Longitudinal associations between insomnia symptoms and all-cause mortality among middle-aged and older adults: a population-based cohort study

Abstract

To date, there is no scientific consensus on whether insomnia symptoms increase mortality risk. We investigated longitudinal associations between time-varying insomnia symptoms (difficulty initiating sleep, difficulty maintaining sleep, early-morning awakening, and nonrestorative sleep) and all-cause mortality among middle-aged and older adults during 14 years of follow-up. Data were obtained from 2004 through 2018 survey waves of the Health and Retirement Study in the United States for a population-representative sample of 15 511 respondents who were ≥50 years old in 2004. Respondents were interviewed biennially and followed through the end of the 2018 survey wave for the outcome. Marginal structural discrete-time survival analyses were employed to account for time-varying confounding and selection bias. Of the 15 511 cohort respondents (mean [±SD] age at baseline, 63.7 [±10.2] years; 56.0% females), 5878 (31.9%) died during follow-up. At baseline (2004), 41.6% reported experiencing at least one insomnia symptom. Respondents who experienced one (HR = 1.11; 95% CI: 1.03–1.20), two (HR = 1.12; 95% CI: 1.01–1.23), three (HR = 1.15; 95% CI: 1.05–1.27), or four (HR = 1.32; 95% CI: 1.12–1.56) insomnia symptoms had on average a higher hazard of all-cause mortality, compared to those who were symptom-free. For each insomnia symptom, respondents who experienced difficulty initiating sleep (HR = 1.12; 95% CI: 1.02–1.22), early-morning awakening (HR = 1.09; 95% CI: 1.01–1.18), and nonrestorative sleep (HR = 1.17; 95% CI: 1.09–1.26), had a higher hazard of all-cause mortality compared to those not experiencing the symptom. The findings demonstrate significant associations between insomnia symptoms and all-cause mortality, both on a cumulative scale and independently, except for difficulty maintaining sleep. Further research should investigate the underlying mechanisms linking insomnia symptoms and mortality.

Statement of Significance.

Despite the scientific evidence of worsened health and lowered quality of life among individuals complaining of insomnia symptoms, there is no consensus concerning causal associations between insomnia symptoms and mortality. In the present study, we used longitudinal data to examine the associations between time-varying insomnia symptoms (difficulty initiating sleep, difficulty maintaining sleep, early morning awakening, and nonrestorative sleep) and all-cause mortality among middle-aged and older adults during 14 years of follow-up. A Marginal Structural Modeling approach was employed to generate average causal effects of the symptoms on mortality and account for time-dependent biological, psycho-cognitive, and lifestyle confounders and selection bias. We found that insomnia symptoms on a cumulative scale and independently (except for difficulties in maintaining sleep) were associated with higher all-cause mortality. Future studies should explore the underlying mechanisms linking insomnia symptoms and mortality.

Introduction

Insomnia is the most prevalent sleep disorder [1–3]. In the United States (US), up to 50% of middle-aged and up to 75% of older adults report experiencing at least one insomnia symptom on a regular basis [4–6]. Insomnia is a substantial public health issue and imposes a significant burden on individuals, society, and the healthcare system [7]. The total direct and indirect costs attributed to insomnia in the US are estimated to range between 92.5 to 107.5 billion dollars annually [3, 8]. There is growing evidence that insomnia symptoms are associated with the onset of major cardiometabolic risk factors such as diabetes and hypertension, along with increased risk of cardiovascular and cerebrovascular diseases [1, 9–11]. Insomnia symptoms are also strongly associated with debilitating mental health problems, including depression, anxiety, alcohol abuse, and psychosis [12].

Despite the scientific evidence of worsened health and lowered quality of life among individuals complaining of insomnia symptoms, there is no consensus about the impact of insomnia symptoms on mortality. Previous studies provide inconsistent and nonconclusive results [13–23]. A recent meta-analysis reported that the risk of mortality did not differ significantly between those with insomnia symptoms and those without. The authors stated that they could not draw any firm conclusions from the current scientific evidence due to substantial limitations in prior studies [24].

Major limitations to prior studies are, namely, the use of a single item survey question to assess insomnia symptoms, and the absence of controls for potential baseline confounding factors such as health behaviors (e.g., smoking, alcohol consumption, physical activity), cognitive functioning, and individual’s depression status [24]. Furthermore, most previous studies evaluated insomnia symptoms at a single point in time (usually at baseline), and did not account for the dynamic, complex, and bidirectional nature of associations between insomnia symptoms and time-dependent cognitive, behavioral, and physical characteristics of the individuals [12]. Therefore, associations between insomnia symptoms and mortality found in prior studies can be explained in part by physical and psychological comorbidities, or lifestyle risk factors, which might themselves worsen sleep, and whose effects may be aggravated over time [25]. Indeed, sleep and its health outcomes are a dynamic condition of the human organism [26], and so is the course of insomnia symptoms, being characterized by a waxing- and waning pattern [1]. Hence, due to these limitations, establishing causality between insomnia symptoms and mortality has been largely contested so far [25, 27].

In the present study, we utilized cohort data from an extensive longitudinal survey in the US to investigate the associations between time-varying insomnia symptoms and all-cause mortality among middle-aged and older adults. We implemented a robust study design to control time-dependent confounding and selection bias and then estimated the direct associations between insomnia symptoms and mortality. Four main symptoms (difficulty initiating sleep, difficulty maintaining sleep, early morning awakening, and nonrestorative sleep) were analyzed. We hypothesized that the symptoms are associated with higher all-cause mortality on a cumulative scale, and that each insomnia symptom evaluated independently is associated with a higher risk of mortality.

Methods

Study design, settings, and participants

Data for the current study were drawn from the Health and Retirement Study (HRS), an ongoing longitudinal, biennial, nationally representative survey on health, employment, income, and family structure of individuals aged 50 years and older, and their spouses (regardless of age), in the US [28]. Participants in the HRS give verbal informed consent at each wave, and the data collection was approved by the Health Sciences and Behavioral Sciences institutional review board at the University of Michigan [29]. The HRS data documentation and detailed information regarding its design and methodology are discussed elsewhere [28, 30, 31]. We used the 2004 HRS wave as baseline because the survey was replenished with new respondents that year, and insomnia symptoms questions had recently been added to the core HRS questionnaire. Proxy respondents were excluded as they do not answer the cognitive and depressive symptoms assessment questions on behalf of the self-respondents. The participants were censored at death, dropout, or at the end of the 2018 survey wave, whichever came first. Figure 1 displays the inclusionary and exclusionary criteria for the study’s participants. Our study follows the reporting guidelines of the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE).

Figure 1.

The inclusion/exclusion flowchart.

All-cause mortality

The outcome of interest in the present study was all-cause mortality. In the HRS, participants’ deaths are ascertained through the National Death Index (NDI) and the postmortem “exit” interviews with a family member or a next-of-kin. Both sources are proven to be reliable in validating approximately 96% of respondent deaths [32]. In our study, all-cause mortalities were treated as aggregated, discrete measures of deaths reported between every two subsequent waves.

Insomnia symptoms

To screen for insomnia symptoms at baseline and each subsequent wave, respondents were evaluated for difficulties in initiating and maintaining sleep, early morning awakening, and nonrestorative sleep. The questions about insomnia symptoms in the HRS ask how often the respondent has trouble with “falling asleep,” “waking up during the night,” “waking up too early and not being able to fall asleep again,” and how often they feel “really rested” when they wake up in the morning. The response options for each question are “most of the time,” “sometimes,” and “rarely or never.” Following prior sleep studies [11, 17, 33–36], we qualified individuals as symptomatic when they answered “most of the time” in the first three questions and when they answered “rarely or never” in the last question. Thus, for analysis, each symptom was represented through a time-varying binary variable, “yes vs. no,” at baseline and each subsequent follow-up wave. Since three of the four symptoms were not evaluated in the 2008 and 2012 survey waves, respondents’ last observation of those symptoms was carried forward for these years. Finally, a time-varying summary variable “number of insomnia symptoms” (i.e., the sum of each respondent’s symptoms) was created at each wave by summing across all four symptoms. Respondents were categorized per their experience of “no symptoms, one, two, three, or four” insomnia symptoms.

Covariates

Prior literature guided the choice of the study covariates based on their associations with insomnia symptoms and mortality [9, 37, 38]. In the present study, time-invariant sociodemographic factors at baseline included respondents’ self-reported sex, race and ethnicity, educational attainment, marital status, respondents’ family poverty level, and respondents’ census region. Time-varying potential covariates included several socioeconomic, biological, functional, psychological, cognitive, psychophysiological, and behavioral factors, as well as factors representing respondents’ health insurance coverage.

The only time-varying socioeconomic factor is paid work performed by the respondent. Biological factors included respondents’ age (measured in years), body mass index (BMI), and the number of comorbid medical conditions. BMI was calculated as weight in kilograms divided by height in meters squared. The number of comorbid conditions was derived from self-reporting a physician’s diagnosis of hypertension, diabetes, cancer (excluding minor skin cancers), chronic lung disease, heart diseases, and stroke. Physical functioning was determined by limitations in any of the five activities of daily living (ADLs; bathing, eating, dressing, walking across a room, and getting in/out of bed) and any of the five instrumental activities of daily living (IADLs; shopping for groceries, managing money, preparing meals, taking medications, and using phone). Potential psychological and cognitive covariates include the modified Center for Epidemiologic Studies Depression (CES-D) scale and cognitive performance scores. The modified CES-D scale [39–41] in the HRS is the sum of 8 items; however, we removed one item from the score (“restless sleep”) to minimize the CES-D scale’s operational confounding with the outcome [42, 43]. The cognitive functioning score was assessed through tests adopted from the modified Telephone Interview for Cognitive Status (TICS) [44, 45]. A higher CES-D score indicates worse depressive symptomatology, while a higher cognitive score indicates better cognitive performance for respondents. An example of a psychophysiological factor is perceived pain, assessed via self-reported trouble with pain. Health behavior characteristics include smoking status, alcohol consumption, and engagement in vigorous physical activity. Finally, possession of health insurance was evaluated through respondents’ coverage by a government-sponsored health insurance plan, and whether they had at least one private health insurance plan. To address the issue of temporal sequencing of events [46], our study was designed to measure time-varying insomnia symptoms and time-varying covariates preceding the subsequent mortality outcomes during follow-up.

Statistical analysis

We first analyzed the respondents’ general characteristics at baseline. The significance of cross-tabulations between the baseline characteristics and mortality status during the 14 years of follow-up was then assessed by the design-adjusted Satterthwaite Rao-Scott chi-square test for categorical variables, and t-test for continuous variables. We also summarized the number and type of insomnia symptoms as well as the time-varying confounders at each follow-up wave with frequencies, proportions, medians, and means (±standard deviation [SD]).

We implemented a marginal structural model (MSM) approach to investigate longitudinal associations between the cumulative number and type of insomnia symptoms and all-cause mortality [47–50]. MSMs are a class of causal models developed by Robins et al., typically used for causal inferencing in epidemiology [47, 48, 50, 51]. MSMs are ideal for estimating the effects of time-varying exposures in longitudinal data when time-varying covariates might simultaneously be confounders and intermediate factors [50]. In longitudinal observational studies where there are repeated measures of exposure, confounding factors, mediators, and colliders, complex and dynamic relationships exist among variables. In these circumstances, bias may occur when a time-varying confounder, conditional on the past exposure history, is a predictor of subsequent exposure [48, 51–53]. Meaning that time-varying confounders could influence subsequent values of a time-varying exposure or an outcome, and themselves could be influenced by prior values of exposure or other confounders. The same holds true when a past, or time-varying, exposure history is an independent predictor of a subsequent time-varying confounder or an outcome [52, 54]. Further, other issues such as differential loss to follow-up, study dropout, and survey nonresponse may arise in longitudinal studies [52, 53]. Individuals may have missing data at baseline and follow-up because they miss a wave interview, or are reluctant to provide information on a study measure. These later issues can be some of the critical sources of selection bias [52]. Therefore, investigations of exposure-outcome relationships should properly account for these biases for accurate or less biased estimation.

In the current study, the values of insomnia symptoms and most of the confounding factors were time-varying and repeated in multiple time intervals at follow-up. Therefore, it is plausible that a time-varying insomnia symptom could be influenced by their own preceding values, those of the baseline and time-varying confounders, and they themselves could affect their subsequent values and those of the confounders. Further, a confounding factor could be the product of insomnia symptoms (e.g., depression) [55] or other time-dependent confounders that could subsequently affect values of insomnia symptoms, or even predict the outcome (i.e., mortality). Moreover, although in this study we had a low and acceptable attrition rate (approximately 5%) during the 14-year follow-up; there were instances of study dropout, loss to follow-up, and missingness due to wave-specific survey nonresponse. To account for all these potential biases, we used an MSM approach in analyses. MSM creates a “pseudo-population” where all confounding and selection bias are, at a best attempt, adjusted for and the prospective association between insomnia symptoms and all-cause mortality becomes direct [51–53]. Hence, the effects of insomnia symptoms on mortality are properly disentangled from the effects of time-dependent confounders [52]. The estimates obtained from our MSM analysis are interpreted as population average causal insomnia symptoms effects on all-cause mortality [47, 48, 52]. Generally, the outcomes obtained from an MSM model in an observational study approximate those obtained from a randomized controlled trial in the best practical way [52]. See Supplementary Figure S1 for the causal diagram illustrating the effects of insomnia symptoms on all-cause mortality.

For the MSM analysis (Supplementary Methods S1), we calculated two inverse probability weights: 1) the inverse probability of treatment weight (IPTW) to account for time-varying confounding, and 2) the inverse probability of censoring weight (IPCW) to account for differential loss to follow up due to study withdrawal, dropout, and wave-specific survey nonresponse. Final MSM weights were derived from multiplying IPTWs by IPCWs and incorporated both time-dependent confounding and censoring adjustments.

We then fit a discrete-time hazard model with a complementary log-log link on the pooled data (88 778 person-waves) and applied the final MSM weights to obtain the hazard ratios (HRs) and 95% confidence intervals (CIs) of all-cause mortality for the number of insomnia symptoms on a cumulative scale. The HRs calculated from this model with the link function resemble those obtained from a continuous-time Cox proportional hazards model [46]. Simply, the log-log link transformation generates a discrete-time statistical model for the hazards that has an innate proportional hazards assumption [46, 56]. A similar model was constructed to obtain HRs for individual insomnia symptoms. Both models adjusted only for the time-invariant factors because, unlike the time-varying confounders, their effects were not directly incorporated into the prediction of the MSM weights [53]. We assessed interaction terms between time-invariant variables and insomnia symptoms, both on the cumulative scale and independently, to check whether any of the time-invariant variables are effect modifiers; none of these interactions were significant.

Analyzing pooled data with repeated measures may bring up a legitimate concern related to dependence within the clusters (i.e., respondents); however, this was not an issue in our models as the outcome, “all-cause mortality,” occurred only once for each respondent during follow-up [56]. Further, this technique naturally handles ties that usually arise from grouping data into intervals [56]. Therefore, there are no violations of the assumptions of independence and the estimated standard errors are unbiased with no inflated test statistics [56]. Using this method in discrete-time survival analysis has other distinctive advantages because its validity requires no misspecification in the MSM weights nor the hazard models [52].

Sensitivity analyses were performed to test the validity of the weights and robustness of the results (Supplementary Methods S2). The analyses were conducted using SAS 9.4 statistical software (SAS Institute Inc., Cary, NC, USA). All obtained P-values from statistical tests were 2-tailed, and the significance threshold was set at ≤.05.

Results

Our analytical sample at baseline included 15 511 respondents who were followed via biennial surveys from 2004 to 2018. The baseline sample represents 67.8 million middle-aged and older adults at the US national level. At baseline, the mean (±SD) of age was 63.7 ± 10.2 years. The respondents were females (56.0%), non-Hispanic whites (81.6%), high-school graduates (30.3%), married or living as married (65.7%), had an annual household income above the poverty threshold (92.2%), and were residing in the southern parts of the US (36.8%). About 41.6% reported experiencing at least one insomnia symptom at baseline. Approximately 13.7% had trouble initiating sleep, 27.4% reported trouble maintaining sleep, 13.4% reported early-morning awakening, and 15.6% reported nonrestorative sleep (see Table 1).

Table 1.

Respondent characteristics at baseline

Respondent Characteristics	All respondents (n = 15 511)a
	Frequency (weighted %)
Time-invariant characteristics
Sex
Male	6364 (44.0)
Female	9147 (56.0)
Race and ethnicity
Non-Hispanic White	11 709(81.6)
Non-Hispanic Black	2126 (9.1)
Hispanic	1334 (6.7)
Non-Hispanic others	339 (2.6)
Level of education
Less than high school or GED	3831 (20.7)
High-school graduate	4884 (30.3)
Some college	3487 (24.2)
College and above	3306 (24.8)
Marital status
Married or living as married	10 104 (65.7)
Otherb	5394 (34.3)
Whether in poverty
Household income above the poverty threshold	14 213 (92.2)
Household income below the poverty threshold	1298 (7.8)
Census region
Northeast	2490(17.3)
Midwest	3948 (25.9)
South	6116 (36.8)
West	2924 (20.0)
Time-varying characteristics
Number of insomnia symptoms
No symptoms	9025 (58.4)
One	3783 (23.8)
Two	1499 (9.7)
Three	836 (5.5)
Four	363 (2.6)
Type of insomnia symptomc
Difficulty initiating sleep	2153 (13.7)
Difficulty maintaining sleep	4252 (27.4)
Early-morning awakenings	2109 (13.4)
Nonrestorative sleep	2228 (15.6)
Working for pay
Yes	6194 (48.9)
No	9300 (51.4)
Age in years, mean (±SD), (range: 50–102)	63.7 (±10.2)
BMI (kg/m2)
Normal or underweight (<25)	5134 (32.9)
Overweight (25–29.9)	5828 (38.5)
Obese (≥30)	4303 (28.6)
Comorbid medical conditions
None	2836 (22.2)
One	4212 (28.7)
Two and more	8462 (49.1)
ADL limitations
Yes	2211(13.1)
No	13 293 (86.9)
IADL limitations
Yes	1827 (10.8)
No	13 674 (89.2)
CES-D scale score, median (IQR), (range:0–7)d	0.00 (0.00–1.20)
Cognitive functioning score, mean (±SD), (range:0–27)e	16.00 (±4.20)
Often troubled with pain
Yes	5034 (32.6)
No	10 465 (67.4)
Smoking status
Current	2314 (16.6)
Former	6667 (41.5)
Never	6421 (41.9)
Alcohol consumption
Yes	7866 (55.5)
No	7642 (44.5)
Vigorous physical activity
Yesf	6110 (43.2)
No	9388 (56.8)
Covered by government health insurance plan
Yes	9324 (47.8)
No	6149 (52.2)
Have at least one private health insurance plan
Yes	10 600 (72.6)
No	4861 (27.4)

Data were obtained from the Health and Retirement Study (HRS), n = 15 511 (age≥50 years), United States, 2004.

SE, Standard Error; GED, General Educational Development; BMI, Body Mass Index (calculated as weight in kilograms divided by height in meters squared); ADL, Activities of Daily Living; IADL, Instrumental Activities of Daily Living; CES-D, Center for Epidemiologic Studies Depression scale; IQR, Interquartile Range.

^aWeighted N = 67.8 million adults of ages≥50 years at the US national level, estimated through adjustments for the HRS complex survey design.

^bOther marital statuses include separated, divorced, widowed, and never married.

^cOnly affirmative responses are included to keep the table short.

^dHigher score indicates worse depression symptoms.

^eHigher score indicates a better cognitive performance.

^fAt least 1 to 3 times per month.

Table 2 illustrates the baseline characteristics of both respondents who died during follow-up and those who stayed alive or were lost to follow-up through the end of the 2018 survey wave. During the 14 years of follow-up, 5878 (31.9%) respondents died.

Table 2.

Associations between baseline characteristics and all-cause mortality at follow-up

Respondent Characteristics	Died during follow-up (n = 5878)	Stayed alive or lost to follow-up through the end of 2018 wave (n = 9633)	P  a
	Frequency (weighted %)	Frequency(weighted %)
Time-invariant characteristics
Sex
Male	2748 (48.2)	3616 (42.0)	<.001
Female	3130 (51.8)	6017 (58.0)
Race and ethnicity
Non-Hispanic White	4562 (83.0)	7147 (80.9)	<.001
Non-Hispanic Black	830 (10.0)	1296 (8.6)
Hispanic	381 (5.0)	953 (7.6)
Non-Hispanic others	103 (2.0)	236 (2.9)
Level of education
Less than high school or GED	1890 (29.7)	1941 (16.5)	<.001
High-school graduate	1947 (33.4)	2937 (28.9)
Some college	1150 (20.7)	2337 (25.9)
College and above	890 (16.2)	2416 (28.7)
Marital status
Married or living as married	3274 (54.2)	6830 (71.1)	<.001
Otherb	2601 (45.8)	2793 (28.9)
Whether in poverty
Household income above the poverty threshold	5241 (89.0)	8972 (93.7)	<.001
Household income below the poverty threshold	637 (11.0)	661 (6.3)
Census region
Northeast	936 (17.1)	1554 (17.4)	.15
Midwest	1484 (26.1)	2464 (25.9)
South	2429 (38.2)	3687 (36.1)
West	1018 (18.6)	1,906 (20.6)
Time-varying characteristics
Number of insomnia symptoms
No symptoms	3138 (52.8)	5887 (61.0)	<.001
One	1590 (26.7)	2193 (22.5)
Two	635 (10.9)	864 (9.2)
Three	351 (6.6)	485 (5.0)
Four	162 (3.0)	201 (2.3)
Type of insomnia symptomc
Difficulty initiating sleep	940 (16.4)	1213 (12.5)	<.001
Difficulty maintaining sleep	1800 (31.4)	2452 (25.5)	<.001
Early-morning awakening	916 (16.0)	1193 (12.1)	<.001
Nonrestorative sleep	905 (16.6)	1323 (15.2)	.06
Working for pay
Yes	1159 (23.4)	5035 (60.4)	<.001
No	4714 (76.6)	4586 (39.6)
Age in years, mean (±SD)	71.2 (±10.9)	60.1 (±7.7)	<.001
BMI (kg/m2)
Normal or underweight (<25)	2275 (38.7)	2859 (30.2)	<.001
Overweight (25–29.9)	2032 (34.5)	3796 (40.3)
Obese (≥30)	1491 (26.8)	2812 (29.5)
Comorbid medical conditions
None	460 (8.9)	2376 (28.4)	<.001
One	1167 (20.9)	3045 (32.3)
Two and more	4251 (70.2)	4211 (39.3)
ADL limitations
Yes	1425 (24.1)	786 (7.9)	<.001
No	4453 (75.9)	8840 (92.1)
IADL limitations
Yes	1239 (20.8)	588 (6.1)	<.001
No	4638 (79.2)	9036 (93.9)
CES-D scale score, median (IQR)d	0.30 (0.00–1.88)	0.00 (0.00–0.87)	----
Cognitive functioning score, mean (±SD)e	14.18 (±4.40)	16.84 (±3.82)	<.001
Often troubled with pain
Yes	2118 (36.9)	2916 (30.6)	<.001
No	3755 (63.1)	6710 (69.4)
Smoking status
Current	1041 (20.5)	1273 (14.7)	<.001
Former	2822 (46.6)	3845 (39.2)
Never	1972 (32.9)	4449 (46.1)
Alcohol consumption
Yes	2420 (43.4)	5446 (61.1)	<.001
No	3456 (56.6)	4186 (38.9)
Vigorous physical activity
Yesf	1522 (27.1)	4588 (50.7)	<.001
No	4349 (72.9)	5039 (49.3)
Covered by government health insurance plan
Yes	4944 (78.1)	4380 (33.7)	<.001
No	919 (21.9)	5230 (66.3)
Have at least one private health insurance plan
Yes	3533 (62.2)	7067 (77.5)	<.001
No	2321 (37.8)	2540 (22.5)

Data were obtained from the Health and Retirement Study, n = 15 511, age≥50 years, United States, 2004–2018.

SE, Standard Error; GED, General Educational Development; BMI, Body Mass Index (calculated as weight in kilograms divided by height in meters squared); ADL, Activities of Daily Living; IADL, Instrumental Activities of Daily Living; CES-D, Center for Epidemiologic Studies Depression scale; IQR, Interquartile Range.

^aDesign-Adjusted Rio-Scott chi-square test, or t-test.

^bOther marital statuses include separated, divorced, widowed, and never married.

^cOnly affirmative responses are included to keep the table short.

^dHigher score indicates worse depression symptoms.

^eHigher score indicates a better cognitive performance.

^fAt least 1 to 3 times per month.

The descriptive statistics of time-varying insomnia symptoms and confounding variables from the 2006 through 2016 survey waves are shown in Supplementary Table S1. There were some variations in the proportions of insomnia experiences and the confounding variables over time among the cohort respondents.

Table 3 depicts the results of the marginal structural discrete-time survival models. Respondents who experienced one (HR = 1.11; 95% CI: 1.03–1.20, p = .005), two (HR = 1.12; 95% CI: 1.01–1.23, p = .02), three (HR = 1.15; 95% CI: 1.05–1.27, p = .003), and four (HR = 1.32; 95% CI: 1.12–1.56, p = .001) insomnia symptoms had, on average, a higher hazard of all-cause mortality compared to those who did not experience any insomnia symptoms. Difficulty maintaining sleep as a standalone symptom was not significantly associated with higher all-cause mortality (HR = 0.95; 95% CI: 0.88–1.02, p = .14). However, respondents experiencing difficulty initiating sleep (HR = 1.12; 95% CI: 1.02–1.22, p = .013), early-morning awakenings (HR = 1.09; 95% CI: 1.01–1.18, p = .04), and nonrestorative sleep (HR = 1.17; 95% CI: 1.09–1.26, p < .001) had a higher hazard of all-cause mortality compared to those not experiencing the symptom. Results from Cox proportional hazards models examining the associations between only baseline (2004) insomnia symptoms and all-cause mortality are provided in Supplementary Table S2 for comparison purposes.

Table 3.

Marginal structural discrete-time survival analyses modeling longitudinal associations between frequency and type of insomnia symptoms and all-cause mortality

Respondent Characteristics	Adjusted HR (95% CI)a	P
Number of insomnia symptoms (ref: no symptoms)
One	1.11 (1.03–1.20)	.005
Two	1.12 (1.01–1.23)	.02
Three	1.15 (1.05–1.27)	.003
Four	1.32 (1.12–1.56)	.001
Individual insomnia symptoms
Difficulty initiating sleep (ref: no)	1.12 (1.02–1.22)	.013
Difficulty maintaining sleep (ref: no)	0.95 (0.88–1.02)	.14
Early-morning awakening (ref: no)	1.09 (1.01–1.18)	.04
Nonrestorative sleep (ref: no)	1.17 (1.09–1.26)	<.001

Data were obtained from the Health and Retirement Study (n = 15 511, ages ≥50 years, United States, 2004–2018).

CI, Confidence Interval; HR, Hazard Ratio; MSM, Marginal Structural Model; ref, Reference group.

^aThe stabilized MSM weights were applied. The models incorporated a general specification for the main effect of time and were adjusted for all baseline respondent characteristics (Sex, race and ethnicity, level of education, marital status, family poverty threshold, and census region).

Sensitivity analysis

We did not find qualitatively different estimates from the performed sensitivity analyses than those obtained from our original selected MSM models (Supplementary Tables S3 and S4). The statistical conclusions for the estimated hazard ratios and directions of the associations remained relatively stable.

Discussion

In a prospective, nationally representative cohort of 15 511 middle-aged and older adults in the US, we found that experiencing more insomnia symptoms is associated with higher all-cause mortality. Similarly, for each insomnia symptom independently, those experiencing difficulties initiating sleep, early-morning awakenings, and nonrestorative sleep had higher hazards of all-cause mortality, compared to those not experiencing the symptom. However, those experiencing difficulties maintaining sleep, compared to those without the symptom, were not different in mortality outcome. In brief, these findings demonstrate that insomnia symptoms are potential risk factors of mortality among middle-aged and older adults. Suffering from any number of insomnia symptoms could be a risk to individuals’ health and well-being and could increase the risks of mortality.

Some of our findings are consistent with those of a few prior prospective studies, despite having differences in baseline sample, design, and methodology. For example, Li et al. [17] followed up with a cohort of 23 447 US men (aged 40 to 75 years) participating in the Health Professionals Follow-Up Study (HPFS). They found that experiencing difficulty initiating sleep, early-morning awakenings, and nonrestorative sleep at baseline were associated with higher hazards of all-cause mortality by 25%, 4%, and 24%, respectively, while experiencing difficulties maintaining sleep was not significantly associated with all-cause mortality [17]. Similarly, in the Hordaland Health Study (HUSK) in Norway [21], Sivertsen et al. followed a cohort of 6236 individuals aged 40–45 at baseline, for 13 to 15 years. The authors screened participants for insomnia symptoms at baseline. Based on the response options, they developed a continuous score (ranged 0–16) for four insomnia symptoms (difficulty initiating sleep, difficulty maintaining sleep, early morning awakenings, and daytime tiredness/sleepiness). Higher scores on the scale represented a higher insomnia symptoms’ burden [21]. Each additional unit increase in the insomnia symptoms score on the scale was associated with a 10% increase in all-cause mortality [21].

Furthermore, in a study of three cohorts of middle-aged and older adults in Finland, Norway, and Lithuania, difficulties initiating sleep was associated with all-cause mortality by 2.5 fold among Finnish, and by 3.4 fold among Norwegian men, although not Lithuanian men, compared to men not experiencing the symptom [16]. No associations were found among women after adjusting for baseline confounding factors. Lastly, Newman et al. followed up with a cohort of 5888 community-dwelling older adults of ≥65 years who participated in the Cardiovascular Health Study (CHS) in the US, to examine the associations between difficulties initiating and maintaining sleep, early morning awakenings, and daytime sleepiness at baseline with all-cause mortality at follow-up [18]. In their study, the associations varied by gender. Age-adjusted hazards were higher by 82% for daytime sleepiness in women, and higher by 29% for men experiencing difficulty initiating sleep, when compared to their symptom-free counterparts.

Compared to the abovementioned studies, ours is more comprehensive because it included a large, population-representative sample of middle-aged and older adults and did not restrict the analysis to any specific segment of the population. The present study also differs from previous scholarly work by using repeated measures of a comprehensive set of insomnia symptoms over 14 years of follow-up while accounting for residual baseline and time-varying confounding, and for selection bias such as loss to follow-up, dropout, and survey nonresponse that all could distort the true nature of the associations between insomnia symptoms and mortality.

In the current study, three out of the four insomnia symptoms were independently associated with all-cause mortality. It is less clear why, unlike other symptoms, difficulty maintaining sleep had a null independent association with mortality. However, it is plausible biologically that reduced sleep maintenance has distinct impacts on neurohormonal activities and sleep-stage distribution [17]. Nonetheless, a comprehensive evaluation of associations between difficulty maintaining sleep and mortality is warranted in future studies. We found nonrestorative sleep to have the strongest association with all-cause mortality. This symptom is currently declared to be distinct from other nocturnal insomnia symptoms and considered an independent component of insomnia [9, 17], because it is believed to have different involved pathophysiologic mechanisms than the other three nocturnal symptoms [1, 57]. Therefore, current diagnostic nosologies of insomnia disorder, such as the Diagnostic and Statistical Manual for Mental Disorders-5 (DSM-525) and the International Classification of Sleep Disorders-3 (ICSD-3), have dropped nonrestorative sleep despite being an important component of insomnia [9, 58]. Research has shown that individuals with nonrestorative sleep usually experience significant daytime impairments including mental and physical fatigue, excessive sleepiness, irritability, and memory impairments [58]. Thus, it is plausible that the effects of nonrestorative sleep on mortality are rather manifested through these consequences, especially since these have been associated with worsened quality of life, adverse health outcomes, and mortality.

The pathways linking insomnia symptoms and mortality are complex, multifactorial, and not yet fully understood. Specifically, the pathophysiological mechanisms underlying the associations between insomnia symptoms and several health and medical conditions need further scrutiny [59]. Nonetheless, since insomnia is considered a state of physiological and emotional hyperarousal throughout the daily sleep-wake cycle, several distinct or overlapping biological mechanisms are plausible [37, 59, 60]. Insomnia causes activation of sympathetic arousal mechanisms that result in heart rate and blood pressure elevation [37, 59, 61]. It could also result in the hypothalamic-pituitary-adrenal stress axis dysregulation, leading to increased cortisol secretion and alterations in the diurnal cortisol profile [37, 61]. Other potential mechanisms include the implications of insomnia symptoms and their role in impaired glucose metabolism and insulin resistance, vascular dysfunction, increased atherogenesis, and increased inflammatory biomarkers such as in C-reactive protein, tumor necrosis factor-alpha (TNF-α), and interleukin-6 (IL-6) [37, 59, 62, 63].

These biological markers can act as crucial risk factors for cardiovascular diseases, diabetes, mental health disorders (e.g., depression, anxiety, suicidality), and cognitive impairments such as Alzheimer’s or other dementias [1, 61–64]. Most of these health and medical conditions, in turn, are considered leading causes of mortality [65, 66]. Therefore, it is plausible that the biological markers and their disease products could act as mediators or moderators on the pathway between insomnia symptoms and mortality.

Insomnia symptoms are also associated with circadian rhythm disruptions and sleep-wake disorders. Specifically, difficulty initiating sleep and early-morning awakening are associated with delayed and advanced sleep-wake phase disorders, respectively; indicating that experiencing these symptoms could sometimes have an underlying circadian-related etiology [67, 68]. Circadian rhythm disruptions or misalignment have been implicated in cardiometabolic and cardiovascular diseases, as well as in immune system dysfunctions and inflammation [69]. The immune and inflammatory changes seen in circadian disruption may play a mediating role in linking these disruptions to many adverse health outcomes including, obesity, diabetes, cancer, cardiovascular diseases, and their associated morbidities and mortality [69–71].

Another potential mechanism that indirectly links insomnia symptoms to mortality is the presence of maladaptive or inadequate health behaviors such as poor diet quality, tobacco smoking, poor physical activity, excessive alcohol consumption, and illicit drug use among individuals with insomnia. Experiencing insomnia symptoms could trigger these health behaviors, and they, in turn, put individuals on the cardiometabolic and cardiovascular diseases pathway [1]. Finally, there is growing knowledge indicating that poor sleep or insomnia results in poor waking alertness, attention, and concentration, as well as impaired psychomotor performance capabilities [72, 73]. These factors could, directly or indirectly, mediate the link between insomnia symptoms and mortality. This potential pathway is partially supported by claims that insomnia symptoms contribute significantly to accidents, unintentional fatal injuries, and fatal motor vehicle accidents [74–77].

Strengths and limitations

The present study has several strengths. First, we extracted 14 years of data from a large, population-representative sample of middle-aged and older adults with a low attrition rate. During follow-up, only 820 (5%) of the respondents were lost to follow-up or dropped out of the sample, except those with confirmed mortality, and the former were properly adjusted for in our analyses. Second, information on insomnia symptoms was obtained from valid and reliable insomnia symptoms questions that are primarily utilized in sleep studies [78]. Third, using detailed and repeated assessments of insomnia symptoms and other time-dependent biological, psycho-cognitive, and lifestyle confounders strengthened the results’ validity. Finally, to our best knowledge, the current study marks the first use of MSM in the context of investigating prospective associations between insomnia symptoms and mortality, and the first to have a life-course approach to this objective. Adopting and implementing an MSM approach facilitated investigating causal associations after strict control for time-varying confounding and selection bias [49, 51].

Nevertheless, there are a few limitations. Insomnia symptoms are self-report measures of sleep. Usually, devices such as actigraphy and polysomnography can provide detailed objective data on sleep [79]. However, objective sleep measures derived from such devices are less sensitive and specific than self-reports in identifying insomnia symptoms and are not feasible in large-scale epidemiological studies [57]. Further, utilizing more standardized insomnia screening instruments such as Insomnia Severity Index (ISI) and Pittsburgh Sleep Quality Index (PSQI) could have been more useful as comprehensive tools to identify insomnia symptoms chronicity, as well as in the evaluation of sleep quality or insomnia as a “sleep disorder” [80]. However, in the current study, we are merely focused on insomnia symptoms and a comprehensive evaluation of insomnia symptoms chronicity and assessing the associations between sleep quality, or insomnia disorder, and mortality were beyond the aims and scopes of our study. Moreover, there was no consistent cohort information in the dataset regarding secondary insomnia complaints and their associated daytime impairments such as excessive daytime sleepiness, fatigue, irritability, and problems of concentration and memory impairments. Similarly, information on sleep duration, or other sleep disorders such as obstructive sleep apnea (OSA), and sleep medication use were not collected consistently across all waves. Considering these measures in our models could have contributed to the analysis of associations between insomnia symptoms and mortality [57, 60, 81]. Nonetheless, although we could not directly account for some factors such as OSA, other respondent characteristics, such as age and BMI, were used as proxies. This was based on results of prior studies indicating that individuals complaining of insomnia symptoms may have OSA as the underlying cause [5, 81], and a positive correlation has been found between increasing age and BMI with the probability of OSA [2]. Thus, we partially controlled for the residual confounding of OSA by incorporating time-varying age and BMI in our analyses.

Lastly, the causal results from an MSM analysis are interpreted under the assumptions of consistency, conditional exchangeability (i.e., no unmeasured confounding), positivity, and correct model specification [47, 51–53]. Satisfying the consistency assumption in our study requires that the outcome (i.e., all-cause mortality), observed for each study participant, be the true causal outcome under their observed insomnia symptoms history. Positivity assumption requires that the probability of being assigned to each insomnia symptoms category in the MSM is positive and greater than zero. Conditional exchangeability means that no additional unmeasured confounding factors are left to be incorporated into the models. Assumptions of consistency, exchangeability, and no unmeasured confounding are not guaranteed to hold in observational studies, cannot be fully proven, and are usually difficult to assess [52, 53].

Nevertheless, necessary steps can be taken during MSM weight estimation to satisfy some of the assumptions. For example, an important condition for correct model specification and positivity is to have a mean stabilized weight closest to one (1) which was achieved in our analysis [51, 52]. Substantial deviations from a mean weight of (one) indicate a possible violation, or near violation, of positivity or correct model specification. Consistency can be assessed by trimming the weights and repeating the analysis to evaluate biased-variance trade-offs [51]. Further, we used expert knowledge and literature review, prior to analysis, to find all relevant available determinants of insomnia symptoms and predictors of mortality and included them in the MSM weight estimation to enhance the plausibility of no unmeasured confounding assumption. Another essential step necessary to address the potential impacts of unmeasured confounding is through conducting sensitivity analyses [47] which were performed for the current study (Supplementary Methods S2). The assumptions for causal inferencing from an MSM are the same underlying assumptions necessary for any observational study [52, 53]. Yet, those underlying MSM are considered to be less restrictive than those of standard methods, as MSM analysis does not require the absence of time-varying confounding [47, 50].

Conclusions

In summary, the current study shows that some insomnia symptoms are potential risk factors for mortality. The impacts of insomnia symptoms on all-cause mortality were seen both on a cumulative scale and independently. This can be translated as that experiencing more insomnia symptoms at any given time might represent a higher burden of insomnia and hence, a higher likelihood of mortality. Expanding public health awareness about insomnia can be an essential prevention strategy. Healthcare providers and policymakers can play a critical role in addressing the issue, specifically among middle-aged and older adult populations. Furthermore, it is paramount for clinicians to properly screen their middle-aged and older patients for insomnia symptoms during patient visits and routine checkups using validated, easy-to-use insomnia questionnaires. Future studies are needed to further understand the complex nature of the relationship between insomnia symptoms and mortality. In this regard, clinical studies investigating the underlying mechanisms that link insomnia symptoms to mortality are warranted.

Funding

The Health and Retirement Study is sponsored by grant U01AG009740 from the US National Institute on Aging (NIA). However, the current study received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. The NIA had no role in the design and conduct of the present study; nor in data collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Disclosure Statement

None declared.

Ethics Approval

The HRS data collection is approved by the institutional review board at the University of Michigan, MI, USA.

Data Availability

The data analyzed in the current study were obtained from the Health and Retirement Study, publicly available to access and download at: https://hrs.isr.umich.edu/data-products.