Associations between night/shift working and late-life brain health

Abstract

Sleep and circadian disturbances are associated with increased dementia risk. The mechanism remains poorly understood. We aimed to examine the relationship between night/shift working at age 31 and biomarkers of late-life brain health and to estimate the extent to which these relationships are mediated by unhealthy lifestyle behaviours. A prospective longitudinal cohort study, Insight 46, recruited participants from the Medical Research Council National Survey of Health and Development (NSHD) 1946 British Birth cohort. All born in the same week in 1946, participants were assessed at age 70 with multi-modal structural and molecular brain imaging and fluid biomarkers, from which whole-brain and hippocampal volumes, white matter hyper-intensity volume (WMHV), ¹⁸F-florbetapir amyloid-β PET Centiloids and plasma phosphorylated tau (p-tau)217 were derived. Prospective data collection included night/shift working at age 31, alongside smoking, alcohol intake, body mass index, exercise, blood pressure, Framingham risk score (FRS) at multiple timepoints from age 20 to 70 and dementia diagnosis or death by age 78. Analyses were adjusted for sex, age, education, socioeconomic position and, where appropriate, total intracranial volume or apolipoprotein E (APOE) genotype. Night/shift working data were available for 431 Insight 46 participants {50% female, mean age 70.7 years [standard deviation (SD) 0.6]}. Night/shift workers had lower whole-brain volume [-19.9 mL, 95% confidence interval (CI) -31.9, -7.9, P = 0.001], lower amyloid PET Centiloids (-9.45, 95% CI -14.7, -4.1, P = 0.0008) and lower plasma p-tau217 concentration (-0.05 pg/mL, 95% CI -0.10, -0.001, P = 0.04), without significant difference in hippocampal volume or WMHV. p-tau217 concentrations were also lower in night/shift workers from a wider sample from the NSHD cohort [n = 1067, mean age 69.9 (SD 0.7), -0.05 pg/mL, 95% CI -0.08, -0.02, P = 0.004]. By age 78, night/shift workers in the NSHD cohort (n = 3040) had lower rates of all-cause (excluding vascular) dementia (hazard ratio 0.33, 95% CI 0.12, 0.92, P = 0.03). Night/shift workers had 0.6% higher FRS (P = 0.01) at age 36, smoked 5.9 more pack-years by age 53 (P = 0.005), consumed 10.7 g/day more alcohol by age 63 (P = 0.006) and had higher rates of APOE ɛ4 allele carriage. Lifestyle behaviours mediated 28% of the lower brain volume in night/shift workers. Despite less healthy lifestyles, higher rates of APOE ɛ4 allele carriage and smaller brains, night/shift workers had lower levels of Alzheimer's disease pathology as measured by amyloid PET and plasma p-tau217 and approximately one-third of the risk of dementia by age 78 compared with non-night/shift workers. Lower brain volume in night/shift workers was partially mediated by unhealthy behaviours. Reduced dementia risk in night/shift workers is unexpected and will require further study.

Keywords: Alzheimer’s disease; circadian rhythms; dementia; shift work; sleep.

Graphical Abstract

Figure 1

Flowchart of recruitment and data acquisition. NSHD participants were invited to participate in Insight 46 where a minimum dataset of life course data was available; this was subsequently relaxed to include those n = 62* where a previous measure of lung function, smoking or physical exercise was unavailable to meet the target sample size. Plasma p-tau217^‡ from participants attending Insight 46 (n = 485) was supplemented by samples taken from the wider NSHD cohort (n = 951) during the same collection period.

Figure 2

Night and shift working is associated with lower brain volume and less Alzheimer’s disease pathology. Forest plots demonstrating lower brain volume, Amyloid PET CLs, amyloid PET positivity and plasma p-tau217 in night/shift workers, without significant difference in hippocampal volume or WMHV by age 70. Plasma p-tau217^‡ from participants attending Insight 46 with night/shift working and covariate data available (n = 414) was supplemented by samples taken from the wider NSHD cohort (n = 653) during the same collection period. Linear/logistic regression was used for continuous/binary outcomes, respectively. The mean difference represents the covariate-adjusted difference in the mean imaging or plasma outcomes in night/shift workers, compared with non-shift workers. Case bootstrapping was used for the relationships with WMHV, Amyloid PET CLs and p-tau217, which are not normally distributed. All analyses are adjusted for sex, adult socioeconomic position, educational attainment, age at imaging/plasma, alongside TIV for volumetric measures and APOE ɛ4 carrier status for Amyloid PET and plasma p-tau217. n, number of participants included in the analysis, having all required data available; CL, Centiloids.

Figure 3

Night/shift working is associated with risk of dementia. Covariate-adjusted cumulative incidence plots demonstrating incidence of all-cause (excluding vascular) dementia (A) and vascular dementia (B) in night/shift workers compared with non-shift workers. Cox proportional cause-specific hazards regression models were used to compare rates of dementia diagnosis in night/shift workers and non-shift workers, censoring for the competing event of death and for age at emigration by age 78.0 years, and controlling for sex, educational attainment and socioeconomic position. n = 3040 for both analyses. Shaded areas represent the 95% CI. Plots are truncated to show only years during which all-cause dementia diagnoses occurred.

Figure 4

Night/shift working is related to life course health and behaviours, and life course risk factors are related to reduced brain volume. Forest plots demonstrating associations between night/shift working and cardiovascular risk factors/alcohol use across the life course (A–F) and the relationship between these factors and brain volume in the eighth decade (G–L). Linear/logistic regression was used for continuous/binary outcomes, respectively. Case bootstrapping was used where FRS, alcohol consumption or smoking are outcomes in the regression model as these are not normally distributed. Smoking is presented as the mean packs smoked per day during the presented time interval, or averaged over 5 years for those smoking prior to the age of 20. Sex, adult socioeconomic position and educational attainment were used as covariates in all analyses, alongside TIV and age at imaging for analyses including brain volume. Sample size varied for each analysis due to data availability: (A) FHS age 36 (n = 408), age 53 (n = 418) and age 70 (n = 428); (B) smoking up to age 20 (n = 408), age 20–25 (n = 407), age 25–36 (n = 403), 36–43 (n = 411), 43–53 (n = 416) and 53–63 (n = 403); (C) alcohol age 36 (n = 347), age 43 (n = 418), age 53 (n = 318) and age 63 (n = 309); (D) hypertension age 36 (n = 412), age 43 (n = 417), age 53 (n = 411), age 63 (n = 430) and age 69 (n = 423); (E) overweight age 36 (n = 412), age 43 (n = 419), age 53 (n = 423), age 63 (n = 431) and age 70 (n = 431); (F) exercise age 36, 43, 54 and 63 (n = 431), age 69 (n = 424); (G) FRS age 36 (n = 423), age 53 (n = 454) and age 70 (n = 463); (H) smoking up to age 20 (n = 427), age 20–25 (n = 421), age 25–36 (n = 391), age 36–43 (n = 430), age 43–53 (n = 437) and age 53–63 (n = 419); (I) alcohol age 36 (n = 360), alcohol age 43 (n = 445), alcohol age 53 (n = 340) and alcohol age 63 (n = 341); (J) hypertension age 36 (n = 426), age 43 (n = 445), age 53 (n = 441), age 63 (n = 467) and age 69 (n = 456); (K) overweight age 36 (n = 429), age 43 (n = 449), age 53 (n = 459) and age 63 and 70 (n = 468); (L) exercise age 36, 43, 53 and 63 (n = 468) and age 69 (n = 461). FRS, Framingham risk score; MD, adjusted mean difference.

Figure 5

Reduced brain volume in shift workers is partially mediated by increased life course cardiovascular risk factors. Causal mediation analysis examined the extent to which the relationship between night/shift working and brain volume was mediated by life course factors. Using a natural-effects model approach, this model included those life course risk factors shown to be most strongly related to both shift working and brain volume (defined as P ≤ 0.15 within 5 years of night/shift working, P ≤ 0.1 at later timepoints, see Fig. 4). FRS was represented by its constituent components (smoking, blood pressure and anti-hypertensive medications use) except for diabetes status, which was not included due to there being only one individual with diabetes at age 36. Mediation analyses were performed on multiply imputed datasets to mitigate potential bias from missing data, using multiple imputation of covariates by SMC-FCS (150 iterations, 150 imputations), including all participants with brain volume data available (n = 468).