Hypnotic drug risks of mortality, infection, depression, and cancer: but lack of benefit

Hypnotic drugs increase all-cause mortality

Use of hypnotic drugs is associated prospectively with a greatly increased risk of all-cause mortality. Some of this mortality has been documented as deaths caused by hypnotics by Medical Examiners, attributed to respiratory arrests resulting from “overdose.” However, it is likely that many deaths from respiratory depression occur among patients never examined by coroners, especially when the death is caused by a combination of hypnotics with other contributing factors, so that the lethal hypnotic dosage may by itself have been within currently-customary dosage ranges. In addition to respiratory depression, hypnotics appear to be causally related to serious illnesses and premature deaths from cancer, serious infections, mood disorders, accidental injuries, suicides and homicides.

The overdose epidemic

Only a small fraction of U.S. deaths are medically evaluated by coroners and Medical Examiners. Thus, it is commonly assumed that overdose deaths have been grossly under-reported. Despite such underestimation, U.S. drug and opioid overdose deaths reported in 2014 reached 47,055, a 137% increase since 2000¹. A higher estimate of 76,227 for 2014 self-injury deaths included overdoses that might have been accidental or cause-undetermined². The overdose rate continued to rise in 2015³. Likewise, U.S. use of hypnotics dramatically increased over most of the same interval until about 2012⁴. Available death certificate reports seem unclear, but perhaps one third of death certificates listing overdose with an opiate as a cause of death also lists a benzodiazepine, Z hypnotic, or barbiturate as a cause of death (retrieved from CDC Wonder). An opioid prescription is more likely to lead to overdose when a hypnotic is also prescribed⁵. Indeed, among women who had received an opioid prescription, the added overdose risk of taking high-dose benzodiazepine or zolpidem was comparable to the risk of high-dose opioid⁶. There have also been several thousand yearly reported overdoses involving a hypnotic in which an opiate was not involved. So great is the recent increase in overdose and suicide deaths that it has lowered overall life expectancy in much of the U.S. adult population^7,8. Overdoses kill more Americans than automobile accidents or murders. On August 31, 2016, the FDA announced “boxed warnings” about the lethality of combinations of opiates and benzodiazepine agonists including the hypnotics.

Aside from overdose deaths, benzodiazepines were involved in a comparably increasing number of emergency room visits over a similar time interval, and over half of these also involved opioids, alcohol, or all three in combination⁹. Combined overdoses of opiates and benzodiazepine agonists had more severe outcomes. Suicides from all causes per capita have been increasing, particularly among women, and particularly since 2007, at which time generic zolpidem became available, increasing zolpidem prescribing¹⁰.

Forty epidemiologic studies

Of 40 epidemiologic studies that provided comparable risk ratios for mortality associated with hypnotics, 39 found that hypnotics were associated with excess mortality, as listed in the Appendix. In addition, a 41^st study of stroke patients showed elevated mortality among those receiving hypnotics and various other psychotropic drugs¹¹. The exception was a small study by Merlo et al. that nevertheless found hypnotics associated with cancer deaths¹². A partial exception was a study from Taiwan that found benzodiazepine hypnotics associated with significant excess mortality, but found zolpidem 10 mg associated with significantly reduced mortality in adjusted models, despite a significantly-increased unadjusted mortality risk for zolpidem and a significantly-increased adjusted risk of cancer mortality for zolpidem¹³. In a comment to this report appearing with it on the internet, I have questioned the statistical methods of adjusting zolpidem risks¹³.

Only that one of the 40 epidemiologic studies of hypnotic drugs reported any association with improved patient survival, and that only after questionable statistical adjustments¹³. None of the other 39 studies found hypnotic drug risk ratios significantly less than 1.0. That is, in 39 studies there was no evidence that hypnotics ever benefit patient survival. To find 39 of 40 studies showing a positive risk ratio was very highly significant, P<0.000001. Also, the evidence of association satisfied all nine Bradford Hill criteria for inferring causality¹⁴, though skepticism despite meeting these criteria may be warranted¹⁵. There remain questions concerning the magnitude of the causality. The randomized placebo-controlled trials I have suggested would help clarify this causality magnitude¹⁴.

Of the 40 epidemiologic studies, 31 individual studies reported statistically significant mortality odds ratios, risk ratios, or hazard ratios exceeding 1.0. All 18 studies reporting on samples of >14,000 people found significant mortality risks, but nine of 22 smaller studies found positive trends that were not significant. Most of the non-significant reports were among the earliest 15 published before 2006. Of studies analyzing follow-ups of 8 years or less, 22 of 26 studies reported a significant association, but of studies with longer follow-ups, only 10 of 14 studies observed significant mortality risks. This may suggest that during long prospective follow-ups, many patients initially taking hypnotics will discontinue hypnotic usage, whereas many controls not using hypnotics at prospective baseline may have begun using hypnotics during a long follow-up, so that the longer the follow-up, the more mixing of hypnotic-consuming and control groups becomes likely. Mixing weakens the risk-ratio contrasts observed. In long follow-ups, one is also studying the selected survivors of the more marked short-term risks that have been recently described^16,17.

Most of the 40 studies reported mortality risk ratios of less than 1.5, but some of the highest quality studies reported among the highest risk ratios. Four of the most recent studies were particularly persuasive, as presented below.

The Geisinger Health System study

From electronic records of the Geisinger Health System in Eastern Pennsylvania, a sample of 34,205 patients was drawn with carefully controlled 1:2 matching of hypnotic users with non-user controls for age, gender, smoking, and various comorbidities. Compared to a reference hazard ratio of 1.0 for non-users of hypnotics, the fully-adjusted mortality hazard ratio for use of 0.4–18 hypnotic doses per year was 3.60 (2.92–4.44, 95% CI), for those using 18–132 doses per year, the hazard ratio was 4.43 (3.67–5.36), and for >132 doses per year, the hazard ratio was 5.32 (4.50–6.30)¹⁸. Each of these associations was significant with P<0.001. Sensitivity studies showed that little of the hypnotic-associated mortality could be explained by known confounders or use of hypnotics before commencement of the study. In this sample, prescriptions for each of the following drugs were found to significantly predict increased mortality with statistical significance: zolpidem, temazepam, eszopiclone, zaleplon, triazolam, flurazepam or quazepam, and barbiturates prescribed to induce sleep. This review is principally concerned with these popular hypnotics for which drug-specific mortality data are available. Barbiturates prescribed at night for sleep considered as a group had about the same empirical hazard ratios as the benzodiazepines and zolpidem, but the observed hazard ratio for eszopiclone was significantly higher than that of barbiturates, possibly biased by the shorter average follow-up intervals for this more-recently introduced drug¹⁸.

The Weich et al. study

In a sample of over 100,000 hypnotic users and matched controls from the representative British General Practice Research Database¹⁹, users of 1–30 defined daily doses (DDD) of hypnotics and anxiolytics within a year had fully adjusted dose-responsive mortality hazard ratios of 2.55 (2.42–2.69, 95% CI) for 1–30 DDD (defined daily doses in the first year); 3.78 (3.54–4.04) for 31–60 DDD, 4.19 (3.84–4.58) for DDD 61–90, and 4.51 (4.22–4.82) for DDD >90. Extensive full adjustment for potential confounders resulted in only very small and inconsistent decreases in the estimated hazard ratios, and many methodological details were focused on minimizing possibilities of confounding. Use of benzodiazepine hypnotics alone was associated with higher hazard ratios than use of “Z” hypnotics alone. These hazard ratios were remarkably similar to those from the Geisinger Health System, considering the many differences in drug characteristics, samples, design, confounder controls, and analyses. Note that as in the Geisinger Health System study, much of the mortality was associated with early deaths after limited doses of hypnotics, perhaps as little as one-two prescriptions filled or refilled.

Norwegian Pharmacy Database

A recent representative study of the Norwegian Pharmacy Database found that benzodiazepine-receptor-agonist use was associated with a mortality odds ratio of 2.30 (2.20–2.40)²⁰. The authors argued that terminal illness caused an upturn in benzodiazepine-receptor agonist use shortly before death (which might be appropriate for hospice care), and therefore they argued that the increased benzodiazepine-agonist use among those who would die was demonstrated as a confound of terminal illness. To the contrary, their data demonstrated an excess of benzodiazepine use even among those who would not die until 22 months or later, so the benzodiazepine use of this population was elevated before the terminal upturn in hypnotic usage that the authors had demonstrated. Also, the upturn in death-associated hypnotic use 6–10 months before subsequent death might be consistent with a causal lethal hazard resulting from only a few short months' exposures to hypnotics. The Norwegian Pharmacy Database did not enable this study to identify terminal illnesses, to analyze comorbidities or to control for other confounders.

The Palmaro et al. study

In this large study, both French and British case-control samples were drawn from reasonably-representative national samples¹⁶. Results had many similarities to those of Weich et al.¹⁹ despite numerous differences in statistical design. Substantially lower overall hazard ratios were found in the French sample (not all significant after adjustment), perhaps because a large number of occasional users were included. An important finding was the much higher hazard ratios associated with the initial 3–6 months of hypnotics-benzodiazepines use, as high as 11.12 (95% CI, 9.91–12.47) for the 3-month analysis of the British sample. This sharpened the evidence, also noted in the three previous studies discussed, that although dose-response is observed over several years, much of the hypnotic-associated hazard is observed during the early months of usage after as little as one or two prescriptions.

It should be noted that Weich et al.¹⁹ and Palmaro et al.¹⁶ found significant hazard ratios associated with diazepam and other benzodiazepines that are not considered hypnotics (though tranquilizer benzodiazepines may often be used for sleep). These more modern data with better drug identification and measurements of prescriptions during follow-up must be considered more reliable, but neither “Valium” nor “Librium” had been associated with excess mortality in the previous large U.S. CPSII study²¹. One might argue that if diazepam has a different hazard from temazepam, for example, this specificity tends to bolster the evidence for causality with temazepam. On the other hand, it would not be clear if the specificity is in the drug’s pharmaceutical effects, in its absorption and half-life, in its usual time of administration, in other aspects of frequency and dosage of administration, or in various associated confounders.

Note that these epidemiologic studies had many limitations¹⁴. However, the limitations that would tend to bias the results towards underestimating the associations of hypnotics and mortality appeared more influential than those that would bias towards overestimation of the risks. In particular, studies with the most careful efforts to control for confounders found that such control made little difference in the estimated risk ratios, but the hazard ratios in these carefully-controlled studies tended to be higher. The risk ratios derived, like the studies themselves, were extremely heterogeneous, probably due to differences in the size, age, gender, and ethnicity of samples and their health status, the nature of the hypnotics studied, the accuracy with which the drugs involved and their dosages were known, the control variables available, and the duration of follow-up observations to ascertain mortality. Meta-analyses attempting to group 40 such heterogeneous studies would not be clarifying.

Short-term hypnotic use is unsafe

Data provided by Palmaro et al.¹⁶ and Chung et al.¹⁷ expanded the hints in the other large epidemiologic studies^18–20 that short-term hypnotic usage has surprisingly high risks: apparently short-term hypnotic use has higher risks than long-term usage on a per dose or per-unit-time basis. It is logical that for a patient with an “overdose” of common contributory factors such as aging, obesity, sleep apnea, alcohol overuse, and opiate use, even a single hypnotic dose could be lethal on the first night of consumption. Depending on the drug and the patient’s metabolic capacity, the hypnotic drug concentration in blood could increase for the first few consecutive nights of dosage, but eventually, developing tolerance might make each dose less risky among those who had survived the initial doses. There would continue to be deaths at a lower rate after tolerance develops because of hypnotic dose-escalation in response to tolerance, addition of other sedatives or opiates, especially-heavy pre-sleep alcohol consumption, body position, altitude, upper respiratory infections, and other contributing factors that could suddenly produce hypnotic lethality even after several years of steady consumption. In addition, new consumption of non-sedative drugs that impair liver drug metabolism and even foods such as grapefruit can suddenly make a patient more vulnerable to a customary dose. Understanding these considerations, limitation of hypnotic prescribing to a small number of doses or a single prescription cannot be considered safe.

Are insomnia and depression explanatory confounders?

Several reports carefully examined insomnia and depression as potential confounders of the association of hypnotics with mortality, finding that insomnia and depression could explain little if any of this association^19,22–24. Note also that the evidence does not permit us to assume that causality between insomnia, depression, and hypnotic usage is a one-way path when contemplating confounder control, as there is reverse causality^25,26.

Summary of mortality risk epidemiology

Altogether, the epidemiologic literature is conclusive that hypnotic use is associated with excess mortality. The better studies tend to show very high dose-response risk ratios suggesting association with a very large number of deaths. A supplement to the Geisinger Health System data showed that the risk ratios demonstrated lead to estimated U.S. deaths associated with hypnotic usage of the same order of magnitude as those associated with cigarette use, around 300,000–500,000 per year¹⁸. Evidence has been presented from several independent studies that most of these deaths cannot be attributed to known forms of confounding, and indeed, adjustment for the major confounders such as smoking and comorbidities produced little change in the estimated associations in most of these studies. Authors acknowledge that their estimates of adjusted association of hypnotics and mortality could be influenced by inadequate ascertainment of confounding factors or lack of control for a very large number of potential confounds with small or rare effects. It is because skeptics may question whether the strong associations of hypnotics with mortality are causal, despite data fulfilling the Bradford Hill criteria for inferring causality¹⁵, that large post-marketing controlled trials of vulnerable patients may still be needed¹⁴.

Hypnotic drugs have a long history of delayed recognition of mortality risks

Despite its well-known risks of lethality, pentobarbital was nevertheless for decades a preferred hypnotic routinely prescribed for patients seeking sleep aids. In the U.S., today, the most notable human application of pentobarbital is in implementing the death sentence. Although it has been believed that the more modern benzodiazepine and benzodiazepine-receptor-agonist hypnotics that replaced barbiturates have higher acute margins of safety and therefore lower risks than pentobarbital, death certificate and epidemiologic data do not confirm that the newer drugs are significantly safer than barbiturates in routine use^14,27.

Hypnotics produce an excess of deaths at night

In the first Cancer Prevention Study, the percentage of deaths at night were found to be increased by 15.6% among those taking hypnotics (P=0.01), presumably due to respiratory suppression²⁸. In that study, the higher percentage of excess deaths at night associated with taking hypnotics accounted for about one third of total excess mortality associated with hypnotics. These nocturnal deaths were attributed to other causes, even though quiet respiratory suppression as a cause would explain the higher percentage of nocturnal deaths observed among those taking hypnotics than among controls.

The mechanisms of dangerous hypnotic respiratory depression are well-understood. The common hypnotics including barbiturates, benzodiazepines, the “Z” drugs and other benzodiazepine-receptor agonists bind to gamma-aminobutyric acid (GABA) receptors. These ligands-agonists alter the configuration of the receptors to allow negative chloride ions to more readily enter the neurons, where the chloride negatively hyperpolarizes the membranes and inhibits the neurons from firing. When they depress neural respiratory center firing, such drugs can acutely suppress respiration and in large enough dosage, or when individuals are particularly sensitive, may effectively arrest respiration, which leads rapidly to cardiac arrest and consequent death^27,29,30. Respiratory depression is accordingly, and accurately, listed among zolpidem’s warnings and precautions³¹. The barbiturates and alcohol bind to different locations on GABA receptors, where they exert additive or perhaps synergistic respiratory depression effects which may add to benzodiazepine-agonist effects³². An antihistamine, diphenhydramine, also binds to GABA_A receptors, but it does not seem known whether the actions of diphenhydramine on GABA receptors are similar to benzodiazepines. Opiates bind to mu (μ) opioid receptors on respiratory neurons, where they hyperpolarize neural membranes by opening potassium channels³³. Thus opiates, benzodiazepine agonists, and alcohol have additive or synergistic effects inhibiting respiratory neurons³². Hypnotics inhibiting respiration would be expected to produce quiet deaths at night.

Human clinical trials strongly suggested that hypnotics cause cancer

A compilation of randomized controlled trials of hypnotics showed 12 cancers or tumors of uncertain malignancy reported among participants randomized to a hypnotic, but none (zero) among those randomized to placebo (P=0.032, two-tailed Fisher Exact Test)⁴⁹. When the FDA repeated this audit of their controlled trials data, they counted 13 cancers among those randomized to hypnotics versus none (zero) from placebo⁴⁹.

The controlled-trials compilations described above did not include indiplon, an unlicensed zaleplon-like benzodiazepine agonist and hypnotic, for which studies published subsequently indicated three incident cancers in the indiplon groups and none in the randomized control groups^50,51. The compilations did include cancers associated with the marketed hypnotic ramelteon that admittedly has a very different molecular mode of action from the benzodiazepine agonists.

The FDA was not persuaded that these human controlled-trials data required regulatory action, because most of the definite cancers were only minor skin cancers, because of heterogeneities in the data, and because the cancers were recognized after such short randomization periods. Nevertheless, the controlled trials data suggested more than skin cancer. There were cancers of organs apart from skin noted among those treated with hypnotics but none among those randomized to placebo. Reconsideration of FDA's deferral of action is now encouraged by new animal testing and new epidemiologic findings: over half of the research referenced in this manuscript appeared after that FDA deferral of action.

Because hypnotics seem to cause cancers to be suddenly recognized during short clinical trials, e.g., from one month to one year, the short-term effects are likely to arise more from hypnotics promoting progression of tiny pre-existing cancers rather than from effects upon microscopic cancer initiation. Such progression may cause a cancer death, whether or not the hypnotics initiated the cancer.

Animal studies proved that hypnotics cause cancer

The animal data in the FDA files for zolpidem indicated that increasing doses of zolpidem fed to rats resulted in increasing numbers of renal liposarcomas and lipomas combined (statistically significant). These data also showed increased thyroid follicular adenomas and carcinomas combined, and increased testicular interstitial cell adenomas, but the latter findings did not reach statistical significance⁵². There were no such tumors – that is, zero tumors – in the placebo groups. These studies were too small, however, to have substantial power for these neoplasms. Expert FDA pharmacy examiners interpreted the data as suggesting an unknown degree of cancer risk for humans.

These experiments, which showed tumors resulting from feeding zolpidem to rats and suggested a dose-dependent relationship, apparently were never extended, clarified, published, or otherwise followed up.

Similarly, the animal data used for eszopiclone evaluation relied largely on old zopiclone data, since eszopiclone is roughly 50% of zopiclone, and eszopiclone is thought to be the active isomer. Along with other issues, the animal evidence that zopiclone caused animal cancers was of great enough concern to FDA’s scientists, that at least five FDA scientists and medical officers recommended against approval of eszopiclone⁵³. Tumors of the lung in rodents were of special concern; these findings also anticipated the human-specific association of hypnotics with lung and esophageal cancers, as will be described below. Eszopiclone was nevertheless approved as a hypnotic.

Since zolpidem and eszopiclone were evaluated, much additional evidence has appeared relating hypnotics to cancer. Amerio et al. systematically surveyed FDA records including much animal data not included in the earlier compilation of hypnotics trials and concluded that hypnotics and sedatives had among the most elevated cancer hazards among psychotropic drugs⁵⁴.

In vitro studies strongly suggest that hypnotics cause cancer

Hypnotics can damage chromosomes. Zopiclone, zaleplon, and ramelteon are clastogenic^53,55,56. Clastogens are potentially mutagenic agents that induce disruption or breakages of chromosomes. This process can lead to carcinogenesis. Cells that are not killed by the clastogenic effect may become transformed to cancer⁵⁷. One of the several formulations of zolpidem was said from in vitro studies not to be clastogenic⁵⁸. Other than the four drugs mentioned, no information has been located that other hypnotic drugs found to be associated with cancer have ever been adequately tested for clastogenicity. Clastogenicity is one mechanism by which hypnotics are likely to be carcinogenic, through either initiating cancers or promoting progression through additional mutations of cancer cells, or both.

The alterations of immune surveillance produced by benzodiazepine agonists, discussed in relation to infection above, suggest additional mechanisms by which cancer initiation and progression might be facilitated or disinhibited⁵⁹. Hypnotic-initiated increases in infections and consequent inflammation is another potential carcinogenic mechanism. These animal-demonstrated and in-vitro mechanisms for carcinogenicity of hypnotics, that have been widely ignored, support evidence that hypnotics cause human cancer.

Human epidemiology studies demonstrate elevated cancer incidence associated with hypnotics

A 2008 paper⁴⁹ listed three prior epidemiologic studies reporting associations of hypnotics with cancer deaths^12,60,61. Analysis of CPSII data found that the elevation in deaths associated with hypnotics was comparable to that associated with cigarettes⁶¹, though not entirely due to cancer. Mallon, Broman, and Hetta found a much higher cancer adjusted hazard ratio for habitual sleeping pill use of 5.3 (95% C.I. 1.8–15.4) than for smoking among males; none of the specific causes of death were individually significant among females⁶⁰. A similar result was shown in a later paper for males, but the simple significant mortality elevation of regular hypnotic use among females was lost after multivariate adjustment in the second study²³. More recently, a number of new studies have appeared reporting that hypnotic usage is related to cancer incidence and mortality. Hartz and Ross found a significant association of hypnotic use with melanoma and close-to-significant associations for lung and breast cancers⁶². Kao et al. found a remarkable 6.24 (4.13–9.43, 95% CI) hazard ratio for cancer incidence among those using at least 300 mg of zolpidem per year without other-benzodiazepine consumption (this would correspond to slightly more than one 5 mg dose per week)⁶³. In this Taiwanese national study, smoking and body mass index (BMI) were not controlled, but the overall cancer hazard ratios for zolpidem users were almost identical among men and women, despite an almost 11-fold greater prevalence of smoking among adult men compared to Taiwanese women at the time⁶⁴. BMI was not controlled, but at that time in Taiwan, although being overweight was more common among women, obesity was more common among men⁶⁵. In a complementary study of benzodiazepines in Taiwan, benzodiazepines were associated with a 1.19 (1.08–1.32 95% CI) cancer incidence hazard ratio, with over twice the benzodiazepine-associated hazard among men as among women⁶⁶. Similarly, a brief analysis of the national data from Taiwan found a significant cancer adjusted odds ratio for two of three benzodiazepine hypnotics⁶⁷.

In the Geisinger Health study using electronic medical records, Kripke et al. found a hazard ratio for cancer incidence of 1.35 (1.18–1.55 95% CI) associated with use of >132 hypnotic doses per year, with specific hazard ratios of 1.28 (1.03–1.59) for high-dose zolpidem and 1.99 (1.57–2.52) for high-dose temazepam¹⁸. There was a significant dose-response. This study was carefully controlled for age, gender, smoking, BMI, and by matching comorbidities among cases and controls. Jiao et al. found no excess of colorectal cancer among those reporting sleeping pill usage <3 times per week versus ≥3 times per week in the Women’s Health Initiative data set⁶⁸, a result consistent with the Hartz and Ross report on the same data set⁶², but since the contrast of frequencies of usage was weak and the type and quantity of hypnotic consumption were not determined objectively, the negative observation was not very persuasive. We would not expect hypnotics to promote all cancers equally. Indeed, selective specificity among cancer types would be anticipated if the mechanisms are causal. Pottegard et al. and Sivertsen et al. found small but significant associations of hypnotic usage with cancer, especially lung cancer^69,70, but since they had not controlled for cigarette smoking, both groups thought their result might have arisen from confounding, albeit confounding was not conclusively demonstrated⁷¹. That investigators failed to control for important confounders is not proof that confounding explains the significant hazard. Several U.S. and European groups^69,70 and also Kao et al.⁶³ found high hazard ratios for lung and esophageal tumors, and the two San Diego studies had carefully controlled for smoking^18,61. We had proposed that effects of hypnotics on weakening the gastro-esophageal sphincter and permitting more gastro-esophageal regurgitation⁷² might account for the high cancer-specific rates of esophageal and lung tumors¹⁸. These multiple studies finding hypnotics associated with human lung cancer were consistent with concerns of FDA scientists about lung cancers found in animal studies of zopiclone. The lung cancer specificity supports causality.

There was one pair of studies that was neither clearly confirmatory nor negative. A large-scale survey screening many drugs with a questionable scheme for reusing controls for multiple tests and incorporating a questionable 2-year drug-to-cancer lag remarked no significant association of cancer with temazepam or zolpidem but did find significant associations with oxazepam and perhaps lorazepam, using P<0.01 and relative risk >1.50 as criteria⁷³. In that study, it was not always possible to control for smoking, and control for other confounders was crude and not well-standardized. A similar study added a possible association for phenobarbital⁷⁴.

To summarize the cancer epidemiology, the available clastogenicity data, animal data, randomized placebo-controlled clinical trials, and human epidemiology studies consistently, if not always conclusively, suggested that hypnotics likely cause human cancers and cancer deaths.

Obesity and aging exacerbate hypnotic risks

Obesity and aging are perhaps the two most important risk factors for sleep apneas, that is, brief cessations of breathing during sleep¹²⁴. Sleep apneas occurs at least a few times per hour in the majority of adults over age 40 years and in a great majority of those over age 65^124,125. If the duration of a sleep apnea before arousal becomes excessively prolonged, e.g., by a hypnotic, death could result. Thus, hypnotic-related hazard ratios are higher among obese patients (see Geisinger Health study supplement Tables 2 and 7¹⁸.) Since there is no evidence that the huge increase in hypnotic hazards among obese patients can be attributed overdoses, it appears that obesity predisposes to covert hypnotic-related deaths, probably by prolonging apneas. It is plausible that among susceptible patients, combinations of aging, obesity, sleep apnea, hypnotics, opiates, other sedatives, and alcohol could produce quiet respiratory cessations followed by cardiac cessation and death even without any ingested doses above common medical practice being taken.

Prescription and non-prescription opiate use increase hypnotic risks

The use of opiates has become increasingly common in recent years¹²⁶. Opiates are respiratory suppressants that (like pentobarbital) in overdose can produce respiratory arrest and cardiac arrest. Among patients taking both benzodiazepines and opiates, a remarkable 75% were found to have sleep apnea, and causality was suggested by significant dose-response correlations both for the opiates and for the benzodiazepines¹²⁷. In some patients, this combination of benzodiazepine and opiate causes hypoxemia (low oxygen)¹²⁸. Our sleep clinic has recorded polysomnographic data from patients who suffered profound almost continuous apnea with severe hypoxemia due to combinations of hypnotics and opiates. Recall that it is understood on a molecular level how benzodiazepine agonists and opiates combine to suppress firing of respiratory neurons that are necessary to breathe. Patients receiving a combination of benzodiazepines and opiates have increased mortality^129–131. The combination of opiates and benzodiazepines has caused a growing overdose problem in emergency rooms¹²⁶. Moreover, the most serious overdose problems are seen when opiates and benzodiazepines are combined with alcohol in older patients, reflecting combined effects of opiate, benzodiazepine, alcohol, and aging^9,131.

It may be relevant that close to 70% of hospice patients were taking an opiate and an anxiolytic or hypnotic in the last week of life¹³². This is not evidence by itself whether this combination influences the survival of hospice patients, nor is the author commenting on the ethics of combining such drugs in a genuine hospice situation. However, most patients given hypnotics and opiates have not consented to hospice management.

Quiet deaths from hypnotics with contributory factors go undetected by medical examiners and unrecorded

In combined-sedative deaths, the individual drug concentrations present in blood may appear within customary therapeutic ranges. Even if a patient is undergoing cardio-respiratory monitoring at the time when respiratory cessation followed by cardiac cessation occurs, there is usually no way of determining whether the fatal respiratory cessation was due to hypnotic drugs in combination with various contributory factors. Especially when death occurs quietly at night (for example, death of an elderly obese patient known to have various comorbidities,) there usually is no autopsy. Physicians signing the death certificates may be tempted to list a cardiac event or a stroke or some long-standing comorbidity as the cause of death without recognizing when hypnotic-induced respiratory suppression was the precipitant.

The press described a highly-distinguished example of how cause-of-death data may be unreliable after U.S. Supreme Court Justice Scalia died unexpectedly at night. According to numerous news reports and sheriff’s documents, Justice Scalia’s appearance was that of a person who had peacefully stopped breathing at night. There was no sign of agitation due to cardiac pain, nor had Justice Scalia complained of cardiac symptoms before going to bed. Justice Scalia might have been taking hypnotics and opiates for the jet lag and pain he was known to be suffering when he arrived at a hunting lodge that routinely gives each guest a free bottle of wine. Without ever viewing the deceased or his bedroom, much less determining what hypnotics, opiates for pain, and alcohol Justice Scalia might have consumed, a local official was guided by Justice Scalia's physician (thousands of miles away) to declare heart attack as the cause of death. Without an autopsy, we will never know if this death was precipitated by hypnotics or opiates and alcohol or if there was a heart attack. Even if a physician suspects that a hypnotic had a role, the physician has little motivation to suggest the hypnotic as a cause of death when it would be hard to prove and may reflect negatively on the physician prescribing that hypnotic.

Along the same lines, when hypnotics cause infection, cancer, depression, falls or other accidents, or murder, hypnotics are rarely listed among the causes of death. These patterns along with quiet respiratory deaths may explain why epidemiology shows much higher risks of death associated with hypnotics than the death certificates document. Nevertheless, even the numbers documented in death certificates are too high to be acceptable.

Commonly-prescribed hypnotics are used in unsafe combinations

Zolpidem, reportedly the most commonly-prescribed hypnotic in the U.S., with an estimated 40 million outpatient prescriptions in 2013¹³³, ranked first for emergency department visits among psychotropic drugs according to CDC data^133,134. According to the Agency for Healthcare Research and Quality (AHRQ) data, 68% of zolpidem patients were sustained users (three or more prescriptions), and of those 22% were also sustained users of opioids¹³³. Note that recent CDC guidelines recommend against use of benzodiazepine agonists with opiates¹³⁵. Although the FDA had recommended that women use only 5 mg or 6.25 mg of zolpidem, only 5% of women and 10% of elderly were prescribed these low doses¹³³. Moreover, 23% of patients with sustained use took another drug targeting the same receptors. A high percentage were depressed, as indicated by 34% of sustained zolpidem users also receiving antidepressants¹³³. Similarly, a 1999–2010 compilation of NHANES data found that 48% of those taking an insomnia medication were ≥60 years of age¹³⁶. Moreover, over half of those who took a pill for insomnia in the past month were alcohol users (most moderate or heavy users), 56% took other sedatives, and 25% used opioids. In the NHANES data, only a minority of the sedatives taken for sleep were insomnia drugs, but most of the remainder were other benzodiazepines. Recall also that the American Geriatrics Society recommended avoidance of any use of hypnotics for elderly patients¹²³, though about half of those receiving hypnotics have been elderly.

In effect, hypnotics may cause death from overdoses of contributory factors and contraindications along with the hypnotic, not from an excessive dosage of a hypnotic considered by itself.

Popular prescribed hypnotics fail objectively to increase sleep significantly even at high doses: new guidelines discourage hypnotic use

In an authoritative National Institutes of Health (NIH)-sponsored meta-analysis of controlled trials including unpublished trials¹⁴⁵, Buscemi and colleagues found that although non-benzodiazepine zolpidem-like drugs [“Z-drugs”] shortened sleep onset latency by an average of 12 minutes (9–17 min, 95% CI), according to objective polysomnograms, these hypnotics increased total sleep time by only 11 minutes (-1 to 23 min, 95% CI, NS). That is, these “Z” drugs produced no substantial statistically-reliable increase in total sleep, even at doses higher than currently recommended. Most of the meta-analyzed studies of zolpidem used doses of 10 mg or more (as high as 30 mg)¹⁴⁵, and most of the studies of zopiclone used 7.5 mg doses or more (containing more eszopiclone than any dose approved in the U.S.) The FDA-approved recommended initial zolpidem dosage for most patients is now 5 mg (6.25 mg for the sustained-release form¹⁴⁶.) Zolpidem and zolpidem-like drugs constitute the bulk of the current U.S. hypnotics market. Based on all available clinical studies, these lower doses would objectively increase sleep trivial amounts if at all¹³³. Indeed, the primary zolpidem manufacturer advised the FDA that the 5–6.25 mg dosages were generally ineffective³⁵. The newly-recommended 1-mg dosage of eszopiclone is similarly ineffective^147,148. Patients typically report more increase in sleep than is measured objectively, but even this self-reported “improvement” at above-recommended doses (which is not supported by objective measurement) is a mere 32 minutes (26–38 minutes, 95% CI)¹⁴⁵. The discrepancies between objective and patient-subjective data may be attributable to the amnesic properties of hypnotics, erasing patients’ memories of how much time they are awake in bed. In conclusion, the FDA-recommended doses of the most popular benzodiazepine agonists are virtually ineffective for objectively increasing sleep. Older benzodiazepines are not much more effective.

A new Comparative Effectiveness Review sponsored by the U.S. AHRQ has recently examined the Management of Insomnia Disorder, largely referring to chronic insomnia¹⁴⁹. As a prepublication Peer Reviewer of this report, I was and still remain very critical of its limitation to mainly-subjective data that are known to give a rosier evaluation of hypnotic effects than objective evaluations, its focus on published reports that are known to be commonly biased towards reporting favorable drug results^150,151, and the AHRQ report's incomplete attention to adverse effects. Nevertheless, it was striking that the AHRQ study found that the strongest evidence for treatment efficacy was with the cognitive-behavioral treatment of insomnia. The evidence for short-term efficacy of zolpidem and eszopiclone in high doses was considered less sufficient, and evidence for efficacy of other hypnotics was judged to be almost entirely insufficient. Moreover, by its clinical trial selection criteria, this Review found essentially no evidence for efficacy of the very low doses of zolpidem and eszopiclone currently recommended by the FDA for most patients, because higher doses appeared unsafe to FDA. In short, the AHRQ study presented no reason why hypnotics are needed, since cognitive-behavioral treatment of insomnia is better. The AHRQ Review found evidence for increased adverse effects with hypnotics compared to placebo, including hypnotic adverse effects of concern (their selection of studies highlighted fractures and dementia)¹⁵². This means that patients randomly treated with hypnotics tended to develop more illness and symptoms, quite the opposite of promoting health. The Review found mention of adverse effects virtually absent for the cognitive-behavioral treatment studies¹⁴⁹. Although the Comparative Effectiveness Review found insufficient studies to estimate the comparative effectiveness of hypnotics versus cognitive-behavioral treatments, when it reviewed potential harms, there was no contest. Moreover, controlled trials reviewed above prove that hypnotics cause comorbidities such as infection and depression and driving impairments, whereas cognitive-behavioral treatment has been found to decrease medical comorbidities such as depression⁷⁶.

The weak evidence for benefits of hypnotics came mainly from carefully selected groups of patients with diagnosed insomnia and few if any comorbidities or contraindications, and who generally did not use opiates or other sedatives or excess alcohol. There are no clinical trials data supporting benefit among patients with multiple comorbidities and contraindications while lacking diagnosed insomnia, but such vulnerable patients are the majority of patients receiving hypnotics.

Derived from the AHRQ report, A Practice Guideline from the American College of Physicians made a still more reserved interpretation of hypnotics’ benefits and risks¹⁵³. This report advised that cognitive-behavioral treatments should always be the initial treatment for insomnia disorder, and if this therapy was unsuccessful, then short-term use of hypnotics would be questionable. This Practice Guideline found the benefits of even short-term hypnotic treatment to be small or trivial and the evidence persuasive for balancing harms. The Practice Guideline did not recommend long-term use of hypnotics at all. Going beyond the AHRQ report, that had not systematically investigated the evidence for severe risks, the Practice Guideline listed depression as a definite risk and cancer and excess mortality as possible risks, listing the evidence for these harms in considerable detail in its supplement. One wonders if the Practice Guideline would have approved use of the particular hypnotics with the most evidence of risks under any circumstances, were the authors aware of the up-to-date severe risk evidence detailed here. The American Geriatric Society and American College of Physicians guidelines were apparently written by experts without substantial financial conflicts.

In contrast, the first author of the 2017 American Academy of Sleep Medicine (AASM) Guideline for Pharmacologic Treatment of Insomnia¹⁵⁴ had previously announced an AASM public awareness of insomnia campaign to “partner with a consortium of industry, from which we anticipate (and have already received) considerable financial support¹⁵⁵.” The latest AASM Guideline did not disclose whether the AASM was receiving financial support from the hypnotics industry during its preparation, but certainly some of its authors were. Despite conflicting interests, the AASM acknowledged that the evidence for efficacy of any hypnotic was “WEAK” or worse¹⁵⁴. The AASM meta-analysis had a biased focus on sleep benefits of hypnotics. The AASM suggested use of the most popular hypnotics with which this review is concerned, even though conceding that they had insufficient evidence to evaluate the less frequent but more serious harms on which this review is focused.

Hypnotics fail to improve next-day performance or general health

Based on manufacturers’ advertising, patients expect that a hypnotic will improve their function and performance the following day. The truth is just the opposite. In 1982, two sleep experts received support from a hypnotics’ manufacturer to survey the daytime performance literature about hypnotics and found, “Drug-related improvement in performance was not found, and, in comparing active drug to placebo, it is clear that all hypnotics, at some doses, produce decrements in performance the next day⁸⁹.” Since 1982, the current author has been looking for objective evidence that hypnotics improve the performance of insomnia patients. Decades later, no evidence that GABA-agonist hypnotics improve objective daytime performance in treating insomnia could be located. When there are proven significant effects, the effects are to make performance worse^103,156. To reiterate, neither the AHRQ Comparative Effectiveness Review nor the AASM documented objective evidence of health or functional benefits from hypnotic drugs¹⁴⁹. On average, most hypnotics make patients sleepier the next day, not more fully awake.

After 35 years, the author is still looking for any evidence of objective functional benefits. In a recent letter to Sleep Medicine, readers were asked to inform us if “any U.S.-licensed hypnotic ever objectively improved any aspect of insomnia patients’ daytime function or any aspect of general health¹⁵⁷.” So far, nobody has informed me of any such evidence.

Hypnotic drugs are prescribed to patients without valid clinical indication

According to the U.S. National Ambulatory Medical Care Survey, insomnia is a stated reason for a patient’s visit in less than one quarter of office visits where a hypnotic is prescribed⁴, but for most hypnotics, insomnia was the only approved indication. Moreover, no diagnosis of any sleep disorder at all is made on 35% of office visits when a hypnotic is prescribed, and of the 65% of such patients who are diagnosed with a sleep disorder (such as hypersomnia and most forms of sleep apnea), often a hypnotic would be contraindicated⁴. Other studies have likewise found that hypnotics are commonly prescribed for patients who have no diagnosis or complaint of insomnia^{136,141,158,159}. Hypnotics are routinely being prescribed without any apparent valid indication in as much as three quarters of the cases. Similarly, 46% of patients receiving polypharmacy of CNS drugs (such as benzodiazepines and benzodiazepine agonists) had no pain, insomnia, or mental health diagnosis¹⁶⁰. From the data reviewed it appears that in most cases, hypnotics are prescribed despite a specific contraindication such as aging. For example, in the 2015 Beers criteria of the American Geriatrics Society, the hypnotics of concern in this presentation are all listed as drugs to avoid¹²³. It would be fanciful thinking to imagine that addicting hypnotics could be generally beneficial as usually prescribed: that is, without indication or despite specific contraindications.

Manufacturers misrepresent hypnotic benefits in direct-to-consumer advertising

An instructive example is a 2006 advertisement representing that “[eszopiclone] provides a full night of sleep (7 to 8 hours).” An equivalent claim was made in a 2007 eszopiclone-hypnotic print advertisement titled “Sleep the night and seize the day…A better tomorrow begins tonight.” In the scientific study cited by both advertisements as evidence¹⁶¹, the average sleep of patients receiving eszopiclone 2 mg was 382 minutes (6 hours, 22 min) and for 3 mg, it was 412 minutes (6 hours, 52 min). The clinical results cited did not support the manufacturer’s claims to “a full 7 to 8 hours of sleep,” even though the 2 mg and 3 mg doses then studied were greater than the currently-recommended starting doses.

As for the manufacturer’s advertised benefits of “seizing the day,” and a “better tomorrow,” the eszopiclone manufacturer’s study demonstrated no significant objective improvement in measured next-day daytime performance or accomplishment. Specifically, an objective morning performance test did not demonstrate significantly better performance with eszopiclone than with placebo¹⁶¹.

It is not my intention to imply that misrepresentation in consumer advertising has come only from a single manufacturer. There have been many examples with other hypnotics.