Benzodiazepine Treatment and Fracture Risk in Young Persons With Anxiety Disorders

We used MarketScan Commercial Claims and Encounters data from January 1, 2007, through December 31, 2016. The database covers individuals with employer-sponsored health insurance and their dependents across the United States. Patient-level details were available on insurance enrollment, outpatient and inpatient services, and outpatient dispensed prescriptions. Service visits contained information on the date, provider type, diagnostic codes, and procedure codes.

We identified privately insured children (6–17 years) and young adults (18–24) newly initiating benzodiazepine or SSRI treatment (Table 1 footnotes). We used records of dispensed prescriptions to identify new benzodiazepine and SSRI treatment episodes, defined as no benzodiazepine or SSRI prescriptions in the year before treatment initiation with continuous insurance enrollment during that year (Supplemental Fig 2). We required an anxiety disorder diagnosis within the 30 days before or the day of treatment initiation. Anxiety disorders were defined with International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) or International Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM) codes (Supplemental Table 4).

We excluded new users with diagnoses for bipolar disorder, schizophrenia, personality disorder, other psychosis, autistic disorder, substance use disorder, or epilepsy and/or convulsions in the previous year and an inpatient admission in the previous 30 days. We excluded persons with any fracture diagnosis in the 3 months before treatment initiation to reduce misclassification of follow-up fracture visits as an outcome. For youth with >1 new benzodiazepine or SSRI treatment episode meeting inclusion criteria, we selected the first episode. Lastly, we excluded youth initiating benzodiazepine and SSRI treatment on the same day.

We identified fracture events using ICD-9-CM or ICD-10-CM diagnostic codes from inpatient and outpatient records beginning the day after benzodiazepine or SSRI initiation and ending at 3 months. The primary outcome was incidence of upper or lower limb factures (shoulder, upper arm, forearm, wrist, femur, lower leg, ankle) because these fractures are more likely than other fracture locations to be caused by falls.^{27,28,42–45}  For a secondary outcome, we included all fracture locations. Fracture definitions are presented in Supplemental Table 5.

We estimated benzodiazepine and SSRI treatment duration with prescription refill dates and days’ supply. If there was no prescription refill within 20 days (grace period) after the days’ supply ended for the previous prescription, the individual was considered to have discontinued treatment. In sensitivity analyses, 10- and 30-day grace periods were used.

Patient characteristics were identified from inpatient and outpatient records (ICD-9-CM, ICD-10-CM, or Current Procedural Terminology codes) and dispensed prescription records in the year before or day of treatment initiation. Broadly, these included age, sex, anxiety disorder diagnosis, provider type of anxiety diagnosis, psychotherapy claims, psychiatric diagnoses, health care use measures, nonpsychiatric medical diagnoses, pediatric chronic complex condition scale,⁴⁶  injuries, and psychotropic and other medications dispensed in the previous 90 days.

We described children and young adults with anxiety disorders initiating benzodiazepine or SSRI treatment and estimated treatment duration and switching between therapies (eg, SSRI initiator filling a benzodiazepine). Primary analyses were as treated given investigation of a safety concern with hypothesized effect during drug use.⁴⁷  Youth contributed person-time beginning the day after treatment initiation until the first of the following: incident fracture, treatment discontinuation, treatment switching, insurance disenrollment, 3 months, or the end of data.

Our primary measures of adverse effect included crude and propensity score–adjusted incident fracture rates, incident rate differences (IRDs), and incident rate ratios (IRRs). We estimated propensity scores of benzodiazepine treatment separately in children and young adults; variables included potential confounders or predictors of fracture, including year of treatment initiation to account for temporal shifts in prescribing and fracture diagnoses. We used inverse probability of treatment weighting (IPTW) and applied asymmetric trimming,⁴⁸  excluding those with a propensity score below the first percentile of benzodiazepine initiators and above the 99th percentile of SSRI initiators, to select youth with greater treatment equipoise. In secondary analyses, we estimated results using stabilized inverse probability of treatment weighting (SIPTW) without trimming, restricting follow-up to 30 days, and by duration of action for the initial benzodiazepine (long acting, short acting).

We completed intention-to-treat analyses with no censoring at treatment discontinuation or switching and as-treated estimates using 10- and 30-day grace periods to define treatment discontinuation. We examined results in youth with an initial benzodiazepine or SSRI days’ supply value ≥3, excluding youth with baseline antidepressant use, and censoring at September 30, 2015, given the shift to ICD-10-CM codes. We stratified results by sex and common comorbid psychiatric diagnoses (attention-deficit/hyperactivity disorder [ADHD], depression) and restricted the data to youth without previous injury diagnostic codes given hypothesized differing baseline fracture rates in these subgroups. We additionally examined results in youth without recent psychotropic prescriptions and we examined results stratified by any psychiatric comorbidity diagnosis. Propensity scores were re-created with trimming applied to each subgroup. For exploratory analyses in benzodiazepine initiators, we examined fracture rates by a high versus lower initial benzodiazepine dose per day. A high dose per day was defined empirically as a dose greater than the third quartile on the basis of the distributions in children (diazepam >10 mg/day, alprazolam >1 mg/day, clonazepam >1 mg/day, lorazepam >2 mg/day) and in young adults (diazepam >15 mg/day, alprazolam >1.25 mg/day, clonazepam >1.1 mg/day, lorazepam >2.1 mg/day). The exploratory analyses included 92% of children and 94% of young adults initiating benzodiazepine treatment; expanded details are in the Supplemental Information.

The final child cohort included 12 840 benzodiazepine initiators and 107 875 SSRI initiators with anxiety disorders, and the young adult cohort included 57 684 benzodiazepine initiators and 122 084 SSRI initiators. Unspecified anxiety disorder and generalized anxiety disorder were the most common anxiety diagnoses, and benzodiazepine initiators were older, more likely to have panic disorder and recent ED visits, and less likely to have comorbid mental health diagnoses (Table 1). Alprazolam and lorazepam were the most common initial benzodiazepine agents (Supplemental Table 6).

Most children (82%) and young adults (77%) initiating benzodiazepine treatment did not have a second prescription fill before discontinuation. Three months after initiation, 8% of child and 11% of young adult benzodiazepine initiators remained on treatment, compared with more than half of SSRI initiators (Supplemental Table 6). In addition, 22% of child and 17% of young adult benzodiazepine initiators filled an SSRI prescription in the 3 months after, and fewer SSRI initiators filled a benzodiazepine during follow-up.

In children, the incident upper and lower limb fracture rate during the first months of treatment was 25.5 per 1000 person-years (PYs), with 33.1 per 1000 PYs in benzodiazepine initiators and 25.1 per 1000 PYs in SSRI initiators (Table 2). After weighting and trimming, benzodiazepine and SSRI initiators were well balanced (Supplemental Table 7). The adjusted fracture rate was elevated in benzodiazepine initiators compared with SSRI initiators: adjusted IRR = 1.53 (95% confidence interval [CI]: 0.94–2.50) and IRD = 13.4 per 1000 PYs (Table 2, Fig 1). The association was stronger for long-acting benzodiazepine initiators versus SSRI initiators: adjusted IRD = 32.9 and IRR = 2.30 (CI: 1.08–4.91). Considering initial benzodiazepine dose, the crude fracture rate was 52.7 (CI: 25.2–96.8) per 1000 PYs in initiators of high dose-per-day benzodiazepine compared to 30.3 (CI: 19.8–44.5) per 1000 PYs in initiators of lower dose-per-day benzodiazepine.

Under the any fracture outcome definition, the incident fracture rate approximately doubled (49.7 per 1000 PYs) and the IRR approached null, whereas the IRD remained similar to our primary fracture definition (Table 2). Results were consistent when restricting follow-up to a maximum of 1 month, in children without a previous injury, and in children without recent psychotropic prescriptions (Table 3). We observed a heightened fracture rate in benzodiazepine initiators when using a stricter 10-day grace period to define treatment discontinuation (IRR = 1.83 [CI: 1.07–3.14]). Fracture rates were higher in boys and children with ADHD; the association between benzodiazepine versus SSRI treatment and fractures was stronger in children with ADHD, although estimates are less precise.

In young adults, the incident upper and lower limb fracture rate was 8.8 per 1000 PYs, approximately one-third of the rate in children. The fracture rate was similar in benzodiazepine (8.7 per 1000 PYs) and SSRI initiators (8.9 per 1000 PYs) (Table 2). After weighting and trimming, benzodiazepine and SSRI initiators were well balanced (Supplemental Table 8). The adjusted fracture rate was 7.6 per 1000 PYs in benzodiazepine initiators and 8.9 per 1000 PYs in SSRI initiators, with an adjusted IRR of 0.85 (CI: 0.57–1.27) and IRD of −1.3 per 1000 PYs (Table 2, Fig 1). The crude fracture rate was similar in young adults initiating benzodiazepine treatment on a high or low dose per day, that of 8.3 (CI: 4.0–15.3) and 8.8 (CI: 6.2–12.1) per 1000 PYs, respectively.

Under the any fracture definition, the young adult incident fracture rate tripled to 26.9 per 1000 PYs with no association between treatment type and fracture (Table 2). Most findings from additional analyses were similar to the primary analysis (Table 3). Incident upper and lower limb fracture rates were lower in male benzodiazepine versus SSRI initiators; however, under the any fracture definition (results not shown), the association was null (IRR = 0.94). Similarly, when restricting follow-up to outcomes under ICD-9-CM fracture codes, the estimate was null under the any fracture definition.

With our findings, we provide novel comparative estimates of fracture rates during benzodiazepine and SSRI treatment in children and young adults with anxiety disorders. In children initiating benzodiazepines, we observed a heightened rate of upper and lower limb fractures compared to SSRI treatment, with no heightened rate in young adult benzodiazepine initiators. Increased caution may be warranted at benzodiazepine initiation in children with anxiety disorders, particularly with initiation of long-acting benzodiazepines. These findings add to the limited understanding of benzodiazepine treatment safety in youth.

The elevated upper and lower limb fracture rate in children initiating benzodiazepine treatment compared to SSRI treatment (IRR = 1.53) is similar to estimates from older adult benzodiazepine users compared with nonusers with pooled meta-analysis effect estimates of risk ratios of 1.25 to 1.52.^15,19,21,22  We focused on short-term fracture risk because benzodiazepine treatment is typically for short durations in youth with anxiety disorders and authors of literature regarding older adults have observed greater fall and fracture risk in the first weeks after initiation.^19,24,25,49  However, not all study authors observed this short-term effect.²⁰ 

Fracture rates were higher in children initiating long-acting benzodiazepines. In adults, there is evidence suggesting benzodiazepines with longer half-lives have a higher relative risk of fractures and motor vehicle crashes than benzodiazepines with shorter half-lives^13,50,51 ; however, findings are mixed.^18,24,25,52  In meta-analyses, authors observed similar fracture risk in short-acting and long-acting adult benzodiazepine users.^15,21  Preferential prescribing of short-acting benzodiazepines to adults at increased fall or fracture risk may have contributed to observations of no elevated risk with long-acting benzodiazepines.²⁵  Relatedly, children prescribed long-acting benzodiazepines for anxiety could have a baseline elevated fracture risk beyond what we controlled for and may also have greater benzodiazepine exposure if prescribed for regular, daily use versus as-needed use. In our exploratory analysis, we observed an elevated fracture rate in children initiating a high benzodiazepine dose per day versus lower dose per day. In older adults, higher benzodiazepine doses are associated with a heightened fracture risk.^25,26,50  It is suggested in our crude estimates that a closer examination of benzodiazepine dose on fracture risk and the interaction of dose and half-life will further inform prescribing decisions.

The estimated rate difference for fractures in children receiving benzodiazepine versus SSRI treatment was moderately small (13 per 1000 PYs), yet notable because alternative treatments are available; these fractures thus represent potentially preventable adverse outcomes. The heightened fracture rate difference in boys and children with ADHD may be related to greater activity levels and thus increased opportunities for treatment-related adverse effects that result in fracture-causing falls. Most falls in children and young adults are without incident; however, falls can result in fracture including severe, life-threatening injuries, particularly if falls occur from elevated heights.^29,53  Clinicians could advise caution in activities that increase the likelihood of a severe injury until it is clear how benzodiazepine treatment will affect the child.

In young adults with anxiety disorders, we did not observe an elevated fracture rate in benzodiazepine initiators versus SSRI initiators. Several factors may contribute to this finding. Young adults are generally less active and experience fewer fall-related injuries than do children.^54,55  Therefore, there may be less opportunity for benzodiazepine adverse effects to cause a fall that could result in fracture. Young adults may receive increased warnings surrounding benzodiazepine side effects given this age group drives and benzodiazepine drug labels state that individuals should not drive or engage in dangerous activities until they know how the medication will affect them. Relatedly, young adults likely have more awareness of symptoms and side effects and autonomy over when medication is taken. Benzodiazepines are more commonly prescribed to young adults than children⁵⁶ ; children who receive benzodiazepines may, on average, have higher anxiety severity than young adults, a potential source of residual confounding. Subthreshold or undiagnosed conduct disorder is another potential source of residual confounding more prevalent in children. Young adults may be more likely to have previous medical or nonmedical exposure to benzodiazepine^4,57  that lessens their risk of benzodiazepine-related fractures. Alcohol and drug use are potential unmeasured confounders that can increase accidental injury and fall risk^58–60  and could increase the likelihood of receiving SSRIs given concerns related to benzodiazepine misuse. This could lead to attenuated fracture risk effect, particularly in young adults for whom substance use is more prevalent; however, there is uncertainty because patient treatment preferences could differ by substance use. In addition, our primary fracture outcome was uncommon in young adults and perhaps less clinically relevant than for children.

As previously recognized, the observed association between benzodiazepines and fractures in older adults could be biased by confounding by indication.^15,18,22  Our restriction to youth with anxiety disorders, active comparator design, and implementation of IPTW sought to reduce the potential for confounding by indication while addressing a clinically relevant question. Despite our study design, unmeasured confounding may persist. Antidepressants (including SSRIs, our active comparator) have been associated with an elevated fracture risk compared to no antidepressant use.^22,61,62  The proposed mechanism of fracture risk with benzodiazepine treatment is related to adverse effects such as sedation or dizziness leading to a fall, whereas for SSRIs, the hypothesized mechanisms are an increased fall risk or decreased bone mineral density.^63,64  However, findings are inconsistent on the association between SSRIs and fractures^26,65  and studies may be susceptible to confounding by indication.^62,66  Whether our observed associations between benzodiazepines and fractures hold across benzodiazepine indications and compared with other alternative pharmacotherapies remains unclear.

Limitations of the work should be considered. We cannot be certain the primary treatment indication was an anxiety disorder and whether, and when, filled prescriptions were consumed. We lack clinical severity measures; it is unclear how much unmeasured differences in psychiatric condition severity exist between youth initiating a benzodiazepine versus SSRI and how anxiety severity impacts fracture risk. Further research is needed to evaluate whether elevated fracture risk with benzodiazepine treatment is concentrated in children with a psychiatric comorbidity or if this observation is a result of differences in prescribing and/or patient characteristics between children with and without psychiatric comorbidities. Benzodiazepine treatment is typically shorter than SSRI treatment; however, our results restricting to 30 days of treatment compare outcomes over a similar denominator. Many subgroup analyses were limited by sample size; future research will determine if findings remain across populations. Although the primary fracture definition intended to capture fractures more likely to be caused by a fall, fractures could be caused by any mechanism and be unrelated to treatment. We do not expect differential misclassification of fracture coding by treatment; however, fracture events were based on diagnostic codes and may not represent true events. The validity of fracture diagnostic codes varies by fracture type, with positive predictive values typically estimated at >80%.^67–69  Our data overlapped the transition to ICD-10-CM codes; authors of future analyses could examine this research question strictly under ICD-10-CM codes.

We observed a heightened rate of fractures in children with anxiety disorders initiating benzodiazepine treatment compared to SSRI treatment but not in young adults. Increased caution in the weeks after benzodiazepine initiation in children with anxiety disorders may be warranted. Moreover, given the lack of efficacy evidence and limited safety data for benzodiazepine treatment of pediatric anxiety disorders, this safety signal is an important consideration in the benefit-harm evaluation for prescribing benzodiazepines to youth with anxiety disorders.

ADHD

attention-deficit/hyperactivity disorder

CI

confidence interval

ED

emergency department

ICD-9-CM

International Classification of Diseases, Ninth Revision, Clinical Modification

ICD-10-CM

International Classification of Diseases, 10th Revision, Clinical Modification

IPTW

inverse probability of treatment weighting

IRD

incident rate difference

IRR

incident rate ratio

PY

person-year

SIPTW

stabilized inverse probability of treatment weighting

SSRI

selective serotonin reuptake inhibitor