What do hypnotics cost hospitals and healthcare?

Hypnotics (sleeping pills) are prescribed widely, but the economic costs of the harm they have caused have been largely unrecognized. Randomized clinical trials have observed that hypnotics increase the incidence of infections. Likewise, hypnotics increase the incidence of major depression and cause emergency admissions for overdoses and deaths. Epidemiologically, hypnotic use is associated with cancer, falls, automobile accidents, and markedly increased overall mortality. This article considers the costs to hospitals and healthcare payers of hypnotic-induced infections and other severe consequences of hypnotic use. These are a probable cause of excessive hospital admissions, prolonged lengths of stay at increased costs, and increased readmissions. Accurate information is scanty, for in-hospital hypnotic benefits and risks have scarcely been studied -- certainly not the economic costs of inpatient adverse effects. Healthcare costs of outpatient adverse effects likewise need evaluation. In one example, use of hypnotics among depressed patients was strongly associated with higher healthcare costs and more short-term disability. A best estimate is that U.S. costs of hypnotic harms to healthcare systems are on the order of $55 billion, but conceivably might be as low as $10 billion or as high as $100 billion. More research is needed to more accurately assess unnecessary and excessive hypnotics costs to providers and insurers, as well as financial and health damages to the patients themselves.


Briefly summarizing the risks of hypnotics
Evidence of hypnotic harms is growing -the American Geriatrics Society has recommended that the popular hypnotic drugs be avoided for older patients, who are almost half of hospital patients 1 . Similarly, the American College of Physicians (ACP) recommended that cognitive behavioral therapy should be the firstchoice treatment for insomnia, and the ACP guideline expressed doubt on whether hypnotics were worth the risks, even as secondary choices for short-term use 2,3 . A 2017 American Academy of Sleep Medicine (AASM) guideline reiterated that cognitive-behavioral therapy is the first-choice treatment for insomnia. For pharmacologic treatment of chronic insomnia, AASM gave "WEAK" positive recommendations for use of certain hypnotics, but qualified those recommendations by conceding inadequate evidence concerning the severe risks of hypnotic drugs discussed below, by conceding that their recommendations were limited to insomnia patients without much comorbidity (probably the minority), and by noting that their recommendations might not be applicable to payer perspectives 4,5 .
The more severe risks of hypnotic drugs are rarely recognized.
• Randomized controlled trials show: a) Hypnotics increase incidence of infections, with a mean 44% increase in controlled trials 6 . Moreover, infections are associated with depression and suicide 7 . b) Hypnotics increase the incidence of major depression by more than two-fold 8 .
• Death certificates list hypnotics and other benzodiazepine agonists among causes of as much as 1 out of 3 U.S. opiate overdose deaths and may be present in about half of suicides 7 .
• Epidemiologic studies demonstrate more risks associated with hypnotics: a) In-hospital falls, e.g. over 3 times as many falls have been observed among patients receiving zolpidem 9 . Outpatient falls are also increased.
b) Hypnotic use is associated with up to double the motor vehicle crash rate 10 .
c) Emergency room visits related to hypnotic ingestions have been increasing 11 . d) Rates of specific cancers, especially lung and esophagus, have multiplied among hypnotic users 7 .
e) According to electronic records systems in 5 countries, overall mortality appeared increased 2-fold to 4-fold among patients receiving hypnotics, after adjustments for comorbid risk factors and confounders 7 .
Hypnotic harms to patients have been documented in more detail elsewhere, with critiques on the strengths and limitations of such evidence 7,12 . It is important to recognize that causality of some of these harms has been supported by randomized controlled trials or death certificates, and for other associated harms, demonstrated association may not be sufficient evidence for causality because of confounding comorbidities. Here the focus is on estimating financial harms to health systems even when lacking precise estimates of the hypnotics-caused components of these harms.
What has been missing from current documentation is a detailed report on the cost of hypnotics to hospitals, insurers, and managed care, where the minimal benefits are weighed against the severe harms. Factual economic data have been so sparse that we must use fragmentary evidence and some speculation to estimate how much hypnotics cost the U.S. medical systems. Additional studies are needed before precise cost estimates can be made.

The benefits of hypnotics are trivial or absent, without documented cost savings
An authoritative systematic review biased towards hypnotics, limited to subjective outcomes of outpatient insomnia patients, restricted to published controlled trials, and including studies of greaterthan-recommended doses, found low-strength evidence for weak benefits for two "Z" hypnotics used mainly in doses higher than recommended 3,7 . Insufficient evidence of any benefit was found for the other benzodiazepine agonists 3 . Moreover, the authors stated that, "it is not known how many minutes' change in SOL, TST, or WASO indicate clinically meaningful improvement 3 ." In other words, it is not known if the weak benefits reported were clinically meaningful even at high doses that are considered unsafe. An older definitive review of objective polysomnographic data that included data from unpublished trials concluded that hypnotics produced

Amendments from Version 1
Version 2 responds to the advice of both reviewers by more carefully highlighting the differences between harms shown to be (at least partly) caused by hypnotics versus those harms inferred mainly from epidemiological evidence of association with hypnotics. Hypnotics are confirmed as being a partial cause of harms demonstrated by randomized controlled trials or recorded in death certificates as causes of death. Harms inferred from epidemiological association might or might not be caused in part by hypnotics, since even in studies controlling as much as possible for possible confounding by comorbidities and other factors, it is impossible to exclude all potential confounding factors of associations. In comments responding to each review of Version 1, the evidence for causality is explained at some length, but even more detail and documentation can be found in several of the references available by quick links. Several of the references also discussed the counterbalancing dangers that epidemiological association may underestimate the harms that hypnotics cause. In any case, the author and reviewers agree that much more research is needed. Hopefully new cost research will be stimulated by this presentation.
In commenting on the reviews, the author also emphasized that clinical decisions need to be made almost daily by physicians and care administrators and payers, even without waiting for more research. That is why the author attempted to present the best data that could currently be assembled, even while frequently reminding readers of the limitations.
Several additional references have been added in Version 2.

REVISED
little or no objective improvements in total sleep in currentlyrecommended doses and no verified overall health benefits 7 .

Available information about the cost of hypnotic harms to hospitals and healthcare
Up to now, medical literature has projected costs of insomnia harms but has hardly mentioned what the harms from treating with hypnotics may cost. The presence of insomnia is obviously confounded by association with prescription of hypnotics, though not as closely as one might expect 7,12 . In a study of 55,000,000 managed care patients, only 31% of those receiving hypnotic medication had a diagnosis of insomnia 13 . The fraction of insomnia patients receiving hypnotics is quite variable depending on the patient samples and definitions of insomnia. Another complication is that using hypnotics may actually cause insomnia 14 , at least following hypnotic withdrawal. For example, in the long run, CBT-treated patients tapered off zolpidem slept better over time than a contrast group randomly provided with continuing zolpidem 15 . Consequences of insomnia such as absenteeism, automobile crashes, and increased medical costs were estimated to be costing the U.S. over $15 billion in 1993 16 . Several more recent cost estimates have been far higher, as high as $216 billion in 2015 dollars 17 . These studies generally made little attempt to differentiate costs caused by insomnia itself from costs of confounding comorbidities and correlated hypnotic harms 18 . Some insomnia cost studies were sponsored by hypnotic manufacturers or others with interests in attributing the costs to insomnia.
Several studies have attributed costs associated with hypnotic prescribing to insomnia, ignoring that less than half of the prescriptions are given to patients with diagnosed insomnia 13,19 . Moloney et al. found that due to recent "medicalization" of sleeplessness, from 1993 to 2007, hypnotic prescribing grew much more rapidly than diagnoses of insomnia, so that in physician office visits, hypnotic prescribing grew to over 3 times the rate of sleeplessness-related complaints 20 . One study used a prescription claim for a hypnotic as an explicit marker for insomnia, in order to compare cohorts with and without insomnia among 87,461 depressed patients. Hypnotic use was associated with more comorbidity-adjusted hospitalization, more frequent ER visits, 12-month healthcare costs that were $3,918 higher, and more short-term disability 21 . For the authors to attribute these cost correlates of hypnotic prescription claims to insomnia (or underlying depression) and not to the hypnotics themselves seemed illogical. A similarly flawed study of a national sample of insured workers found yearly health costs were $936 higher among those with insomnia, but 2/3 of the insomnia cohort were defined by receiving hypnotics without having received a recorded diagnosis of insomnia 22 . Another nationwide study found that insomnia was correlated with prolonged hospital stay, but lacked data to determine whether length of stay was more closely correlated with hypnotic use rather than with insomnia diagnoses 23 . A study of insomnia patients both before and after treatment versus controls found an 85% increase in health costs of insomnia patients treated with sedatives/hypnotics as opposed to insomnia patients that were not treated with hypnotics 24 . The authors attributed this difference to more serious underlying conditions amongst those treated, without considering the possibility that the treatment itself was increasing costs. A study in Taiwan found that in contrast with a cohort without insomnia that did not use sedatives or hypnotics, a comorbidity-matched cohort with a diagnosis of insomnia suffered more acute myocardial infarction and stroke, but only among those taking hypnotics or other sedatives 25 . This may suggest that after control for insomnia, it was the sedative/hypnotics causing myocardial infarctions and stroke. Other studies relating insomnia to health care costs have explicitly found greater healthcare costs among those given prescription treatment for insomnia 26,27 .
A Mayo Clinic study found that hospital patients who received zolpidem had a 2% longer length-of-stay (not statistically significant), possibly due to their triple hazard of falls 9 . Although a 2% average increase in length of stay might appear small, such small mean increases would cost billions of dollars if pervasive throughout the United States. Another study found that in-hospital benzodiazepine prescriptions were associated with 23% higher readmissions 28 . We must recognize that without randomized placebo controls, none of these studies can offer definite proof on whether hypnotics or insomnia cause increased health costs.
It is ironic that several of the studies that document hypnotic prescriptions as being associated with increased healthcare costs were sponsored by hypnotic manufacturers, when they had intended to attribute these costs to insomnia.

In-hospital hypnotic cost benefits and risks
Fifty years ago, it was routine to prescribe an "as-needed" hypnotic with almost every hospital admission. In 1982, Perry and Wu reviewed 331 charts of a distinguished teaching hospital and reported that, "Most surgical patients (96%) and a large number of medical patients (46%) had hypnotic agents prescribed on admission without a recorded reason, without the patient's request or knowledge, and without a statement in the medical chart indicating whether the therapeutic objectives were met 29 ." Personal communications indicate that routine hypnotic prescribing without evidence of benefit is still a common practice in many of the most renowned academic medical centers. The recent Mayo Clinic report is an example that listed only 32% of patients given zolpidem as having an insomnia diagnosis 9 .
An up-to-date systematic review of 15 in-hospital controlled trials of hypnotic drugs going back to 1983 found that only one of the included trials (of intravenous dexmedetomidine) showed a convincing advantage for sleep efficiency 30 , even though several of the studies involved such intravenous drugs. Out of the 15 studies, 5 showed some evidence that oral benzodiazepines reduced sleep latency, but most of the treated patients still had abnormal sleep latencies exceeding 30 minutes. The review concluded with "insufficient evidence to suggest that pharmacotherapy improves the quality or quantity of sleep in hospitalized patients suffering from poor sleep 30 ," and no other health or cost benefits were documented.
The controlled hospital trials were not designed to assess the costs of hypnotic harms. Indeed, I know of no formal studies on the health cost of harms produced by in-hospital administration of hypnotics, whether randomized or not. It is hard to imagine how drugs that are known to increase the incidence of infection and depression and are strongly associated with in-hospital falls could fail to increase hospital costs.

Hypnotic risks cannot be justified when given to patients without diagnosed insomnia
As previously mentioned, most prescriptions for hypnotics are given to patients without diagnosed insomnia, even though insomnia is the sole approved indication for most hypnotics.

What more can be learned?
With the recent expansion of electronic medical records, many hospital systems and insurance systems already have sufficient data in their existing electronic records to estimate the outcomes References and costs associated with hypnotics prescribing, including hospital admissions, infections, falls, and incident delirium and dementia, lengths of stay, and readmissions. Such available data could give us a much clearer idea of costs associated with in-hospital hypnotic prescribing, but control for comorbidities and other confounders could not assure an accurate estimate of the causal component of associated costs.
Fortunately, it is becoming increasingly possible to utilize genetic data and "Mendelian randomization" to effectively compare groups who received hypnotics due to random genetic propensities with those who did not 45 . With the increasingly widespread development of personalized medicine, involving genotyping and whole-genome analyses, an increasing number of hospital systems will have accumulated sufficient genetic data to isolate the causal contribution of hypnotics to infection, hospitalization, depression, hospital readmissions, cancer and mortality.
Because of ethical and practical difficulties and manufacturer liability concerns, it appears unlikely that anyone will ever do large enough randomizing hypnotic vs placebo drug trials to demonstrate the costs of hypnotic harms accurately. Fortunately, an alternative randomizing strategy relying on patient choice after education and patient-empowerment has been suggested: such studies might be integrated into the wellness-promotion and costreduction programs of managed care organizations 46 .

Conclusion
It might be years before more reliable data are assembled on the harm costs and cost-benefits of hypnotics. Meanwhile, hospital leaders and managed care and insurance administrators would be wise to infer from available evidence that the costs of hypnotic harms exceed any cost benefits of hypnotics. A decision to protect patients from hypnotics would also protect payer budgets.

Competing interests
The author has no financial interests or conflicts to declare.

Grant information
The author(s) declared that no grants were involved in supporting this work. The author has done considerable work in assembling data relating the use of sedatives and hypnotic drugs to harmful outcomes and increased healthcare costs. The main thrust of this article is, assuming that hypnotic use causes harmful health outcomes, what would be the economic costs of this harm?
However, the question about hypnotic use by itself causing harm and health morbidity is still frustratingly not completely resolved. The author amasses a wealth of evidence associating the two, lending strength to the notion of causality but not confirming it. There is good experimental evidence showing negative effects of hypnotics including slowed reactions, cognitive and memory impairments, impaired balance arising from their sedating effects both before sleep and, for longer acting hypnotics, after sleep. However, the evidence for the direct effects of hypnotics on physical health measures is not as strong since these are largely correlation studies.
The greater increase in health morbidity and mortality in those prescribed hypnotics may arise from the possibility that hypnotics are more likely to be prescribed by health professionals in cases with greater and more serious co-morbidity and those likely to have increased mortality risk. This self selection of hypnotic prescriptions to those likely to have higher morbidity and mortality may account for all or some of the subsequent elevated risks. The author, himself, admits that this may not be resolved without randomized control trials (with some diagnosed insomniacs not receiving hypnotics or receiving non-drug therapies instead). However, it may be difficult to obtain ethics approval for a study allocating insomnia sufferers to a non-treatment control over a long period of time (e.g 2-3 years to measure long term health outcomes). The only opportunity might be to compare pharmacotherapy alone with cognitive/behavior therapy with long term follow-ups of sleep and many health outcomes. This should be acceptable ethically.
The difficulty with drawing causal conclusions from epidemiological studies is nicely illustrated by one of the cited references with some of the most compelling evidence, the study by Anderson, . In their et al study they reported on an insomnia subset of 5,773 of which 75% were treated with a prescription medication and the other 25% were not and all were followed up after a 12 month period. The treated subset was more likely to have a mental health diagnosis and anti-depressant medications. These differences were significant between the groups at baseline which supports the notion that it is the higher morbidity patients more likely to receive hypnotic medication. These initial morbidity differences were then amplified at the follow-up and the medicated group showed greater increases in health costs ($4276 vs $2309) over that time. This is strong circumstantial evidence that the hypnotic medications were causing an increase in medical costs. At least one could say with greater certainty that the medications were not curing their insomnia and reducing health costs. The only lingering doubt, however, is not knowing the outcome had they not been medicated. Would they have been even worse off, no different, or, as Dr. Kripke is suggesting, better off? We simply cannot be certain until randomized controlled trials are conducted.
A crude but good analogy to the point I am making is the effect of hospitalization. Should a 70 year old with a severe chest infection be hospitalized? Those admitted to hospital certainly have a higher morbidity and mortality risk. Does that mean someone with these symptoms will be better advised to stay out of the hospital? In the mean time, the weight of the evidence presented by the author is compelling. My only suggestion is for the author not to slip into the use of terminology that the causal connection is proven. By doing so he is in jeopardy of reducing the strength and consistency of the correlation data. Let the data be presented and just percolate in the reader's mind. It will thus be more effectively taken up than if the author takes the one step too far to insist upon the causal link. Already there is too much research being presented implying causal links between sleep variables (e.g. reported total sleep time) and health outcomes from epidemiological studies. Let's not contribute to this inappropriate trend. Dr. Lack's very thoughtful emphasis on the problem of causality highlights a crucial issue for readers of this manuscript about hypnotic costs. His review contributes to the discussion already in the manuscript calling for more studies including studies of causality. Likewise, the manuscript specifically acknowledged the problem of associational studies potentially being confounded by comorbidities. The problem of comorbidities is emphasized a bit more in Version 2. Until more reliable data become available, the manuscript argues that managed care and insurance administrators should weigh the best available evidence of whether the costs of hypnotic harms are likely to exceed the value of hypnotic benefits.
As in most areas of medicine, timely decisions about hypnotic treatments and reimbursements must be made based on incomplete and imperfect evidence. Decisions need to be made daily and must be made based on incomplete and imperfect evidence. Decisions need to be made daily and cannot be deferred for years more study. We still do not have definitive randomized controlled trials about whether cigarettes cause cancer in humans. After millions of people had died from cancers "associated with" cigarette smoking, it became time to advise against cigarette smoking based on available evidence that association probably indicates a large element of causation.
The reader may be sure that this manuscript on costs never implies causation without causal evidence, but uses the word "association" whenever there is insufficient evidence of an element of causation. Some added emphasis on this distinction is included in the Version 2 revision. This response to both reviews is a welcome opportunity to elaborate on the issues of causation versus association that were not presented at length in a manuscript about costs, since that would repeat much detail previously published.
However, just as it would be an error to slip into language implying causation when only association has been demonstrated, it would be an error not to recognize existing human trials and meta-analysis confirming evidence of causation. Moreover, in the case of hypnotics, as with cigarettes, there is much animal causal data and in vitro data supporting the evidence of association studies. The sleep disorders community has been slow to recognize and replicate the existing findings as the causal evidence of serious hypnotic risks has accumulated. Joya et al. performed formal meta-analysis of 38 randomized controlled trials of 8828 participants randomized to 4 hypnotics and 4383 participants randomized to placebos, in whom the overall infection causation hazard ratio was 1.44 (1.25-1.64, confidence interval, P<0.00001). If any manufacturer could do a meta-analysis showing that this analysis was wrong, would they not have done so? How much more evidence is needed? As a matter of fact, Sanofi subsequently did an independent analysis of their Ambien data and informed the FDA that they could confirm that Ambien (zolpidem CR) causation of infections such as influenza was "probably real." The FDA expert agreed. Would skeptics nevertheless question that hypnotic causation of infections such as influenza creates medical costs and sometimes deaths?
For an invited lecture, Kripke did an informal analysis of incident depressions in randomized trials of 4 hypnotics (including data available through the FDA that manufacturers did not publish). There were 5535 participants randomized to hypnotics with 2% incidence of depression and 2318 participants randomized to placebos with 0.9% incidence, yielding a depression causation risk ratio of 2.1 (P<0.002). If any manufacturer could do a larger, more formal meta-analysis showing that hypnotics do not cause depression, would one not think they would have published? In an important study of 593 participants randomized to eszopiclone and 195 randomized to placebo, the rate of discontinuation for depression was 2.0% in the eszopiclone group and 0.0% in the placebo group, a result that appears statistically significant. In their modesty, the authors did not even mention that the study had proven that eszopiclone causes infection. Note that these were new, incident depressions, not exacerbations, since the study excluded patients with any DSM-IV Axis I psychiatric diagnosis at baseline. Would skeptics nevertheless question that depression elevates medical costs and sometimes causes suicides?
The previous detailed review also discussed at length the many death certificates that mention hypnotics as one of the causes of death and likewise discussed why these are probably underestimates. For 2015, the U.S. CDC's WONDER data base listed over 10,000 deaths with causes of death that included a barbiturate, benzodiazepine, or a Z-drug (the Z-drug category included anti-epileptics), but for many reasons, the 10,000 reported deaths is believed to be an underestimate. Would skeptics nevertheless argue that death certificates listing a hypnotic as a underestimate. Would skeptics nevertheless argue that death certificates listing a hypnotic as a cause of death are evidence for association, not causality?
To summarize, there are certainly adequate data confirming that hypnotics cause some of the harms discussed in this manuscript and some of costs resulting. There are also controlled trials showing that hypnotics cause impaired balance and poor on-road driving performance, though controlled trials with automobile accidents or falls as endpoints have not yet reached adequate size. Further research is needed to better define the magnitude of costs from hypnotic harms, but some causal role of hypnotics in several of the harms resulting has already been confirmed.
Dr. Lack's comments are very helpful in reiterating the need for expanded studies. The problems of ethical approval are not insurmountable. If it is ethical to administer hypnotics to patients in short-term hospitalization, would it not be ethical to perform randomized trials focusing on whether hypnotic prescribing increases or decreases hospital costs? Short-term hospital studies might also provide useful evidence regarding causality of serious infections and depression. Even for early death as an endpoint, short-term randomizing clinical trials could be ethical and have adequate power. Since the epidemiology suggests that the greatest risk-ratio of hypnotics may be within the first 30 administrations, mainly among frail and elderly patients, randomized placebo-controlled clinical trials might only need to last 30 days among frail and vulnerable participants (who currently receive much of the hypnotics prescribed). To be sure, large numbers of participants would be needed for adequate power in relatively brief randomization designs. Another alternative would be the randomized education-patient-empowerment technique initiated by Dr. Tannenbaum. Another alternative would be Mendelian randomization studies. Randomizing studies of causality are feasible if those who advocate and provide hypnotics recognize their responsibilities to assess the risks and costs.
No financial competing interests with any entities that may benefit or lose Competing Interests: financially from this work. No papers or active projects with the author of this work for over 3 years. However, I have worked with the author previously as a trainee.
I have read this submission. I believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.
Author Response 26 May 2017 , University of California, San Diego, USA Daniel F. Kripke Dr. Grandner's has provided several very helpful suggestions which will be incorporated in a revision of the manuscript after the initial reviews. He mentioned two points of interest deserving comment here.
First, there is the question of whether there is evidence that hypnotics cause suicide, going beyond the extensive evidence for association. This is discussed in more detail in my recent more detailed review of the evidence for hypnotic harms: doi: . There 10.12688/f1000research.8729.2 are several kinds of evidence. Controlled trials and meta-analysis have shown that hypnotics cause incident depressions and cause incident infections. Controlled trials are our best tests of causality. There is much new evidence that infection causes depression and suicides, and depression itself is widely believed to be the main cause of suicide. Second, in a narrower sense, when a medical examiner reports a hypnotic drug as a cause of intentional overdose death (and there are thousands of such examples), that is an expert finding of causality. Since no specific costs of suicides and suicide attempts could be included in the hypnotic cost estimates offered, causality of suicide is not a key issue for the manuscript being reviewed. More study is certainly needed. Incidentally, my petition to the FDA specifically mentioned suicides as one endpoint which should be incorporated in expanded Phase IV randomized safety trials of hypnotics.
Second, there is the question of why people keep taking hypnotics. This is a question for the ages, since throughout the history of magical beliefs and shamanism, people have consumed countless remedies including sleep remedies that are scientifically considered non-beneficial. Moreover, benzodiazepine-agonist hypnotics are known to be addicting drugs that are sometimes sought for pleasure, but equally important, these hypnotics cause withdrawal insomnia. There is much evidence that people persist in using hypnotics because of the withdrawal insomnia and anxiety that occur when consumers try to stop. As the reviewer suggested, people may take hypnotics seeking the amnesia that hypnotics produce, just as alcoholics drown their sorrows. That amnesia has not been recognized and advertised as a medical benefit. I am not aware of evidence that hypnotic-induced amnesia is medically beneficial, though it does often elicit a favorable subjective report. In hospitals, there might be an argument for helping patients to forget their anxieties, pains, and discomforts, but if the same amnesic effects contribute to falls, confusion, and dementia, it may be a costly strategy.
Please see my statement of competing interests in the main article. Competing Interests: