Significant outcomes
-
• Between 27,280, 65,529, and 17,703 individuals initiating treatment with sertraline, citalopram, and escitalopram, respectively, there was no material or statistically significant differential risk of osteoporosis during follow-up.
-
• The results were not indicative of sertraline, citalopram, and escitalopram having a dose-response-like effect on osteoporosis risk.
-
• Sertraline, citalopram and escitalopram do not appear to differentially affect the risk of osteoporosis. The lack of clear dose-response-like relationships suggest that they do not have a causal effect on osteoporosis risk.
Limitations
-
• The registers providing data for this study do not hold information on the severity of the depression being treated with SSRIs, which means that this cannot be considered in the analyses and may have resulted in residual confounding.
-
• Data on psychopharmacological treatment during inpatient hospital stays are not available in the registers, which may affect estimation of incidence and calculation of cumulative doses.
-
• Despite the target trial emulation, this was not a randomised controlled trial and the results should, therefore, be interpreted with caution with regard to causal inference.
Introduction
Characterised by increased fracture risk associated with low bone mineral density and strength, osteoporosis is the most commonly occurring bone disease (Föger-Samwald et al., Reference Foger-Samwald, Doviak, Azizi-Semrad, Kerschan-Schindl and Pietschmann2020). It is estimated to affect 41.5 million worldwide, with the incidence as well as burden predicted to increase with the ageing population (Zhu et al., Reference Zhu, Yu, Wu, Wu, Tan, Ling, Ma, Zhang, Zhu and Liu2023). Certain biological, environmental and lifestyle factors are known to be implicated in onset, with female sex, increasing age and genetic factors among the major known non-modifiable risk factors. Secondary causes of osteoporosis include certain medical conditions and medications, with more recent data suggesting antidepressants, in particular selective serotonin reuptake inhibitors (SSRIs), could be implicated (Fernandes et al., Reference Fernandes, Hodge, Pasco, Berk and Williams2016).
SSRIs, the first line pharmacological treatment for mood and anxiety disorders—approved for use in both adult and paediatric population—are now among the most widely used drugs worldwide (Statista, 2023, National Health Service, 2023).The widespread use of SSRIs reflects the fewer known side effects, and better tolerability than prior commonly used antidepressants (Cipriani et al., Reference Cipriani, Furukawa, Salanti, Chaimani, Atkinson, Ogawa, Leucht, Ruhe, Turner, Higgins, Egger, Takeshima, Hayasaka, Imai, Shinohara, Tajika, Ioannidis and Geddes2018). While the primary mechanism of action is the inhibited reuptake of serotonin, hence increasing serotonin extracellular activity, each SSRI has unique pharmacokinetics, pharmacodynamics, efficacy and side effect profiles (Edinoff et al., Reference Edinoff, Akuly, Hanna, Ochoa, Patti, Ghaffar, Kaye, Vishwanath, Urits, Boyer, Cornett and Kaye2021). Curiously, serotonin receptors, the transporter and the tryptophan hydroxylase enzyme have been identified on each major bone cell indicating their ability to respond to, uptake and produce serotonin (Warden et al., Reference Warden, Bliziotes, Wiren, Eshleman and Turner2005, Hodge et al., Reference Hodge, Wang, Berk, Collier, Fernandes, Constable, Pasco, Dodd, Nicholson, Kennedy and Williams2013).
Epidemiological studies have generally shown use of SSRIs to be associated with reduced bone mass density, bone strength and increased fracture risk in clinical and population-based samples of men and women (Mercurio et al., Reference Mercurio, De Filippis, Spina, De Fazio, Segura-Garcia, Galasso and Gasparini2022, Kumar et al., Reference Kumar, Bajpai, Shaik, Srivastava and Vohora2020, Tamblyn et al., Reference Tamblyn, Bates, Buckeridge, Dixon, Girard, Haas, Habib, Iqbal, Li and Sheppard2020, Vestergaard et al., Reference Vestergaard, Rejnmark and Mosekilde2008, Ak et al., Reference Ak, Bulut, Bulut, Akdağ, Öter, Kaya, Kaya, Şengül and Kısa2015, Vestergaard et al., Reference Vestergaard, Prieto-Alhambra, Javaid and Cooper2013, Agarwal et al., Reference Agarwal, Germosen, Kil, Bucovsky, Colon, Williams, Shane and Walker2020, Diem et al., Reference Diem, Ruppert, Cauley, Lian, Bromberger, Finkelstein, Greendale and Solomon2013, Rauma et al., Reference Rauma, Honkanen, Williams, Tuppurainen, Kröger and Koivumaa-Honkanen2016, Williams et al., Reference Williams, Berk, Hodge, Kotowicz, Stuart, Chandrasekaran, Cleminson and Pasco2018, Williams et al., Reference Williams, Henry, Berk, Dodd, Jacka, Kotowicz, Nicholson and Pasco2008, Saraykar et al., Reference Saraykar, John, Cao, Hnatow, Ambrose and Rianon2018). Furthermore, both in vitro and in vivo evidence provides further support to the epidemiological findings (Tsapakis et al., Reference Tsapakis, Gamie, Tran, Adshead, Lampard, Mantalaris and Tsiridis2012, Fernandes et al., Reference Fernandes, Hodge, Pasco, Berk and Williams2016). Utilising a human in vitro bone cell model, Hodge et al found SSRIs to inhibit osteoclast formation and bone mineralisation by osteoblasts differentially with sertraline possibly more likely than citalopram to cause bone loss (Hodge et al., Reference Hodge, Wang, Berk, Collier, Fernandes, Constable, Pasco, Dodd, Nicholson, Kennedy and Williams2013), a phenomenon yet to be explored in depth in humans.
While a randomised controlled trial (RCT) examining the potentially differential effects of SSRIs on the risk of osteoporosis would be ideal, such a study is likely not feasible due to demand for a very large number of participants and long follow-up time to obtain sufficient power. In this situation, an epidemiological study may be the best possible alternative (Köhler-Forsberg et al., Reference Köhler-Forsberg, Rohde, Nierenberg and Østergaard2022, Frank, Reference Frank2000). Epidemiological studies are, however, predisposed to methodological weaknesses that limit causal inference. Therefore, the target trial emulation (TTE) design has been developed specifically in an attempt to counter these weaknesses (Hernán et al., Reference Hernán, Wang and Leaf2022). TTE aims to infer causality from observational data by mimicking (emulating) an RCT. Specifically, by organising the observational data in a manner similar to those obtained with an RCT, TTE helps to avoid common methodological pitfalls such as prevalent user bias and immortal time bias (Hernán et al., Reference Hernán, Sauer, Hernández-Díaz, Platt and Shrier2016, Hernán et al., Reference Hernán, Wang and Leaf2022). Here, we used TTE, which has previously been used to study the effectiveness of SSRIs (Lagerberg et al., Reference Lagerberg, Matthews, Zhu, Fazel, Carrero and Chang2023, Rohde et al., Reference Rohde, Hieronymus and Østergaard2024), with comprehensive data from nationwide registers to determine whether sertraline is associated with a risk of developing osteoporosis over and above the risks associated with the use of citalopram or escitalopram. Furthermore, we estimated whether sertraline, citalopram, or escitalopram affected the risk of osteoporosis in a cumulative dose-response-like manner. If this was not the case, it would be a strong argument against these drugs having a tangible effect on the risk of osteoporosis in the real-world setting (Hill, Reference Hill1965).
Materials and methods
The study protocol was preregistered and is available on the Open Science Framework (OSF): https://osf.io/ntcr2/ (and available as Supplementary Table 1). The protocol describes the target trial specification and emulation, which are further accounted for below.
Target trial specification
First, we specified the hypothetical trial we sought to emulate. In brief, this target trial would include patients with depression in need of antidepressant treatment between January 1, 2007, and March 1, 2019. The trial would assess eligibility on the index date (the date of antidepressant treatment initiation), using the following eligibility criteria: i) Age >= 40 on the index date, ii) no previous use of antidepressants, iii) no use of antipsychotics, lithium, valproate, or lamotrigine in the five years preceding the index date, iv) no contact with a psychiatric hospital (inpatient-, outpatient-, or emergency room contact) in the five years preceding the index date, and v) no diagnosis of osteoporosis or use of medication for osteoporosis before the index date. Anatomical Therapeutic Chemical (ATC) (World Health Organization, 2022) and International Classification of Diseases, Tenth Revision (ICD-10) (World Health Organization, 1993) codes operationalising psychopharmacological treatment and osteoporosis are available in Supplementary Table 1. The target trial would then randomly assign the included patients to one of three open-label treatment arms: i) sertraline, ii) citalopram or iii) escitalopram. Follow-up would begin at the index date and continue until the development of osteoporosis, death, or end of follow-up (March 1, 2019), whichever occurred first. The primary outcome of the target trial would be development of osteoporosis (see definition below). In all analyses, the outcome would be compared between the three treatment arms (sertraline, citalopram, or escitalopram) in an intention-to-treat analysis via Cox proportional hazards regression.
Target trial emulation
Registers: We emulated the target trial described above using data from the following Danish nationwide registers (Munk-Jorgensen and Ostergaard, Reference Munk-Jorgensen and Ostergaard2011): The Danish Civil Registration System (DCRS) (Pedersen, Reference Pedersen2011), which contains information on date of birth, date of death, sex, marital status and a 10-digit personal identifier that allows linkage between all registers in Denmark; The Danish Psychiatric Central Research Register (Mors et al., Reference Mors, Perto and Mortensen2011), which contains data on all contacts to Danish psychiatric hospital services, including admissions, outpatient contacts, and emergency visits; The Danish National Patient Register (DNPatR) (Lynge et al., Reference Lynge, Sandegaard and Rebolj2011), which contains information on all contacts to the non-psychiatric hospital system in Denmark; The Danish National Prescription Register (DNPreR) (Wallach Kildemoes et al., Reference Wallach Kildemoes, Toft Sørensen and Hallas2011), which contains data on redemptions of prescribed medication from Danish pharmacies; The Population Education Register (PER), and the Employment Classification Module (ECM) (Thygesen et al., Reference Thygesen, Daasnes, Thaulow and Brønnum-Hansen2011), which hold data on education and occupational status, respectively. The Danish Health Data Authority and Statistics Denmark approved the use of these data for the present study. In Denmark, register-based studies are exempt from ethical review board approval. The project is registered with the Danish Data Protection Agency (Registration number: 2022-0368974).
Study subjects and intervention:Via the DNPreR, we identified all individuals that redeemed their first prescription (treatment indication specified as depression) for sertraline, citalopram, or escitalopram between January 1, 2007, and March 1, 2019 (individuals could only be included once). We used the date of this redeemed prescription as the index date. We excluded individuals aged <40 years (to focus on a period in which the risk of osteoporosis was greater) (Compston et al., Reference Compston, McClung and Leslie2019), with previous use of antidepressants, use of antipsychotics, lithium, valproate, or lamotrigine, or with a psychiatric hospital contact in the five years preceding the index date, and individuals that had developed osteoporosis prior to the index date. As sertraline, citalopram, and escitalopram have been considered equal first-line treatments for depression in Denmark according to national guidelines (Danish Health Authority, 2007), we assumed that the study subjects were fairly randomly allocated to the three treatment arms. However, to reduce the impact of potential confounding, we used information from the DCPR, PER, ECM, DNPatR and DnPreR to adjust for the following baseline covariates: age, sex, calendar year, educational level, marital status, occupational status, somatic comorbidity (the Charlson Comorbidity Index – based on data from the 5 years prior to the index date), redemption of at least one prescription for a sedative/hypnotic in the year preceding the index date and redemption for at least one prescription for a systemic corticosteroid in the year preceding the index date. The ATC and ICD-10 codes operationalising psychopharmacological treatment, osteoporosis and the Charlson Comorbidity Index are available in Supplementary Table 2. Notably, we deliberately chose not to include other SSRIs or antidepressants from other classes in this study as this would introduce confounding by indication beyond what can be adjusted for with the data at hand (Østergaard, Reference Østergaard2023, Østergaard and Rohde, Reference Østergaard and Rohde2023, Psaty et al., Reference Psaty, Koepsell, Lin, Weiss, Siscovick, Rosendaal, Pahor and Furberg1999). We refrained from using a control group (not exposed to an SSRI) for the same reason.
Follow-up and outcome definition: We followed the included individuals until the development of osteoporosis, death or end of follow-up (March 1, 2019), whichever came first. The outcome of interest was the development of osteoporosis, operationalised as either receiving a hospital diagnosis of osteoporosis (ICD-10 codes available in Supplementary Table 1), identified via the DNPatR or redeeming a prescription for a medication used in the treatment of osteoporosis (ATC-codes available in Supplementary Table 2), identified via the DNPreR (Köhler-Forsberg et al., Reference Köhler-Forsberg, Rohde, Nierenberg and Østergaard2022).
Statistical analysis
Figure 1 illustrates the conducted analyses. First, we compared the hazard rate of osteoporosis between the three treatment arms (sertraline, citalopram, and escitalopram) using a Cox proportional hazard model (see Figure 1A), analysing for an effect of treatment allocation (intention-to-treat effect). The sertraline treatment arm was used as the reference group. We calculated and reported partly adjusted (adjusted for age, sex, and calendar year) and fully adjusted (adjusted for all covariates listed above) hazard rate ratios (aHRR).
Figure 1. Illustrations of the analyses. (A) Comparison of the rate of osteoporosis between individuals receiving sertraline, citalopram and escitalopram, respectively. (B) Examination of potential cumulative dose-response-like relationships.
Following the target trial emulation, we estimated whether sertraline, citalopram, and escitalopram affected the risk of osteoporosis in a cumulative dose-response-like manner (see Figure 1B) (Köhler-Forsberg et al., Reference Köhler-Forsberg, Rohde, Nierenberg and Østergaard2022). Here we only included individuals with an index date in the period from January 1, 2007, and March 1, 2014. From the index date and five years ahead, we calculated the cumulative dose of sertraline, citalopram, and escitalopram using the following formula: dose × number of pills per package/Defined Daily Dosage (DDD) (World Health Organization, 2022). As in a prior study employing a similar design (Köhler-Forsberg et al., Reference Köhler-Forsberg, Rohde, Nierenberg and Østergaard2022), we chose the 5-year period to get sufficient variance in drug exposure to allow for the dose-response-like analysis. Individuals who developed osteoporosis or died in these five years were excluded. The remaining individuals were followed from the date five years after the index date until the development of osteoporosis, death, or the end of follow-up (March 1, 2019), whichever came first. We then assessed the association between SSRI treatment (separate analyses for sertraline, citalopram, and escitalopram) and osteoporosis using Cox proportional hazards regression with adjustment as described for the primary analysis, while stratifying by cumulative dose (<=100 DDDs (reference group), >100-365 DDDs, >365-730 DDDs, >730-1460 DDDs, and >1460 DDDs). We used the log-rank test to test for cumulative dose-response-like relationships. The analyses were conducted using Stata version 15 (StataCorp) via remote access to Statistics Denmark. All tests were 2-sided, and the significance level was set at .05.
Sensitivity analyses
First, we stratified the primary analysis on sex. Second, as the escitalopram patent, according to our knowledge, expired in Denmark in late 2011 (much later than sertraline and citalopram), we reran the primary analysis based only on data from those with an index date from January 1, 2013 and onwards. This analysis was performed, as price differences before 2011 may have introduced confounding by indication. Third, we repeated the primary analysis, while restricting the maximum follow-up to 1-year after the index date. Fourth, we conducted a ‘per protocol’-like version of the primary analysis restricted to the study subjects which had at least 6 months of follow-up and a medication possession ratio ≥90 (for the SSRI they were allocated to) over the course of these 6 months (the full follow-up time was used for outcome assessment). Six months was chosen as this represents a reasonable duration of SSRI treatment of the first pharmacologically treated depression. The medication possession ratio was calculated as follows: ((dose × number of pills per package/DDD)/(number of days in observation period)) × 100 (Karve et al., Reference Karve, Cleves, Helm, Hudson, West and Martin2009). Fifth, to account for death as a potential competing risk, we also assessed the association between the three SSRIs and osteoporosis via competing-risks regression based on Fine and Gray’s proportional subhazards model, using the same adjustment as in the primary analysis. Fifth, we repeated the cumulative dose-response-like analysis, using information on the adjustment variables from the end of the collection of the 5-year cumulative dose of sertraline, citalopram and escitalopram (instead of using information on the adjustment variables from the beginning of the 5-year period).
Test of proportional hazards
The proportional hazards assumption for the Cox proportional hazards regression analyses was tested by inspection of log-log survival functions.
Results
Study population characteristics
Figure 2 shows the recruitment and treatment allocation for the target trial emulation. We identified 248,630 individuals redeeming their first prescription for sertraline, citalopram, or escitalopram in the period from January 1, 2007 to March 1, 2019. Of these, 138,118 were excluded (108,840 were <40 years old at the index date, 10,757 had used antipsychotics, lithium, valproate or lamotrigine in the five years preceding the index date, 7,327 had a contact with a psychiatric hospital in the five years preceding the index date, and 11,194 had osteoporosis preceding the index date). Among the remaining 110,512 individuals meeting the eligibility criteria, 27,280, 65,529, and 17,703 initiated treatment for depression with sertraline, citalopram, or escitalopram, respectively.
Figure 2. Flowchart showing the recruitment and treatment allocation for the target trial emulation. *Redemption of the first prescription for sertraline, citalopram or escitalopram in the period from January 1, 2007 to march 1, 2019.
Table 1 lists the characteristics of the individuals included in the target trial emulation. Those initiating treatment with sertraline tended to be younger, have less medical comorbidity, and have a longer education than patients initiating treatment with citalopram or escitalopram.
Table 1. Baseline clinical and demographic characteristics of the individuals included in the target trial emulation
Comparison of risk of osteoporosis with sertraline, citalopram and escitalopram treatment
The cumulative incidence proportion of osteoporosis after 5 years of follow-up (just below the mean duration of follow-up) was 3.5% for sertraline (n = 965 cases), 5.5% for citalopram (n = 3615 cases), and 4.9% for escitalopram (n = 860 cases). Table 2 lists the incidence rate of osteoporosis during follow-up and the results of the Cox proportional hazards regression comparing the risk of developing osteoporosis with sertraline, citalopram andescitalopram treatment, respectively. There was no material difference in the risk of osteoporosis when comparing patients treated with sertraline to those treated with citalopram (aHRR = 0.98, 95%CI = 0.92–1.05) or escitalopram (aHRR = 0.94, 95%CI = 0.92–1.05). Similar results were found in the analysis stratified by sex.
Table 2. The association between sertraline, citalopram and escitalopram treatment and osteoporosis
a age, sex (not for the sex-stratified analyses), and calendar yearb age, sex (not for the sex-stratified analyses), calendar year, marital status, occupation, education, Charlson comorbidity index, prior use of a sedative/hypnotic, and prior use of a corticosteroid.
Examination of cumulative dose-response-like relationships
Figure 3 shows the results of the cumulative dose-response-like analyses for sertraline, citalopram, and escitalopram, respectively.
Figure 3. Examination of potential cumulative dose response like relationships.
The p-values from the log-rank tests testing cumulative dose-response-like relationships were 0.94 for sertraline, 0.01 for citalopram and 0.87 for escitalopram. While the p-value for citalopram, as opposed to those for sertraline and escitalopram, seems supportive of a cumulative dose-response-like relationship, Figure 3 shows that such a relationship is unlikely to be present. Specifically, the figure shows that the estimates for the strength of the association between the individual dosing strata and osteoporosis are very narrowly centred around the null, and not even the highest dose stratum (>1460 DDDs of citalopram) is statistically significantly associated with subsequent development of osteoporosis (aHRR = 1.08, 95%CI = 0.93–1.24).
Sensitivity analysis
When rerunning the primary analysis only including patients with an index date from January 1, 2013, and onwards, the results were analogue to those from the primary analysis (see Supplementary Table 3). The results of the analysis using a maximum follow-up of 1 year are similar to those from the primary analysis and are listed in Supplementary Table 4. The results from the per protocol-like analysis align with those from the primary analysis and are available in Supplementary Table 5, although this analysis suggested that the risk of osteoporosis was lower for escitalopram than for sertraline (aHRR = 0.89, 95%CI = 0.79-1.00). The results of the competing-risks regression based on Fine and Gray’s proportional subhazards model align with those from the primary analysis and are available in Supplementary Table 6. The results of the cumulative dose-response-like analysis using information on the adjustment variables from the end of the collection of the 5-year cumulative dose of sertraline, citalopram and escitalopram are almost identical to those from the primary analysis and are shown in Supplementary Figure S1.
Test of proportional hazards
The proportional hazards assumption was met. For representative examples of the log-log survival functions, see Supplementary Figure S2.
Discussion
In this TTE, we used data from Danish nationwide registers to examine if the first line SSRI antidepressants sertraline, citalopram and escitalopram differentially affect the risk of osteoporosis.
Furthermore, we estimated whether sertraline, citalopram, or escitalopram affect the risk of osteoporosis in a cumulative dose-response-like manner. The findings suggest that these SSRIs i) do not have a differential effect on the risk of osteoporosis, and ii) do not have a dose-response-like effect on the risk of osteoporosis. The latter points against these SSRIs having a causal effect on the risk of osteoporosis.
That the ‘per protocol’ sensitivity analysis suggested a higher risk of osteoporosis with sertraline treatment than with escitalopram may be due to a detection bias driven by the fact that sertraline (used with the stability required by the per protocol analysis) is prescribed more often for anxiety than is the case for escitalopram in Denmark (Therman Soerensen et al., Reference Therman, Ishtiak-Ahmed, Gasse and Sparle2023). Accordingly, while the present study focused exclusively on SSRIs prescribed for depression, sertraline is probably prescribed more commonly for patients with anxious depression, who may be more likely to have potential osteoporosis detected, due to increased anxiety-driven healthcare utilisation (Horenstein and Heimberg, Reference Horenstein and Heimberg2020). Unfortunately, the register data available for this study does not contain information on depressive subtype or (anxiety) symptom level, so this potential bias cannot be adjusted for.
The results of this study are at odds with those reported by Hodge et al., who found that SSRIs inhibit in vitro osteoclast formation and bone mineralisation by osteoblasts differentially with sertraline being most potent and citalopram being least potent (Hodge et al., Reference Hodge, Wang, Berk, Collier, Fernandes, Constable, Pasco, Dodd, Nicholson, Kennedy and Williams2013). There are two main possibilities for this discrepancy. First, even though Hodge et al, used human precursor (from umbilical blood) and primary (from trabecular bone) cells (Hodge et al., Reference Hodge, Wang, Berk, Collier, Fernandes, Constable, Pasco, Dodd, Nicholson, Kennedy and Williams2013), the observed effects of SSRIs may not translate to man. This preclinical to clinical translational gap in studies of aetiological mechanisms and drug effects has been widely discussed (Seyhan, Reference Seyhan2019, Parrish et al., Reference Parrish, Tan, Grimes and Mochly-Rosen2019). Second, as described in the limitation section below, given its observational design, we cannot exclude that the results of the current study may be biased or confounded.
The results of the present study are also at odds with most epidemiological studies on the association between use of SSRIs and poor bone health, which have generally pointed to a positive association (Mercurio et al., Reference Mercurio, De Filippis, Spina, De Fazio, Segura-Garcia, Galasso and Gasparini2022, Kumar et al., Reference Kumar, Bajpai, Shaik, Srivastava and Vohora2020, Tamblyn et al., Reference Tamblyn, Bates, Buckeridge, Dixon, Girard, Haas, Habib, Iqbal, Li and Sheppard2020, Vestergaard et al., Reference Vestergaard, Rejnmark and Mosekilde2008, Ak et al., Reference Ak, Bulut, Bulut, Akdağ, Öter, Kaya, Kaya, Şengül and Kısa2015, Vestergaard et al., Reference Vestergaard, Prieto-Alhambra, Javaid and Cooper2013, Agarwal et al., Reference Agarwal, Germosen, Kil, Bucovsky, Colon, Williams, Shane and Walker2020, Diem et al., Reference Diem, Ruppert, Cauley, Lian, Bromberger, Finkelstein, Greendale and Solomon2013, Rauma et al., Reference Rauma, Honkanen, Williams, Tuppurainen, Kröger and Koivumaa-Honkanen2016, Williams et al., Reference Williams, Berk, Hodge, Kotowicz, Stuart, Chandrasekaran, Cleminson and Pasco2018, Williams et al., Reference Williams, Henry, Berk, Dodd, Jacka, Kotowicz, Nicholson and Pasco2008, Saraykar et al., Reference Saraykar, John, Cao, Hnatow, Ambrose and Rianon2018). In comparison to the present study, these have, however, been based on smaller/selected samples and designs that are weaker from a causal inference perspective (e.g., using cross-sectional designs, including prevalent users of SSRIs, not considering the indication for SSRI treatment, using fractures rather than osteoporosis as outcome, having insufficient data for confounder adjustment etc.) (Hernán et al., Reference Hernán, Wang and Leaf2016, Hernán et al., Reference Hernán, Wang and Leaf2022).
While the use of register-based data allows for study of long-term outcomes (e.g., osteoporosis) of psychopharmacological treatment (e.g., with SSRIs) at nationwide level, it is also associated with limitations. First and foremost, the registers do not hold information on the severity of the depression being treated with SSRIs, e.g., as quantified by the Hamilton depression rating scale or other quantitative measures, which means that this cannot be considered in the analyses and may have resulted in residual confounding. Second and relatedly, the register data do not readily allow for operationalisation of treatment response to SSRI treatment. Third, data on drug treatment prior to 1995 is not available in the registers, entailing that some of the included SSRI users are not completely ’incident’. Fourth, data on psychopharmacological treatment during inpatient hospital stays are not available in the registers, which may affect estimation of incidence and calculation of cumulative doses. This is, however, somewhat unlikely to have a materially differential impact on the three SSRIs studied here, which reduces the risk of bias. Fifth, the dose-response-like analysis does not distinguish between exposure to, e.g., 10 mg of escitalopram for four consecutive years vs. exposure to 20 mg of escitalopram in year one and year five. These exposures may not have the same impact on the risk of osteoporosis. Future studies should examine the potential role of the relationship/interaction between the duration of SSRI exposure and SSRI dose in the context of osteoporosis risk more closely. Sixth, we deliberately chose not to include other SSRIs or antidepressants from other classes in this study as this would introduce confounding by indication beyond what can be adjusted for with the data at hand (Østergaard, Reference Østergaard2023, Østergaard and Rohde, Reference Østergaard and Rohde2023, Psaty et al., Reference Psaty, Koepsell, Lin, Weiss, Siscovick, Rosendaal, Pahor and Furberg1999). We refrained from using a control group (not exposed to an SSRI) for the same reason. Seventh, while we compare drugs that are used for the same indication and have adjusted for a series of potential confounders, we cannot exclude residual confounding. Sixth, quantitative measures of bone health (e.g., bone mineral density) to operationalise osteoporosis were, unfortunately, not available to us. However, by defining osteoporosis via hospital discharge diagnoses and prescribed medications, the study outcome should have high specificity, as osteoporosis is typically diagnosed by clinical experts in public hospital settings in Denmark and since the medications in question are used exclusively for osteoporosis. The sensitivity is, arguably, less optimal as many cases will go undetected. This underdiagnosis is, however, unlikely to be differential between the three SSRI groups and will, therefore, not infer bias. Eight and finally, regarding generalizability, it should be borne in mind that Denmark has universal health coverage and is one of the countries where SSRIs are most widely used. Hence, the results of this study may not necessarily generalise to healthcare systems with different levels of coverage or different SSRI use patterns.
In conclusion, the results of this TTE study using nationwide data suggest that sertraline, citalopram and escitalopram, three of the most widely used SSRIs, do not appear to affect the risk of osteoporosis in a dose-response-like manner, and that these three drugs do not differentially affect the risk of osteoporosis. It would be ideal to follow-up on these findings with an RCT but given the required sample size and long follow-up time, such a study may not be feasible and affordable.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/neu.2025.3.
Acknowledgements
None.
Author contributors
The study was designed by all authors. SDØ procured the data. The statistical analyses were carried out by CR. All auth authors contributed to the interpretation of the results. CR and SDØ wrote the first draft of the manuscript, which was subsequently revised for important intellectual content by LJW and MB. All authors approved the final version of the manuscript prior to submission.
Financial support
There was no specific funding for this study. CR reports funding from the Danish Diabetes Academy, the Novo Nordisk Foundation (grant number NNF17SA0031406), and the Lundbeck Foundation (grant number R358-2020-2342). LJW is supported by a National Health and Medical Research Council (NHMRC) Emerging Leader Fellowship (1174060). MB is supported by MRFF, NHMRC, Congressionally Directed Medical Research Programs USA, AEDRTC Australian Eating Disorders Research and Translation Centre, Patient-Centered Outcomes Research Institute, Baszucki Brain Research Fund, Danmarks Frie Forskningsfond. Psykiatrisk Center Kobenhavn, Stanley Medical Research Institute, Victorian Government Department of Jobs, Precincts and Regions, Wellcome Trust, Victorian Medical Research Acceleration Fund, ControversiasPsiquiatria Barcelona, CRE, and the Victorian COVID-19 Research Fund. SDØ reports funding from the Lundbeck Foundation (grants R358-2020-2341 and R344-2020-1073), the Novo Nordisk Foundation (grant NNF20SA0062874), the Danish Cancer Society (grant R283-A16461), the Central Denmark Region Fund for Strengthening of Health Science (grant 1-36-72-4-20), the Danish Agency for Digitisation Investment Fund for New Technologies (grant 2020-6720), and Independent Research Fund Denmark (grant 7016-00048B and 2096-00055A). These funders played no role in the design or conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Competing interests
CR received the 2020 Lundbeck Foundation Talent Prize. MB has acted as consultant (Last 3 years) for Lundbeck, Sandoz, Servier, Medisquire, HealthEd, ANZJP, EPA, Janssen, Medplan, RANZCP, Abbott India, ASCP, International Society of Bipolar Disorder, Precision Psychiatry, Penn State College of Medicine, Shanghai Mental Health Centre. SDØ received the 2020 Lundbeck Foundation Young Investigator Prize. Furthermore, SDØ owns/has owned units of mutual funds with stock tickers DKIGI, IAIMWC, SPIC25 KL and WEKAFKI, and owns/has owned units of exchange traded funds with stock tickers BATE, TRET, QDV5, QDVH, QDVE, SADM, IQQH, USPY, EXH2, 2B76, IS4S, EUNL and SXRV. LJW reports no conflicts of interest.
Open access
