The extent to which antidepressants are effective for pediatric depression continues to stir controversy, as discussed in this issue of CCPR. Parents, the media, and even our young patients themselves sometimes tell us that our medications don’t work and may well be dangerous. And yet, as clinicians, we know of many children who have apparently benefited from using them. So we continue to prescribe, if sometimes with ambivalence or ill-conscience. This article puts the question of antidepressant efficacy in some perspective.

What does it means to say that a medication “works”? An intuitively understandable yet scientifically respectable answer is to use the Number Needed to Treat (NNT). (For a good explanation of the concept, see Citrome L, Acta Psychiatr Scand 2008;117:412–419.)

NNT is the number of patients who must take the drug (or undergo the studied intervention) to obtain one more favorable outcome than the alternative treatment. For placebo-controlled studies, NNT can be calculated by taking the reciprocal of the absolute risk—that is, the reciprocal of the percentage of people who got better when treated, minus the percentage of people who got better with placebo.

For example, the TADS study found that after 12 weeks of treatment, 60.6 percent of adolescents remitted with fluoxetine and 34.8 percent remitted with placebo (March J et al., JAMA 2004;292(7):807–820). The NNT would be the reciprocal of 0.606 minus 0.348, or 1/0.258=3.87, which rounds up to 4. This means that for every four patients treated with fluoxetine, one more would respond than in the placebo group.

A relatively effective drug has a low NNT, while a relatively ineffective drug has a high NNT. However, it may well be worth treating many patients, even if just a few benefit, when the cost of the medication is low and the risks are small. For example, the NNT of phototherapy for newborn girls with jaundice is 222 (333 for newborn boys), but the risks of photo-therapy are very low (Newman T et al., Pediatrics 2009;123(5):1352–1359). Consequently, phototherapy is recommended for all newborns with jaundice even though only a few will avoid a blood transfusion they would have otherwise required.

Number Needed to Harm (NNH) is an analogous statistic that measures the risks of an intervention. It is calculated in exactly the same way as NNT (one over the absolute risk), but the outcome used is different. Instead of a beneficial outcome (eg, remission from depression), it is based on an undesired outcome, such as suicidal ideation. The NNH is the number of patients who have to take the drug to have one more suffer the bad outcome than if they had not been treated. The ideal drug or intervention has a low NNT and a very high NNH.

The FDA meta-analysis that generated the black box warning for antidepressants in children showed an absolute risk of one percent for serious suicidal ideation or attempt and two percent for a broader definition of suicidal ideation. This means that for every 100 patients treated, one to two of them are likely to experience suicidality that they would not have without drug treatment. The NNH is, therefore, 100 for serious suicidality and 50 (the reciprocal of two percent) for broad suicidality.

There have been published studies of antidepressant medications that showed no benefit over placebo, but the literature is not entirely devoid of studies showing a benefit. Furthermore, it is hard to know what to make of the negative studies—are they insufficiently powered to detect a difference? Or perhaps the placebo effect is so high as to be insurmountable, as posited by Bridge et al., who noted that the placebo response in 12 antidepressant studies examined averaged 44 to 58 percent (Bridge JA et al., Am J Psychiatry 2009;166(1):1–3). The clinical trials registry (found at http://clinicaltrials.gov), which addresses the problem of unpublished negative studies, will eventually allow better meta-analyses by making both negative and positive findings more available.

In the meantime, it’s important to look at the data we do have in a larger context. When compared with medications in other medical specialties, you may be surprised at how the antidepressants, even when used with adolescents and children, stack up.

In the accompanying chart, I have listed the NNTs of some antidepressants for comparison with some common medical interventions. For example, statins: The NNT (for the outcome of prevention of a cardiovascular event) is 37 for women and 33 for men. These numbers appear much less “impressive” than the lower NNTs for antidepressants. And yet, there is little question among physicians or patients that this represents enough of a benefit to warrant treatment.

What about the statins’ NNH? The NNH for the development of cataracts due to statins is 33 for women and 52 for men. These numbers are comparable or lowerthan the NNH of 50 to 100 for suicidal events in the FDA database. Furthermore, there were no episodes of actual suicide reported.

How about penicillin for strep throat? The American Heart Association recommends the treatment of strep to decreasethe risk of rheumatic heart disease, even though throat strep resolves without treatment in most cases. Check the table for some other surprising statistics.

The bottom line? While we would love to see more impressive remission numbers for antidepressants in children, let’s keep in mind that the NNTs and NNHs of these meds are in many cases more impressive than the comparable numbers for nonpsychiatric medications whose value we rarely question.

Number Needed to Treat Among Antidepressants and Common Medical Interventions
ANTIDEPRESSANT	NUMBER NEEDED TO TREAT (NNT)	OUTCOME MEASURE	REFERENCE
Fluoxetine (10 mg to 40 mg)	4	Improvement in CDRS-R and CGI	March J et al., JAMA 2004;292(7):807-820 (TADS)
Paroxetine (20 mg to 40 mg)	6	HAM-D score less than or equal to 8	Keller MB et al., JAACAP 2001; 40(7):762-772*
Imipramine (200 mg to 300 mg)	25	HAM-D score less than or equal to 8	Keller, ibid
Sertraline (50 mg to 200 mg)	10	Improvement in CDRS-R and CGI-I	Wagner K et al., JAMA 2003;290(8):1033-1041
Citalopram (20 mg to 40 mg)	8	CDRS-R (no significant difference in CGI-I)	Wagner K et al., Am J Psychiatry 2004;161(6):1079-1083
MEDICAL INTERVENTION	NNT	OUTCOME	REFERENCE
Statins for adults at high risk for cardiovascular disease	35	Prevention of cardiovascular events	Hippisley-Cox J, Coupland C, BMJ 2010;340:c2197-c2197
Penicillin for acute strep pharyngitis	35	Prevention of rheumatic fever	Catanzaro et al., Am J Med 1954;17(6):749-756
Phototherapy for newborns with jaundice	281	Prevention of blood transfusion	Newman et al., Pediatrics 2009;123(5):1352-1359
Tissue Plasminogen Activator for stroke (in adults, within 3 to 4.5 hours of symptoms)	7	Improvement on the Rankin scale of stroke symptoms	Saver et al., Stroke 2009;40(7):2433- 2437
Number Needed to Harm Among Antidepressants and Common Medical Interventions
INTERVENTION	NUMBER NEEDED TO HARM (NNH)	OUTCOME	REFERENCE
SSRIs	50	Suicidality	Hamm d et al., Arch Gen Psychiatry 2006;63(3):332-339
Statins, as above	42	Cataract	Hippisley-Cox, op.cit
Penicillin (as for strep)	20	Self-reported allergy	Macy E, Poon K-Y T, Am J Med 2009;122(8):778.e1-778.e7
Tissue Plasminogen activator, as above	37	Hemorrhagic event	Saver, op.cit
**To be fair, although Keller found a statistically significant difference, several other studies of paroxetine have been negative, and the consensus is against both paroxetine and imipramine being effective in kids. This table is not meant to dismiss the controversy in the field regarding antidepressants, but to demonstrate a means of considering the data across studies and across medical specialties. Note: I have glossed over certain statistical sophistications, such as the difference between “persons” and “person-years” and also that I have rounded to whole persons, as is conventional.
GLOSSARY OF TERMS
CDRS-R	Children’s depression rating scale-revised
CGI	Clinical global impression scale
CGI-I	Clinical global impression improvement scale
HAM-D	Hamilton depression scale