For ten years or more, doctors have been debating whether the new antidepressants, those that alter the way the body handles serotonin, are associated with gastrointestinal bleeding. A study in the current Archives of General Psychiatry suggests that certain of the selective serotonin reuptake inhibitors do cause upper GI bleeds, with (the researchers and the press report) a “number needed to harm” of 2000. That is, if you treat 2000 patients with SSRIs, you will see one excess bleed — not an insubstantial risk. But as is often the case, the fuller story is more complex.

In this blog, we have looked at length at the statistics “number needed to treat” and “number needed to harm,” or NNH. I thought it might be fun to think through the recent findings (they are from Francisco de Abajo and Luis A. García-Rodríguez, of the Spanish equivalent of the FDA), in terms of NNH.

But first, an overview: There is theoretical reason to believe that SSRIs should cause bleeding. The drugs alter the functioning of platelets, the circulating blood cells that help start the clotting process. This change may be one reason that SSRIs are useful for patients who have had heart attacks; the medications don’t just mitigate depression, they also thin the blood. But studies of SSRIs and GI bleeding have been contradictory. Most find an effect, while some show none at all.

The new research looked at 1321 patients with upper GI bleeds and 10,000 controls. The scientists then compared the two groups in terms of medication use — antidepressants and non-steroidal anti-inflammatory drugs, or NSAIDs, like Motrin. You might think that the sample size would be large enough to lead to clean contrasts, but patients with GI bleeds look different from the general population. Compared to the control group, the subjects with bleeds were older and were more likely to be smokers and heavy drinkers. They also took more medications (such as NSAIDs) that can cause bleeding. And they had more prior ulcers and other negative GI events.

In a coarse analysis, it looked as if SSRIs more than doubled the risk of GI bleeding. But factoring out other sources of harm, like alcohol use, that risk became 1.6 times, and it was yet less for Prozac and Paxil.

That’s right: the study reports that for Paxil, Prozac, and some other antidepressants, “the risk was decreased and rendered nonsignificant” after confounding factors were taken into account. I’m not at all sure what that finding means; the SSRIs may all be risky, in terms of GI side effects. (My own math finds a NNH for Prozac of 6666. How vague is this finding? At the 95% confidence interval, the margin of error includes the possibility that Prozac protects against GI bleeds.) Still, I am always curious to see whether journalists attend to details such as “decreased and rendered nonsignificant.” The Bloomberg wire report, as run by Newsday, is headlined “Prozac, antidepressants linked in study to stomach bleeding” and reads “Drugs known as selective serotonin reuptake inhibitors, such as Eli Lilly & Co.'s Prozac, Forest Laboratories Inc.'s Celexa and Lexapro, GlaxoSmithKline Plc's Paxil and Pfizer Inc.'s Zoloft, could trigger gastrointestinal bleeding . . .”

Actually, the press failed to pick up a more important detail. The researchers used their unadjusted figures to come up with the number needed to harm. Their formula for the NNH is: NNH = 1/[(OR – 1) UER], where OR is the odds ratio and UER is the rate of bleeds in people not exposed to medications, in this case 0.5 per 1000 people, or .0005. The adjusted odds of a bleed on SSRIs were 1.6, giving an NNH of 3333, or 50% less risk than the figure quoted in the press. As I say, the Prozac risk is half that again.

The greater risk was for Effexor, which affects a broader range of neurotransmitters and which had, by my calculation, an NNH of about 1000, or more than three times the risk of the more selective drugs, the SSRIs. The risk increases substantially for patients taking both one of the newer antidepressants (SSRIs or Effexor) and NSAIDS. For patients taking both sorts of medications at once, the NNH was, again by my math, something over 500. The real risk arises for patients on SSRIs and NSAIDs who do not take acid-suppressing agents, like Prilosec; the NNH may be as low as 250. (It is hard to know what to use for comparison; for adverse metabolic events in response to the newer antipsychotics, the NNH appears to be about 15.) But for patients on SSRIs and NSAIDS who took acid suppressors, the NNH was 20,000. That is to say, acid suppressors offered basically complete protection.

The NNH figures tell a story with clear morals. If you are prone to GI bleeds, or if you are on medications like Motrin, then when you take an antidepressant like Zoloft or (especially) Effexor, you should also take an acid-suppressor. Otherwise, the risk of bleeding for antidepressant users appears to be reasonably low. Still, with millions of patients on these medications, even an NNH over 6500 is worth thinking about. In their annual physicals, antidepressant users might well have their internists test a stool sample for occult blood.