Short‐ and long‐term mortality following hypnotic use

Potential long‐term consequences of hypnotics remain controversial. We used the prospective Swedish National March Cohort, a study based on 41,695 participants with a mean follow‐up duration of 18.9 years. Logistic regression models and Cox proportional hazards models with attained age as timescale were used to assess associations of hypnotic use with short‐ and long‐term mortality. The proportion of subjects who initiated or discontinued hypnotic use during follow‐up was substantial. All groups of hypnotics were associated with increased mortality within 2 years after a first prescription, with an overall OR of 2.38 (95% CI, 2.13–2.66). The association was more pronounced among subjects younger than 60 years (OR, 6.16; 95% CI, 3.98–9.52). There was no association between hypnotic use and long‐term mortality. The association between hypnotic use and increased mortality was thus restricted to a relatively short period after treatment initiation, and may be explained in terms of confounding by indication.

not distinguished between short-and long-term mortality. Beside the association with all-cause mortality, hypnotic use has been associated with specific causes of death, such as cardiovascular disease (Mallon, Broman, & Hetta, 2009) and cancer (Kripke, 2009;Kripke, Langer, & Kline, 2012). Using a large Swedish cohort with a mean follow-up time of 18.9 years, we aimed to study the association of hypnotics with short-and long-term all-cause and cause-specific mortality.

| ME THODS
We used the Swedish National March Cohort (Trolle Lagerros, Hantikainen, & Mariosa, 2017), designed to investigate associations between lifestyle factors and chronic diseases. The study was established in September 1997 during a 4-day nationwide fundraising event for the Swedish Cancer Society. Nearly 3,600 Swedish cities and villages took part in the event. All participants were invited to fill out a 36-page questionnaire regarding demographic, lifestyle and medical information. They also provided their national registration number, an individually unique identifier assigned to all Swedish residents, which enables follow-up by linkage to multiple nationwide, continuously updated and essentially complete databases.
In total, 43,863 participants completed the questionnaire. Those with incorrect national registration numbers were excluded (n = 11), as were those who were younger than 18 years (n = 1,732) or had emigrated or died (n = 55) before the start of follow-up. We also excluded subjects who did not provide information regarding hypnotic use (n = 370). Our final study population thus included 41,695 subjects followed prospectively for all-cause and disease-specific mortality until the end of April 2018. The study was approved by the Regional Ethics Committee in Stockholm and all participants provided written informed consent.
The cohort was followed from baseline on October 1, 1997 until date of death, emigration or April 30, 2018, whichever occurred first. Using the national registration numbers, mortality data were obtained by linkage to the Swedish Cause of Death Register held by the National Board for Health and Welfare. A total of 9,093 deaths occurred during the follow-up period. Information regarding diagnoses of cardiovascular disease (I00-I99), cancer (C00-C97) and psychiatric disorders (F00-F99) was obtained from the Swedish National Patient Registers and the Swedish Cancer Register.
Hypnotic use at baseline was assessed by asking the participants to estimate how often they took sleeping pills. The response alternatives were never, seldom, sometimes, mostly or always. The reference group was those who never used hypnotics. Information regarding hypnotic use during follow-up was obtained from the Swedish Drug Registry, in which all prescriptions dispensed in Swedish pharmacies in 2005 or later are registered. Hypnotics were categorized into benzodiazepines (N05CD),  and other hypnotics (N05CM).

| Statistical analyses
Differences in baseline variables across categories of hypnotic use frequency were assessed using one-way analysis of variance (ANOVA) for continuous variables and the Kruskal-Wallis test for categorical variables.

| Short-term hypnotic use
We used logistic regression models to study the association between hypnotic use at baseline and all-cause and cause-specific mortality during the first 4 years of follow-up. Among subjects who were alive in January 2005, we used logistic regression to assess mortality risk during 2005-2006 among those who collected prescriptions of hypnotics in 2005, compared with those who never collected a prescription of hypnotics. We next assessed mortality during 2007-2008 among those who collected their first prescription of hypnotics in 2007 compared to those who never collected a prescription of hypnotics. In the same way, we assessed mortality within 2 years after F I G U R E 1 Assessment of short-term mortality Assessment of short-term mortality 1997 2007 Exposure period ( ) Time-period 1   2009  2011  2013  2015  2017  2019   Time-period 2  Time-period 3  Time-period 4  Time-period 5  Time-period 6 Exposed ; those who collected their first prescripƟon of hypnoƟcs during the exposure period.
Unexposed; those who had never collected a prescripƟon of hypnoƟcs before or during the exposure period.
Mortality was assessed during a 2-year period ( ) Results from the six logisƟc regression models were pooled together (supplementary tables 1-6).
collecting a first prescription of a hypnotic during 2009, 2011, 2013 and 2015. We then pooled the results from these six logistic regression models (Figure 1). The analyses were stratified by age (<60 years or 60+ years) and gender.

| Long-term hypnotic use
We used Cox proportional hazards models with attained age as timescale to estimate hazard ratios (HRs) with 95% confidence intervals (CIs) for hypnotic users at baseline, compared to those who never used hypnotics. We assessed the proportionality hazard assumption, based on the Schoenfeld residual plots and statistical tests. We also assessed mortality risk using a logistic regression model in which subjects were categorized based on their hypnotic consumption in 2005-2008, 2009-2012 and 2013-2016. The reference group comprised those who never collected a prescription of hypnotics during follow-up. The analysis was restricted to subjects who were alive in January 2017.

| Potential confounding variables
The Cox regression models with self-reported hypnotic use as the exposure were adjusted for potential confounding variables, including sex, occupation, education, body mass index (BMI), physical activity, smoking and alcohol consumption. The Cox regression models were also adjusted for a diagnosis of cardiovascular disease and cancer, and for psychiatric conditions.
The logistic regression models, in which subjects who had died before 2005 were excluded, were adjusted for sex, education, BMI at baseline, ever smoking at baseline, cardiovascular disease, cancer, psychiatric conditions, and medication with antipsychotics, sedatives and antidepressants.
The proportion of missing data in the potential confounding variables was 4.6% for BMI, 1.4% for ever smoking, and less than 1% for occupational level, educational level, physical activity and alcohol consumption. We therefore conducted supplementary analyses after imputing missing data using the multiple imputation chained equation procedure. We also conducted sensitivity analyses in which we further adjusted for habitual sleep duration and insomnia. Insomnia was defined as mostly or always experiencing any of the nocturnal insomnia symptoms (difficulties initiating sleep, difficulties maintaining sleep and early-morning awakenings) in combination with mostly or always experiencing symptoms of non-restorative sleep (not rested at awakening and daytime sleepiness). All analyses were performed using Statistical Analysis System 9.4.

| RE SULTS
Characteristics of participants at baseline, overall and by frequency of hypnotic use, are presented in Table 1. Hypnotic use was highly correlated with sleep duration and insomnia. Generally, hypnotic users were older, less educated and less physically active. They were more often diagnosed with cardiovascular disease, cancer or psychiatric conditions, compared to subjects who never used hypnotics.
In Table 2, we illustrate how hypnotic use changed between time periods. The proportion of subjects who initiated or discontinued hypnotic use during follow-up was substantial.

| Short-term hypnotic use
Self-reported hypnotic use was associated with a significantly increased mortality confined to the first 2 years of follow-up (p for Follow-up 2 Follow-up 3 Follow-up 4 Follow-up 5 Exposed ; those who collected their first prescripƟon of hypnoƟcs during the exposure period. Unexposed; those who never collected a prescripƟon of hypnoƟcs before or during follow-up ( ).
Mortality was assessed in 2-year intervals, from start of each follow-up unƟl Dec 2018.
Corresponding 2-year intervals from each follow-up were pooled together (supplementary table 7).
trend 0.02) ( Table 3). Compared to subjects who never collected prescriptions of hypnotics, short-term mortality was increased after collecting a first prescription of hypnotics.
Within 2 years after collection of a first prescription of hypnotics, all-cause mortality was increased 2.38-fold; benzodiazepines increased risk 5-fold, whereas Z-drugs and other hypnotics doubled the risk (Table 4). Mortality due to cancer was increased 4-fold, whereas mortality due to cardiovascular disease and other causes was increased by 30% and 60%, respectively. Mortality due to external causes, respiratory diseases, neurologic diseases and psychiatric diseases (mainly Alzheimer's disease and other kinds of dementia) was more common among hypnotic users.
The mortality rate for each of these causes of death during each 2-year period was less than 0.1% among non-users, whereas the mortality rate was more than doubled among hypnotic users (data not shown).

| Long-term hypnotic use
Overall, self-reported hypnotic use was associated with long-term increased mortality during a mean follow-up time of 18.9 years (SD 4.1). Mortality increased with increasing frequency of hypnotic use (Table 5). However, when we studied mortality risk in 2-year time intervals after collecting a first prescription of hypnotics, the increased mortality risk was confined to a relatively short period following the initiation of hypnotic use, with no association between hypnotic use and long-term mortality (Table 6).
Similarly, when subjects were categorized based on history of collected prescriptions of hypnotics between 2005 and 2012, mortality was increased only in subjects who had recently started using hypnotics (Table 7).
Our results remained stable in analyses including multiple imputed data (data not shown). Our results also remained similar after adjustment for habitual sleep duration and insomnia (data not shown).

| D ISCUSS I ON
In our prospective cohort study, comprising 41,695 participants, initiation of hypnotic use was associated with increased short-term mortality, mainly among patients with prevalent cancer, whereas no association with long-term mortality was observed. All pharmacologic groups of hypnotics were associated with increased mortality within 2 years after collecting a first prescription. The association between hypnotic use and short-term mortality was more pronounced among males, and more pronounced among younger participants.
Investigating the association between hypnotic use and mortality is challenging due to the substantial proportion of subjects who initiate or discontinue hypnotic use in a relatively short period This may be because the longer the follow-up period, the greater the overlap between hypnotic users and non-users. Absence of impact of hypnotic use on long-term mortality is another potential explanation.
An association of hypnotics with excess mortality may arise due to confounding by indication because chronic illness increases the need for hypnotic use. Impaired sleep has indeed been suggested to represent an early sign of cancer and often precedes a cancer diagnosis (Garland, Irwin, Posner, & Perlis, 2018). Symptomatic treatment, including insomnia treatment, is also one of the primary goals in palliative care. This is in agreement with our finding of a strong association between hypnotic use and short-term mortality due to cancer and cardiovascular disease. All categories of hypnotics, with differing mechanisms of action, were associated with increased short-term mortality. This lack of specificity supports the view that the association between hypnotics and mortality is non-causal and mainly due to confounding by indication. However, adverse effects of hypnotics, such as respiratory suppression and impairment in psychomotor performance, may contribute to increased short-term mortality due to less common causes of death, such as respiratory diseases and external causes.
Whether an association between hypnotics and cancer is causal or due to confounding or reverse causation has been debated (Neutel & Johansen, 2015), and is of great importance considering the extensive use of these drugs. We found no association between hypnotic use and long-term mortality, which contradicts the biologic impact of hypnotics on cancer induction or progression of pre-existing cancers.
Non-pharmacologic interventions are the first-line therapy for adults with insomnia. Numerous clinical trials provide evidence that cognitive behavioural therapy, traditionally delivered in either individual or group settings, is an effective treatment for insomnia (Trauer, Qian, Doyle, Rajaratnam, & Cunnington, 2015). Internetbased cognitive behavioural therapy seems to be equally effective and could make the treatment more available in order to better meet population needs (Seyffert et al., 2016 is unresponsive to other approaches (Winkelman, 2015). In order to prevent and relieve unnecessary suffering, hypnotic medication may also be part of palliative care.
Insomnia is associated with a chronic physiologic arousal (Bonnet & Arand, 2010) and may increase the risk of medical disorders such as hypertension (Fernandez-Mendoza, Vgontzas, & Liao, 2012), cardiac disease (Laugsand, Vatten, Platou, & Janszky, 2011) and diabetes (Vgontzas et al., 2009). Although insomnia often co-occurs with psychiatric disorders such as depression and anxiety (Baglioni et al., 2011), a recent meta-analysis TA B L E 3 Logistic regression analyses with estimated odds ratio (OR) and 95% confidence interval (CI) for all-cause and cause-specific mortality among subjects who reported hypnotic use at baseline, compared to those who never used hypnotics at baseline The p value for trend was calculated with hypnotics as an interval variable (0 = never, 1 = seldom, 2 = sometimes, 3 = mostly/always).

TA B L E 4
Pooled results from logistic regression analyses with estimated odds ratio (OR) and 95% confidence interval (CI) for all-cause and cause-specific mortality within 2 years after collecting a first prescription of hypnotics between January 2005 and December 2015, compared with subjects who never collected prescriptions of hypnotics concluded that insomnia increases the risk of incident depression (Fang, Tu, Sheng, & Shao, 2019) and suicide (Bjorngaard, Bjerkeset, Romundstad, & Gunnell, 2011;Lin, Lai, & Perng, 2018).
Whether hypnotic medication reduces this excess risk deserves further study.
The strengths of this prospective cohort study are the large sample size, and the long and almost complete follow-up ascertained by linking baseline information with nationwide, continuously updated registers. A limitation is that self-reported information regarding lifestyle factors was only measured at baseline, and potential changes in lifestyle habits during the follow-up period would go undetected. The baseline measurement of self-reported hypnotic use was crude and did not allow distinction between prescribed and non-prescribed hypnotics.
However, information was available for all prescriptions dispensed in Swedish pharmacies in 2005 or later, and we had the possibility to study the association between hypnotics and mortality in detail.
In conclusion, the association between hypnotic use and increased mortality was restricted to a relatively short period after treatment initiation and may to be explained by confounding by indication.

CO N FLI C T O F I NTE R E S T
The authors report no conflict of interest.
TA B L E 5 Cox regression analysis with estimated hazard ratio (HR) and 95% confidence interval (CI) for all-cause and cause-specific mortality among subjects who reported hypnotic use at baseline, compared to those who never used hypnotics at baseline

TA B L E 6
Pooled results from logistic regression analyses with estimated odds ratio (OR) and 95% confidence interval (CI) for all-cause and cause-specific mortality in 2-year time intervals after collecting a first prescription of hypnotics, compared with those who never collected prescriptions of hypnotics Adjusted for gender, educational level, ever smoking at baseline, body mass index at baseline, cardiovascular disease, cancer, psychiatric disorders, neuroleptic medication, sedatives and antidepressants. Subjects who initiated hypnotic use after 2012 were excluded from these analyses.