MONTREAL, Canada—A study by the Canadian Multicentre Osteoporosis Study (CaMos) Research Group published this week in the Archives of Internal Medicine concludes that daily use of selective serotonin reuptake inhibitor (SSRI) antidepressants doubles the risk of fragility (low-trauma) fractures in the community-dwelling elderly.¹ The conclusion was widely reported in the consumer media but may not be as firm as it seems, according to pharmacoepidemiologist Sebastian Schneeweiss, MD, ScD.

J. Brent Richards, MD, lead author of the study, concluded, "Daily SSRI use in adults 50 years and older remained associated with a 2-fold increased risk of clinical fragility fracture after adjustment for potential covariates. Depression and fragility fractures are common in this age group, and the elevated risk attributed to daily SSRI use may have important public health consequences."

Problems with the report

The study used a population-based, randomly selected, prospective cohort of 5008 community-dwelling adults 50 years and older, who were followed over 5 years for incident minimal-trauma fractures. Of these, 137 subjects reported daily use of SSRIs, including citalopram (Celexa®), fluoxetine (Prozac®), fluvoxamine (Luvox®), paroxetine (Paxil®), and sertraline (Zoloft®).

"The authors do not report the number of fractures they observed in a cohort of 137 SSRI users and 4871 nonusers. From the figure, it can be guessed that there were very few events in the SSRI group," said Dr. Schneeweiss, who reviewed the study for CIAOMed.

Dr. Schneeweiss, associate professor of medicine and epidemiology at Harvard Medical School and Harvard School of Public Health and is in the division of pharmacoepidemiology at Brigham and Women's Hospital, in Boston, Massachusetts, has a long-standing interest in the problem of how to use database information to draw clinically valid conclusions. In 2004, he conducted a major review of claims data studies of the suspected association between hip fracture and SSRIs.² That analysis showed that epidemiological claims data studies tended to overestimate the relation between antidepressant use and hip fractures, but that a significant association (RR 1.8) with hip fracture persisted even after correction for this bias.

In the current study, Richards et al consider a number of potential confounders, including physical activity, calcium and vitamin D intake, comorbidities, and medication use. The researchers conclude that daily SSRI use is associated with an adjusted hazard ratio of 2.1 for fragility fractures, and that the risk increases with dose.

Dr. Schneeweiss told CIAOMed that the Richards study has several limitations, including lack of data on the number of fractures observed in the cohort of SSRI users versus nonusers, a problematic method used to adjust for potential confounders, and an "exposure" group composed of prevalent SSRI users rather than new users.

"The report provides too little information to evaluate the validity of its findings, but raises some concerns," Dr. Schneeweiss said. He suggested that ideally such a study would have to be 10 to 100 times larger to produce clinically useful information. "Any clinical recommendation on the basis of the present study may be too early," he cautioned.

Bigger study needed

Dr. Schneeweiss also noted that the method used to adjust for confounders is problematic because of the very small number of fractures observed. "Propensity score techniques are the preferred method in pharmacoepidemiology when events are rare but exposure is frequent," he explained. "Although exposure to SSRIs is not exactly frequent (137 patients), such propensity score estimation would have allowed the inclusion of all covariates in one score." He said that the preferred use of such a propensity score would then be to match treated and untreated patients by the score.

"This would have reduced the study population to 137 times 2 equals 274 patients. Although this would underpower the present study, the advantage of the method is that these patients have overlapping propensities for treatment given all measured confounders. This avoids parametric extrapolations of the data by the regression model into areas of the covariate space not supported by data. The method used by the authors may have provided a reasonable model fit, but it is impossible to evaluate given the sparse information they provided," Dr. Schneeweiss said.

References

1. Richards JB, Papaioannou A, Adachi JD, et al. Effect of selective serotonin reuptake inhibitors on the risk of fracture. Arch Intern Med. 2007;167:188-194.
2. Schneeweiss S, Wang PS. Association between SSRI use and hip fracture and the effect of residual confounding bias in claims database studies. J Clin Psychopharmacol. 2004;24:632-638.