Answers are grounded in this paper and cite their sources.
This was not an industry supported study. The authors have indicated no financial conflicts of interest. We appreciate the opportunity to respond to the critique entitled “Death risk from insomnia unclear” posted in Choices,1 an official site of the National Health Service (NHS) of United Kingdom, and Dr. Kripke's letter2 on our recently published study on mortality and insomnia.3 The NHS reviewers of our study state that an important limitation is that the participants were not randomly selected. Since 1998, many publications resulting from our adult cohort have described in great detail our sampling method and how we corrected for selecting people who, for example, had higher BMIs.4 Indeed, our population is a stratified random sample from a well-defined general population. We have consistently applied weighted analysis in all of our publications of the data generated from this sample. Thus, our findings are applicable to the general population. Another point of criticism is that objective sleep was measured only once in the lab. Indeed, one night in the sleep lab may not reflect the persons' habitual sleep pattern. However, in our previous studies the association between objective sleep duration and markers of the activation of the stress system were based on a 4 consecutive night sleep laboratory protocol.10 The consistency on the role of objective sleep duration as a marker of insomnia severity between the physiological studies with multiple night recordings and the current epidemiological study based on a single night recording increases our confidence about the replicability and generalizability of the present findings. Furthermore, we cautioned in our paper that we used a six-hour cut-off point only for the statistical evaluation of the severity of insomnia and not as the recommended optimum sleep duration for the general population. Finally, our findings on the average objective and subjective sleep duration are identical to that reported by other major epidemiologic studies that have included thousands of subjects such as the Sleep Heart Health Study11 and the CARDIA12 study. The consistencies between our cohort and these studies support the validity of our measurements and the generalizability of our findings. These points are thoroughly discussed in the second to last paragraph in the discussion section. The issue of the large confidence intervals for some of our results is due to the relatively small sample. This does not, however, negate the significance and importance of our findings based on data that were analyzed in the most rigorous statistical way. We state, in the discussion section of our paper, this potential limitation. In our study we controlled for many confounders such as age, race, education, body mass index, smoking status, alcohol use, depression, sleep apnea, hypertension, and diabetes. In fact, we are the only published study to control for sleep apnea based on objective sleep lab recording. We controlled for almost all of the known major factors that could have influenced our results. The reviewers' comment that “insomnia may be related to numerous medical and psychological conditions that can also affect mortality risk” implies that there are other factors that were not included in our model. However, they did not specifically suggest any new confounders for us to consider. We thank Dr. Kripke for acknowledging the strength of our large population-based sample, and the objectively measured sleep duration using the standardized PSG approach. The primary focus of our paper was to examine the mortality impact of insomnia associated with objective short sleep duration. It was suggested that the sleep duration should be analyzed by dividing a continuous variable into three categories, instead of the binary approach that we used in our paper which would have reduced our power. Throughout our paper, sleep duration was analyzed as an effect modifier, a priori. The data presented in Table 3 of our paper clearly suggested that insomnia without objective short sleep duration is not associated with a higher risk of mortality, whereas insomnia with objective short sleep duration is associated with significantly higher mortality risk. If the goal of our paper were to examine the effect of sleep duration on mortality, both approaches would be appropriate and both share the same disadvantages of dividing a continuous variable into a categorical variable. We were asked to summarize the association between sleep duration and mortality in our data. To avoid any disadvantages associated with categorizing a continuous variable, we actually analyzed sleep duration as a continuous variable, which has a median of 5.99 hours (our previous cut-off for short vs. normal sleep duration), a mean of 5.75 hours and SD = 1.23 hours. We entered sleep duration as an original scale and a cubic term to detect any “U” shape association. Table 1 below summarized a model identical to the Model 1 in Table 3 of the original paper. These detailed analyses clearly suggest lack of “U” shape association between sleep duration and mortality in our data, regardless of insomnia status. However, the interpretation among insomnia men should be made with caution, because the sample size for this group is only 55. Regression coefficients of mortality associated with objective sleep duration in men: multivariable adjusted logistic regression model Adjusted for age, race, education, BMI, smoking status, alcohol use, depression, SDB, and sampling weight. Regression coefficients of mortality associated with objective sleep duration in men: multivariable adjusted logistic regression model Adjusted for age, race, education, BMI, smoking status, alcohol use, depression, SDB, and sampling weight. Dr. Kripke suggested that we use a half-hour below the median as our cut-off point to define “short” vs. “normal” sleep duration.13 First, we have chosen the median as a cut-off point based on standard mathematical approaches, i.e., the 50th percentile. Second, in our previous studies based on the same cohort, insomnia associated with lower than the median sleep duration, i.e. five hours, was associated with a stronger risk of hypertension and diabetes.7,8 Due to small sample size, we cannot pursue a similar approach here; however, one can reasonably expect a similar trend. Dr. Kripke suggested that we control for sleep duration divided into three categories to see if insomnia is still associated with mortality. Since sleep duration is a significant effect modifier in our data, it would not be appropriate to control for sleep duration, whether treating it as a binary, ordinal, or continuous variable. The appropriate approach is to examine the insomnia effect at different levels of sleep duration, which is what we presented in our original paper. In regard to the multiple comparisons concern, our a priori hypothesis was that insomnia is associated with mortality only if it is accompanied with objective short sleep duration. This hypothesis is supported by our previous publications on the association between insomnia and diabetes and hypertension.7,8 The entire paper is focused on this central hypothesis, as presented in Table 3. Furthermore, as we have clearly indicated in the original paper, the stratified analysis according to hypertension and diabetes status was “exploratory” (Table 4). Therefore, in our opinion, in both analyses there is no need for adjusting for multiple comparisons. There was an additional concern about our decision to adjust for major confounding factors, and sampling weight. In a model without adjusting for any of the covariables adjusted in Table 3, the odds ratio for mortality due to insomnia and less than 6 hours of sleep is 10.47 (95 %CI 4.23-25.89). Comparing the adjusted odds ratio from the same group in Table 3, the differences in the odds ratios clearly suggested that these covariables are important confounding factors in the data, thus, must be adjusted for to make a valid interpretation of the associations. We have consistently adjusted for the sampling weight in all of our publications from the Penn State Cohort. It is only with the appropriate sampling weight adjustment that we can make inference to the general population from which this sample is drawn, and is one of the main advantages of population-based sampling. Finally, the associations remained the same when we controlled for hypnotic use. We would like to note that our study on mortality and insomnia is the fourth study published by our group within a year on the topic of insomnia and its association with morbidity and mortality. In our three previous papers using the same cohort and the exact same methodology, we have demonstrated that this subtype of insomnia, i.e., insomnia with objective short sleep duration, is associated with a higher risk for hypertension, diabetes, and neurocognitive deficits.7–9 Clearly, what sets our findings apart from previous studies is that we have defined insomnia in a novel way, i.e., combining a subjective complaint and a physiological marker, i.e., objective sleep duration. This novel definition is based on our and other investigators previous findings that this type of insomnia is associated with an activation of the stress system, i.e., hypothalamic-pituitary-adrenal axis and catecholaminergic system,10,14 which is known to be associated with cardiometabolic diseases. We are thankful to the critiques and the Editor of SLEEP who gave us the opportunity to clarify some important issues and hence, to promote scientific dialogue. However, we are disappointed with the conclusion of the NHS critique that “this research does not provide strong evidence that insomnia is linked to an early death, and it sheds no light on possible reasons behind a link.” In our opinion, their conclusion is not based on any evidence, neither from our data or any other published data, and tends to trivialize a significant health problem. The study referred to in this letter was funded in part by the National Institutes of Health grants RO1 51931, RO1 40916, and RO1 64415.
No takes yet. Share an insight, caveat, or question.
Vgontzas et al. (2011) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: