Ordinarily Well: The Case for Antidepressants

3

Random Thoughts

BECAUSE RANDOMIZED TRIALS have become essential to medical practice, we may forget that they are of recent origin. Developed in agriculture, where farming methods were tested through being applied to random sections of fields, the approach entered medicine in the late 1940s, after the discovery of the first antibiotics. The pioneering work concerned a treatment for tuberculosis.

Tuberculosis that reaches into the nervous system, tuberculous meningitis, was almost always fatal. When meningitis responded to streptomycin, it was clear, without recourse to sophisticated evaluation methods, that the drug worked.

Tuberculosis in its more frequent location, the lung, behaved differently. Some infected patients recovered on their own. The question of whom to treat was critical, since pulmonary tuberculosis was common and streptomycin was in short supply. In 1946, the British Medical Research Council, guided by the pioneering statistician Sir Austin Bradford Hill, planned a trial with a new design. Participants would be randomly assigned to receive the usual treatment, strict bed rest, or the experimental treatment, bed rest supplemented by injections of the antibiotic.

The study’s main outcome measure was progress of the disease as it appeared in chest X-rays. Streptomycin performed well. In the first weeks, patients on streptomycin were much likelier to see marked radiologic improvement—and fewer died. Looking more broadly, twice as many patients on streptomycin (69 percent, versus 33 percent of those on bed rest only) showed some clearing on their chest films.

Antibiotic treatment rarely eliminated infections. By the four-month mark, new strains of the bacillus emerged. In the following two months, bed rest matched the antibiotic. Using unsystematic observation, doctors had already reached the right conclusion: streptomycin helps in the early going. Even though the experiment only confirmed clinical wisdom, the research design was seen to enact the scientific spirit of the age. The study established the randomized trial as a standard in medicine, especially to evaluate treatments for conditions like pulmonary tuberculosis, which wax and wane on their own, with remedies like streptomycin, whose reach is limited.

Because the antidepressant controversy turns on the interpretation of randomized trials, we will need to gain comfort with them. The design has three elements intended to keep assessments objective: a control conditionrandomization, and blinding.

In an outcome trial, a promising intervention is contrasted to a comparison intervention, the control, in which participants receive all the elements of patient care except the active treatment under study—in this case, streptomycin. The testing setups are called arms. The streptomycin trial had a medication-plus-bed-rest arm and, as a control, a bed-rest-only arm. Sometimes, the control condition will include a placebo, or sham treatment.

In patient care, lots of things happen at once. Doctors prescribe, but they also offer support and attention. Time passes, allowing for what commentators on the streptomycin trial called vis medicatrix naturae, the healing power of Mother Nature.

Including a control arm helps researchers to estimate the inherent efficacy of an intervention, the contribution that it makes through its special properties—with streptomycin, the ability to inhibit the growth of bacteria. When we ask whether a medicine is inherently effective, we are asking whether it offers benefits beyond those of time, circumstance, and the doctorly encounter. A simple assumption—not always justified, but a fair starting point—is that if we subtract the progress patients make in the control arm from the greater progress their counterparts make in the treatment arm, the remainder will be the contribution made by the active ingredient, the treatment’s inherent efficacy.

A placebo is a treatment that has the same form as the active intervention but lacks the ingredient understood to be effective. In a test of acupuncture, a sham version, with needles stuck in the wrong spots, where they ought not to promote healing, will serve as a placebo. In tests of antidepressants, inert pills, containing, say, sugar, act as placebos.

Hearing placebos, people may think of the magic feather that lends Dumbo confidence and so allows him to fly—talismans that work through inspiring hopeful expectancy. As we shall see, scientists disagree about how commonly this specific phenomenon, the classic placebo effect, operates in clinical medicine. In practice, the placebo arm of a trial covers a host of sins and virtues. Some patients will recover naturally, over time. Others will, on entering the trial, have been misdiagnosed; later, when diagnosed correctly, they will be counted as free of the disease under study. These recoveries, real or apparent, are not due to the inherent efficacy of the treatment.

The control condition packages what statisticians call nuisance variables: the noise that researchers want to filter out—as with noise-canceling headphones a passenger might use during an air flight—so that they can hear the pure music of the active treatment. Placebo is the equivalent of static, engine hum, and murmured conversation—everything other than the signal we’re listening for.

The statisticians who designed the streptomycin trial were unconcerned about classic placebo effects. The antibiotic was given by injection into a muscle, four times a day; participants in the control arm were not subjected to that pain. The control arm took into account improvement due to bed rest, nursing care, and natural waning of the infection.

All well and good—we will contrast bed rest plus medicine to bed rest alone. But the comparison will be valid only if, when our study begins, the patients receiving each treatment are similar: equally ill, equally resilient, and so on.

In some early controlled trials, experimenters alternated assignments, giving every other subject the medicine under test. Practicing doctors could game the system, holding back promising patients, timing their entry so as to afford them the presumed benefit of the active treatment. If research participants on the drug did better, the difference might be due to their having been healthier to start with.

This and other hidden sources of confusion are called confounds. Often, confounds result from bias, a systematic flaw, such as the pattern of assigning promising patients to receive the drug. This particular problem, the one caused here by meddlesome doctors, is called susceptibility bias; participants in one arm are more susceptible to further illness and death than those in the other arm. Bias can lead to spurious findings—in this instance, an inflated estimate of a drug’s powers.

In a randomized trial, researchers use something like a coin toss—typically they turn to a table of random numbers—to pair subjects and interventions. In the streptomycin study, once a patient had been accepted, a numbered envelope was randomly selected and then opened to reveal a card with the letter C, assigning that patient to the control condition, or S, for streptomycin.

The streptomycin trial suggests an image of how randomization works: A triage nurse sits at a desk. A line of participants approach her. Without regard to any trait of the person before her, she reaches into a box and hands over a sealed envelope, to be delivered to the trial pharmacist, containing instructions that consign the research subject to the trial’s active-treatment arm or control arm.

We need the nurse-and-envelope image because in our exploration of the antidepressant controversy we will encounter experiments that get randomization wrong. They have the nurse do something different, say, handing young patients an envelope with one sort of instructions and old patients an envelope with another. We will need to watch for violations of randomization.

The purpose of randomization is to avoid confounds, especially those arising from susceptibility bias. As the editorialist in the British Medical Journal noted in response to the streptomycin trial, “The random allocation has not only removed personal responsibility from the clinician and possible bias in his process of choosing patients, but has on the whole effectively equated the groups—fundamental, of course, to the general comparisons…”

The beauty of the randomized trial is that the researcher does not need to understand all the factors that influence outcomes. Say that an undiscovered genetic variation makes certain people unresponsive to a medication. The randomizing process will ensure—or make it highly probable—that the arms of the trial contain equal numbers of subjects with that variation. The result will be a fair test.

double-blind trial includes another safeguard against bias: neither the patients under study nor the observers measuring outcomes know which subject has received which intervention. This uncertainty protects against patients’ and doctors’ favoring the promising treatment and reporting overly encouraging results in the active arm of the trial.

The streptomycin-versus-bed-rest study had this protection in partial form. Radiologists reading a film were not told whether it came from the file of a patient on medication. But clinicians and patients knew who was getting injections.

Doctors had long believed that morale affected recovery from pulmonary tuberculosis. Group psychotherapy traces its origin to instructive and inspirational meetings for tubercular patients conducted with the support of a church in Boston in the early 1900s. The Boston Medical and Surgical Journal, precursor to The New England Journal of Medicine, reported that the method outperformed sanatorium treatment. By the 1940s, the understanding of tuberculosis had changed. If getting a shot induced optimism, and if optimism revved up the immune system, no one worried that this potential confound would invalidate the findings.

Assumptions about the nature of an illness affect the design of research. With streptomycin, doctors accepted that antibiotic action, and not hope aroused by injections, would determine outcomes. Different beliefs would make researchers conduct the trial differently.

Scientists favor double-blind, randomized, placebo-controlled trials, but there is no doubt about which element matters most. The British Medical Research Council could do with partial blinding and without a placebo. If a trial is to be run, the essential element, the one that statisticians most hate to forgo, is randomization. Randomization is a great medical advance of the twentieth century, on a par, historians say, with penicillin. Randomization minimizes confounds. The midcentury epidemiologist Archie Cochrane, an advocate of evidence-based medicine avant la lettre, has become the movement’s hero and icon. Cochrane had little respect for doctors’ wishes to spare their patients exposure to placebo and to offer them promising, unproven therapies. Challenged by a colleague who demanded to know how far Cochrane was prepared to push “this randomizing game,” he shot back, “You should randomize until it hurts [the clinicians].”



If you find an error or have any questions, please email us at admin@doctorlib.info. Thank you!