Introduction

Osteoporosis remains a significant cause of morbidity and mortality, with the estimated 2 million osteoporosis-related fractures in the USA in 2005 forecasted to exceed 3 million per year by 2025 [1]. Most osteoporotic fractures are attributed to primary osteoporosis, which refers to the reduced bone mass, skeletal microarchitectural deterioration, and increased bone fragility associated with aging and aging-related changes in sex steroid hormones. However, there is increasing acknowledgement of the contribution of secondary osteoporosis to the overall fracture burden, with secondary causes for osteoporosis being identified in up to two thirds of men, more than half of premenopausal women, and one fifth of post-menopausal women with osteoporosis [2].

Secondary osteoporosis refers to bone changes and a heightened risk of low-trauma fractures due to age independent mechanisms. Sample causative factors include lifestyle choices (such as smoking and chronic heavy alcohol use), injury (such as neurological insult), disease (such as systemic inflammatory disorders, malignancy and chronic kidney disease), surgical interventions (such as solid organ transplantation), and the use of specific medications. In terms of the latter, off-target negative skeletal effects have been reported for corticosteroids, aromatase inhibitors, androgen deprivation therapy, thiazolidenediones, and proton pump inhibitors, to name a few [3]. Most recently, there have been suggestions to add selective serotonin reuptake inhibitors (SSRIs) to the list of drugs contributing to secondary osteoporosis [4–8].

SSRIs antagonize the serotonin transporter to potentiate extracellular serotonin activity and effectively alleviate the symptoms of depression and other affective disorders. They are the most commonly prescribed antidepressant and one of the most commonly prescribed drugs overall, with 8.5 % of adults in the USA reporting use of an SSRI in 2011–2012 [9]. This value represents a doubling in the prevalence of SSRI use since 1999–2000 [9], which likely reflects expanded recommendations for SSRI use in conditions other than affective disorders [10]. Of concern with regard to potential negative skeletal effects of SSRIs is the fact that the population most at risk of developing primary osteoporosis (i.e., Caucasian White females) is also the most frequent users of antidepressants [11]. It is possible that skeletal changes induced by SSRIs exaggerate osteoporotic fracture risk in already at-risk individuals.

The call to add SSRIs to the list of potential causes of secondary osteoporosis has been stimulated by an apparent increase in fracture risk in those taking the class of agents. In particular, two recent meta-analyses of case-control and cohort studies independently reported a relative risk (RR) of fracture in SSRI users compared to non-users of 1.7 (95 % confidence interval [CI], 1.5–1.9) [12•, 13•]. These observations were confirmed by a third more recent meta-analysis [14]. However, debate continues as to whether to formally add SSRIs to the list of secondary causes of osteoporosis as the currently available data are observational and limited in their ability to establish causality.

The causality of the observed association between SSRIs and fractures may be assessed using the Bradford Hill criteria [15]. While these criteria were not specifically created as a checklist [16], they describe a set of conditions that can be considered in determining causality of observed associations and consist of the following: (1) strength—what is the magnitude or effect size of the association?; (2) consistency—how reproducible is the association?; (3) specificity—are there alternative explanations for the association?; (4) temporality—is there a temporal relationship between cause and effect?; (5) biological gradient—is there a dose response?; (6) biological plausibility—is there a plausible biological mechanism for the association?; (7) coherence—is there coherence between epidemiological/observational and laboratory findings?; (8) experimental evidence—is there controlled evidence supporting the association?; and (9) analogy—has a causal relationship been accepted for an analogous scenario?

The current paper will discuss the relationship between SSRIs and fracture according to the first seven of these criteria. There is very limited experimental evidence for SSRIs and fractures, although some evidence points to an effect of SSRI effects on bone strength. This evidence will be discussed under the plausibility criterion. Analogy will not be discussed as we are not aware of a causal relationship being accepted for a similar scenario.

Strength

There is a strong relationship between SSRIs and fractures, as shown by an approximately 70 % increase in fracture risk in SSRI users compared to non-users in meta-analyses [12•, 13•]. While there is no accepted scale for determining the significance of a RR based solely on its magnitude, significance can be determined by knowing the level of risk in the control group (i.e., non-SSRI users). The lifetime risk for osteoporotic fracture in White women over the age of 50 has been reported at 39.7 % [17]. Conservatively assuming a SSRI usage rate of 12 % in elderly women [9] and approximately one million White women reaching menopause annually in the USA [17], the increased fracture risk associated with SSRI use suggests that approximately 30,000–35,000 White women in the USA who reach menopause annually will ultimately suffer an SSRI-related fracture during their lifetime. This is a significant number of potentially preventable fractures.

Consistency

A consistent positive relationship exists between SSRIs and fractures. By our count of currently available studies, all 7 case-control studies [18–24] and 10 out of 12 cohort studies [25, 26••, 27–34] reported a significant positive association between SSRIs and fractures. The two cohort studies failing to demonstrate a significant association reported RRs similar to the overall meta-analysis-derived RRs but had large CIs that impeded reaching statistical significance [35, 36].

The consistency of the association between SSRIs and fractures is also evident by the narrow CI for the association in pooled meta-analysis data, as well as the robustness of the meta-analysis outcomes to sensitivity analyses [12•, 13•, 14]. In terms of the latter, the magnitude of the calculated association between SSRIs and fractures was not influenced by study design (case-control vs. cohort study), sample type (population-based vs. convenience), study methodological quality (low vs. high quality), study geographic location, or sex of the study participants (female vs. male) [12•, 13•, 14].

Specificity

Specificity is a concern regarding the association between SSRIs and fractures. The strong and consistent relationship between SSRIs and fracture has been derived from observational data, which is susceptible to the influence of confounders. Available case control and cohort studies exploring the relationship between SSRIs and fractures used a variety of confounding variables in their models to account for factors contributing to the association. The importance of controlling for potential confounders was evident in two meta-analyses associating SSRIs with fractures. Wu et al. [13•] observed in eight studies reporting both unadjusted and adjusted RRs that adjusting for confounders reduced the overall pooled RR from 2.25 (95 % CI, 1.90–2.68) to 1.76 (95 % CI, 1.51, 1.95). Similarly, Eom et al. [12•] observed that studies adjusting for more than four potential confounders reported a lower association between SSRIs and fractures than studies adjusting for less confounders.

Although the published studies exploring the association between SSRIs and fractures have adjusted for varying degrees of confounding, the potential for residual confounding by unknown or unaccounted factors is unavoidable using observational study designs. With regard to the relationship between SSRIs and fractures, the biggest concern is confounding by indication. This occurs when a treatment and the condition for which it is being administered both exhibit increased risk for the outcome under study [37]. With regard to SSRIs, the increased fracture risk associated with their use may be due to the skeletal consequences of their primary indication—depression.

Most studies exploring the relationship between SSRI use and fractures compare SSRI users with a diagnosis of depression to non-depressed, non-users. There is general consensus from multiple meta-analyses that depression is independently associated with fracture and the major determinants of fracture—low bone mass and falls [38–43]. For instance, depression was found to be associated with a 17 % increase in fracture risk in six studies reporting outcomes as hazard ratios (HRs) (1.17; 95 % CI, 1.00–1.36) and a 52 % increase in fracture risk in four studies reporting outcomes in terms of RR (1.52; 95 % CI, 1.26–1.85) [41]. Reduced bone mass may contribute to the heightened fracture risk associated with depression, with individuals exhibiting major depressive disorder having 4.7, 3.5, and 7.3 % lower DXA-derived BMD at the spine, total femur, and femoral neck, respectively [38]. Similarly, falls may be a contributing factor, with a meta-analysis of 17 prospective studies finding an odds ratio for the association of depression and falls of 1.63 (95 % CI, 1.36–1.94) in community-dwelling older people [39].

As depression appears to be independently associated with fracture, it is possible that the increased rate of fractures in individuals taking SSRIs simply reflects confounding by indication. Some studies attempted to control for confounding by indication by adjusting for depression and/or the existence of mental disorders [19, 21–24, 26••, 32, 33]. In one meta-analysis, the pooled RR for four studies that adjusted for depression was equivalent to the pooled RR for eight studies that did not adjust [13•], suggesting that SSRIs are independently associated with fracture. However, as the methods used to define depression have been varied or not reported and usually did not take into account disease chronicity or severity, concern for confounding by indication remains.

To control for confounding by indication, a number of studies compared fracture rates between SSRI users and users of tricyclic antidepressants (TCAs) [21, 23–25, 26••, 28], with the assumption that SSRIs and TCAs are prescribed for the same indications. Although TCA users have an increased risk of fracture compared to non-users (RR = 1.45; 95 % CI, 1.31–1.60) [44], the risk appears higher in SSRI users. In particular, case-control and cohort studies directly comparing fracture risk between TCA and SSRI users observed higher fracture rates in the latter [21, 23–25, 26••, 28]. As TCAs and SSRIs are both prescribed for affective disorders, direct comparison between treated groups suggests an association between SSRI use and fracture that is not due to confounding by indication. In support of this hypothesis, a recent comprehensive cohort study by Coupland et al. [26••] demonstrated an elevated fracture risk in SSRI users with a diagnosis of depression compared to patients with a similar diagnosis that elected to remain non-medicated.

The contrasting fracture risk in patients with similar diagnoses but differing treatment approach reduces concerns regarding confounding by indication; however, this approach presents a new concern with regard to channeling bias. Channeling bias occurs when groups of patients are prescribed therapies based on different baseline prognostic factors [45]. It is possible that SSRIs were selectively prescribed to patients who possessed baseline characteristics that ultimately placed them at a greater risk of fracture than patients who were alternatively managed. For instance, SSRIs may have been preferentially prescribed over TCAs to individuals with a greater perceived risk of falls. Investigators typically adjust their RRs for various differences in baseline characteristics in their cohorts and cases; however, unassessed or unidentified characteristics likely contributed to treatment selection resulting in some degree of channeling bias.

An alternative approach that better establishes a causal relationship between SSRIs and fracture is performance of a self-controlled case series. During a self-controlled case series, the incidence of fracture during SSRI exposure periods is compared to the incidence during baseline non-risk periods within the same individual. This effectively controls for between-person confounding. To our knowledge, two published studies have used a self-controlled case series approach to explore the association between SSRIs and fracture [20, 26••]. Both studies reported significantly elevated fracture risk throughout SSRI treatment periods compared to baseline observation periods, with risk remaining elevated for up to 3 months following treatment cessation (Fig. 1). These data provide relatively strong observational evidence for a causal relationship between SSRIs and fracture that is not due to confounding by indication.

Fig. 1
figure 1

Self-controlled case series suggest a causal relationship between SSRIs and fractures. Using an approach whereby fracture risk was assessed within-subject during periods of SSRI use and non-use, both Hubbard et al. [20] and Coupland et al. [26••] observed an increased relative risk of fracture within the first 2–4 weeks of SSRI use. Risk remained elevated during ongoing SSRI use and for up to 3 months after SSRI use was stopped. Asterisk indicates significantly elevated fracture risk compared to baseline observation period

Full size image

One additional piece of evidence has been presented as providing proof that the association between SSRIs and fracture is not due to confounding by indication. In a recent cohort study, Sheu and colleagues [31] compared fracture rates in perimenopausal women without a diagnosis of any psychiatric disorder that were treated with either an SSRI or a H2 antagonist (H2A)/proton pump inhibitor (PPI). SSRIs are used in this population for the management of vasomotor menopausal symptoms (VMS) [46], with H2A/PPI users chosen as a comparator group in the study by Sheu and colleagues [31] on the assumption that H2As have a trivial association with fracture risk and PPIs are associated with a slightly higher risk [47]. HRs were higher in SSRI users than H2A/PPI users, suggesting that the relationship between SSRIs and fractures is independent of depression; however, the finding is still limited by confounding by indication. In this case, the indication for SSRI use was VMS, which is itself associated with low bone mass and a heightened risk of fracture [48, 49].

Temporality

There is a clear temporal relationship between SSRIs and fractures, which is most clearly and convincingly observed in the self-controlled case series [20, 26••]. Using within-person analyses, these studies show that fracture risk increases within the first 2–4 weeks of treatment and remains elevated throughout treatment (Fig. 1). Once treatment ceases, there appears to be a residual fracture risk for 3 months before risk returns to the same level as during baseline observation periods.

The rapid onset of the association between SSRIs and fracture in self-controlled case series [20, 26••], as well as in cohort [26••, 32, 33] and case-control [20, 21] studies reporting on risk within differing treatment windows, suggests that SSRIs cause an early increase in fracture risk due to an increase in falls. The two leading causative factors for fracture are altered bone properties and falls, and an increase in fracture risk in less than 1 month would appear to be too rapid to be caused by an alteration in underlying bone properties.

SSRI use has been linked to an increased risk of falls [50, 51]. As a result, the American Geriatrics Society has suggested that SSRIs be avoided in older adults with a history of falls or fractures [52]. However, there remains debate as to the association between SSRIs and falls [53]. Confounding by indication may contribute to the increased risk of falls in SSRI users in case-control and cohort studies, with depression independently elevating fall risk [39]. However, acute SSRI use has been independently associated with gait impairments associated with an increased risk of falls [54]. Also, Coupland et al. [26••] in their self-controlled case series found a 2.6-fold increase in fall risk in the first 4 weeks of commencing SSRI use. Thus, there are strong indications that SSRIs are associated with an early increased risk of falls, with these falls likely contributing to the rapid increase in fracture risk with SSRI use.

The rapid increase in fracture risk with acute SSRI use appears related to falls; however, alterations in bone properties may contribute to the increased risk of fracture with ongoing SSRI use. There is biological plausibility for a link between serotonin transporter inhibition and bone changes (see “Biological Plausibility and Coherence” section below), and numerous studies have associated SSRI use with reduced bone mass [5]. In particular, Diem et al. [55] found SSRIs to double the rate of bone loss at the hip among elderly women when assessed longitudinally over an average period of 4.9 years. It is plausible that the persistent increase in fracture risk with ongoing SSRI use shown in self-controlled case series is due to an ongoing increased risk of falls [26••] combined with underlying changes in the ability of bone to resist mechanical loads. However, there is a need to confirm the bone changes associated with SSRI use as Diem et al. [56] were unable to replicate their bone loss findings in middle-aged women tested over an average period of 5.9 years, and confounding by indication due to bone changes associated with depression is ever present [38, 40–43].

Biological Gradient

There is an apparent dose-response relationship between SSRIs and fractures. Case-control and cohort studies demonstrate that fracture risk increases with both increasing SSRI defined daily dose [19, 24, 26••] and heightened compound affinity for the serotonin transporter [23], although the presence of a dose-response was not confirmed by others [21, 28, 33]. Increasing risk of falls with increasing SSRI dose may be one contributing factor [26••, 57]; however, confounding by indication is a concern as increasing SSRI dose is likely in response to the presence of greater depressive symptoms. Increasing severity of depressive symptoms is associated with greater longitudinal bone loss [58, 59•] and more falls [60]. To our knowledge, there are no self-controlled case series data exploring SSRI dose and fracture risk.

Biological Plausibility and Coherence

Beyond the studies associating SSRI use with falls and bone changes (see “Temporality” section above), a growing body of preclinical evidence provides preliminary biological plausibility for negative skeletal effects of SSRIs, with findings being generally coherent with clinical observations in terms of directionality (for a complete review of preclinical evidence, see Warden et al. [61, 62]). We initially demonstrated that genetic disruption of serotonin transporter function in mice resulted in a skeletal phenotype of reduced mass, altered architecture, and inferior mechanical properties [63]. Subsequent work demonstrated that pharmacological inhibition of the transporter using clinically relevant daily doses of a SSRI (fluoxetine hydrochloride) in growing and mature Swiss-Webster and C57BL/6 mice resulted in a similar low bone mass, size, and strength phenotype [63–67]. These changes appeared to result from a reduction in bone formation in growing animals treated with the SSRI or possessing a null mutation in the gene encoding the serotonin transporter and reduced bone formation and increased resorption in mature animals treated with the SSRI [63, 66].

While preclinical data has demonstrated skeletal effects of SSRIs, the full coherency and biological plausibility of the association remain questioned as the mechanism for the effect has not been established. Putatively, the mechanism involves serotonin as SSRIs block uptake of serotonin from the extracellular space to prolong its activation of a family of dedicated receptors. However, serotonin exhibits separate central (central nervous system [CNS]) and peripheral (non-CNS) functional identities due to impermeability of the blood-brain barrier to the monoamine and differential regulation of central and peripheral serotonin synthesis [68]. The potential for duality of serotonin effects impedes identification of the mechanism for the skeletal effects of SSRIs, as these agents cross the blood-brain barrier to inhibit the serotonin transporter both centrally and peripherally. As a result, SSRIs may have skeletal effects via multiple overlapping mechanisms, with the directionality and/or magnitude of the ultimate skeletal effect representing the summation of each contributing mechanism.

Karsenty and colleagues [69–71] recently supported a duality of serotonin effects on the skeleton, with central and peripheral serotonin promoting and inhibiting bone formation, respectively. Although others have been unable to independently confirm some of the observations (particularly relating to the skeletal effects of peripheral serotonin) [72–74], the identification of independent and contrasting skeletal effects of serotonin may have implications for understanding the mechanism/s for the apparent skeletal effects of SSRIs.

It is possible that SSRIs have direct skeletal effects by acting on resident bone cells. Each major bone cell type (osteoblasts, osteocytes, and osteoclasts) possesses functional serotonin receptors [71, 75–84], as well as a serotonin transporter that is highly specific for serotonin uptake [75, 76, 79, 80, 85]. In addition, each major bone cell type expresses the gene for tryptophan hydroxylase-1, the rate-limiting enzyme for peripheral serotonin synthesis, and may produce autocrine- or paracrine-acting serotonin [71, 75, 79]. Alternatively, SSRIs may induce peripheral upstream changes that impact serotonin delivery to the skeleton.

Approximately 95 % of total body serotonin is produced by enterochromaffin cells in the gastrointestinal tract where it functions as a paracrine factor to stimulate peristalsis and mucus secretion [86]. A proportion of this serotonin reaches the cardiovascular system where most (>95 %) is taken up by platelets using the serotonin transporter [87]. The remaining <5 % of circulating 5-HT stays free in the plasma where it may conceivably act as a hormone influencing bone cells, as suggested by Karsenty and colleagues [69, 71]. SSRI-induced inhibition of the serotonin transporter reduces platelet uptake of the monoamine and may theoretically increase plasma serotonin concentration resulting in its greater delivery to resident bone cells. Increased local serotonin due to heightened delivery from the circulatory system or inhibition of the serotonin transporter within bone cells themselves could be inhibitive of bone formation and explain the negative skeletal effects of SSRIs. However, this putative mechanism has contradictory support [72–74], with Robling and colleagues [74] most prominently reporting that peripheral serotonin had no measurable skeletal effect.

Central effects may also contribute to the observed skeletal effects of SSRIs. In the adult brain, the serotonin transporter is located presynaptically on tryptophan hydroxylase 2-expressing (i.e., CNS serotonin producing) neurons in the raphe nuclei where it regulates serotonin activity to influence a range of behavioral, physiological, and cognitive functions. Karsenty and colleagues [70] demonstrated that serotonin may activate receptors in the ventromedial hypothalamus to modulate sympathetic outflow and increase bone mass. How this latter observation relates to the skeletal effects of SSRIs remains to be established.

Centrally mediated reductions in physical activity levels may also contribute to the bone changes observed with 5-HTT inhibition in preclinical studies. An established central effect of gene- and pharmacological-mediated inhibition of the 5-HTT in mice is heightened anxiety-like behavior, which manifests in a hypoactive locomotor behavioral phenotype [88, 89]. However, physical inactivity does not explain why mice treated with an SSRI exhibit a skeletal phenotype compared to animals treated with a TCA as the latter mice show equivalent physical inactivity but do not show a consistent skeletal phenotype [64, 65]. Further, SSRI reduced bone formation in tail-suspended mice, with tail suspension effectively removing the physical inactivity phenotype induced by these agents and normalizing skeletal loading between treated and control animals [65].

Conclusions

Currently available data associating SSRIs with fractures are observational and limited in their ability to establish causality. Using the Bradford Hill criteria to explore the causal relationship between SSRIs and fractures, we found a strong, consistent, and temporal relationship between SSRIs and fractures, which appears to follow a biological gradient. However, specificity and plausibility remain concerns. In terms of specificity, the majority of currently available data has limitations due to either confounding by indication or channeling bias. Self-controlled case series address some of these limitations and provide relatively strong observational evidence for a specific relationship between SSRIs and fracture. The latter studies suggest that falls contribute to fractures in SSRI users, as evidenced by a rapid rate of fracture onset within individuals initiating SSRI use. Whether SSRIs also induce changes in underlying skeletal properties remains unresolved. Initial studies provide some biological plausibility for a skeletal effect of SSRIs; however, the pathways involved need to be established before plausibility can be readily accepted.

As the link between SSRIs and fractures is based on observational data and not evidence from prospective trials, there is insufficient evidence to definitively determine a causal relationship and it appears premature to label SSRIs as a secondary cause of osteoporosis. However, self-controlled case series do suggest that SSRIs cause fracture-inducing falls. Addressing any fall risk associated with SSRIs may be an efficient approach to reducing SSRI-related fractures. As fractures stemming from SSRI-induced falls are more likely in individuals with compromised bone health, it is also worth considering bone density testing and intervention for those presenting with additional risk factors for osteoporosis (i.e., increased age, female gender, family history, smoking, history of falls, etc.).