INTRODUCTION

As many as 10% of US adults meet diagnostic criteria for insomnia1,2,3 and about one-third have difficulty falling or staying asleep.4,5 Persistent insomnia is associated with increased suicide risk and is a modifiable risk factor.6,7,8 Agents approved by the Food and Drug Administration (FDA) for the treatment of insomnia include benzodiazepines, benzodiazepine receptor agonists, ramelteon, and suvorexant. Off-label medications include antihistamines and trazodone, a serotonin reuptake inhibitor/agonist. The 2017 American Academy of Sleep Medicine (AASM) guidelines for the pharmacological treatment of chronic insomnia in adults recognized that trazodone and other off-label medications are commonly prescribed despite poor evidence. Similarly, in the Department of Veterans Health Affairs (VA), the national formulary identifies zolpidem as the preferred agent for insomnia, yet high fill volumes of inexpensive sedating antihistamines and older antidepressants at sub-therapeutic doses for depression (i.e., trazodone = < 200 mg) suggest routine off-label use for insomnia.9

Some insomnia agents carry warning labels for risk of adverse events of suicide risk (Table 1).9 Yet, little is known about the comparative safety of these agents. The literature suggests a possible association between insomnia medications and suicidal ideation and behavior. Among 2245 elderly suicide decedents in Europe, toxicology panels identified benzodiazepines as the means of suicide in about half of poisoning deaths.11 In the National Comorbidity Survey Replication (N = 5692), sedative-hypnotic use was significantly associated with self-reported suicide attempt12 after adjusting for insomnia, mood and anxiety disorders, general health, and demographics. In a smaller (N = 525), cross-sectional community study, self-reported hypnotic use was associated with suicidal ideation after controlling for depression and insomnia.13 In prospective studies with different follow-up periods, medications for the treatment of insomnia were associated with suicide and/or suicide attempts.14,15,16 Sedating antidepressants and antihistamines are used to induce or maintain sleep, although they are not recommended in current clinical practice guidelines.17,18,19 Finally, a 2017 review of the studies of hypnotic medication exposures and suicide found that none of the studies had adequately controlled for depression or other psychiatric conditions associated with suicide.20

Table 1 Medications with ≥ 1% of Total VA Insomnia Medication Utilization in VA in 2011–2012
Full size table

The purpose of this study is to compare the risk of suicide attempt associated with medications commonly prescribed to treat insomnia in the VA over a 2-year period, controlling for days’ supply of medication exposure, and history of psychiatric disorders, comorbidities including pain and substance or alcohol abuse, and use of analgesics and central nervous system (CNS) medications.

METHODS

Study Population

Subjects and Data Sources

This cohort consists of patients aged 18 and older receiving care in the Veterans Health Administration (VHA) who filled a first prescription for an insomnia medication in 2011 or 2012. Subjects were defined as active patients if they received at least one prescription medication for an indication other than insomnia or had an inpatient or outpatient visit in the 12 months prior to the index insomnia prescription. Subjects were excluded if they had a possible or confirmed suicide attempt in the 12 months prior to their first insomnia medication prescription. Also excluded were subjects in hospice or skilled nursing facility care or with more than 30 inpatient days at any time during the study period. The observation period was the 12 months after the first insomnia prescription fill date (index date) or until death or suicide attempt, whichever came first. Data were drawn from the VA Corporate Data Warehouse (CDW).21 The CDW includes prescription fill data and electronic medical records linked by unique patient ID numbers.

Measures

Study Endpoint

The endpoint was suicide attempt as recorded in emergency department encounter records or inpatient discharge summaries in the 12 months of first use of any monotherapy for sleep (ICD-9 external cause of injury codes (E950-E959, E980.6, E980.8, E981-E984, E-988)).22

Insomnia Medication Exposure

We reviewed medication labelling, clinical guidelines, and VA formularies (2011–2012) to identify medications used for insomnia, dosing and administration instructions, and any labeled warning for suicidal ideation or behavior risk (Table 1). To ensure adequate sample size to detect the rare event of suicide attempt, we included only agents comprising at least 1% of the fills for sleep medications in 2011 or 2012. These included low-dose (< 200 mg) trazodone, zolpidem, two first-generation antihistamines (hydroxyzine, diphenhydramine), and three benzodiazepines (lorazepam, diazepam, temazepam) (Table 1). Antihistamine fills were included only if administration instructions specified use at bedtime or for sleep or insomnia. The index date for insomnia monotherapy was defined as the date of first fill in 2011 or 2012 following 12 months of no fills for any insomnia medications. Days’ supply for the 12-month period following the first fill were categorized as follows: 2–30 days, 31–120 days, and more than 120 days. Fills of 1-day supplies were excluded as these are indicated for procedures (e.g., colonoscopy). We excluded subjects with polypharmacy regimens for insomnia and those who changed sleep monotherapies (i.e., cross-over regimens) during the study period. Only medications filled by VA pharmacies and received by patients were included.

Covariates

The 12-month period before the index prescription date was used to identify potential confounding variables including Gagne comorbidity score, inpatient days, and comorbidities including major depression, post-traumatic stress disorder (PTSD), bipolar disorder, schizophrenia, other psychoses, other anxiety, alcohol or substance use disorder (SUD), personality disorders, sleep disorder, chronic pain, and headache. The presence of a comorbidity was defined by any of the following: two or more outpatient service dates for the same diagnosis or one inpatient diagnosis. To control for differential risks of suicide in the 12-month observation period based on subject history, we included comorbid psychiatric conditions in our propensity score models: major depressive disorder, bipolar affective disorder, schizophrenia, other psychosis, PTSD, anxiety disorders, and personality disorders. We also included history of fills for analgesic and CNS drug classes (opioids, non-opioid analgesics, antidepressants, antipsychotics, anticonvulsants, anti-Parkinson’s and other CNS agents). Additional covariates included demographics (age, sex, race, Hispanic ethnicity). Age was categorized as follows: 18–49, 50–60, and over 60 because insomnia treatment selection is confounded with age. Benzodiazepines and sedating antihistamines are contraindicated in older adults.10 Risk of suicide also varies by age and is non-linear. To account for the degree of exposure to insomnia monotherapy, we categorized days’ supply as follows: 2–30 or 31–120 or 121–365 days based on the distribution of days’ supply and guidelines for treatment.

Data Analysis

Propensity score matching (with a greedy matching algorithm with a caliper width set at 0.2 of the standard deviation of the logit of the propensity score) was used to control for potential confounding.23 We estimated the same model to determine the propensity scores for each drug-drug comparison. Only one set of population baseline characteristics was used for all comparisons (Tables 3,4, and 5). Logistic regression was used to estimate the propensity score (i.e., the probability of receiving zolpidem), conditional on the aforementioned covariates. Comparison subjects for trazodone, the benzodiazepines, and the antihistamines were identified by performing a 1:1 match separately with zolpidem on propensity score and age category. The standardized difference of less than 10% was used to examine the balance of the baseline covariates between subjects who filled zolpidem prescriptions and comparator groups in the matched samples.24 Cox proportional hazards regression models were used to estimate hazard ratios and 95% confidence intervals (CI) on the matched data. Covariates included categorical variables for days’ supply: 31–120 days and 121–365 days.

Sensitivity Analyses

To ensure that our models were unaffected by somewhat different populations in each model, we repeated the Cox proportional hazard models using only subjects who filled index prescriptions for zolpidem and also matched to every comparator (antihistamines, benzodiazepines, trazodone) (n = 70,525). We then included only the antihistamine, benzodiazepine, and trazodone patients who matched those 70,525 zolpidem patients for a total sample of 141,050 for each of the three models. This approximates the multinomial propensity score model. Covariates were those used in the original models and were identical across all three models. The hazard ratios and confidence intervals were similar to those reported in our primary analyses.

This study was approved by the Syracuse VA Institutional Review Board.

RESULTS

A total of 1,742,860 subjects filled at least one prescription for an agent routinely used to treat insomnia in VHA in 2011 and 2012. Of those, 348,449 met the study criteria. The most common insomnia medication fills were for the antidepressant trazodone, followed by benzodiazepines, antihistamines, and zolpidem (Table 1). Subjects were typically white men in their late 50s or early 60s (Table 2). About 1 in 4 had a VA disability rating of 50% or more. Hypertension was the most common comorbidity (35.6–39.8%) followed by arthritis (20.6–22.8%), back pain (16.5–19.4%), depression (14.8–23.5%), and PTSD (11.4–20.2%).

Table 2 Characteristics of Veterans in VHA Who Filled a New Insomnia Medication Prescription in 2011–2012
Full size table

Propensity score modeling matched 76,247 patients with trazodone (< 200 mg), 72,641 patients with benzodiazepines, and 74,229 patients with antihistamines prescribed for insomnia (Table 3). The matched cohorts were well-balanced by demographics, disability rating, inpatient days, comorbidity burden, history of psychiatric, cardiovascular, oncologic and pain conditions, and prescription fills for pain and CNS medications in the 12 months prior to the first insomnia prescription (Tables 3, 4, and 5). Standardized differences were all below 0.10. Cox proportional hazards regressions on the propensity score-matched samples with days’ supply as covariates identified the risk of suicide attempt as 61% higher for trazodone than for zolpidem (HR = 1.61, 95% CI 1.07–2.43) (Table 6). Interaction terms (days’ supply by trazodone) were insignificant. The risk of suicide attempt for any insomnia medication days’ supply beyond 120 days (compared to 30 days or less) was 70% lower than for days’ supplies of 30 days or less (HR = 0.30, 95% CI 0.17–0.52). The model was robust to the assumption that suicide attempt risk was constant over time.

Table 3 Selected Patient Characteristics of Propensity Score-Matched Samples of Zolpidem and Trazodone
Full size table
Table 4 Selected Characteristics of Matched Sample of Zolpidem and Benzodiazepines
Full size table
Table 5 Selected Characteristics of Matched Sample of Zolpidem and Antihistamines
Full size table

The same Cox regression models for benzodiazepines (HR = 1.31, 95% CI 0.85–2.08) and sedating antihistamines (HR = 1.37, 95% CI 0.90–2.07), respectively, identified no difference in suicide attempt risk associated with either class compared to zolpidem (Table 6). In both models, days’ supply beyond 120 days compared to days’ supply of 30 days or less was associated with lower risk of suicide attempt (benzodiazepines or zolpidem days’ supply > 121 days HR = 0.42, 95% CI 0.23–0.78; and sedating antihistamines or zolpidem days’ supply > 121 days, HR = 0.53, 95% CI 0.29–0.94). Yet, in both models, interaction terms for days’ supply and medication (benzodiazepines or zolpidem, sedating antihistamines for sleep or zolpidem) were insignificant, indicating that risk of suicide attempt was significantly lower for subjects with insomnia monotherapy days’ supply of more than 120 days compared to 30 or fewer days’ supply, regardless of the specific monotherapy. The Cox regression model assumption of constant suicide rates over time was robust in both models.

Table 6 Adjusted Incidence of Suicide Attempts by Index Prescription: Zolpidem Versus Trazodone or Benzodiazepines or Antihistamines
Full size table

DISCUSSION

Compared to zolpidem, trazodone (< 200 mg) was associated with clinically and statistically significant increased risk of suicide attempt on average in propensity-matched samples over time while controlling for history of inpatient days, psychiatric and substance or alcohol use history, comorbidities, pain, central nervous system medications, and demographics. We found no differences in risk of suicide attempt for benzodiazepines or sedating antihistamines prescribed for insomnia compared to zolpidem, respectively. For each insomnia monotherapy, days’ supply of 30 days or less was associated with significantly higher risk of suicide attempt than days’ supply of 120 days or more. This effect was independent of the specific insomnia monotherapy.

To the best of our knowledge, this study provides the first comparative safety evidence to support the 2017 clinical practice guideline17 for the pharmacologic treatment of insomnia from the AASM which reject the use of trazodone for insomnia treatment. These findings extend the results of the meta-analysis on which the guideline was based through the inclusion of frail, complex patients routinely excluded from the randomized trials that comprised the meta-analysis.

The finding that suicide risk was significantly lower for days’ supply of greater than 120 of 365 days across all monotherapies may be due to several reasons. Refills of insomnia medications typically require follow-up and continued low risk of misuse. A typical prescription of zolpidem and benzodiazepines for sleep are both for 30 days or less. Response to treatment may lag by several days or weeks. Similarly, extended days’ supply of newly prescribed monotherapies for insomnia are likely associated with frequent follow-up and treatment response. That is, insomnia monotherapy days’ supply beyond 120 of 365 days is likely associated with treatment adherence, which is associated with improved outcomes.

The findings of this study are an important contribution because of the widespread use of both off-label trazodone for the management of insomnia and the chronic use of insomnia medications indicated for only short-term use.17,25,26 Trazodone is the most commonly prescribed medication for insomnia27; its use for insomnia eclipses its use as an antidepressant. In our sample, the typical patient prescribed low-dose (< 200 mg) trazodone picked up more than six 30-day supplies during the 12-month observation period. Similarly, guidance for the use of zolpidem is that any sleep disturbance not resolved within 7 or 10 days be reevaluated for medical or psychiatric illness.17,26 Yet, the typical patient with newly prescribed zolpidem picked up more than seven 30-day supplies over 12 months. Addressing questions of the appropriateness of extended prescribing are beyond the scope of this study, yet our findings suggest that suicide attempt risk is comparatively lower for patients who are successfully maintained on insomnia monotherapies beyond 120 days compared to those with supplies of 30 days or less, regardless of the specific medication.

Suicide attempt is a strong predictor of suicide28 suggesting that avoiding use of low-dose trazodone for the treatment of insomnia might delay or deter progression to suicide. It is not clear, however, whether untreated insomnia presents a greater risk of suicidal thought and behavior than treatment with medications on a short- or long-term basis. Although not significantly associated with suicide attempt risk compared to zolpidem, benzodiazepines and sedating antihistamines are contraindicated in older adults.10 The study subjects’ older ages resemble that of the typical VA patient as well as 65% of Veterans who are suicide decedents (ages 50 and over).

The most recent clinical guidelines and evidence reviews strongly support non-pharmacologic interventions as first-line treatments for insomnia.17,29,30 A recent clinical practice guideline by the American College of Physicians has unequivocally made cognitive-behavioral therapy for insomnia (CBT-I) the only first-line treatment for insomnia; pharmacotherapy is a suggested second-line treatment after failure or inadequate response to behavioral treatment.30 That recommendation supports an earlier consensus statement from the British Association for Psychopharmacology in 2010.29 Direct comparisons between pharmacotherapy and non-pharmacologic interventions were beyond the scope of this study.

Limitations

The diagnosis of PTSD was well-balanced in our propensity score models. Yet, the cohorts could not be balanced by nightmares and trazodone is first-line therapy for nightmares associated with PTSD. There is the possibility that the PTSD patients taking trazodone were more likely to be experiencing nightmares before the first prescription than those taking zolpidem. Similarly, healthcare that is supportive for suicidal patients but unmeasured in electronic medical records may include social support, active listening, and motivational interviewing. Yet, these limitations apply to any secondary data analysis.

Sedating antihistamines may be purchased without a prescription; however, all of the fills observed in this study were provided by VA pharmacies and received by patients. It is possible that some patients may have been using non-prescription insomnia medications at the same time as their VA-provided monotherapy against provider recommendations. Yet, this limitation is true of all secondary data analyses using VA prescription fill data.

At least one VA regional guideline recommended low-dose trazodone for the treatment of insomnia during the study period.31 Yet, it is possible that these low doses represent suboptimal prescribing for depression. This is unlikely because trazodone is not a first-line therapy or depression and the VA pharmacy has access to the electronic medical record and is charged with checking prescriptions and addressing suboptimal doses prior to filling prescriptions.

After the period of observation of this study, zolpidem underwent a dose form change. The maximum dose was reduced from 10 to 5 mg for women after evidence emerged of auto crashes and other adverse events related to extended sedation. Yet, the effect of dosing on the comparative safety of zolpidem and suicide attempts is beyond the scope of this study.

Lapses in medication exposure are difficult to characterize as insomnia medications are typically prescribed “PRN” (as needed) such that 30 doses may cover more than 30 days. (For example, a 30-day supply of zolpidem in VA may contain only 20 doses.) Identifying the unique administration instructions for each fill was beyond the scope of our study. Yet, a strength of our data are that they include only medications filled by VA pharmacy and received by the patient.

Only subjects with a new fill of only one insomnia monotherapy (with no additional agents or cross-overs) were included. This approach isolates the comparative effects of monotherapies and is consistent with the drug safety literature, but a necessary limitation is that it does not allow comparisons of polypharmacy combinations or medication switching.

Finally, as with most observational studies of drug safety, a variety of other exposures and therapies may affect suicide attempt risk that are not controlled in this study such as loss of a job or relationship, cognitive behavioral therapy for insomnia, and exposure to any number of other pharmacotherapies possibly associated with suicide risk.

CONCLUSIONS

Nearly 350,000 Veterans started insomnia monotherapy in VA in 2011–2012, with suicide attempt rates reported at VA emergency departments and hospitals in excess of 50 attempts per 100,000 person-years within 12 months of the first fill. Compared to zolpidem, the incidence of suicide attempts was significantly higher among those with first fills of trazodone at doses indicated off-label for insomnia. These findings are among the first in support of the 2017 AASM clinical guidelines discouraging the use of trazodone for insomnia. To the best of our knowledge, this study is the first to describe the comparative safety of insomnia monotherapies and suicide attempts.