https://orcid.org/0000-0001-9512-7407
Figures
Abstract
Background
Long-term benzodiazepine use has been associated with increased risk of morbidity and mortality. Preventing long-term use through safer prescribing practices has received little attention to date. We sought to better understand associations between initial prescription characteristics and duration of benzodiazepine use.
Methods and findings
This was a retrospective population-based cohort study of 1,820,808 adults in Ontario with incident benzodiazepine prescriptions between January 1, 2013 and December 31, 2020, with follow-up to December 31, 2021. The primary exposure was duration of the index prescription (≤7 days—referent group, 8–14 days, 15–30 days, or >30 days). Secondary exposures were: (a) duration of action of index benzodiazepine(s) prescription (short-acting, long-acting or both); (b) number of benzodiazepine dispensed on index (1 or 2+); and (c) mean daily dose of the index prescription in Diazepam Milligram Equivalents (DMEs). The primary outcome was time to benzodiazepine discontinuation in days. Multivariable models were adjusted for age, sex, anxiety, insomnia, and substance use disorders as well as other important comorbidities and socio-demographic characteristics. The median age at index was 53 years (Interquartile Range (IQR) 38–67), and 62.6% were women. The median time to discontinuation in women was 16 days (IQR: 6–29) while the median time to discontinuation in men was 19 days (IQR: 6–29). Lorazepam was the most commonly prescribed benzodiazepine on index (63.9%), followed by clonazepam (17.3%) and diazepam (5.8%). In multivariable Cox Proportional Hazards Models, longer index prescriptions were associated with a lower likelihood of benzodiazepine discontinuation (adjusted Hazard Ratio (aHR) 0.54 (95% Confidence Interval (CI) [0.54,0.54]) for 8–14 days; aHR 0.26 (95% CI [0.25,0.26] for 15–30 days and aHR 0.14 (95% CI [0.14,0.14]) for >30 days, compared to ≤7 days, respectively). Being prescribed two or more benzodiazepines versus 1 was also associated with a reduced likelihood of discontinuation (aHR 0.59 (95% CI [0.57,0.61])), as was being prescribed long-acting benzodiazepines (aHR 0.80 (95% CI [0.80,0.80])) or a combination of short and long acting benzodiazepine (aHR 0.84 (95% CI [0.80,0.88])) versus short-acting benzodiazepines alone. Mean daily doses of >5 to ≤10 DME and >10 to ≤20 DME were associated with an increased likelihood of discontinuation (aHR 1.03 (95% CI [1.03,1.03]); aHR: 1.03 (95% CI [1.03,1.04])), whereas doses >20 DME were associated with a reduced likelihood of discontinuation (aHR 0.98 (95% CI [0.97,0.98])) compared with ≤5 DME. Findings may be subject to bias from unmeasured confounding.
Conclusion
This large population-based cohort study found that prescribing shorter courses of benzodiazepines, use of a single benzodiazepine, use of a short-acting agent, were associated with reduced likelihood of long-term benzodiazepine use. Findings suggest that simple changes to prescribing practices could reduce prolonged benzodiazepine use and the morbidity and mortality associated with long-term use of these medications.
Author summary
Why was this study done?
- Benzodiazepines are a class of commonly used anti-anxiety medication that are disproportionately prescribed to women.
- Long-term use of benzodiazepines is generally not recommended because of known harms associated with their use, including addiction, memory problems, increased risk of falls, overdose, and death.
- Previous research suggested that some characteristics of initial prescriptions were associated with prolonged use; however, results were not always consistent, and it was unclear if these factors were the same in men compared with women.
- We therefore sought to better understand the association between initial prescription characteristics and duration of benzodiazepine use, and whether that relationship differed by sex.
What did the researchers do and find?
- This was a retrospective population-based cohort study of 1,820,808 adults in Ontario with new benzodiazepine prescriptions between January 1, 2013 and December 31, 2020 and follow-up to December 31, 2021.
- We examined the association between initial prescription characteristics including duration of the prescription, duration of action of the prescribed benzodiazepine(s) (short-acting, long-acting, or both), number of benzodiazepine dispensed (1 or 2+) and the mean daily dose of the prescription in Diazepam Milligram Equivalents (DMEs), and time to benzodiazepine discontinuation.
- Across multiple analyses, we found that prescribing shorter courses of benzodiazepines, use of a single benzodiazepine and use of a short-acting agent, were associated with reduced likelihood of long-term benzodiazepine use, and these results were consistent across sex.
What do these findings mean?
- Findings suggest that simple changes to prescribing practices including limiting initial prescriptions to <7 days and choosing a single, short-acting agent could reduce prolonged benzodiazepine use and associated morbidity and mortality in both men and women.
- Results may be limited by factors that were not measured in our datasets or considered in our analyses.
Citation: Bozinoff N, Hauck TS, Kleinman RA, Sloan ME, Sproule BA, Vigod SN, et al. (2026) Association between initial benzodiazepine prescribing patterns and time to benzodiazepine discontinuation: A population-based retrospective cohort study. PLoS Med 23(6): e1005126. https://doi.org/10.1371/journal.pmed.1005126
Academic Editor: Louisa Degenhardt, University of New South Wales, AUSTRALIA
Received: September 11, 2025; Accepted: May 19, 2026; Published: June 18, 2026
Copyright: © 2026 Bozinoff et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The dataset from this study is held securely in coded form at the Institute for Clinical Evaluative Sciences (ICES). While data sharing agreements prohibit ICES from publicly releasing a minimal deidentified dataset, access can be granted to those who meet pre-specified criteria for confidential access through the Data & Analytics Service (DAS). More information on how to access this data is available at https://www.ices.on.ca/DAS.
Funding: This study was funded by a Womenmind Grant to co-principal investigators NB and TG https://www.camh.ca/en/get-involved/join-the-cause/womenmind. The Funders did not play any role in the study design, data collection, analysis, decision to publish or preparation of the manuscript.
Competing interests: I have read the journal’s policy and the authors of this manuscript have the following competing interests: N.B. reports receiving consulting fees from the Ontario College of Family Physicians (for development of continuing professional development materials) and Mentoring, Education, and Clinical Tools for Addiction - Partners in Health Integration (for guideline review) unrelated to this work. She is supported by a New Investigator Award from the Department of Family and Community Medicine at the University of Toronto. M.E.S. has received research support from the Canadian Institutes of Health Research, the Ontario Ministry of Health and Longterm Care, the CAMH Foundation, and the Academic Health Sciences Centre AFP Innovation Fund. M.E.S. is also supported in part by an Academic Scholar Award from the Department of Psychiatry, University of Toronto. M.E.S. has received travel support from the Society of Biological Psychiatry and the International Society for CNS Clinical Trials and Methodology. He has received an honorarium for giving a talk at the McGill University Health Centre. R.A.K has received support from the Canadian Institutes of Health Research, the CAMH Foundation, and the Academic Health Sciences Centre AFP Innovation Fund. R.A.K. is also supported in part by an Academic Scholar Award from the Department of Psychiatry and reports holding stock in Pfizer and Novo Nordisk.
Abbreviations: ACG, Adjusted Clinical Groups; ADGs, Aggregated Diagnosis Groups; aHR, adjusted Hazard Ratio; CI, Confidence Interval; DAD, Discharge Abstract Database; DMEs, Diazepam Milligram Equivalents; IQR, Interquartile Range; NACRS, National Ambulatory Care Reporting System; NMS, Narcotic Monitoring System; OAT, opioid agonist therapy; OHIP, Ontario Health Insurance Plan; RPDB, Registered Persons Database; STROBE, Strengthening the Reporting of Observational Studies in Epidemiology
Introduction
Benzodiazepines are commonly prescribed medications—in 2023, over 23 million individuals in the United States reported use of benzodiazepines [1]. Benzodiazepines have a number of therapeutic uses including in the acute management of seizures, alcohol withdrawal and panic, however, chronic use has been implicated in negative health outcomes including increased risk of mortality, falls, cognitive impairment, overdose (particularly when combined with opioids and other sedating agents), motor vehicle collisions, and substance use disorder [2–4]. Women are particularly affected—across multiple countries, women are prescribed benzodiazepines at 1.5–2 times the rate with which they are prescribed to men [5–10]. In this context, efforts to reduce long-term benzodiazepine use have emerged, and include motivational interventions and tapering guidelines/protocols [11–13]; however, in 2018 approximately 50% of individuals prescribed benzodiazepines in the US received them for 2 months or longer [1], and as many as 15% of individuals exposed to benzodiazepines remain on these medications in the long-term [14]. Prevention of long-term use via safer prescribing practices has received relatively little attention to date, and it remains unclear if these safer prescribing factors are the same in men compared with women [14]. Past work has demonstrated an association between ‘heavy’ use during the initiation period (higher frequency, daily use, etc.) and prolonged use [15–17]. A larger number of tablets dispensed initially [18], a longer initial prescription duration in special populations (older adults, individuals with depression) [19,20], and concurrent use of multiple benzodiazepines [21,22] have all been associated with long-term use. Studies examining the association between the type of benzodiazepine prescribed at initiation (short versus long-acting) and long-term use, have had mixed results with some suggesting longer-acting benzodiazepines may be associated with prolonged use [23], others finding short-acting benzodiazepines are associated with prolonged use [17,21], and still others finding no association between benzodiazepine half-life and time to discontinuation [24,25]. In the opioid literature, research investigating initial prescriptions have revealed that longer initial prescriptions, higher initial doses, and use of long-acting versus short-acting formulations are associated with long-term use [26,27], leading to recommendations that clinicians prescribing opioids for acute pain prescribe the lowest effective dose of immediate-release opioids, and in “no greater quantity than needed for the expected duration of pain severe enough to require opioids” [28]. Current benzodiazepine prescribing guidance generally recommends “short-term” prescribing, typically defined as prescriptions lasting less than 4 weeks [29]. However, four weeks remains a relatively long prescription duration, and even shorter durations may be preferable. Further understanding the impact of initial benzodiazepine prescribing factors on sustained benzodiazepine use could help prescribers reduce harm by preventing long-term use and dependence. We therefore set out to understand the association of initial prescription characteristics with time to discontinuation of benzodiazepines in adults 18 years or older in Ontario, Canada, and given the disproportionate impact of benzodiazepine prescribing on women, whether this relationship differed by sex. We hypothesized that longer initial prescription durations would be associated with a reduced likelihood of discontinuation.
Methods
Study design and setting
We conducted a retrospective population-based cohort study using linked administrative health data from the province of Ontario, Canada. Ontario is home to over 14 million people who receive publicly-funded healthcare via a single payer, the Ontario government. This study was approved by the Centre for Addiction and Mental Health Research Ethics Board, #048/2022. The dataset creation plan for this study can be viewed in the supplement (S1 Data). The study’s reporting followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline (S1 STROBE Checklist).
Data sources
Data for this study were held at ICES, an independent, non-profit research institute. We utilized the Narcotic Monitoring System (NMS), a mandatory prescription monitoring system that captures all controlled substances dispensed by a community pharmacy, regardless of payer. This includes all benzodiazepines, opioids, and stimulant prescriptions. Demographic and vital statistics for all Ontarians were captured via the Registered Persons Database (RPDB), physician services via the Ontario Health Insurance Plan (OHIP) and physician characteristics via the Corporate Provider Database and the ICES Physician Database. We characterized inpatient hospitalizations via the Discharge Abstract Database (DAD) [30], mental health hospitalizations using the Ontario Mental Health Reporting System, and emergency department visits via the National Ambulatory Care Reporting System (NACRS). These databases were linked using unique encoded identifiers and analyzed at ICES. More information about the included databases are available in S1 Appendix, Table A.
Cohort creation
We included adults aged 18 years or older with a new prescription for oral benzodiazepines (excluding z-drugs which are not uniformly available in the NMS database) between January 1, 2013 and December 31, 2020, defined as no benzodiazepine prescription in the 182 days prior to cohort entry. Episodes where midazolam or clobazam were prescribed were excluded as these benzodiazepines are typically used for specific indications related to palliative care or seizure disorders. We also excluded individuals who had received palliative care services (including hospice services) in the 6 months prior to cohort entry as this population may have unique considerations with respect to medication continuation at end-of-life. We excluded individuals with missing age or sex, age <18 or >105 years, those who were not Ontario residents, and episodes with death date prior to index date. We then constructed episodes of continuous benzodiazepine use on the basis of receiving a refill within 1.5 times the days supply of the prior benzodiazepine prescription, with a minimum look-forward window of 30 days. If two benzodiazepine prescriptions were overlapping, we assumed they were taken concurrently rather than sequentially. The prescription end date was the date on which the last prescription was dispensed plus the days supplied, subtracting 1 (to account for us of medication on day of dispense). As more than one episode per individual could meet inclusion criteria, and we randomly included one episode per individual for the primary analysis. In a first-episode sensitivity analysis, we included only the first episode in the accrual period. In recurrent event sensitivity analyses, we utilized a 2-year look back for no benzodiazepine use, and accrued episodes from July 1, 2014 to December 31, 2020.
Exposures
The primary exposure was the duration of benzodiazepine use in days categorized as ≤7 days (referent), 8–14 days, 15–30 days, and >30 days. If more than one prescription was dispensed on index, we categorized the prescription with the longest duration. Secondary exposures were duration of action of benzodiazepine(s) dispensed on index defined as short-acting, long-acting or both short- and long-acting (see Table B in S1 Appendix for classification used); whether one or more than one benzodiazepine was dispensed on index; and the mean daily dose of the index prescription, in Diazepam Milligram Equivalents (DMEs) [31]. To calculate the mean daily dose of the index prescription in DMEs, we first calculated the total DME based on drug name and strength of benzodiazepine (benzodiazepine strength × Conversion Factor = DMEs). We then calculated the mean daily dose as strength in DME * quantity dispensed/days supplied. If more than one prescription was dispensed at index, their mean daily DMEs were summed (See Table C in S1 Appendix for conversion factors used). In all cases, categories were defined to align with standard clinical practice as it relates to benzodiazepine prescribing.
Outcome
The outcome of interest was time to benzodiazepine discontinuation as defined above (no new benzodiazepine dispense within 1.5× the days supplied of the last prescription, or a minimum of 30 days). We censored on the first of death, 3 years from cohort entry, or maximum follow up (December 31, 2021).
Covariates
Covariates were measured at baseline and were selected ‘a priori’ based on clinical expertise and their known or hypothesized relationship with the exposure and outcome. We included demographic variables (age, sex, rurality of residence, and neighborhood income quintile), concurrent mental health and substance use disorders (alcohol use disorder [32], harmful sedative-hypnotic use or dependence, receipt of opioid agonist therapy (OAT) in the last 6 months, as well as psychotic disorders [33], anxiety and mood disorders [33] and insomnia) [34] and medical comorbidity by using the Johns Hopkins Adjusted Clinical Groups (ACG) System Version 10 [35]. Patients were assigned a weighted score based on presence or absence of 32 ACG System Aggregated Diagnosis Groups (ADGs) characterizing medical conditions based on their use of inpatient and outpatient healthcare services in the preceding 2 years. ADGs were summed and grouped into quintiles [35]. Finally, we also included the medical discipline of the prescriber of the index prescription (see Table D in S1 Appendix for full variable definitions).
Analysis
Baseline characteristics (see Table E in S1 Appendix for full variable definitions) were stratified by sex and compared using standardized differences. A standardized difference >0.10 was considered meaningful [36]. Kaplan–Meier curves were used to present crude survival probabilities for the primary outcome, and the Log-Rank test was used to test for differences in survival functions between strata. We used multivariable Cox proportional hazards regression modeling to characterize the association between the exposure variables and our primary outcome, while adjusting for the covariates described above. We tested the proportional hazards assumption via visual inspection of log-negative-log survival curves (Figs A–D in S1 Appendix). In a secondary analysis, we stratified models by sex. Two pre-specified sensitivity analyses were conducted in order to assess the robustness of our findings to methodologic assumptions made in the primary analysis related to the definition of continuous use episodes and the calculation of DMEs. In the first sensitivity analysis, we relaxed the definition of discontinuation by looking forward 2 times the days supplied, with a minimum look forward of 60 days for the next benzodiazepine prescription. In a second sensitivity analysis, we used an alternative published conversion factor to calculate DMEs [37] (Table C in S1 Appendix). Notable differences in the conversion factors for DME calculations included differences in the equivalence of lorazepam (1 mg lorazepam = 10 mg diazepam in the main analysis versus 2 mg lorazepam = 10 mg in the sensitivity analysis), alprazolam (0.5 mg alprazolam = 10 mg diazepam in the main analysis versus 1 mg alprazolam = 10 mg diazepam in the sensitivity analysis) and clonazepam (0.5 mg clonazepam = 10 mg diazepam in the main analysis versus 1 mg clonazepam = 10 mg diazepam in the sensitivity analysis). In order to ensure there were not systematic changes in prescribing practices over the study period that would impact study results, we also undertook a post-hoc sensitivity analysis stratifying the results by time period (2013–2016 versus 2017–2020) and a sensitivity analysis including only first episodes in the accrual period. Finally, in order to understand the conditional, event-ordered-average effect across recurrent episodes, and whether prescribing factors associated with prolonged benzodiazepine use changed with recurrent prescription episodes, we conducted post-hoc recurrent event analyses using the Prentice–Williams–Peterson Total Time model overall, and stratified by episode number. This model estimates the exposures’ effects of the hazard of a recurrent event, conditional on prior event history [38]. The analyses were completed using SAS version 9.3, and a 2-tailed p value of <0.05 was deemed significant. There was missingness in only one variables (specialty of physician prescriber), and this was grouped separately.
Results
Overall, 3,248,734 benzodiazepine index prescriptions met inclusion criteria. Following the random inclusion of one index prescription per individual, 1,820,808 episodes were included in the analysis (Fig 1).
Baseline characteristics overall and stratified by sex are presented in Table 1. Overall, the median age at index was 53 years (Interquartile Range (IQR) 38–67 years), and 62.6% of included episodes occurred in females. Males were more likely to have a diagnosis of alcohol use disorder (6.7% versus 2.3%) and any substance use disorder (5.4% versus 2.0%) while females were more likely to have a mood or anxiety disorder diagnosis (46.5% versus 41.0%). Lorazepam was the most commonly prescribed benzodiazepine on index overall (63.9%), followed by clonazepam (17.3%) and diazepam (5.8%) (Table 1). Males were more likely to be prescribed diazepam (9.2%) while females were more likely to be prescribed lorazepam (66.9%). Median time to discontinuation overall was 19 days (IQR: 6–29). In women median time to discontinuation was 16 days (IQR: 6–29) while the median time to discontinuation in males was 19 days (IQR: 6–29).
Among the 1,820,808 included episodes in the primary analysis, 17,973 (0.99%) were censored due to death, 14,488 (0.80%) were censored at end of follow-up, and 1,788,347 (98.22%) experienced the outcome of interest (discontinuation). In crude models, median time to discontinuation differed by duration of index prescription (median 3 days for initial days supply ≤7 versus 88 days for initial days supply >30). Kaplan-Meier curves are presented in Fig 2 (log-rank test, p < 0.0001).
After multivariable adjustment, longer index prescriptions were associated with a reduced hazard of benzodiazepine discontinuation (8–14 days versus ≤7 days adjusted Hazard Ratio [aHR] 0.54 (95% Confidence Interval [CI] [0.54,0.40]; 15–30 days versus ≤7 days aHR 0.26 (95% CI [0.26,0.58]); > 30 days versus ≤7 days, aHR 0.14 (95% CI [0.14,0.14])) (Table 2). Being dispensed both a short-acting and long-acting benzodiazepine together (aHR 0.84 (95% CI [0.80,0.88])) and being prescribed long-acting benzodiazepines alone (aHR 0.80 (95% CI [0.80,0.80])) were associated with a reduced hazard of discontinuation compared to being dispensed short-acting benzodiazepines alone. Being initially prescribed two or more benzodiazepines compared with one was also associated with a reduced hazard of discontinuation (aHR 0.59 (95% CI [0.57,0.61])). Finally, mean daily doses between 5 and ≤10 DME (aHR 1.03 (95% CI [1.03,1.03])) and between 10 and ≤20 DME (aHR 1.03 (95% CI [1.03,1.04])) were associated with an increased hazard of discontinuation compared to doses ≤5 DME, while doses >20 were associated with a reduced hazard of discontinuation (aHR 0.98 (95% CI [0.97,0.98])). In the analysis stratified by sex, results were consistent with the primary analysis with respect to the impact of days supplied at index, the type of benzodiazepine dispensed at index, and the number of benzodiazepines dispensed at index although there were some differences in the direction of the association with the mean daily dose (Table 2). Among females, receipt of all dose categories had an increased hazard of discontinuation compared with ≤5 DME (aHR 1.03 (95% CI [1.03,1.04]), aHR 1.07 (95% CI [1.06, 1.07]), and aHR 1.01 (95% CI [1.00, 1.01]) for >5 - ≤10 DME, > 10-≤20 DME, and >20 DME, respectively, compared with ≤5 DME). Among males, receipt of >5 to ≤10 compared with ≤5 DMEs was associated with an increased hazard of discontinuation (aHR 1.03 (95% CI [1.02, 1.03])), while receipt of doses >10 to ≤20 and >20 DME were associated with a reduced hazard of discontinuation compared to receipt of <5 DMEs (aHR 0.98 (95% CI [0.92, 0.98]), aHR 0.94 (95% CI [0.93, 0.95]), respectively) (Table 2).
In the sensitivity analysis loosening the definition of discontinuation to allow a longer look forward, results were consistent (Table 3) with the exception that doses between 10 and ≤20 DME were associated with reduced hazard of discontinuation (aHR 0.96 (95% CI [0.96,0.97])) compared with ≤5 DME.
In the sensitivity analysis using a different DME conversion factor, the direction of the effects were similar except for daily doses between 10 and ≤20 DME were associated with a reduced hazard of discontinuation (aHR 0.95 (95% CI [0.95,0.96])), while daily doses between 5 and ≤10 and >20 DME were associated with higher hazard of discontinuation (aHR 1.06 (95% CI [1.05–1.06]); aHR 1.02 (95% CI [1.01,1.03]), respectively; Table 4).
Similarly, in both post-hoc sensitivity analyses stratifying the results by time period (2013–2016 versus 2017–2020) (Table F in S1 Appendix) and the sensitivity analysis including the first episode of benzodiazepines per individual during the accrual period (Table G in S1 Appendix), results were consistent with the primary analysis with respect to the impact of days supplied at index, type of benzodiazepine at index, and the number of benzodiazepines dispensed at index, although there were some differences in the direction of the association with the mean daily dose.
Finally, in the overall recurrent event analysis which included multiple episodes per individual (Table 5, N = 2,380,122 episodes), results were consistent with the primary model, except that compared with mean daily doses ≤5 DME, all higher dose categories were associated with a reduced likelihood of discontinuation (aHR 0.99 (95% CI [0.98,0.99]), aHR 0.98 (95% CI [0.98, 0.99]), aHR 0.98 (95% CI [0.97,0.99]), for doses >5 to ≤10 DME, >10 to ≤20 DME, and >20 DME, respectively). In the recurrent event model stratified by episode number (Table 6), the effect of the exposure variables were of the same direction and similar magnitude of effect across episode number except for mean daily dose, where the protective effect of lower mean daily doses was consistent in earlier episodes, but not for an individuals’ 4th or greater episode.
Discussion
In this analysis of nearly 2 million benzodiazepine treatment episodes, after adjusting for potential confounding variables, several initial benzodiazepine prescribing patterns were associated with time to benzodiazepine discontinuation. Specifically, longer initial index prescriptions were strongly associated with a reduced likelihood of benzodiazepine discontinuation, a finding that was replicated in sex stratified analyses, and in all sensitivity analyses. Furthermore, receipt of long-acting benzodiazepines at initiation, multiple benzodiazepine types initiated, two or more benzodiazepines initiated, were consistently associated with prolonged benzodiazepine use across the analyses. Across the various models, the aHR of mean daily dose was close to 1, suggesting mean daily dose is a less important initial prescribing factor. Effect sizes were largest for longer initial prescriptions, and 2 or more benzodiazepines (compared with 1) initiated suggesting these may be clinically more important factors associated with long-term use.
Our findings replicate what has been found in the opioid literature [26], which has led to recommendations of shorter initial prescription durations for managing acute pain following surgery or ED visits. Existing epidemiologic studies in specific populations (older adults [19], individuals with depression and concurrent anti-depressant use [20]) have also found an association between initial benzodiazepine prescription duration and long-term use. Current benzodiazepine prescribing guidance generally recommends “short-term” prescribing, however, this is typically defined as prescriptions lasting less than 4 weeks [29]. Our findings suggest that the risk of sustained benzodiazepine use is increased with prescriptions lasting greater than 7 days, which suggests that this recommendation is too liberal.
Unsurprisingly, provision of 2 or more benzodiazepines at time of initiation was associated with a reduced hazard of discontinuation, a finding that was also consistent across males and females and robust to changes in definition of discontinuation. Similar findings were observed in a population-based observational cohort of individuals prescribed benzodiazepines in Taiwan which found that being prescribed 2 or ≥3 benzodiazepines increased the odds of long-term use [21]. Similarly, prescription of a short-acting plus a long-acting agent compared with short-acting alone was associated with a reduced likelihood of discontinuation. Given that there are few clinical scenarios where simultaneous prescription of more than one benzodiazepine is necessary or indicated, prescription of more than one agent should be avoided.
As in the opioid literature, we also found that the use of long-acting agents at initiation was associated with a reduced likelihood of discontinuation compared with short-acting only [26]. Some observational studies of people who use benzodiazepines have concurred that longer-acting benzodiazepines may be associated with prolonged use [23], although elsewhere, short-acting agents have been associated with prolonged use [17,21], or no association has been found between benzodiazepine half-life and prolonged use [24,25]. Tolerance may be achieved more quickly with long-acting formulations leading to physiologic dependence and therefore continued use. High-potency and rapid-onset benzodiazepines, can enhance reinforcing subjective effects and have been associated with dose escalation and long-term use [14,16,18]. Further research to disentangle the effect of index prescription potency, time to onset, and half-life would be an important future direction. Overall, our findings suggest that the use of a single, short-acting agent may reduce the risk of prolonged benzodiazepine use.
In our primary analysis, among males, and in a sensitivity analysis with a longer look forward, we found that those dispensed index prescriptions with daily doses >20 DME compared with ≤5 DME, were more likely to remain on benzodiazepines, although this was not consistent across the other sensitivity analyses. Interestingly, in the recurrent event analysis stratified by episode number, doses ≤5 DME were protective against prolonged use, but this effect disappeared by episode ≥4, suggesting that lower doses may be more important for an individual’s earlier exposures to benzodiazepines. It should be noted however, that all estimates were close to 1.0, limiting clinical significance of these findings, and suggesting that future work should continue to explore the relationship between dose, indication for benzodiazepine use, sex and duration of use. It may be helpful for prescribers to familiarize themselves with DMEs in the same way that morphine equivalents have been widely utilized clinically in opioid prescribing and deprescribing, and they should be aware that higher doses, particularly during an individual’s first benzodiazepine exposures, may be associated with prolonged use.
Our study also adds to the literature, by demonstrating that several prescribing factors (longer days supplied, receipt of two or more benzodiazepines at index, receipt of long-acting or both short- and long-acting benzodiazepines (compared with short-acting alone)), were associated with prolonged use, even as the number of prior episodes of benzodiazepine use increased, suggesting prescribers should keep these prescribing characteristics in mind, even in individuals who have a history of benzodiazepine use.
In previous studies, other factors associated with prolonged benzodiazepine use include older age, multi-morbidity, psychiatric co-morbidity, low-income, initial prescription by a psychiatrist, and prescription on discharge from hospital [5,14,21,39,40]. Future investigations should determine whether some of these sociodemographic and diagnostic factors interact with initial prescription characteristics to determine length of use. Future studies could also examine whether these results are generalizable to other sub-populations at higher risk of harms related to benzodiazepine use including individuals with cognitive impairment, or a history of falls.
This was a large population-based study in Canada’s most populous province, capturing all outpatient prescriptions for benzodiazepines across the province, regardless of payer, and linked administrative data reflecting all health service utilization in a publicly-funded system which supports the generalizability of our findings. There are, however, several limitations. Most importantly, our study relies on pharmacy dispensing records and assumes that individuals took medications as prescribed. It is possible that medications were stock-piled or not used as prescribed, which could result in misclassification of exposure variables and outcomes. Similarly, since z-drugs are not all captured in the NMS database, we were unable to include them in the analysis, and this may have resulted in misclassification of the exposure and outcome variables. However, the consistency between our primary analysis and the sensitivity analyses are reassuring. Other psychiatric medications such as SSRIs are also not captured in the NMS and therefore could not be included as covariates in the model. We were, however, able to include psychiatric diagnoses. Readers should note that we have presented hazard ratios (a commonly used statistic in time-to-event analyses) and that the results represent relative rate reductions rather than relative risk reductions [41]. Although we attempted to control for important confounders, there may be residual confounding by indication. Finally, as with all observational studies, our study is subject to bias from unmeasured confounding. Disease severity, use of non-prescribed benzodiazepines and social stability, for example, are not captured in administrative health data.
In this large population-based cohort study, initial benzodiazepine prescribing patterns were associated with time to benzodiazepine discontinuation. Prescribing shorter courses of benzodiazepines, use of a single benzodiazepine, use of a short-acting agent were associated with reduced likelihood of long-term benzodiazepine use. These findings suggest that simple changes to prescribing practice including limiting initial prescriptions to <7 days and choosing a single, short-acting agent could reduce prolonged benzodiazepine use and associated morbidity and mortality.
Supporting information
S1 Data. Dataset creation plan used for study planning.
https://doi.org/10.1371/journal.pmed.1005126.s002
(DOCX)
S1 STROBE Checklist. Completed STROBE checklist.
STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) Statement – checklist of items that should be included in reports of observational studies, licensed under CC BY 4.0. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP; STROBE Initiative. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE)statement: guidelines for reporting observational studies. PLoS Med. 2007 Oct 16;4 (10):e296. PMID: 17941714.
https://doi.org/10.1371/journal.pmed.1005126.s003
(DOCX)
Acknowledgments
This study was supported by ICES, which is funded by an annual grant from the Ontario Ministry of Health (MOH) and the Ministry of Longterm Care (MLTC). Parts of this material are based on data and information compiled and provided by the Ontario Ministry of Health and CIHI. Data adapted from the Statistics Canada Postal Code Conversion File was used, which is based on data licensed from Canada Post Corporation, and/or data adapted from the Ontario Ministry of Health Postal Code Conversion File, which contains data copied under license from Canada Post Corporation and Statistics Canada. The analyses, conclusions, opinions and statements expressed herein are solely those of the authors and do not reflect those of the funding or data sources; no endorsement is intended or should be inferred. The authors would like to thank our community advisor for her thoughtful contributions throughout the research process.
References
- 1.
SAMHSA. National Survey on Drug Use and Health. Center for Behavioral Health Statistics and Quality; National Survey on Drug Use and Health.
- 2. Murphy Y, Wilson E, Goldner EM, Fischer B. Benzodiazepine use, misuse, and harm at the population level in Canada: a comprehensive narrative review of data and developments since 1995. Clin Drug Invest. 2016;36(7):519–30.
- 3. Soyka M. Treatment of benzodiazepine dependence. N Engl J Med. 2017;376:2399–400.
- 4. Weich S, Pearce HL, Croft P, Singh S, Crome I, Bashford J, et al. Effect of anxiolytic and hypnotic drug prescriptions on mortality hazards: retrospective cohort study. BMJ. 2014;348:g1996. pmid:24647164
- 5. Taipale H, Särkilä H, Tanskanen A, Kurko T, Taiminen T, Tiihonen J, et al. Incidence of and characteristics associated with long-term benzodiazepine use in Finland. JAMA Netw Open. 2020;3(10):e2019029. pmid:33119104
- 6. Andersen ABT, Frydenberg M. Long-term use of zopiclone, zolpidem and zaleplon among Danish elderly and the association with sociodemographic factors and use of other drugs. Pharmacoepidemiol Drug Saf. 2011;20(4):378–85. pmid:21259378
- 7. Zheng D, Brett J, Daniels B, Buckley NA, Pearson S-A, Schaffer AL. Potentially inappropriate benzodiazepine use in Australian adults: a population-based study (2014-2017). Drug Alcohol Rev. 2020;39(5):575–82. pmid:32391624
- 8. Davies SJC, Jacob B, Rudoler D, Zaheer J, de Oliveira C, Kurdyak P. Benzodiazepine prescription in Ontario residents aged 65 and over: a population-based study from 1998 to 2013. Ther Adv Psychopharmacol. 2018;8(3):99–114. pmid:29492258
- 9. Weir DL, Samanani S, Gilani F, Jess E, Eurich DT. Benzodiazepine receptor agonist dispensations in Alberta: a population-based descriptive study. CMAJ Open. 2018;6(4):E678–84. pmid:30591546
- 10. Olfson M, King M, Schoenbaum M. Benzodiazepine use in the United States. JAMA Psychiatry. 2015;72(2):136–42. pmid:25517224
- 11. Pottie K, Thompson W, Davies S, Grenier J, Sadowski CA, Welch V. Deprescribing benzodiazepine receptor agonists. Can Fam Physician. 2018;64(5):339.
- 12.
ASAM. Joint clinical practice guideline on benzodiazepine tapering: considerations when benzodiazepine risks outweigh benefits. 2025.
- 13. Le TM, Campbell S, Andraos A, Ahlmark P, Hoang H, Isserman S, et al. Implementation of an intervention aimed at deprescribing benzodiazepines in a large US healthcare system using patient education materials: a pre/post-observational study with a control group. BMJ Open. 2024;14(4):e080109. pmid:38569687
- 14. Kurko TAT, Saastamoinen LK, Tähkäpää S, Tuulio-Henriksson A, Taiminen T, Tiihonen J, et al. Long-term use of benzodiazepines: definitions, prevalence and usage patterns – a systematic review of register-based studies. Eur Psychiatry. 2015;30(8):1037–47. pmid:26545257
- 15. Isacson D. Long-term benzodiazepine use: factors of importance and the development of individual use patterns over time--a 13-year follow-up in a Swedish community. Soc Sci Med. 1997;44(12):1871–80. pmid:9194248
- 16. Kjosavik SR, Ruths S, Hunskaar S. Use of addictive anxiolytics and hypnotics in a national cohort of incident users in Norway. Eur J Clin Pharmacol. 2012;68(3):311–9. pmid:21928085
- 17. van Hulten R, Teeuw KB, Bakker A, Leufkens HG. Initial 3-month usage characteristics predict long-term use of benzodiazepines: an 8-year follow-up. Eur J Clin Pharmacol. 2003;58(10):689–94. pmid:12610746
- 18. Simon GE, VonKorff M, Barlow W, Pabiniak C, Wagner E. Predictors of chronic benzodiazepine use in a health maintenance organization sample. J Clin Epidemiol. 1996;49(9):1067–73. pmid:8780618
- 19. Gerlach LB, Maust DT, Leong SH, Mavandadi S, Oslin DW. Factors associated with long-term benzodiazepine use among older adults. JAMA Intern Med. 2018;178(11):1560–2. pmid:30208384
- 20. Bushnell GA, Stürmer T, Gaynes BN, Pate V, Miller M. Simultaneous antidepressant and benzodiazepine new use and subsequent long-term benzodiazepine use in adults with depression, United States, 2001-2014. JAMA Psychiatry. 2017;74(7):747–55.
- 21. Fang S-Y, Chen C-Y, Chang I-S, Wu EC-H, Chang C-M, Lin K-M. Predictors of the incidence and discontinuation of long-term use of benzodiazepines: a population-based study. Drug Alcohol Depend. 2009;104(1–2):140–6. pmid:19515515
- 22. Isomura K, Wang X, Chang Z, Hellner C, Hasselström J, Ekheden I, et al. Factors associated with long-term benzodiazepine and Z-drug use across the lifespan and 5-year temporal trajectories among incident users: a Swedish nationwide register-based study. Eur J Clin Pharmacol. 2023;79(8):1091–105. pmid:37294340
- 23. Yeh H-H, Chen C-Y, Fang S-Y, Chang I-S, Wu EC-H, Lin K-M. Five-year trajectories of long-term benzodiazepine use by adolescents: patient, provider, and medication factors. Psychiatr Serv. 2011;62(8):900–7. pmid:21807829
- 24. Egan M, Moride Y, Wolfson C, Monette J. Long-term continuous use of benzodiazepines by older adults in Quebec: prevalence, incidence and risk factors. J Am Geriatr Soc. 2000;48(7):811–6. pmid:10894322
- 25. Nomura K, Nakao M, Sato M, Yano E. The long-term prescription of benzodiazepines, psychotropic agents, to the elderly at a university hospital in Japan. Tohoku J Exp Med. 2007;212(3):239–46. pmid:17592211
- 26. Béliveau A, Castilloux A-M, Tannenbaum C, Vincent P, de Moura CS, Bernatsky S, et al. Predictors of long-term use of prescription opioids in the community-dwelling population of adults without a cancer diagnosis: a retrospective cohort study. CMAJ Open. 2021;9(1):E96–106. pmid:33563639
- 27. Shah A, Hayes CJ, Martin BC. Characteristics of initial prescription episodes and likelihood of long-term opioid use – United States, 2006-2015. MMWR Morb Mortal Wkly Rep. 2017;66(10):265–9.
- 28. Dowell D, Ragan KR, Jones CM, Baldwin GT, Chou R. CDC clinical practice guideline for prescribing opioids for pain – United States, 2022. MMWR Recomm Rep. 2022;71(3):1–95.
- 29. Brandt J, Bressi J, Lê M-L, Neal D, Cadogan C, Witt-Doerring J, et al. Prescribing and deprescribing guidance for benzodiazepine and benzodiazepine receptor agonist use in adults with depression, anxiety, and insomnia: an international scoping review. EClinicalMedicine. 2024;70:102507. pmid:38516102
- 30.
Juurlink D, Preyra C, Croxford R, Chong A, Austin P, Tu J. Canadian Institute for Health Information Discharge Abstract Database: a validation study. Toronto; 2006.
- 31. Brandt J, Alessi-Severini S, Singer A, Leong C. Novel measures of benzodiazepine and Z-drug utilisation trends in a Canadian provincial adult population (2001-2016). J Popul Ther Clin Pharmacol. 2019;26(1):e22–38. pmid:31002486
- 32.
Nickel NC, Bolton J, MacWilliam L, Ekuma O, Prior H, Valdivia J. Health and social outcomes associated with high-risk alcohol use. Winnipeg, MB; 2018.
- 33.
Team MR. The mental health of children and youth in Ontario; 2017 scorecard. Technical appendix. Toronto, ON: Institute for Clinical Evaluative Sciences; 2017.
- 34. Thorpy MJ. Classification of sleep disorders. Neurotherapeutics. 2012;9(4):687–701. pmid:22976557
- 35.
Weiner EI. The John Hopkins ACG Case-Mix System Version 6.0 Release Notes. The John Hopkins University; 2003.
- 36. Austin PC. Using the standardized difference to compare the prevalence of a binary variable between two groups in observational research. Commun Stat Simul Comput. 2009;38(6):1228–34.
- 37. Borrelli EP, Bratberg J, Hallowell BD, Greaney ML, Kogut SJ. Application of a diazepam milligram equivalency algorithm to assess benzodiazepine dose intensity in Rhode Island in 2018. J Manag Care Spec Pharm. 2022;28(1):58–68. pmid:34949119
- 38. Amorim LDAF, Cai J. Modelling recurrent events: a tutorial for analysis in epidemiology. Int J Epidemiol. 2015;44(1):324–33. pmid:25501468
- 39. Morgan SG, Weymann D. Patterns, predictors and persistence of chronic sedative use: a population-based observational study of older adults in British Columbia, Canada. Eur J Clin Pharmacol. 2017;73(8):1001–8. pmid:28435984
- 40. Brandt J, Janzen D, Alessi-Severini S, Singer A, Chateau D, Enns M, et al. Risk of long-term benzodiazepine and Z-drug use following the first prescription among community-dwelling adults with anxiety/mood and sleep disorders: a retrospective cohort study. BMJ Open. 2021;11(11):e046916. pmid:34725071
- 41. Verbeeck J, Saad ED. Rethinking survival analysis: advancing beyond the hazard ratio?. Eur Heart J Acute Cardiovasc Care. 2024;13(3):313–5. pmid:38330167