Use of hypnotic‐sedative medication and risk of falls and fractures in adults: A self‐controlled case series study

To evaluate the risk of falls and fractures in users of benzodiazepines, Z‐drugs, or melatonin.

Worldwide, approximately 171 million fall accidents occur every year. 1 With more than 110,000 new cases annually, fall accidents are ranked among the commonest causes of acute hospitalization in Denmark. 2 Falls in young populations may have lifelong detrimental effects on functional capacity due to traumatic brain injury or complicated bone fractures.In older populations, the most common severe outcome of fall accidents is hip fractures, which are associated with significant mortality and loss of independence. 3lthough falls typically result from a complex interaction of predisposing and precipitating risk factors, ranging from illicit drug use and alcohol intoxication in younger populations to frailty and comorbidity in older populations, the use of sedative-hypnotic medication is often singled out as a risk factor.][6][7][8][9][10][11] Adverse effects of sedative-hypnotic medications include impaired coordination and concentration, muscle weakness, somnolence, and balance problems, which presumedly underlie the association between sedative medication and increased risk of falls. 12However, an important limitation of the studies, assessing the risk for falls and fractures of sedative-hypnotic medication, is the issue of confounding by indication.Some studies have argued that insomnia, rather than sedative medication, drives the increased risk of falls. 13Moreover, it has been suggested that persons with a higher risk of injurious falls are more likely to receive a new prescription for benzodiazepines than people with a lower baseline risk. 14n this nationwide study, we aimed to evaluate the risk of falls and fractures in adults who initiated treatment with benzodiazepines, Z-drugs, or melatonin.In an attempt to differentiate between the risk conferred by the medications and the risk from the underlying disorder that required treatment, we assessed the risk of falls or fractures before and after treatment initiation.An approach that has recently been used to explore the association between the use of antipsychotic drugs and cholinesterase inhibitors and the risk of falls and fractures. 15| METHODS

| Study population
We used information on the entire Danish population from nationwide registers.The study population was identified in the Danish Civil Registration System using the unique personal identification number (CPR), which enables accurate linkage of recorded individual-level information in all registers. 16We first included all 4,598,819 individuals above 15 years of age living in Denmark at the start of the followup period (January 1st, 2003) and excluded those who currently or previously used benzodiazepines, Z-drugs, or melatonin (n = 1,002,208).Of the remaining 3,576,111 individuals, 699,335 persons purchased benzodiazepines, Z-drugs, or melatonin for the first time between 2003 and up to and including 2016 (the source population).The study was approved by the regional Data Inspection.No individual-level consent was required, and all data used were pseudo-anonymized.

| Study design
We applied a self-controlled case series design, 17,18 in which follow-up time is divided for each individual into exposed and unexposed periods, after which the risks for the outcome of interest in the different periods are compared.A significant strength of the applied case-series approach is that it implicitly controls for the effects of the confounding factors, assuming that these factors remain unchanged during the period of observation. 18We classified the available follow-up time for each individual into the following six predefined periods: (1) Baseline unexposed to the hypnotic-sedative drugs (24-12 months before initiation); (2) Early pre-treatment:

Significant outcomes
• Older persons treated with sedative-hypnotic medication had the highest risk for falls immediately before treatment start.• In the period under treatment, the risk gradually declined but was consistently higher than in the non-treatment period.• Other factors beyond sedative-hypnotic medication should be considered when evaluating an older person's risk for falls.

Limitations
• Our outcomes only included falls and fractures diagnosed at a hospital and thus represent the most severe events.• In the self-controlled design, time-invariant confounders were efficiently controlled for but it may be subject to confounding by timevarying within person confounding.

| Outcome events
The main outcome was fall accidents ascertained in the Danish National Patient Registry using the Nordic accident classification system (NAC) 23,24 codes EUS0, EUS1, and EUHE from 1995 through 2018.This registry includes information on the diagnoses, assigned by the physician at discharge for all visits (on an emergency, in-, or outpatient basis) to Danish hospitals. 25Information on the second outcome, fractures, was retrieved from the Danish National Patient Registry using International Classification of Diseases (ICD)-10 codes: S222-S225; S321-S325; S420-S429; S500-S529; S620-S628; S720-S729; S820-S829; S920-S920.The register has almost 100% completeness and high precision concerning fracture diagnoses. 25,26Since many previous studies have focused on hip fractures, we conducted a supplementary analysis with hip fractures as the outcome (ICD-10 codes S720-S724; S729).In the main analyses, we only included the first occurrence of an outcome, since recurrence of falls or fractures is not independent.We repeated the analyses with multiple falls as the outcome in a supplementary analysis, where we regarded admissions within 30 days of each other as the same event.

| Covariates
From the Danish National Patient Registry and the Danish National Prescription Registry, we selected several morbidities and medication types present in the year preceding inclusion in the analyses to characterize the source population (see Table S2 for ICD and ATC codes).
From the Danish Civil Registration System, we obtained information on sex and birth date.Data on the highest achieved educational level was obtained from the Population's Education register and was categorized into four groups (low: primary school; middle: high school, vocational education (education aimed toward manual work), and high: higher education, and higher advanced education and missing (patients born before 1920, older immigrants, or individuals still receiving education)).

| Statistical analyses
The distribution of covariates in relation to the use of the exposure drugs was examined using descriptive statistics.We used a Poisson regression model conditioned on the individual to calculate the incidence of falls and fractures for the six predefined periods (cf. the section "study design" above) and to derive incidence rate ratio (IRR)s with 95% confidence intervals (CI), using the baseline period as a reference.This analysis only included participants who experienced the outcome of interest during one of the predefined periods.The design of the self-controlled case series eliminates time-invariant confounders but is still susceptible to confounding by timevarying factors, such as age, which we adjusted for. 17ndividuals who died were censored at the time of death.A total of 12,064 (19.4%) of the persons who experience a fall (n = 62,105) died during the study period, and we assumed that competing mortality risk might be present because early death would necessarily preclude the later occurrence of falls and fractures. 27Analyses were repeated both using any sedative-hypnotic medication or using the specific drug classes as described in the exposure section.We explored the potential impact of competing mortality in a sensitivity analysis, where we additionally included data on fatal fall accidents obtained from the Danish Cause of Death Registry (ICD-10 codes W00-W19) in the outcome measure falls.Dementia, frailty, and the use of antipsychotics and antidepressants have been associated with the risk of falls or fractures.Therefore, we performed analyses stratified for prior hospital diagnosis or concurrent use of these drugs defined by at least one prescription within the year before treatment with benzodiazepines, Z-drugs, or melatonin was initiated.As the etiology of falls and fractures varies with age, all analyses were stratified by age (15-39, 40-69, ≥70 years).As analyses showed only slightly higher estimates in ≥80 years-olds than in 70-to 79-yearold, these categories were combined (Table S3).Furthermore, as several previous studies found that the risk of falls varies by sex, 28 we chose to also stratify analyses on sex.Analyses were repeated where individuals who discontinued were censored 90 days after the last prescription.This yielded similar results as the main analysis without censoring.The senior author wrote the statistical analysis plan and conducted the statistical analysis.

| RESULTS
Out of 3,576,611 adults, we identified 699,335 (19.6%) users of benzodiazepines, Z-drugs, or melatonin in the period 2003-2016.Of these, 53.6% were women, 50.2% used Z-drugs, 44.4% used benzodiazepines, and 5.4% used melatonin.At the study start (January 1st, 2003), users were older than the general population and had more comorbidities, with heart disease and stroke being the most common.Antidepressants were the most common co-medication (9.9%) followed by antipsychotics (3.4%) (Table 1).The median age of initiation of a sedating drug was 53 years but was higher for those who experienced a fall (59 years) or a fracture (61 years) (Tables S4 and S5).In total, 62,105 (8.7%) of the study population experienced a fall accident, while 36,808 (5.2%) had a fracture.Incidence rates are shown in Table S6.At the initiation of a sedating drug, the prevalence of comorbidity and comedication was higher than at baseline (Table 1 and Tables S7 and S8).
When examining risks of falls stratified for age and sex, we found similar patterns in men and women, but with slightly higher risk estimates for men (Figure 1 and Table S9).The incidence rate ratio (IRR) of falls was higher among older adults (≥70 years) compared to younger men and women (<40 years).For older adults, compared to the baseline (>1 year before initiation), the risk of falls was highest during the 3 months before initiation (pre-treatment period, IRR men + 7 , 4.22 (95% CI, 3.53-5.05),IRR women + 70 , 3.03 (95% CI, 2.59-3.55)),followed by the 3 months after treatment initiation (treatment period, IRR men + 70 , 2.96 (95% CI, 2.41-3.65),IRR women + 70 , 1.72 (95% CI,1.43-2.06)).Similar was seen when the pre-treatment period was used as a reference (Table S10).Sensitivity analyses reducing the pretreatment period to 1 or 2 months showed that the incidence of falls was the highest closest to treatment initiation (Table S1).The risk continued to be slightly higher in the later treatment periods.Contrarily, in adults aged 40-69 years, the risk was only higher in the 3-month pre-treatment period.The incidence of falls among young men and women seemed even slightly lower after initiation of sedating medication (treatment period, IRR men15-39 , 0.66 (95% CI, 0.50-0.86),IRR women15-39 , 0.65 (95% CI, 0.51-0.83))(Figure 1 and Table S8).When examining fractures, the IRR of fractures was similarly highest in T   older men and women during the pre-treatment period and decreased during treatment (Figure 2 and Table S11).The IRRs were in general higher than for falls.In these analyses, the risk was also slightly higher in adults aged 40-69 years, but not in young adults.
For both falls and fractures, analysis of the individual medication types in adults aged +70 years showed the strongest association in the pre-treatment period for all drugs.The only exception was alprazolam, which was not associated with fractures (Figure 3 and Tables S9 and S11).Estimates for the younger individuals were lower, but with the highest risk estimates in the pre-treatment period (Tables S8 and S9).The subgroup analyses showed a similar pattern of risk in the subgroups with and without dementia, antipsychotic or antidepressant medication, but tended toward a slightly higher risk of fractures in patients with dementia (Figure 4, Tables S9 and S11).One hundred and two persons experienced a fall on the same day they redeemed hypnotic-sedative medication.Exclusion of these persons did not materially change the results.
The results of the supplementary analyses for multiple falls as the outcome were not markedly different from the analyses with the first occurrence of a fall accident as the outcome (Table S12).Moreover, 11,869 (1.7%) fractures were registered as hip fractures.Analyses using this outcome separately showed a similar pattern of associations to the analyses for fractures, yet IRRs tended to be even higher (Table S13).The results from the sensitivity analysis-to explore any impact of competing mortalityin which both fatal and non-fatal falls were included as the outcome were generally consistent with the main analysis (Table S14).

| DISCUSSION
In this nationwide, population-based self-controlled case series, we found that the use of benzodiazepines, Z-drugs, and melatonin, respectively, were associated with a higher risk of falls and fractures, including hip   F I G U R E 3 Risk of falls (A) and fractures (B) in different study periods by drug types in first-time users ≥70 years of age of benzodiazepines, z-drugs, or melatonin with either a fall or a fracture.*Total number of registered falls or fractures in the period.BZ = benzodiazepines.Chlordiazepoxide was not included in the figure due to lack of power and consequently very unprecise estimates (see Tables S7 and S8).
fractures, compared with a non-treatment period from 24 to 12 months before initiation of the medication.The incidence of falls and fractures was highest in the 3 months before the initiation of the medications.The higher risk in the pre-treatment period was most prominent among older men, whereas there was no association in adults below 40 years.][8][9] Most studies showed higher risks of falls and fractures in persons using these medications.Nearly all previous studies have used a traditional cohort or case-control design, which is vulnerable to confounding and selection bias.A few studies have used a self-controlled design similar to the present study, but these studies focused on other psychotropic drugs and/or outcomes. 15,18,21In agreement with our findings, a case-series analysis from the United Kingdom, concerning benzodiazepines and Z-drugs, showed that the risk of traffic accidents for persons using this medication was highest during the 4 weeks up to the date of the prescription.This increased risk persisted for diazepam and temazepam during the 4 weeks following the first prescription.Similarly, a more recent case-series study from Taiwan demonstrated that the risk of falls and fractures in adults aged 65 years and above using antipsychotics, which have sedative properties, was highest in the period before treatment compared to a baseline period. 15This might imply, that factors beyond medication use affect the risk of accidents.Studies exploring the influence of melatonin on the risk of falls and fractures are scarce.In a randomized clinical trial among 81 postmenopausal women, 1-year treatment with melatonin did not affect postural balance or risk of falls compared to placebo. 29Notably, in the main analyses of this study use of Z-drugs, benzodiazepines, and melatonin were combined.However, only 5.4% of those using sedative drugs, used melatonin.In separate analyses (Figure 3), the pattern of falls and fractures did not differ for melatonin users.
The higher risk in the 3 months before treatment and the relatively lower risk after initiation, can be due to the indication for which the medication was prescribed.The only indication for Z-drugs in Denmark is insomnia, Early post-treatment (3-12  while insomnia and jetlag are indications for melatonin use.The indication for benzodiazepines is anxiety or sleep disorders. 30Thus, presumably most prescriptions are due to insomnia or anxiety.We were able to adjust for treatment with antidepressants as a proxy for anxiety or depression, but we had no information on insomnia.A large number of studies have shown that poor sleep, including objective 31,32 and subjective [33][34][35][36][37][38] measures of short sleep duration and many awakenings per night, is associated with fall accidents.Insomnia is the most common sleep disorder and defined as persistent difficulty initiating or maintaining sleep despite adequate opportunities and circumstances. 39Poor sleep may cause fall accidents because of slower gait speed and poorer cognitive performance during a dual-task test.Sleep disturbances may be linked with fractures either indirectly through falls, or directly through altered bone turnover and hypoxia. 40It is thus conceivable that the increased risk of falls before initiating benzodiazepines, Z-drugs, or melatonin is due to untreated insomnia, and that initiation of treatment gradually diminishes this risk.
Benzodiazepines and Z-drugs are among the most prescribed medications globally.Research has extensively implicated benzodiazepines and Z-drugs, particularly with falls and fractures. 5,8A general limitation of the studies demonstrating this risk is the possibility of confounding by indication: the underlying condition for which sedative-hypnotic medication is prescribed may render the individual more susceptible to fallsindependent of the effects of medication.In the present study, we minimized this source of confounding, by assessing the risk shortly before and after treatment within each individual, thereby assuming that the underlying condition had not materially changed.Our findings implicate that the increased risk of falls and fractures observed in patients during treatment with sedative-hypnotic medication are better explained by factors that necessitated the prescription of this medication rather than its adverse effects.The higher risk during the treatment periods, compared with the baseline period, could be due to residual symptoms of the condition which led to the prescription of hypnoticsedative medication, combined with its adverse effects.This may imply that the patient's condition has not improved, despite receiving treatment.
An advantage of this study is the large sample size enabling analyses stratified for sex and age.Moreover, as we employed nationwide population-based registers in a country with free access to health care, we had access to data from a large, unselected population.The Danish person identification numbers allowed us to link individual health data with different registries and, thus, obtain complete follow-up information on falls and fractures from hospital registers.The large sample size guaranteed a high number of users for all the specific medication types.Since hypnotic-sedative medication, including melatonin, can only be obtained by prescription in Denmark, our study has nearly complete coverage of users of hypnotic-sedative medication.Purchase without prescription and/or illicit use of hypnotic-sedative medication was assumedly negligible.
A number of limitations of our study should be acknowledged.First, as this concerns an observational study, no conclusions regarding the causality of the observed associations can be drawn.Secondly, data on physical performance the (severity of) neurocognitive disorders, and symptoms were not available, which might have acted as time-variant confounders by indication.For example, patients with symptoms of neurocognitive disorders might manifest agitation, and insomnia that could lead to the prescription of sedative-hypnotic medication and both the symptoms as well the treatments could influence the risk of subsequent falls and fractures.This confounding effect is especially likely when the events occur shortly before the diagnosis and treatment of the disorder.
Thirdly, the Danish National Prescription Registry does not include medication dispensed at hospital departments, yet this concerns less than 1% of all prescribed medication.Moreover, we did not explore how long patients continued treatment after the first prescription.Fourthly, the outcomes only included falls diagnosed at a hospital and thus represent the most severe events.Finally, when studying outcomes in older persons, among whom mortality is higher, mortality should be considered as a competing risk since mortality precludes the later occurrence of the event of interest.Although an ancillary analysis including fatal falls did not materially impact the IRR, we cannot preclude that a larger proportion of individuals would have experienced a fall in the posttreatment period, if they had survived.
In conclusion, we found that the risk of falls and fractures was highest among individuals in the middle-aged and old age group in the period just before initiation of treatment with benzodiazepines, z-drugs, or melatonin.Although we do not intend to trivialize the potentially harmful effects of sedative-hypnotic medication in terms of falls and fractures, our results emphasize the need to take other factors beyond sedative-hypnotic use into account, when evaluating a patient's risk for falls.

8 F
I G U R E 1 Risk of falls in different study periods in men and women in 62,105 first-time users of benzodiazepines, z-drugs, or melatonin who experienced a fall.*Total number of registered falls in the period.
Distribution of the study population characteristics at the start of benzodiazepines, Z-drugs, or melatonin (BZM) use in Danish adults in the period 2003-2016.
T A B L E 1 months after) F I G U R E 4 Risk of falls (A) and fractures (B) in different study periods by co-morbidity or co-medications in first-time users ≥70 years of age of benzodiazepines, z-drugs, or melatonin with either a fall or a fracture.*Total number of registered falls or fractures in the period.