Association of Mental Disorders and Related Medication Use With Risk for Major Osteoporotic Fractures

Key Points

Question  How do mental disorders and psychotropic medications affect fracture risk estimates with FRAX, a tool developed for clinical use in the general population?

Findings  In this longitudinal cohort drawn from a population-based registry, FRAX scores were calculated for 68 730 individuals. FRAX significantly underestimated 10-year hip fracture risk in those with depression by 51% and in those using selective serotonin reuptake inhibitors by 57%, those using mood stabilizers by 98%, those using antipsychotics by 171%, and those using benzodiazepines by 31%.

Meaning  Depression and some psychotropic medications are associated with an increased risk of fracture that is not fully captured by FRAX scores.

Importance  Osteoporotic fractures are a leading cause of disability, costs, and mortality. FRAX is a tool used to assess fracture risk in the general population. Mental disorders and medications to treat them have been reported to adversely affect bone health, but, to date, they have not been systematically studied in relation to osteoporotic fractures.

Objective  To examine the association of mental disorders and psychotropic medication use with osteoporotic fracture risk in routine clinical practice.

Design, Setting, and Participants  In this population-based cohort study, bone mineral density and risk factors were used to calculate FRAX scores using data from the Manitoba Bone Density Program database of all women and men 40 years of age or older in Manitoba, Canada, referred for a baseline dual-energy x-ray absorptiometry scan from January 1, 1996, to March 28, 2013. Population-based health services data were used to identify primary mental disorders during the 3 prior years, psychotropic medication use during the prior year, and incident fractures. Cox proportional hazards regression models estimated the risk for incident fractures based on mental disorders and use of psychotropic medications. Data analysis was conducted from November 25, 2013, to October 15, 2016.

Main Outcomes and Measures  Incident nontraumatic major osteoporotic fractures (MOFs) and hip fractures.

Results  Of the 68 730 individuals (62 275 women and 6455 men; mean age, 64.2 [11.2] years) in the study, during 485 322 person-years (median, 6.7 years) of observation, 5750 (8.4%) sustained an incident MOF, 1579 (2.3%) sustained an incident hip fracture, and 8998 (13.1%) died. In analyses adjusted for FRAX score, depression was associated with MOF (adjusted hazard ratio [aHR], 1.39; 95% CI, 1.27-1.51; P < .05) and hip fracture (aHR, 1.43; 95% CI, 1.22-1.69; P < .05) before adjustment for medication use, but these associations were not significant after adjustment for medication use. In contrast, the use of selective serotonin reuptake inhibitors (aHR for MOF, 1.43; 95% CI, 1.27-1.60; P < .05; aHR for hip fracture, 1.48; 95% CI, 1.18-1.85; P < .05), antipsychotics (aHR for MOF, 1.43; 95% CI, 1.15-1.77; P < .05; aHR for hip fracture, 2.14; 95% CI, 1.52-3.02; P < .05), and benzodiazepines (aHR for MOF, 1.15; 95% CI, 1.04-1.26; P < .05; aHR for hip fracture, 1.24; 95% CI, 1.05-1.47; P < .05) were each independently associated with significantly increased risk for both MOF and hip fracture. FRAX significantly underestimated the 10-year risk of MOF by 29% and of hip fracture by 51% for those with depression. It also underestimated the 10-year risk of MOF by 36% for use of selective serotonin reuptake inhibitors, by 63% for use of mood stabilizers, by 60% for use of antipsychotics, and by 13% for use of benzodiazepines. FRAX underestimated the 10-year risk of hip fracture by 57% for use of selective serotonin reuptake inhibitors, by 98% for use of mood stabilizers, by 171% for use of antipsychotics, and by 31% for use of benzodiazepines. FRAX correctly estimated fracture risk in people without mental disorders and those not taking psychotropic medications.

Conclusions and Relevance  Mental disorders and medication use were associated with an increased risk for fracture, but in simultaneous analyses, only medication use was independently associated with fracture. Depression and psychotropic medication use are potential risk indicators that are independent of FRAX estimates.

Mental disorders affected 18% of adults in the United States in 2013, including 15% of those 50 years of age or older.¹ Psychotropic medications are among the most highly prescribed agents in the United States, with 330 million prescriptions for antidepressants and anxiolytics in 2009.² In Canada, 17% of women 45 to 64 years of age take antidepressant medications, making these agents the most commonly used in that demographic subgroup.³ Mental disorders such as depression have been associated with an increased risk of osteoporosis and fracture.⁴ Likewise, specific psychotropic medications have been consistently associated with lower bone mineral density (BMD) and increased risk of fracture, independent of the mental disorders they are treating.^5-9

Fractures are associated with increased morbidity, mortality, disability, and cost and thus represent an important target for public health prevention efforts.^10-12 Osteoporosis, a disorder manifested by reduced density and quality of bone with susceptibility to fragility fractures, is also highly prevalent; a 50-year-old white woman has a 15% to 20% lifetime risk of hip fracture and a 50% risk of any osteoporotic fracture.¹³ The fracture risk assessment tool FRAX (Centre for Metabolic Bone Diseases, University of Sheffield, Sheffield, UK) is based on easily assessed risk factors identified in meta-analyses of population-based prospective cohorts and estimates 10-year probability for a major osteoporotic fracture (MOF) (ie, hip, vertebral [clinical], humerus, or forearm) or for a hip fracture.^14,15 FRAX has been incorporated into more than 100 clinical guidelines worldwide for osteoporosis management.^16,17 However, there remain clinical factors that contribute to fracture risk that are not included in the FRAX algorithm.¹⁸ Some examples include types 1 and 2 diabetes, heart failure, falls, and certain medications that contribute to the risk of falls such as antihypertensive and psychotropic medications.^19-23 Identifying additional factors that influence the ability of FRAX to estimate fracture risk within specific populations is therefore important to inform practitioners how to interpret FRAX results within these populations.

Mental disorders and the psychotropic medications used to treat these disorders represent another potentially important set of risk factors that have not previously been examined in the context of FRAX. Therefore, using a large clinical dual-energy x-ray absorptiometry (DXA) scan registry linked to administrative health data sets that contain physician-diagnosed mental disorders and detailed medication information, our objective was to provide, to our knowledge, a first-ever assessment of the performance of FRAX in routine clinical practice for men and women with mental disorders and/or using psychotropic medications, and to determine whether these disorders and medications contribute to fracture risk beyond the risk estimated by the FRAX tool.

Data for this study comprised the Manitoba Bone Density Program database, which contains information on all individuals in the province of Manitoba, Canada (population, 1.2 million), who have received a DXA assessment of BMD since 1990. This database is linked through an encrypted personal health number to data sets in the population-based Manitoba Centre for Health Policy Data Repository. Specific repository databases used in this study included physician billing claims and hospital discharge abstracts (which together provided data on mental disorders, fractures, and medical illnesses contributing to the FRAX score), vital statistics (mortality), and the provincial drug database (containing information on medication type, dose, and date of dispensation for the entire population). Patient demographic information was obtained through the population registry file. The accuracy and completeness of these data sets are excellent and have been described in detail.^24-27 The study was approved by the University of Manitoba Human Research Ethics Board. Participant consent is waived for administrative data studies under the Personal Health Information Act of Manitoba.

The study included men and women 40 years of age or older at the time of baseline BMD assessment during the period from January 1, 1996, to March 28, 2013. The cohort consisted of 68 730 individuals (62 275 women and 6455 men; mean [SD] age, 64.2 [11.2] years). Individuals were included in the study cohort if they had continuous medical coverage in Manitoba.

The mental disorders examined included depression, anxiety disorders, and schizophrenia. Disorders were ascertained using International Classification of Disease, Ninth Revision, Clinical Modification (ICD-9-CM) and International Statistical Classification of Diseases and Health Related Problems, Tenth Revision, Canada (ICD-10-CA) codes. Disorders were coded as present if 2 or more diagnostic codes for the same disorder (eAppendix 1 in the Supplement) occurred in either hospital discharge abstracts or physician billing claims in the 3 years prior to DXA scan. For individuals with codes for different mental disorders, a diagnosis was assigned based on the most frequently coded diagnosis using definitions based on previous work from the same databases.²⁸ Mental disorder severity scores were created based on increasing numbers of outpatient visits (2, 3-5, or ≥6) or hospitalizations (1, 2, or ≥3) associated with mental disorders.

Exposure to psychotropic medications during the year prior to the DXA scan was ascertained using the World Health Organization Anatomical Therapeutic Chemical Classification System (eAppendix 2 in the Supplement). A 1-year time frame was chosen to capture a more proximate medication exposure. Categories included selective serotonin reuptake inhibitors (SSRIs), tricyclic antidepressants, other antidepressants, lithium, nonlithium mood stabilizers, antipsychotics, and benzodiazepines. Exposure was arbitrarily defined as dispensed prescriptions covering more than 50% of days during the year (medication possession ratio, >0.5).

The primary study outcome was an incident hip fracture or MOF occurring between the date of the baseline BMD assessment and the end of the observation period (March 31, 2013, unless coverage was terminated by death or migration out of province). Major osteoporotic fractures included hip, vertebral (clinical), humerus, or forearm fractures. Fracture definitions were based on ICD-9-CM and ICD-10-CA codes adapted from algorithms validated in Canadian administrative databases (eAppendix 1 in the Supplement).²⁹ Fractures indicating a high level of trauma were excluded using relevant ICD-9-CM and ICD-10-CA codes.

Dual-energy x-ray absorptiometry scans of the hip and spine were performed and analyzed in accordance with manufacturer recommendations using one of the Manitoba Bone Density Program’s cross-calibrated DXA instruments (the Lunar DPX prior to 2000 and the GE Lunar Prodigy in 2000 and thereafter; both General Electric). Femoral neck T scores were calculated using US National Health and Nutrition Examination Survey III³⁰ white female reference values as required by FRAX.

Ten-year probabilities of incident hip fracture and MOF, including femoral neck BMD, were calculated using the FRAX tool calibrated for the Canadian population (FRAX Desktop Multi-Patient Entry, version 3.7) from nationwide data.³¹ Age and sex at BMD assessment were determined using the population registry file. Height and weight were obtained by self-report prior to 2000 and objectively in 2000 and thereafter using a wall-mounted stadiometer for height and a scale for weight. A diagnosis of rheumatoid arthritis was ascertained using ICD codes from hospital separation abstracts and outpatient physician visits. Chronic obstructive pulmonary disease and alcohol or substance abuse diagnoses were identified from the same sources and used as proxy measures for smoking and high alcohol use. Similar prevalence estimates and effects on fracture have previously been shown using these proxy definitions as was seen in population-based data.^32,33 Parental hip fracture was self-reported by participants from 2005 onward; prior to 2005, we applied an age- and sex-specific adjustment for the effect of missing parental hip fracture, as previously described.³² Prolonged use of glucocorticoids (≥90 days) in the year prior to BMD assessment was ascertained using the provincial drug database. FRAX input variables constructed using these definitions have been shown to be adequate surrogates in terms of prevalence and risk estimation.^32,33 Although not a component of FRAX, we also ascertained the use of osteoporosis medication (>180 days) in the year prior to BMD assessment (bisphosphonates, calcitonin, systemic estrogen products, raloxifene, or teriparatide) as done in previous studies.¹⁹ Under Canadian clinical practice guidelines, a T score in the osteoporotic range might not require osteoporosis treatment unless fracture probability is also high.³⁴

Statistical analysis was conducted from November 25, 2013, to October 15, 2016. The prevalence of FRAX risk factors, mental disorders, and the use of psychotropic medications was compared between individuals with incident fractures and those without incident fractures using analysis of variance and χ² tests of independence. All analyses were conducted separately for MOF and hip fracture. Cox proportional hazards regression analysis yielded adjusted hazard ratios (aHRs) and 95% CIs for fractures associated with mental disorders (compared with no mental disorder) and the use of psychotropic medications in the year prior to the index BMD (compared with no use). Results were adjusted for FRAX score and prior use of osteoporosis medications. Two models were fit to the data: model 1 examined mental disorders and the use of psychotropic medications separately, while model 2 examined mental disorders and the use of psychotropic medications simultaneously. Validation of a prognostic model should include an evaluation of both discrimination and calibration.³⁵ Discrimination is the ability of a predictive model (FRAX) to distinguish between individuals who do or do not experience the outcome (fracture). Fracture discrimination using FRAX was estimated for the subgroups of mental disorders and psychotropic medications and was expressed as the aHR per SD increase. Within the same subgroups, calibration of the FRAX model was assessed, where calibration was determined from the difference between the observed cumulative fracture probability at 10 years and the expected 10-year fracture probability estimated by FRAX. Assessment of calibration considered the effect of competing mortality on the 10-year cumulative incidence for fracture (ie, a fracture outcome cannot be incurred following death), which is a component of the FRAX method. Expected probability within 10% of observed probability was considered to indicate good calibration.³⁶

Supplementary analyses examined the association between mortality, mental disorders, and use of psychotropic medications with multivariable Cox proportional hazards regression adjusted for FRAX and use of osteoporosis medications; C statistics for fracture risk stratification from FRAX scores among mental disorder and psychotropic medication subgroups; the influence of mental disorder severity on associations between fractures and psychotropic medications; adjusting for prior use of antihypertensive medication; adjusting for use of osteoporosis medication subsequent to BMD assessment (rather than prior use); the effect of excluding participants who used multiple psychotropic medications and for FRAX calculated without BMD. Supplementary 2-way interaction analyses were performed for mental disorders and use of psychotropic medications vs (1) sex, (2) age, (3) FRAX probability, (4) World Health Organization category or prior hip or spine fracture, (5) osteoporosis treatment prior to BMD, and (6) osteoporosis treatment subsequent to BMD. Statistical analyses were performed with Statistica, version 10.0 (StatSoft Inc). A 2-sided P < .05 was set as the threshold for assessing statistical significance.

The overall rate of mental disorders was 18.9% (n = 12 982), with the most common diagnosis being depression (6604 of 12 982 [50.9%]). Psychotropic medication use was recorded in 17.4% of the sample (n = 11 938), and 2468 of these individuals (20.7%) were taking multiple psychotropics.

During 485 322 person-years (68 730 persons; median, 6.7 years) of observation, 5750 (8.4%) sustained an incident MOF, 1579 (2.3%) sustained an incident hip fracture, and 8998 (13.1%) died. The characteristics of the sample according to mental disorder or psychotropic medication exposure are shown in Table 1. Incident MOFs were higher among participants with mental disorders compared with those without mental disorders (1188 of 12 982 [9.2%] vs 4562 of 55 748 [8.2%]; P < .001). Incident MOF (1170 of 11 938 [9.8%] vs 4580 of 56 792 [8.1%]; P < .001) and hip fracture (354 of 11 938 [3.0%] vs 1225 of 56 792 [2.2%]; P < .001) were higher among those with psychotropic medication exposure compared with those without such exposure.

Table 2 presents aHRs for the association between fractures, mental disorders, and use of psychotropic medication adjusted for FRAX probability and prior use of osteoporosis medication. In model 1, which analyzed disorders and medications separately, depression (aHR, 1.39; 95% CI, 1.27-1.51), anxiety disorders (aHR, 1.19; 95% CI, 1.09-1.30), and schizophrenia (aHR, 1.82; 95% CI, 1.16-2.85) were all significantly associated with MOF. Selective serotonin reuptake inhibitors, antipsychotics, and benzodiazepines were associated with both MOF and hip fracture: for MOF, SSRIs had an aHR of 1.47 (95% CI, 1.32-1.63), antipsychotics had an aHR of 1.48 (95% CI, 1.21-1.81), and benzodiazepines had an aHR of 1.16 (95% CI, 1.05-1.28), and for hip fracture, SSRIs had an aHR of 1.51 (95% CI, 1.22-1.85), antipsychotics had an aHR of 2.18 (95% CI, 1.58-3.02), and benzodiazepines had an aHR of 1.23 (95% CI, 1.04-1.45). Similar results were observed for psychotropic medications when mental disorders and use of psychotropic medications were analyzed simultaneously (model 2), but aHRs for mental disorders with MOF and hip fracture became nonsignificant.

Table 3 presents measures of fracture discrimination for FRAX scores within the subgroups of mental disorders and psychotropic medications used. For each 1-SD increase in FRAX score for incident MOF, there was approximately a doubling in the aHR for MOF in all subgroups, ranging from 1.87 (95% CI, 1.55-2.25) for antipsychotics to 2.69 (95% CI, 1.48-4.88) for lithium. The gradient of risk for incident hip fracture was even steeper, with aHRs that ranged from 3.28 (95% CI, 2.00-5.38) for mood stabilizers to 5.60 (95% CI, 4.38-7.15) for diagnosed anxiety disorder. No statistically significant interactions were detected between FRAX scores in association with both mental disorders and use of psychotropic medications, suggesting that the fracture discrimination for those with mental disorders and the use of psychotropic medications are similar to those in the general population.

Figure 1 presents the observed vs expected 10-year fracture probability for MOF, and Figure 2 presents the observed vs expected 10-year fracture probability for hip fracture. In the absence of mental disorders or psychotropic medication exposure, the observed fracture probability was within 10% of expected for both MOF and hip fracture, indicating good calibration. FRAX underestimated the 10-year risk of MOF by 29% for those with depression, by 36% for those exposed to SSRIs, by 63% for those using mood stabilizers, by 60% for those using antipsychotics, and by 13% for those using benzodiazepines. Similarly, FRAX underestimated the 10-year hip fracture risk by 51% for those with depression, by 57% for those using SSRIs, by 98% for those using mood stabilizers and those with schizophrenia, by 171% for those using antipsychotics, and by 31% for those using benzodiazepines.

eTable 1 in the Supplement shows the characteristics of the sample presented by fracture outcome. eTable 2 in the Supplement provides the aHRs for the association between death, mental disorders, and use of psychotropic medications. Exposure to every class of psychotropic medication was associated with an increased risk of death. eTable 3 in the Supplement presents C statistics showing that FRAX stratified fracture risk similarly among individuals with mental disorders, those with exposure to psychotropic medications, and the referent population. In the additional supplementary analyses, adjusting models for an index of disease severity did not alter the statistical significance of any of the psychotropic medication results from model 2. Supplementary analysis that adjusted for prior use of antihypertensive medication or use of osteoporosis medication subsequent to BMD testing did not alter the results. The sensitivity analysis that excluded users of multiple psychotropic medications revealed results that were unchanged in significance. eTable 4 in the Supplement presents the interaction analyses, none of which were statistically significant, with the possible exception of SSRI use and osteoporosis treatment subsequent to BMD testing for MOF, although this association may be spurious because no significant interaction was seen for hip fracture or osteoporosis treatment prior to BMD testing. Analyses based on FRAX probability computed without BMD testing are presented in eTables 5 and 6 and eFigures 1 and 2 in the Supplement.

This study adds to the existing literature on the use of psychotropic medications and the risk of fracture.³⁷ The primary finding is that FRAX significantly underestimated fracture risk among people with mental disorders and among those taking a wide range of psychotropic medications. Our results are based on all persons referred for osteoporosis screening with DXA, with linkage to physician diagnoses of mental disorders and a population-based database capturing detailed information on medication dispensed and incident fractures. Accordingly, our results are most applicable to older individuals, especially women, who are considered to be at risk of osteoporosis. Health care professionals should be aware of these effects when assessing osteoporosis in people with mental illness.

Although fracture stratification with the FRAX tool was preserved among those with mental disorders and those using psychotropic medications compared with the general population, calibration was significantly affected. FRAX scores underestimated the risk of MOF and hip fracture among persons with depression and those using SSRIs, mood stabilizers, antipsychotics, and benzodiazepines. Although age did not affect the relative risk associated with a specific mental health diagnosis or medication, the absolute risk would be larger among older individuals and those with higher baseline probability of fracture.

Several meta-analyses have found an association between fractures and the use of SSRIs, benzodiazepines, and antipsychotics.^8,38-40 The mechanism of this association may be in part due to the bone-mediated effects of these medications (median duration of use, >3 years for all medications), but it also may be secondary to falls.^40,41 Falls are not a FRAX input variable, which remains a plausible explanation for the increased risk of fracture associated with some psychotropic medications.²³

There are limitations to consider in the interpretation of these results. The study sample consisted of people referred for DXA testing; therefore, these results are most applicable to older adults, predominantly female, who are considered by primary care professionals to be at risk for osteoporosis. This potential selection bias may be offset by the fact the Manitoba Bone Density Program is the single centralized program for DXA testing in the province. A second limitation relates to the ascertainment of mental disorders. Given the low rates of mental health service use among people with mental disorders surveyed in many population-based studies,⁴² the rates in administrative data may underestimate the true rate in the sample. A related issue is disease severity, which was based on a proxy measure of the use of health services. Clinical assessment of severity using standardized measures would be preferable but was not an option in this data set. A third factor is the assessment of psychotropic medications, which is limited to prescriptions dispensed and raises uncertainties as to whether an individual adhered to treatment or what the specific indication for treatment was. The examined medications could have been used for conditions unrelated to mental disorders, such as neuropathic pain or a seizure disorder. Furthermore, certain classes, specifically other antidepressants and mood stabilizers, have intraclass variability with respect to the adverse effects of weight, which in turn makes it difficult to interpret their effect on fracture risk. Exposure was defined by an arbitrarily chosen medication possession ratio of more than 0.5, given that there is no accepted cutoff for high exposure. Nevertheless, the medication data sets used in this study still represent a distinct advantage over previous studies, given their population coverage and accuracy of medications, dates of prescription, and dose recordings.

The lack of assessment of falls is an important limitation. The menopausal status of most of the women was unknown but was recorded during the last data year, and more than 99% of the women (6578 of 6604) were postmenopausal. The results were not stratified by sex given the low percentage of men in the study. Men qualifying for DXA testing had a high prevalence of prior fracture, glucocorticoid use, or other secondary cause of osteoporosis. Finally, pharmacologic treatment for osteoporosis has not been shown to reduce the risk of traumatic fractures.

Diagnosed mental disorders and psychotropic medications identify subgroups at increased risk for fractures, and where this excess risk is not fully captured by the FRAX tool. Results are similar for FRAX scores calculated with and without BMD testing, and would therefore be applicable in the absence of DXA testing. Osteoporosis and mental disorders are highly prevalent conditions, and, as such, these findings have substantial implications for mental health care workers and primary care practitioners. Broader public health initiatives and advocacy may be required to achieve better recognition and management of this elevated risk of fracture given that patients with psychiatric illness may not receive optimal treatment for comorbid physical health conditions.⁴³

Accepted for Publication: February 19, 2017.

Corresponding Author: James M. Bolton, MD, Department of Psychiatry, University of Manitoba, PZ430-771 Bannatyne Ave, Winnipeg, MB R3E 3N4, Canada (jbolton@hsc.mb.ca).

Published Online: April 19, 2017. doi:10.1001/jamapsychiatry.2017.0449

Author Contributions: Dr Leslie had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Bolton, Majumdar, Sareen, Odén, Leslie.

Acquisition, analysis, or interpretation of data: Bolton, Morin, Majumdar, Lix, Johansson, McCloskey, Kanis, Leslie.

Drafting of the manuscript: Bolton, Sareen, Leslie.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Bolton, Lix, Odén, Leslie.

Administrative, technical, or material support: Sareen, Johansson, McCloskey, Leslie.

Study supervision: Kanis, Leslie.

Conflict of Interest Disclosures: Dr Morin reported receiving research grants from Amgen and Merck for unrelated work and receiving honoraria from Amgen for participation on an advisory board. Dr Sareen reported serving as a consultant for UpToDate and holding stocks in Johnson and Johnson. No other conflicts were reported.

Disclaimer: The results and conclusions are those of the authors, and no official endorsement by the Manitoba Centre for Health Policy, Manitoba Health, or other data providers is intended or should be inferred. This article has been reviewed and approved by the members of the Manitoba Bone Density Program.

Additional Contributions: We acknowledge the Manitoba Centre for Health Policy for use of data contained in the Population Health Research Data Repository (HIPC Project HIPC 2011/2012-31).