Ending Medical Reversal

by Vinayak K Prasad

  Similar data regarding extended-release niacin were published a year later. In this trial, more than 3,000 patients with persistently low good cholesterol (HDL) were randomized to niacin or placebo. As in the previous trial, all patients were also given a statin. At two years, niacin had raised good cholesterol by seven points but did not improve the study’s primary end point, a smorgasbord of bad things, “death from coronary heart disease, nonfatal myocardial infarction, ischemic stroke, hospitalization for an acute coronary syndrome, or symptom-driven coronary or cerebral revascularization.” In 2014 yet another study cast doubt on niacin.

  If these two studies looked only at surrogate end points, like so many earlier studies did, they would have concluded that these drugs impact cholesterol favorably and therefore both should work to reduce heart attacks and strokes. But because these trials went further, and asked directly whether the medications improved clinical end points, they were able to give us true, albeit disappointing, answers. The lesson when it comes to cholesterol is the same as the HbA1c lesson—improving the numbers is just that, improving numbers. It does not necessarily translate into what we care about, helping people live longer or better.

  We should add that the examples we have highlighted do not show that niacin and fenofibrate do not work under any circumstances. They just show that they do not work in the circumstances under which they were tested. These circumstances mirrored how the drugs are used in the real world, and it is under these circumstances that proponents assert the drugs should work. Like millions of other chemical compounds in the world, these medications may improve some outcomes under some conditions (in some patients, with a certain array of risk factors), but as of 2015, just how they can help remains to be seen. And, as should be true for all medicines, until a treatment is proven to work, there is no reason to assume it does and use it blindly.

  The problem of surrogate end points goes beyond blood tests like HbA1c and cholesterol. In chapter 1 we saw how a common medication, atenolol, successfully lowered blood pressure, a surrogate outcome, but did not reduce deaths.* Two more examples extend the argument to an even more diverse set of surrogates. The first is progression-free survival in the treatment of cancer.

  BEVACIZUMAB IN BREAST CANCER

  Metastatic breast cancer is a terrible disease. It is a cancer that, either because of its biology or its time of detection, has spread beyond the breast to other organs of the body. That is what we mean by metastatic. We treat metastatic breast cancer with chemotherapy drugs that slow the growth of the cancer cells. The goal is not to cure the disease—at this stage that is impossible—but to control it and allow patients to live longer. In the very early 2000s, a new drug called bevacizumab (Avastin) was added to the menu of chemotherapy options for metastatic breast cancer. Bevacizumab is actually not a chemotherapy drug but a monoclonal antibody that changes the way the blood vessels inside cancer grow.* Some experts believe bevacizumab can decrease blood-vessel growth inside tumors, thus robbing them of the nutrients and oxygen that they need to grow. Others believe that the drug may work by improving the delivery of chemotherapy to the cancer. Regardless of the precise mechanism of action, it has become one of the most studied and used drugs in cancer medicine.

  Patients and doctors came to believe that bevacizumab, when added to more traditional chemotherapy, could help women with metastatic breast cancer. In 2008 the U.S. FDA granted bevacizumab accelerated approval for this purpose, based on evidence that it improved progression-free survival in breast cancer.

  What is progression-free survival? Progression-free survival, or PFS in oncologists’ lingo, is an end point that means the patient has survived and that her cancer has not progressed. Progression is defined as growth of a tumor by more than 20 percent of its initial size as measured by a CT scan. If the patient is alive and the tumor has grown by 19 percent or shrunk by 16 percent, that outcome counts as progression-free survival. In studies, the survival portion of the end point is typically of lesser importance, as it usually makes up just a fraction of the total PFS: of all the people who either progress or die, most have progressed, but only a few have died.

  PFS is a surrogate end point. Do patients feel progression? Not really. Often patients feel bad as their cancer returns, but no one can tell you, “OK, I have crossed the arbitrary 120 percent threshold and now I have progressed.” Sometimes patients feel bad at 101 percent, and sometimes they feel fine up to 160 percent.

  Does PFS predict improved overall survival—a longer life span? If it did, it would be a very good surrogate end point. Turns out, the answer to this question is complicated. A British group called the National Institutes for Clinical Excellence commissioned an analysis of this question. The analysis looked at all the studies that compare PFS to actual survival. There was huge variability among the studies—depending on which study you picked, PFS and actual survival either correlated very well or not at all. The outcome of the National Institutes for Clinical Excellence’s analysis: “We need more research.” But for the specific question of metastatic breast cancer, the answer is clearer. The correlation between PFS and survival is poor in nearly all published analyses.

  In the case of bevacizumab, the value of PFS as an end point was critical. The U.S. FDA approved bevacizumab in breast cancer based largely on a trial showing that, when added to chemotherapy, bevacizumab improved PFS. It improved it a lot, nearly doubling PFS, from 5.9 to 11.8 months. In this particular cancer, that is a sizable improvement. What was unexpected in this study was that survival barely changed. It was around 26.7 months for those who received bevacizumab and 25.2 months for those who did not. But at the time, since the PFS benefit was so large, many doctors assumed that meant we would eventually see a true survival benefit or, at a minimum, a benefit in terms of quality of life. The drug became widely used.

  Just three years later, however, the FDA reversed its decision, voting against bevacizumab’s use in metastatic breast cancer. After looking at two more trials, the FDA found that bevacizumab may increase PFS but not actual survival. Tumors hit the 20 percent growth mark later, but patients did not live a day longer. The women who took this drug suffered side effects, had better CT scans (for a while), and ultimately did not benefit. No matter whether PFS and overall survival correlate well on average, bevacizumab taught us that PFS is still a surrogate that can be misleading.

  Surrogate end points generally have two qualities. First, they are well correlated (or we think they are well correlated) with a clinical end point. We know that people with lower blood sugar, better cholesterol profiles, and slower cancer progression do better than those with higher blood sugar, worse cholesterol, and more rapidly progressive cancers. Second they make perfect biological sense. Of course, bringing blood sugars close to normal will improve outcomes in diabetes, especially since we know that it is the high sugars that do damage. It is also these two facts that make it so surprising when surrogates turn out not to be predictive—when they improve with an intervention but the real, clinical end points do not. A final example of a misleading surrogate end point is one that was especially surprising because it made such perfect sense: improving blood flow in people who have suffered a heart attack.

  BLOOD FLOW AS A SURROGATE MARKER

  Some heart attacks are mild. They cause only brief discomfort, injure a small portion of the heart, and leave its function nearly intact. Other heart attacks are massive, leaving the heart unable to pump normally. This life-threatening situation is called cardiogenic shock. The patient has dangerously low blood pressure and a very high risk of death. Doctors take cardiogenic shock very seriously. We use medicines and intravenous fluids to improve blood pressure, we do everything we can to open the blocked artery that caused the heart attack, and, often, we place an intra-aortic balloon pump (IABP).

  The IABP was invented in the 1960s and was truly ingenious. The IABP is a long balloon that is inserted in the aorta, the large artery into which the heart pumps all of its blood. The balloon is c
onnected to a computer that controls when the balloon inflates and deflates. The balloon inflates when the heart is filling with blood and deflates when it is pumping. IABPs are clever because both their inflation and their deflation assist the heart’s function. When the heart pumps into the aorta, the balloon deflates rapidly, creating a vacuum and drawing the blood forward into the peripheral circulation—toward the rest of the body where the oxygenated blood is needed. The deflating balloon acts like someone sucking on a straw, in this case, the aorta. When the heart is not pumping, the part of its cycle when it fills and rests, the balloon inflates. This not only pushes blood forward into peripheral circulation, but it also pushes some blood back to the heart, potentially giving this same oxygen-rich blood to the arteries that supply the heart muscle itself. Early studies supported this theoretical physiological benefit.

  Over the years, the IABP steadily gained popularity. The idea was that because it helped a surrogate outcome, cardiac output—the volume of blood being pumped by the heart—it would help patients with a low cardiac output. Recent international guidelines (recommendations for doctors) said that placing an IABP should be part of the care of patients with a heart attack and cardiogenic shock. Because of this, as of 2010, up to half the patients who fit this description had an IABP placed. You know where this story is going, don’t you.

  In 2012 the IABP SHOCK II trial enrolled 600 patients with cardiogenic shock caused by a heart attack. All the patients received the best care available. Half also got an IABP. The results were surprising. At 30 days, 40 percent of the patients had died (this is a very serious condition) and the pump made no difference at all. Not only did IABPs not save lives, they also did not improve any of the end points the researchers examined. They did not reduce the rate of second heart attacks in the hospital, or stroke, or complications of the procedures to open the blocked artery that caused the first heart attack. The IABP is another example of a therapy, adopted because it made really good sense and improved a surrogate outcome, that failed to improve the end points that really matter. It is also worth noting that the IABP is a costly intervention that can have significant complications (as you might expect if you imagine inflating and deflating a balloon in the aorta).

  There are some interesting points about IABPs. First, we include IABPs in our discussion of surrogate markers because the procedure improved cardiac output, a stand-in for a real clinical end point, preventing death.* Second, the IABP is an intervention instituted not only because it improved a surrogate end point but because it seemed like it should work. The mechanism of the treatment was just so darn elegant. We discuss this phenomenon in great detail later when we examine the causes of reversal. Lastly, there was also some hint that doctors were suspicious of the effectiveness of IABPs. Although guidelines supported their use, only 25 to 40 percent of patients with cardiogenic shock actually received them. Make no mistake, that is a lot of balloons, but for some reason 60 to 75 percent of doctors were not following this guideline.

  HOSPITALIZATION IS, AMONG OTHER THINGS, A SURROGATE

  We end this discussion with a final type of surrogate end point, one that is both a surrogate and an end point of importance in and of itself. A small group of unfortunate people in America are frequently hospitalized. These people often have conditions that get better and worse, such as heart failure or emphysema. When things are going well, these conditions can be managed at home. But when things are going poorly, these are common reasons for admission to the hospital. Since patients are hospitalized because they are doing poorly, many doctors believe that the end point that matters most for these people is a decrease in hospitalization.

  Hospitalization is tricky. All things being equal, it is better to avoid being hospitalized. Anybody who has spent any time as a patient in a hospital knows that it is not where you want to be. Putting aside that you usually feel terrible while you are hospitalized, there are the issues of the food, the noise, and those gowns—not to mention the risk of infection, medication errors, and potential adverse effects of all of our 21st-century interventions. So, in part, avoiding hospitalization is an objective end point. However, many doctors read more into it. They argue that it is also a surrogate end point. The fewer times you are hospitalized, the less likely you are to die. Used this way, hospitalization is a surrogate for death.

  Lars Hemkens and colleagues wanted to test the second part of this statement. Of course, all things being equal, it is better not to be hospitalized. But is it possible that some intervention could decrease the rate of hospitalization while paradoxically increasing the death rate? This was the puzzle they set out to solve: was hospitalization also a good surrogate for death?

  To do so, the group studied both hospitalization and death across a large collection of trials where both were reported. They found that one-third of the time, when hospitalizations went down, mortality went up, or vice versa, meaning that a treatment that prevented hospitalization made you more likely to die; something that seemed like a bad intervention (after all, it increased hospitalization) actually made you live longer. This analysis is thought-provoking because few people would have even considered hospitalization to be a surrogate marker. But, of course, in part, it is. Findings like this allow doctors to give more nuanced guidance to patients as they choose what they consider to be the most important outcomes. Their conclusion was that mortality and hospitalization are both important and both should be reported in studies. It is a mistake to assume that one implies the other. In this sense, hospitalization—like so many other measurements we have looked at—is not a reliable surrogate end point for the most important outcome: living longer.

  CONCLUSION

  Surrogate end points come in many forms, some obvious (HbA1c, hypercholesterolemia) and some less so (hypertension, hospitalization). They are attractive because they are relatively easy and inexpensive to study. They also usually make intuitive sense. But sometimes, maybe often, they do not correlate with the outcome that we care about.

  Throughout this book, we make the case against surrogate end points. We argue that the way to prevent reversal is to have good evidence that medical practices improve hard end points—the ones that we care about. It is entirely possible that we will someday have perfectly reliable surrogate end points, but that achievement will involve a great deal of hard work. To prove that a surrogate is perfectly reliable will require many studies of hard end points, demonstrating that the surrogate always gets it right. We have yet to see such assurance for any surrogate end point, and recent history has given us ample reason to be skeptical for the time being.

  4 SCREENING TESTS

  AFTER USING THE BATHROOM, Christopher D’Amico carefully folded a square of clean toilet paper and tucked it in his underwear. This was routine—just a precaution—a perennial reminder that he was not a young man anymore. Ever since his prostate cancer treatment, Christopher had struggled with a little incontinence. It was not too bad, nothing a little toilet paper couldn’t handle, and, all in all, it was a small price to pay to be alive.

  Three years earlier, at age 63, Christopher had gone to see a urologist. For months before this visit, he had been waking two or three times a night to use the bathroom. From the drug ads on TV, he knew these were probably symptoms of benign prostatic hyperplasia, BPH. BPH occurs in up to half of middle-aged men. Among other things, this enlargement of the prostate can cause incomplete bladder emptying and nighttime trips to the bathroom. Christopher did not have a regular doctor, and a friend suggested that he go straight to a specialist.

  The urologist took Christopher’s history, examined his prostate, and, as expected, diagnosed him with BPH. Before Christopher left the office, the doctor drew blood to screen him for prostate cancer. A few days later Christopher got a call telling him that his prostate specific antigen (PSA) was elevated, meaning that in addition to BPH, he might also have prostate cancer. A week later, Christopher was having biopsies of the prostate. A few days after that, the urologist called to say
it was cancer.

  Cancer. The big C. Just hearing the word gave Christopher shivers. His father, an inveterate drinker, had died from esophageal cancer. His older brother, a longtime smoker, had lost his battle with oral cancer the previous year. And, now, he thought, it’s my turn. His children were still in college; he was still working to support them, and his wife needed him. I’ve got to beat this, he said to himself.

  A few weeks later, Christopher underwent surgery. The surgeon was a robot. Well, not exactly—his surgeon was actually a nice guy, but he controlled a robot during the surgery. The robot did the actual cutting.*

  Christopher recovered well from the surgery, and his PSA went to zero—the prostate cancer was cured. The only lingering effects were the incontinence and impotence. Although he still desired sex, Christopher really could not get any sort of useful erection. This was an unfortunate side effect, but Christopher’s doctor told him his life was saved because the cancer was caught early. It all seemed worth it.

  A few years later, Christopher began to have mixed feelings about the surgery and the PSA blood test that led to it. Some of his friends told him that they had had long conversations with their doctors about the risks and benefits of prostate cancer screening and ultimately decided to forgo the PSA test. Christopher asked, “Didn’t the test save lives?” Well, it wasn’t so clear, his friends told him. Then Christopher read in the newspaper that a major task force had announced that before routinely performing the PSA test, doctors and patients should have a conversation about it. A few years later, Christopher turned on the radio to hear that the same group of experts now advised men not to get a PSA at all. Each time he tucked the toilet paper into his pants, Christopher was reminded of how little he understood about that blood test, which had either saved his life or made it worse.