Skip to main content

Archived Comments for: Hypnotics and mortality in an elderly general population: a 12-year prospective study

Back to article

  1. Over-adjustment leads to underestimation of hypnotic risks

    Daniel Kripke, UCSD

    14 October 2013

    This excellent paper of Jaussent and colleagues is indeed among the best studies which have been reported concerning hypnotic-associated risks. As the manuscript asserts, an outstanding strength of the study is the number of potential confounders which have been included in the model, some of which were measured with excellent methods. An important limitation is the absence of hypnotics dosage and dose-response data. Much of the statistical approach is admirable. Also, as the manuscript emphasizes, the results¿if accurate--do imply that mortality associated with hypnotics consumption might result primarily from confounding factors.

    Unfortunately, the manuscript seriously underestimated the possibility that overadjustment of intermediate variables could obscure the causal pathway between hypnotics intake and mortality. Several of the covariate measures of morbidity are not independent risk factors, but have been proven by controlled trials to be caused or exacerbated by hypnotics. Prominent among these are depression (and by inference, antidepressant use), impairments of cognition (e.g., Mini Mental State), and excessive daytime sleepiness [1-3]. Depression, impaired cognition, and excessive daytime sleepiness, in turn, are likely contributors to confinement. Considering that 21.7% of participants were taking at least one hypnotic at baseline, the great majority for more than 5 years, it is virtually a certainty that hypnotic usage was a causal factor in these comorbidities at baseline, and that adjustment for these intermediate variables produced a element of over-adjustment. Thus, hypnotic use might have appeared a significant mortality risk if over-adjustment had been avoided.

    There are several sensitivity analyses which might allow us to appraise the importance of over-adjustment. First, what were the mortality hazard ratios associated with hypnotic usage when participants who had ever used hypnotics at baseline were excluded, so that the hazard estimates were based entirely upon prospective usage of hypnotics after baseline comorbidities had been employed to control for confounding? Second, what were the mortality hazard ratios among participants who used hypnotics at all follow-ups before death? Were participants who only took 1-2 DDD per month included as users?

    Perhaps the authors could clarify whether the Cox proportional hazard models were time-dependent models in which the timing of incident comorbidities in relationship to the timing of hypnotic usage might be evaluated. Were stepwise models employed, and if so, what were the criteria for entry or removal of each covariate from the stepwise models? Inclusion of excessive covariates which are not independently significant or collinearities could also cause an element of over-adjustment.

    Incidentally, it is interesting that insomnia complaints were NOT significant predictors of deaths during follow-up, although insomnia has often been claimed to be a confounder of hypnotic mortality studies.

    Because of the complexity of causal interactions between hypnotics, comorbidities, and various behavioral elements, it will never be possible to be confident that an epidemiologic design is safe both from confounding and from over-control. Obviously, hypnotics manufacturers such as the authors¿ sponsor, Sanofi-Synthélabo, have no intention of doing large enough controlled trials to assure us that their products do not cause excess mortality or cancer. Controlled trials are already available showing that certain hypnotics do cause cancers in experimental animals. Perhaps Mendelian randomization studies will prove the only realistic method of being reasonably certain whether hypnotics do or do not cause excess mortality and cancer.

    Reference List

    1. Kripke DF. Greater incidence of depression with hypnotics than with placebo. BMC Psychiatry 2007, 7:42.
    2. Boyle J, Groeger JA, Paska W, et al. A Method to Assess the Dissipation of Residual Hypnotics: Eszopiclone Versus Zopiclone. J Clin Psychopharmacol 2012, 32:704-709.
    3. Johnson LC, Chernik DA. Sedative-hypnotics and human performance. Psychopharmacology (Berlin) 1982, 76:101-13.

    Competing interests

    DFK is a long-time critic of hypnotic drugs via his web site,, now available in Kindle. The Kripke family own shares in a mutual investment corporation which holds a small percentage of its capital in Sanofi-Aventis.

  2. Reply:Comment on D Kripke "Over-adjustment leads to underestimation of hypnotic risks"

    Yves Dauvilliers, Universitary hospital

    24 October 2013

    Jaussent I, Ancelin ML, Berr C, Pérès K, Scali J, Besset A, Ritchie K, Dauvilliers Y
    We are pleased that Daniel Kripke recognizes the value of our prospective study and we would like to thank him for his comment on possible over-adjustment bias in epidemiology. However, another potential drawback of epidemiological studies which has rarely (or not exhaustively) been considered in previous studies is the bias of indication that we also intended to address in our manuscript justifying the design of our analyses. This was only briefly discussed and we agree that this warrants further explanation. To this end, as suggested by D Kripke, we performed a series of additional sensitivity analyses.
    To rule out a potential over-adjustment due to the inclusion in the same model of variables which may not be independent, we performed a multivariate analysis adjusted for center, age, gender, educational level, history of cardio and cerebro-vascular disease, respiratory disease and diabetes mellitus. In this model which did not consider confinement as well as any underlying indication, i.e. depression and antidepressant use, cognitive impairment and excessive daytime sleepiness, the association between hypnotic use and the risk of mortality was not significant HR=1.12 95%CI=[0.98;1.28]. This was even less significant after further adjustment for depression, anxiety and insomnia, which are recognized risk factors for mortality and the main indication for hypnotic use (which cannot be considered as ¿a causal factor of these comorbidities¿).
    In addition, we studied the relationship between persistent use of hypnotics during a period twice as long, i.e. the first 8 years, and all-cause mortality. A total of 1070 (10.8%) reported hypnotic use both at baseline and at all three follow-ups (persistent users), 447 (25.9%) were taking hypnotics at one of three follow-ups (intermittent users) and 2618 participants (63.3%) did not report hypnotic use at any time. The risk of mortality for the last 4 years of the follow-up was not significantly associated with the persistent use of hypnotics (when compared with non-users, HR=1.17 95% CI=0.79-1.73, in the minimally adjusted model 1 only adjusted for center, age and gender) and the same non-significant results were observed after further adjusting for education, history of cardio and cerebrovascular disease, respiratory disease and diabetes mellitus. However this method focusing on a long period (eight years) may be limited by the attrition due to mortality.
    Lastly, we examined the relationship between hypnotic use at the 2-year follow-up and all-cause mortality after excluding participants taking hypnotics at baseline in order to exclude any possible confounding effect with baseline comorbidities. At the two-year follow-up, 587 participants were taking hypnotics and 4172 were not. In this case also, the association was not significant after adjustment for center, age, gender, as well as level of education, history of cardio and cerebrovascular disease, respiratory disease, diabetes mellitus (HR= 1.13 95% CI= 0.92-1.40), suggesting that our results are unlikely to have been strongly affected by over-adjustment (see also the constancy of the HR value regardless of the analysis and adjustment).
    These additional sensitivity analyses thus confirm our findings that in our sample of community-dwelling elderly persons, hypnotic intake may not be significantly and independently associated with an increased risk of 12-year mortality. Of course, we cannot definitely rule out a negative effect of hypnotics on mortality. In the absence of large, time-consuming, and costly randomized controlled trials and given that experimental animals are probably not the most relevant to address this complex question, we think that well-designed prospective studies in general population capable of controlling for indication bias are a realistic and reasonable way to address this question.

    Competing interests

    Professor Dauvilliers has consulted for UCB Pharma, JAZZ, and Bioprojet.