By Philip R. Alper,MD Dr Alper is Clinical Professor of Medicine, University of California, San Fransisco, Visiting Scholar, Hoover Institution, Stanford University, and Practices of Internal Medicine in Burlingame, Calif. e-mail:

Prescribing medication is supposed to be straightforward and evidence based. But there is a lot to consider, and relevant evidence is often lacking. First, is it necessary or preferable to use medication at all? If it is, which medication best fits the condition being treated? And if the answer is positive, does it fit the particular patient being treated? (I'm glossing over cost and insurance.)

Allergies and incompatibility with other drugs may also send us back to the drawing board. This whole process is instinctive and automatic and seems to go on all at the same time. But it isn't enough. Perhaps the most important question to ask patients almost never gets asked—"Are you lucky?"

Just take a few moments with the Physicians' Desk Reference or, more conveniently, with the ePocrates Rx program, and have a glance at the listed side effects seen with drugs in everyday use. They are horrendous: Stevens-Johnson syndrome, malignant cardiac arrhythmias, anaphylaxis, and potentially fatal bone marrow disturbances, among many others.

True, the most serious reactions are infrequent. If they weren't, the Food and Drug Administration would never have approved the drugs for general use. But most drugs carry with them the potential for at least one "serious" adverse effect. In addition, the intended therapeutic effect may also cause harm, either due to incautious prescribing or unusual patient sensitivity. Blood pressure or blood sugar may drop too far. Sedation may be excessive. And so forth.

It has been said that physicians who encounter a fatal penicillin reaction early in their career will never again prescribe penicillin appropriately. Professional preparation doesn't matter. Men and women of science are still human and are subject to guilt and fear and may unconsciously favor errors of omission over those of commission once they have been traumatized. Like it or not, we play the odds when we prescribe, and along with all our risk/benefit analysis, past experience plays a role. It can make us unduly cautious…or the opposite.

And this is before we consider personality—upon which experience is superimposed. What induces us to pick a drug that may be more effective but slightly more risky over another that is safer but less effective? What gets us to recommend surgery for something that may be uncomfortable (like arthritis or a hernia) rather than continue to "just live with it"? We may quibble over whether the choice represents personality or values, but we'll have to agree that there are no black and white answers.

Bad choices, and bad luck, confer risk to the physician for every choice that is made. There is a risk to reputation and possibly legal liability. But even when there is no public accusation of fault, there is still a personal awareness of having harmed a patient. Sadness and guilt are worse when a less-than-ideal drug has been picked. But these negative emotions are not altogether absent when bad luck alone (as in an initial anaphylactic penicillin reaction) has left patients worse off for having seen us.

The trade-offs in selecting pharmaceuticals get even more complicated when paradoxical risks and benefits appear. For example, a drug to treat diabetes may increase the incidence of heart problems but protect against breast cancer. The question then becomes: What kind of heart problems, and how severe are they—and, just how much protection against breast cancer is there? Here too, there is no gold standard to use for weighing given harms against given benefits.

Then there is patient variation. My father had hay fever. A single over-the-counter antihistamine would make him drowsy for the next 24 hours. I could take double the dose and suffer no ill effects. That kind of variation occurs all the time in practice. Educated guesses based on patient characteristics may help a little in predicting who is likely to respond to treatment (eg, when choosing among classes of antihypertensives). Unfortunately, there are very few clues, apart from the history, to predict who will develop side effects, or atypical sensitivity, or resistance to drugs. There is hope that the human genome may help, but the potential complexity and cost of such "personalized medicine" are staggering, as we are currently learning with the controversy over pretreatment genetic testing for unusual warfarin sensitivity.

Perhaps the greatest challenge in prescribing occurs when drugs produce opposite effects in different patients. The problems may be relatively innocuous (eg, constipation in one patient and diarrhea in another). More problematic are paradoxical responses opposite to the intended effect (ie, agitation in response to a tranquilizer). Such paradoxes may reach tragic proportions with the use of antidepressants.

I am personally aware of 2 cases in which a highly popular antidepressant was prescribed in the lowest possible dose to different adults. Within 2 days, each patient developed tremors, agitation, and insomnia. Both stopped the drug, and one recovered uneventfully. The other patient took an overdose of sleeping pills the following day. Fortunately, she survived, and when she awoke from coma she described a sudden overwhelming impulse to end the awfulness of the way she felt. This phenomenon is known to occur in adolescents and much less often in adults. It is thought to be due to "disinhibition" that precedes onset of the antidepressant effect. The risk is low but real. Yet $13 billion were spent worldwide on antidepressants in 2006. The class action lawsuits are just beginning.

Paradoxical drug effects, extreme variation in patient response, and selective response to some drugs (or to just one) but not others, even within the same class, aren't unique to antidepressants. But what is one to do—patient or doctor—when antidepressants lower the overall incidence of suicide but induce suicidal behavior in an unpredictable subset of depressed patients?

This conundrum sparked an editorial citing suicidality and 45 additional pairs of paradoxical reactions to antidepressants (Lieb J. Variations: Darwin's finches, sea barnacles and the side effects of antidepressants. Med Hypotheses. 2007 Aug 6. Online before print). The author believes that adverse effects are manifestations of human variation and that "paradoxical reactions suggest that many people are more unlike each other at the molecular level than alike."

Considering how blind we are flying when prescribing, it is clear that in an ideal world, patients would be consulted about risks, benefits, and preferences before every prescription. But as Dr Lieb asks, "Given variation, how does one respond to a patient who asks, 'What are the side effects?' when there are two hundred?" Should every patient be warned about the rarest of undesirable reactions? How much time should be devoted to such discussions, and where will the time be found?

These are new questions. In the bad old days of paternalism, physicians didn't dwell on these issues and did the best they could to choose wisely on behalf of patients. It was an imperfect solution. Now we live in an age of informed consent, and we must deal with the power of suggestion—creating side effects by warning about them. We can follow strict protocols to protect ourselves and please the most zealous of consumer advocates, but that hardly seems like humane medical care. What we say and how we say it tend to vary with our assessment of the needs of individual patients.

But in the end, asking, "Are you lucky?" isn't all that crazy. Still, the question may be too broad. One woman I spoke with considers herself very lucky to have surmounted great adversity in her life. Medications are a different story, however; she gets bad reactions to "all of them!" It took her 2 years to "get over" taking verapamil for an irregular heartbeat—surely a record.

Unfortunately, "lucky" does not appear in the lexicon of evidence-based medicine. It should.