high cholesterol and heart disease, saves lives. The success of statins is especially satisfying because earlier medications, some tested in the 1980s, seemed to improve cholesterol numbers without improving people’s overall health.

  This last fact suggests that cholesterol level might not be a perfect surrogate for cardiovascular disease. This has been borne out recently. Extended-release niacin was approved in 1997 based on its ability to raise good cholesterol (HDL), and lower bad cholesterol (LDL). Fenofibrate received Food and Drug Administration (FDA) approval in 2001 based on data showing that it could lower bad cholesterol (LDL). Despite the success of both of these drugs in improving people’s cholesterol levels, neither has proved to actually help people.

  In 2010 fenofibrate was tested in a trial of more than 5,000 people with type 2 diabetes. All these people were given a statin medication and then randomized to fenofibrate or a placebo pill. This was a beautiful study because it mimics precisely how doctors use this medication in the real world—to lower the cholesterol levels of an at-risk patient who is already on a statin. After 4.7 years, 2.4 percent of patients in the placebo group had had a heart attack, stroke, or death related to the cardiovascular system. In the fenofibrate group, 2.2 percent of the people experienced one of these outcomes. These probabilities were statistically indistinguishable. Fenofibrate improved cholesterol levels but failed to improve the end points we care about most; the drug improved a surrogate end point but not an objective, clinical one.