Introduction

In a lifetime, more than 50% of women and 25% of men experience at least one fragility fracture [1]. The mortality rates are higher for the fracture of the hip and vertebrae in both sexes [2]. Osteoporosis is a condition in which the bones become porous and their mineral density and quality are reduced, so there is an increased fracture. Osteoporosis remains undetected until the occurrence of fracture [3]. Out of the many causes of osteoporosis, the use of medication is one of the important factors leading to bone loss [4].

A recent study on the prescribing trends of antidepressants revealed that the prescription of antidepressants has been increased many folds and that selective serotonin reuptake inhibitors (SSRIs) constitute 51% of the total antidepressant prescriptions [5]. This is because of their better safety and efficacy profiles [6]. It has been evidenced that the chronic administration of SSRIs for the treatment of psychiatric conditions in humans is associated with osteoporosis [7,8,9]. It has also been reported that SSRIs play a dual role in bone primarily due to opposing effects of serotonin on the bone turnover where gut-derived serotonin is reportedly associated with bone loss while brain-derived serotonin causes bone formation [10].

The three meta-analyses evaluating fracture risk with the use of SSRIs in adults concluded a significant risk associated with these drugs with the possibility of a major clinical impact [11,12,13]. However, this meta-analysis was published in 2012 and 2013 and studies included were carried out until April 2011. There has been a lot of observational studies reported on this aspect after 2010, thus necessitating an updated analysis of data. Further, the most recent meta-analysis incorporated studies on two categories of antidepressants including both serotonin-norepinephrine reuptake inhibitors (SNRIs) and SSRIs up to the period of November 2016. The meta-analysis concluded that SSRIs are associated with an increased risk of fracture irrespective of age [14]. Contrary to the results on fracture risk, a recent meta-analysis on four studies on woman, with bone mineral density as outcome, concluded that antidepressants including tricyclic antidepressants (TCAs) or SSRIs do not have any impact on bone mineral density at all three measured sites including lumbar spine, femoral neck, and total hip [15]. Based on this contradiction and the fact that various high-quality studies have been added after 2016, there was a need for an updated meta-analysis.

In the present systematic review and meta-analysis, we have evaluated the association of the SSRI uses and the fracture risk for case-control and cohort studies (as the randomized clinical trial could not have been possible with this kind of outcome) carried out from inception until April 2019. The study is expected to provide a better picture of the possible association between SSRI use and risk of fracture with updated literature and can help guide the physician in selecting antidepressant for those patients with existing risk factors.

Material and methods

We followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines for reporting of data [16]. The protocol of the review was registered in PROSPERO reference no CRD42018086090.

Literature search

We have performed computational data search of the electronic libraries on the search engines like PubMed, Cochrane library, and Google Scholar for relevant studies by using individual keywords or combination of the keywords like selective serotonin reuptake inhibitor, fracture, osteoporosis, osteopenia, bone loss, SSRIs, SSRI’s, and fracture risk. An extensive data search was done from the first known literature of SSRI use with fracture outcome until April 2019 of the data published in the English language. The first search was started on 15 September 2018 and then updated on 30 April 2019. In addition to the above search plan, the references of the relevant literature were checked manually for any missing eligible studies.

Eligibility criteria

We have included studies if they fulfill the following criteria:

  1. 1.

    Population - the adult population.

  2. 2.

    Intervention - observational study design with SSRIs (sertraline, fluoxetine, escitalopram, citalopram, paroxetine, and fluvoxamine) as treatment regardless of indication, dose, and duration of the usage.

  3. 3.

    Comparison - SSRI non-users as controls.

  4. 4.

    Outcome - fractures as the primary outcome regardless of the site of fracture which is self-reported, recorded, or diagnosed.

We have excluded animal studies, any duplicate studies, studies with other adjuvant therapies that interfere with bone turnover, abstracts, and non-English literature.

Study selection and assessment of the quality

The data were independently reviewed by two impartial reviewers (MK, ARS) for the inclusion of the studies as per the eligibility criteria. The literature was first looked for the title and abstract followed by a full article for relevant literature. In case of any discrepancy, it was resolved by mutual consensus of both the reviewers.

The quality of the eligible studies was assessed by the ROBINS-I scale [17] as applicable for the case-control and cohort study by the two 2 reviewers (MK, ARS) independently. The study was defined as low, moderate, serious, critical, or no information.

Data extraction

The data of the eligible studies were further summarized in a tabular form with the information regarding author detail, year, country, study design, sample type, study size, age, number of females, overall risk of bias, site of fracture, follow-up period (in cohort studies), clinical risk factor adjustment, adjustment of physical activity, adjustment of calcium intake, adjustment of depression, number of exposed cases, and number of exposed cases in control, as shown in Tables 1 and 2.

Table 1 Characteristics of case-control studies included in the analysis
Full size table
Table 2 Characteristics of cohort studies included in the analysis
Full size table

Apart from other factors known to interfere with bone loss, depression itself is a major confounder. Hence, the “adjustment for depression” was extracted from various studies and on this basis, we divided the studies into two groups: studies that have considered depression as a biasing parameter and studies that have not considered depression as a biasing parameter.

Data synthesis and analysis

Data were extracted in a pre-designed excel template and the reference was updated in endnote version X9. We computed pooled relative risk and 95% confidence interval (CI) from confounder adjusted ORs/RRs/HRs and corresponding 95% CIs as reported in the studies. We considered the odds ratio (OR) as a surrogate measure of the corresponding risk ratio (RR)/hazard ratio (HR) in longitudinal studies because the absolute risk of fracture is low. To stabilize the variance and normalize the distributions, we transformed ORs, RRs, and HRs into their natural logarithms before pooling the data (and therefore, a variation could be possible when converting back to relative risk; however, it did not change any interpretation of results) [48]. The standard error (SE) of the natural logarithm of OR/HR/RR was derived from the corresponding CI, which was either provided in the study or calculated with standard formulas [49]. To estimate the overall effect size, each study was weighted by the reciprocal of its variance. In studies where only subgroup estimates were reported for the outcome, the overall effect size across subgroups in each individual study was estimated with meta-analysis. Random-effects meta-analysis, using DerSimonian and Laird method [50], was employed on individual study estimates to obtain a pooled summary estimates for relative risk. Heterogeneity between studies was assessed using the Cochrane Q statistic (P < 0.1 considered as the presence of heterogeneity) and I-squared (I2) statistics (> 50% representing moderate heterogeneity) [51], and a number of subgroup analyses were conducted to identify potential sources of heterogeneity. A 95% prediction interval for the random-effects distribution was also calculated to understand the possible range of relative risk if a new study is conducted as suggested by Higgins and Thompson [52]. Publication bias was assessed by funnel plot and its asymmetry was tested by the Begg and Mazumdar rank correlation test (P < 0.10 was considered as an indication of publication bias) [53]. To determine whether there is a relation between fracture risk and subgroup variables (i.e., study design, with defined daily dose, year of reporting, number of adjusted risk factors, other key risk factors such as depression, physical activity, osteoporosis, and bone mineral density (BMD)), we used univariable and multivariable meta-regression analysis using the maximum likelihood method (P < 0.10 considered significant given the low power of these tests). We were able to add age into the regression analysis due to high disparity in reporting. Further, a sensitivity analysis was also carried out by adjusting the risk-factors such as BMD and osteoporosis. All statistical analyses were conducted on Stata statistical software (version 15.2, StataCorp LLC, College Station, TX, USA) using user-written admetan, metafunnel, metabias, and metareg commands. A P value of < 0.05 was considered statistically significant for the effect of study-level covariates on the estimated relative risks.

Results

A total of 1324 studies were identified; out of which, we have assessed the full text of 69 articles to further include 37 eligible studies in the analysis as shown in Fig. 1. The studies extracted were published from the period 1998 to 2019 (Tables 1 and 2). Out of 37 eligible studies, 14 were case-control and 23 were cohort studies. Among the eligible case-control studies, 2 were nested case-control [20, 21] and one was retrospective in nature [22], whereas others were prospective case-control studies. In the cohort studies, 4 were prospective cohort [39,40,41, 43], 3 were retrospective cohort [30, 45, 46], and others were classical cohort studies.

Fig. 1
figure 1

PRISMA flowchart of the studies selection

Full size image

Quality of the studies

The ROBINS-I tool assesses risk of bias in seven domains and an overall risk of bias according to the highest level of risk in any one domain. If a study is assessed to have a serious risk of bias in one domain, but low risk of bias in all others, the overall risk of bias for the study will be serious. Risk of bias within the seven domains, and overall, is displayed for all 37 studies in Fig. 2.We deemed the overall risk of bias to be critical for 8 studies and serious for the remaining 29 studies. We deemed all studies to have a serious risk of bias in the measurement of outcomes and critical or serious risk of bias in confounding because of the study designs.

Fig. 2
figure 2

Overall quality assessment of studies using ROBINS-I scale

Full size image

Meta-analysis for fracture risk

The main outcome of the meta-analysis of both case-control and cohort studies is that SSRIs are significantly associated with the increase in the fracture risk with a relative risk of 1.62 (95% CI 1.52–1.73; P < 0.000; I2 = 90.8%). In case-control studies, when considered alone, the fracture risk was significantly associated with SSRIs with a relative risk of 1.80 (95% CI 1.59–2.03; P < 0.000; I2 = 93.2%), while cohort studies also show the same trend of increased fracture risk with a relative risk of 1.52 (95% CI 1.40–1.65; P < 0.000; I2 = 88.0%). Figure 3 shows the forest plot of the combined effect of 14 case-control and 23 cohort studies. As shown, heterogeneity between groups was significantly associated; hence, random-effects analysis was carried out for the pooled analysis.

Fig. 3
figure 3

Risk of fracture associated with the selective serotonin reuptake inhibitor (SSRI) use according to the study design using random-effects meta-analysis

Full size image

Subgroup analysis

The subgroup analysis, as shown in Table 3, indicated that the fracture risk remained consistent after taking into consideration the geographical location (Australia (P < 0.001), Asia (P < 0.038), Europe (P < 0.001), the USA and Canada (P < 0.001)) and study design (case-control (P < 0.001), cohort (P < 0.001)) and after adjusting for clinical risk factors (< 5 (P < 0.001), ≥ 5 (P < 0.001)); studies with defined daily dose (Yes (P < 0.001), No (P < 0.001)), SSRI use duration (≤ 6 months (P < 0.001), > 6 months (P < 0.001)); anatomical site of fracture (hip (P < 0.001); all sites (P < 0.001); hip/femur (P < 0.001); hip, humerus, radius, and ulna (P < 0.002)); and the period of study (before 2011 (P < 0.001) and 2011 or after (P < 0.001)). Further, the fracture risk also remained significant after adjusting for depression (P < 0.001), physical activity (no (P < 0.001), yes (P = 0.034)), gender males (P < 0.001), % of females < 60% (P < 0.001), % of females ≥ 60% (P < 0.001), and mean age < 50 (P < 0.001), mean age ≥ 50 (P < 0.001), age ≥ 50 (P < 0.001), and age ≥ 18 (P < 0.001).

Table 3 Relative risk of fracture associated with the use of SSRIs in subgroups defined by study characteristics using the random-effects model
Full size table

The overall association between the fracture risk and the reported study characteristics was assessed by univariable and multivariable mixed-effect meta-regression analysis. We found no independent statistically significant association on fracture risk in the multivariable meta-regression for study design (P = 0.405), with defined daily dose (P = 0.919), the total number of adjusted variables (P = 0.420), year of reporting (P = 0.787), and other key factors (such as depression (P = 0.142), physical activity (P = 0.525), osteoporosis (P = 0.241), and BMD (P = 0.698)).

Publication bias

We used a funnel plot (Fig. 4) to assess publication bias. In the figure, the vertical line represents the summary estimate, i.e., RR of the risk of fracture due to SSRI treatment. The diagonal lines represent the 95% confidence limits around the summary treatment effect. These show the expected distribution of studies in the absence of heterogeneity or selection biases. The funnel plot was almost symmetric and indicated none of the missing potential studies. The funnel plot asymmetry was assessed by Begg and Mazumdar’s rank correlation test for publication bias that showed no significant publication bias (|z|corrected = 0.92, P = 0.360). Similar results were also found for case-control (|z|corrected = 0.77, P = 0.443) and cohort (|z|corrected = 1.27, P = 0.205) studies.

Fig. 4
figure 4

Funnel plot of relative risk with 95% pseudo-confidence limits according to the study design

Full size image

Sensitivity analysis

A sensitivity analysis is carried out to estimate the risk of fracture by adjusting risk factors such as bone mineral density (BMD) (Fig. 5) and osteoporosis (Fig. 6). Studies adjusted for BMD showed a 17% lower risk of fracture compared with unadjusted studies (for adjusted, RR 1.47, 95% CI 1.19–1.82; for unadjusted, RR 1.64, 95% CI 1.53–1.76). similarly, studies adjusted for osteoporosis showed a 19% lower risk of fracture compared with unadjusted studies (for adjusted, RR 1.54, 95% CI 1.39–1.70; for unadjusted, RR 1.73, 95% CI 1.57–1.90).

Fig. 5
figure 5

Sensitivity analysis showing the risk estimate by adjusting bone mineral density (BMD)

Full size image
Fig. 6
figure 6

Sensitivity analysis showing the risk estimate by adjusting osteoporosis or risk factors for osteoporotic fractures

Full size image

Discussion

This pooled meta-analysis shows that the SSRIs are significantly associated with fracture risk. We reported a 1.62-fold increase in fracture risk (95% CI 1.52–1.73) for SSRI users as compared with non-users for the combined case-control and cohort studies. Our results are in agreement with previous meta-analyses conducted in 2012 and 2013 showing an increase in fracture risk with SSRI users [11,12,13]. Randomized clinical trials cannot be possible for fracture as an outcome; hence; we included observational studies in our analysis. Among the studies included in the analysis, though the quality of most of the studies was found to be serious (29) while others were critical (8) as per the ROBINS-I tool, we adjusted for various risk factors that may bias the results. We reported that the risk of fracture remained consistent on subgroup analysis when adjusted for geographical location, study design, number of clinical risk factors adjusted, anatomical site of the fracture, defined daily dose, SSRI use duration, period of study, adjustment for depression, adjustment for physical activities, gender, and age group of the population included in the groups. Additionally, no previous meta-analysis has performed sensitivity analysis to adjust studies for osteoporosis and BMD which elucidate that studies adjusted for both the parameters show lesser fracture risk. Hence, the history of BMD and osteoporosis must be taken into consideration while interpreting fracture risk with SSRIs. Our study did not find any statistical evidence for publication bias. However, we cannot rule out that there are some small studies that found no harm with SSRIs and in the same may not have been published.

The previous meta-analysis conducted in 2012 and 2013 included studies from Western countries only and hence, the results could not be generalized to all other populations [11,12,13]. Our study, however, showed that a significant risk persisted across geographical locations with higher fracture risk reported in the case of Australia, Europe, the USA, and Canada as compared with Asia. This could also be due to fewer studies available from Asia as compared with other continents. We also observed that cohort studies showed lesser fracture risk as compared with case-control design. The reason could be due to differences in the study design. The trend of increase in the fracture risk was also seen in case-control studies by previous meta-analysis [11, 12, 14], but only one of them [11] actually reported this observation that case-control studies are significantly associated with the fracture risk as compared with the cohort study design.

The strength of the present meta-analysis is that it consists of 37 studies that accommodate most of the recent literature for SSRIs and fracture risk. Our study has limitations. We observed that adjustment for depression did not show any lesser risk of fracture as compared with studies that were not adjusted which shows that depression was not the confounder in the analyzed studies. However, previous studies have reported that depression itself causes bone loss leading to a reduction in bone mineral density [54]. The reason could be that we could not adjust for depression at an individual or patient level as this information was not available to us. Depression was mentioned in studies for the entire population but not individually at a patient level. Further, the majority of the studies did not report adequate data for sun exposure or vitamin D status or concomitant medications such as glucocorticoids that may have significant effects on bone. Another important limitation of all available studies in this area is that fracture risk could not be ascertained for individual SSRI and most of the studies report effects as a category. This is important as it was earlier shown by Hodge et al. that different drugs of SSRI class behave differently on bone cell lines with sertraline being the most potent to inhibit the bone cell line while citalopram did not have any effect [55]. In addition to the above, a placebo randomized clinical trial conducted on one of the SSRI, escitalopram, demonstrated that 8 weeks of treatment of the drug did not alter the serum bone turnover markers when compared with the placebo group [56]. The same was also seen in our preclinical study showing how fluoxetine and escitalopram, when given orally for 40 days to rats, differ in altering the bone micro-architecture with fluoxetine deteriorating the bone micro-architecture and escitalopram having no effect on the same [57]. The above evidence clearly points towards the need to have future research focus on how different SSRIs behave on the bone which may have clinical implications of showing one drug to be safer than another drug.

To conclude, the results from this meta-analysis suggest the SSRI users may have an increased risk of fractures as compared with non-SSRI users; hence, bone health should be taken into consideration while prescribing this class of drugs particularly for those having existing risk factors for the same. However, the included studies were at serious or critical risk of bias and therefore, the conclusions on fracture risk must be interpreted in the context of any potential bias. Further, the lack of a clear mechanistic effect of SSRIs on BMD and opposing effects of gut and brain serotonin on bone makes the interpretation less certain. It is possible that the SSRI patients may have more fractures as the drug makes them fall over and sustain trauma as serotonin syndrome by SSRIs at higher doses manifests as ataxia. Future research could investigate these aspects and can target on determining the effect of individual SSRIs on fracture risk and bone health in general.