Objective but not subjective sleep predicts memory in community-dwelling older adults

Authors

  • Marina G. Cavuoto,

    1. School of Psychology & Public Health, La Trobe University, Melbourne, Victoria, Australia
    Search for more papers by this author
  • Ben Ong,

    1. School of Psychology & Public Health, La Trobe University, Melbourne, Victoria, Australia
    Search for more papers by this author
  • Kerryn E. Pike,

    1. School of Psychology & Public Health, La Trobe University, Melbourne, Victoria, Australia
    Search for more papers by this author
  • Christian L. Nicholas,

    1. Melbourne School of Psychological Sciences, The University of Melbourne, Parkville, Victoria, Australia
    2. Institute for Breathing & Sleep, Heidelberg, Victoria, Australia
    Search for more papers by this author
  • Bei Bei,

    1. School of Psychological Sciences, Monash University, Clayton, Victoria, Australia
    2. Centre for Women's Health, Royal Women's Hospital, Melbourne, Victoria, Australia
    Search for more papers by this author
  • Glynda J. Kinsella

    Corresponding author
    1. School of Psychology & Public Health, La Trobe University, Melbourne, Victoria, Australia
    2. Caulfield Hospital, Caulfield, Victoria, Australia
    • Correspondence

      Professor Glynda Kinsella, School of Psychology & Public Health, La Trobe University, Melbourne, Victoria, Australia.

      Tel.: (+613)-9479-1381;

      fax: (+613) 9479 1956;

      e-mail: g.kinsella@latrobe.edu.au

    Search for more papers by this author

Summary

Research on the relationship between habitual sleep patterns and memory performance in older adults is limited. No previous study has used objective and subjective memory measures in a large, older-aged sample to examine the association between sleep and various domains of memory. The aim of this study was to examine the association between objective and subjective measures of sleep with memory performance in older adults, controlling for the effects of potential confounds. One-hundred and seventy-three community-dwelling older adults aged 65–89 years in Victoria, Australia completed the study. Objective sleep quality and length were ascertained using the Actiwatch 2 Mini-Mitter, while subjective sleep was measured using the Pittsburgh Sleep Quality Index. Memory was indexed by tests of retrospective memory (Hopkins Verbal Learning Test – Revised), working memory (n-back, 2-back accuracy) and prospective memory (a habitual button pressing task). Compared with normative data, overall performance on retrospective memory function was within the average range. Hierarchical regression was used to determine whether objective or subjective measures of sleep predicted memory performances after controlling for demographics, health and mood. After controlling for confounds, actigraphic sleep indices (greater wake after sleep onset, longer sleep-onset latency and longer total sleep time) predicted poorer retrospective (∆R2 = 0.05, = 0.016) and working memory (∆R2 = 0.05, = 0.047). In contrast, subjective sleep indices did not significantly predict memory performances. In community-based older adults, objectively-measured, habitual sleep indices predict poorer memory performances. It will be important to follow the sample longitudinally to determine trajectories of change over time.

Introduction

Reductions in the length and quality of sleep are common in normal ageing (Ohayon et al., 2004), and while recent research has evaluated the importance of sleep for general cognition (Yaffe et al., 2007) and memory consolidation post-learning (see Scullin and Bliwise, 2015 for comprehensive review), the relationship between habitual sleep patterns and memory in older adults has not been sufficiently studied (Cochrane et al., 2012; Naismith et al., 2011; Seelye et al., 2015; Wilckens et al., 2014). Deterioration in various domains of memory is common with age, and evaluating the relationship between habitual sleep patterns and domains of memory (such as retrospective memory, working memory and prospective memory) will contribute to strategies for improving cognitive health in ageing populations.

In studies using polysomnography, which examines cerebral sleep–wake states in addition to other physiological changes that occur with sleep, and is the gold standard of objective sleep measurement, indices of good sleep quality have been associated with better subsequent memory performance (Lafortune et al., 2014; for review, see Scullin and Bliwise, 2015). However, because polysomnography studies are usually conducted in sleep laboratories, both the monitoring equipment and sleeping in a novel environment can lead to disruption of a person's normal sleep, particularly on the first night of recording, known as first night effects (Edinger et al., 2001). Furthermore, in-laboratory studies generally restrict sample sizes, reducing the reliability of any found relationships. Actigraphy, which uses activity-based algorithms to infer sleep/wake states, offers an alternative form of objective sleep measurement. Whereas actigraphy cannot measure sleep stages or microarousals, it involves unobtrusive collection of successive nights of sleep data in naturalistic settings, thereby providing a measure of habitual sleeping patterns, and lends itself to the possibility of much larger sample sizes. Although the existing actigraphy literature suggests no relationship between habitual sleep duration and memory recall or working memory in older adults (Cochrane et al., 2012; Wilckens et al., 2014), consensus is lacking about whether indices of habitual poor sleep quality are associated with poorer memory (Cochrane et al., 2012; Naismith et al., 2011; Seelye et al., 2015; Wilckens et al., 2014). Interpretation of study findings is limited by the use of mixed-age samples, small sample sizes, possible inclusion of people with sleep apnoea, and the use of clinical samples that provide atypical age-related data. The relationship between habitual sleep patterns and memory in non-clinical, older-aged samples, therefore, requires further evaluation.

Research on habitual sleep patterns and memory in older adults has also been conducted using subjective measures of sleep; however, the area is again marked by significant inconsistencies in outcomes. Self-reported sleep quality, with the exception of sleep-onset latency (SOL; time taken to fall asleep), has not generally been associated with poor memory in older adults, after controlling for depression (Gamaldo et al., 2008; Jaussent et al., 2012; Nebes et al., 2009; Schmutte et al., 2007; Sutter et al., 2012; Tworoger et al., 2006; Waller et al., 2015). However, there are exceptions (Miller et al., 2014); and others have found self-reported poor sleep quality at age 70 years (but not age 50 years) to be associated with greater lifetime risk of developing dementia, in particular Alzheimer's disease (Benedict et al., 2015).

Shorter subjective sleep duration has been associated with poorer verbal memory performance in a large cohort study of 50–85 year olds (Xu et al., 2011). However, others have found short sleep duration to be associated with poorer memory in middle aged-adults (50–64 year olds), but not in older adults aged 65 years and above (Miller et al., 2014). In addition, a decline in sleep duration from a baseline of 7–8 h per night has not been associated with lower retrospective memory performance (Loerbroks et al., 2010); and while short subjective sleep duration in a sample spanning middle-age to older adulthood was associated with greater beta-amyloid burden (Spira et al., 2013), short sleep in midlife was not associated with subsequent increased risk of Alzheimer's disease in late life in a much larger cohort (Virta et al., 2013).

In contrast, longer subjective sleep duration has more consistently been associated with poorer memory and the development of dementia of the Alzheimer's type (Benito-León et al., 2009; Loerbroks et al., 2010; Miller et al., 2014; Schmutte et al., 2007; Virta et al., 2013; Xu et al., 2011), although not all findings have been in agreement (Nebes et al., 2009; Tworoger et al., 2006; Waller et al., 2015). A difficulty in explaining these discrepant findings is due to most studies employing subjective measures only, and the degree to which subjective report reflects actual sleep is questionable (Buysse et al., 2008). While the mechanisms for the observed relationships between long subjective sleep duration and poorer memory remain contended, a number of possibilities exist; for example, degeneration of cholinergic neurons in the basal forebrain (Stern and Naidoo, 2015) or underlying poor health (Fang et al., 2014; Ramos et al., 2014) could contribute to both. Another possible mechanism is that longer sleep may cause poor memory in older adults through a process of over-downscaling of synaptic weights (Scullin, 2013). Downscaling of synapses that accrue during wakefulness is a process that has been hypothesised to occur during slow-wave sleep to promote neural efficiency (Tononi and Cirelli, 2003). It has, however, also been proposed that a relative over-downscaling of synaptic weights can occur in the case of older adults in whom slow-wave sleep is preserved (albeit at lower levels), but the accrual of synaptic weights during the waking day is overly diminished as a result of ageing (Scullin, 2013). Finally, poorer memory may lead to longer sleep through a process of compensation, whereby older adults with poorer memory function are cognitively exhausted as a result of increased mental effort required for daily function, and thus sleep longer as a result. Further research, however, is required to substantiate these varied explanations.

Existing research is further limited in only examining memory performance through tests of retrospective memory (the retrieval of explicit, episodic information; Moscovitch, 1992) and working memory (the temporary storage and manipulation of information; Baddeley, 1992), even though researchers are now recognising that prospective memory (the capacity to remember to perform an intended action in the future; Einstein et al., 2012) is an important aspect of everyday memory. In younger adults, sleep, in particular slow-wave sleep, is involved in the consolidation of prospective memory intentions (Diekelmann et al., 2013; Scullin and McDaniel, 2010). The relationship between sleep and prospective memory has not been examined in older adults, but is important to evaluate as prospective memory has been identified as critical for the maintenance of independence in activities of daily living (Schmitter-Edgecombe et al., 2009).

The objective of this study, therefore, was to investigate whether habitual sleep patterns (using objective and subjective measures) predicted various domains of memory performance in community-living older people. Sleep measures were collected prior to memory assessment, although collection of prospective memory data and actigraphic sleep data were completed in parallel. The effects of potential confounds were mitigated by controlling for mood and health disorders, and people with reported sleep disorders were excluded. Although objective and subjective measures of sleep are often not associated, there is evidence that both may predict memory performance. It was, therefore, expected that both objective and subjective measures of good sleep quality would be associated with better memory, whereas objective and subjective measures indicating long sleep duration would be associated with poorer memory.

Materials and methods

Participants

Two-hundred older adult volunteers from Victoria, Australia were recruited through experimenter networks and advertisements in community-based organisations. Inclusion criteria were: (i) ≥ 65 years; (ii) fluent in English; (iii) independent in activities of daily living. Exclusion criteria were: (i) presence of diagnosed dementia or mild cognitive impairment; (ii) history of neurological or psychiatric disorder that may affect cognition; (iii) low cognitive status (Mini Mental State Examination score of < 24); (iv) uncorrected impairment of vision, hearing or communication that would interfere with study participation; (v) post hoc exclusion of those who self-reported a diagnosed sleep disorder. The research was approved by La Trobe University human ethics committee, and all participants provided written informed consent.

Assessments

As part of a larger study on memory in ageing, participants completed questionnaires prior to two neuropsychological assessment sessions, with measures of objective sleep, prospective memory and a subjective sleep diary collected between the two sessions, spaced 2 weeks apart.

Objective sleep

Actigraphy is a widely used instrument that objectively estimates sleep quantity and quality based on wrist movement measured using wristwatch-like devices. It is light-weight, non-intrusive and has been validated in older adults (van Hilten et al., 1993). Activity data were sampled in 1-min epochs using a medium (default) threshold for sleep/wake determination (Actiwatch 2 Mini-Mitter, Phillips-Respironics, OR, USA) over a fortnight [mean = 13.38 days (SD = 1.79)]. Bedtimes and rise-times were determined by concordance between light, activity and sleep diary data. Nocturnal sleep quality was indexed by: (i) wake after sleep onset [WASOActi (min)]; and (ii) SOL [SOLActi (min)]. Nocturnal sleep quantity was indexed by total sleep time (TSTActi; minutes in bed scored as sleep). These variables relate to distinct aspects of sleep measurement, have been used in other related research (Cochrane et al., 2012; Wilckens et al., 2014) and were not highly correlated (WASOActi and SOLActi: = 0.46, < 0.001; WASOActi and TSTActi: = −0.19, P = 0.006; and SOLActi and TSTActi: = −0.19, P = 0.006), thus would not lead to multicollinearity in regression.

Subjective sleep

The widely used Pittsburgh Sleep Quality Index (PSQI; Buysse et al., 1989) was completed by participants prior to the first cognitive assessment session (test–retest = 0.87; Backhaus et al., 2002). The PSQI assesses sleep over the month prior using 19 questions that are converted into seven subscale scores, ranging from 0 to 3 (with higher numbers indicating poorer sleep), the sum of which provides a global score. In order to use subscales that appear to provide a similar construct to the actigraphic sleep variables, and to avoid multicollinearity in regressions, the following variables were extracted for analysis: sleep quality was indexed by: (i) the subscale score for Sleep DisturbancePSQI; and (ii) the raw score for SOLPSQI. Sleep quantity was indexed by the raw score for TSTPSQI. The subscale score Daytime DysfunctionPSQI was also included to indicate the subjective impact of sleep on daily function. Incomplete responses to the questionnaires led to missing data on some of the PSQI variables (resulting in SOLPSQI = 164; TSTPSQI = 169; Daytime DysfunctionPSQI n = 171). While wearing the Actiwatch, participants completed a sleep diary, with bedtime and rise-time used to aid scoring of the actigraph.

Memory performance

The Hopkins Verbal Learning Test – Revised (HVLT-R; Brandt and Benedict, 2001) was administered, using the standard procedure, in the second assessment session. This task is a word-list memory task in which the experimenter orally presents a 12-item word-list from three semantic categories. There are three learning trials, and participants are asked to recall as many words as they can after each trial. This was followed by a 20–30 min delay with a subsequent delayed free-recall trial without representing the word-list; and the total number of correct words in the delayed free-recall trial was used as the index of retrospective memory. In the delay between the learning and delayed recall trials, the n-back task was administered (conditions 0-, 1- and 2-back). Working memory was assessed by the n-back task, 2-back condition (accuracy). This computer-based task consisted of two blocks of 30 trials of random lower case letters presented for 500 ms, with a fixed inter-stimulus interval of 2 s. Participants indicated, on a numeric keypad, whether or not the currently presented letter was the same as the one presented two previously, which occurred 10 times in each of the 30 trial blocks.

Prospective memory was assessed using the event marker button on the Actiwatch that had been given to participants in the first assessment session to objectively record sleep patterns. The event marker button records time-stamped information when pressed. At the end of the first assessment session, in addition to the general instructions about wearing the watch, participants were asked to press the event marker button daily when getting into bed and intending to go to sleep (i.e. at ‘lights out’), and upon rising from bed in the morning over the 2-week assessment period. A typed instruction sheet about the use of the Actiwatch was also provided to each participant, but they were not informed that the instructions contained a memory test. The Actiwatch was returned at the second assessment session. Successful prospective memory performance was determined by the proportion of button presses within 10 min of rise-time over the 2-week assessment period (Cavuoto et al., 2015). Rise-time prospective memory was selected for the study variable as it was relevant to investigating the relationship between sleep and prospective memory performance.

Due to logistic reasons, one participant was unable to complete the second assessment session, and data were lost for four participants following n-back administration, which led to = 172 for the HVLT-R and = 168 for the n-back.

Vigilance

Vigilance was objectively measured using a 5-min psychomotor vigilance task (PVT; Thorne et al., 2005). This was administered in the same testing session as the HVLT-R and n-back task (approximately 20 min prior to these tasks). Due to logistic reasons, data were only collected on 162 participants. The total number of lapses (i.e. response times > 500 ms) was included as a covariate.

Health

Three indices of health were used: (i) the number of self-reported vascular medical conditions (sum of transient ischaemic attack, heart problem, high cholesterol, pulmonary disease, high blood pressure and diabetes), an approach similar to that used by others (McKinnon et al., 2014); (ii) the number of self-reported, current medications; and (iii) the PSQI subscale for frequency of sleep medication use (dichotomised into those who reported medication use versus no medication use over the past month). For the latter question there were missing data for two participants.

Mood

The 21-item version of the Depression, Anxiety, and Stress Scale (DASS-21) is a well-validated measure of mood (Lovibond and Lovibond, 1993). Items are rated 0–3, with higher numbers indicating greater severity. Each scale is multiplied by two to reflect the full (42-item) version of the scale. Indices of depression, anxiety and stress were derived, with Cronbach's alpha levels of 0.72, 0.62 and 0.80, respectively.

Statistical method

Outliers were replaced with a value just outside of the median ± 3 × interquartile range (‘O’). Skewed or kurtotic variables were square-root (‘SQ’) or natural-log transformed (‘LN’) and, where necessary, were reflected before transformation (‘RSQ/RLN’). Analyses were conducted using transformed variables. The assumption of linearity was considered upheld. In order to determine the contribution of sleep variables to memory performance, over and above that accounted for by potentially confounding variables (i.e. demographics, health and mood), the following analyses were conducted. Two hierarchical regressions were performed to determine whether: (i) WASOActi, SOLActi and TSTActi; and (ii) Sleep DisturbancePSQI, SOLPSQI, TSTPSQI, Daytime DysfunctionPSQI predicted HVLT-R delayed recall, 2-back accuracy and rise-time prospective memory after controlling for age, gender, vascular conditions, number of medications, use of sleep medications, depression, anxiety and stress. Additional analyses were conducted to determine whether vigilance accounted for any observed relationships between sleep measures and retrospective and working memory performances. Associations between each variable are presented in a correlation matrix.

Results

Participant characteristics

From 200 participants who completed the study, actigraphy data on seven were lost due to recording malfunction, one participant chose not to wear the Actiwatch, and 19 participants were excluded due to sleep disorders (sleep apnoea: 13; restless legs syndrome: 4; insomnia:1; delayed sleep phase syndrome: 1). The mean age of the final sample (= 173) was 73.78 years (SD = 5.73 years), 65.3% were female, 55.5% had more than 12 years of education, and the average estimated intelligence (Test of Premorbid Function; Wechsler, 2009) was in the average range at 109.37 (SD = 11.26).

Retrospective memory performance was average (Table 1), and levels of depression, anxiety and stress were comparable to normative data (Crawford and Henry, 2003). The average actigraphically derived bedtime was 22:55 hours (SD = 54.14 min), and the average rise-time was 07:22 hours (SD = 48.07 min). Normal levels of sleep complaint were reported by 58% of the sample (i.e. global PSQI ≤  5). Other sleep parameters are listed in Table 1. Because the dependent variables (HVLT-R delayed recall, 2-back accuracy and rise-time PM) were not correlated more than 0.7 (i.e. HVLT-R delayed recall and 2-back accuracy = 0.27; HVLT-R delayed recall and rise-time PM = 0.22; and 2-back accuracy and rise-time PM = 0.16), the following analyses were not corrected for multiple comparisons.

Table 1. Descriptive statistics on memory, sleep, health and mood
 Mean/median (SD/IQR)*
  1. Acti, actigraphy; HVLT-R, Hopkins Verbal Learning Test – Revised; IQR, interquartile range; PM, prospective memory; PSQI, Pittsburgh Sleep Quality Index; SD, standard deviation; Sleep medications (% Y), proportion of sample who endorsed use of sleep medication over the past month; SOL, sleep-onset latency; TST, total sleep time; WASO, wake after sleep onset.

  2. Note. The following transformations were applied to variables used in the analysis: O = outlier transformed; R = reflected prior to transformation; SQ = square root transformed; LN = natural log transformed.

  3. *Medians and interquartile ranges reported for skewed variables.

HVLT-R total trials 1–324.83 (5.28)
HVLT-R total t-score51.47 (10.38)
HVLT-R delayRSQ9.00 (4.00)*
HVLT-R delay t-score52.00 (18.00)*
2-back accuracyO.RLN (total)48.00 (9.00)*
PM % accuracy57.16 (26.20)
WASOActiO.SQ (min)47.86 (22.04)*
SOLActiO.LN (min)6.14 (7.93)*
TSTActiO (h: min)7:16.79 (1:12.36)*
Sleep DisturbancePSQI (subscale score)1.23 (0.50)
SOLPSQIO.SQ (min)15.00 (20.00)*
TSTPSQI (min)418.58 (65.98)
Daytime DysfunctionPSQI (subscale score)0.59 (0.58)
Global PSQI5.00 (5.00)*
Number of vascular conditions1.29 (1.19)
Number of medicationsSQ3.00 (4.00)*
Sleep medications (% Y)(20.8%)
DepressionO.LN2.00 (4.00)*
AnxietyO.LN0.00 (4.00)*
StressLN4.00 (8.00)*

Prediction of memory performances

Retrospective memory

For HVLT-R delayed recall, in Step 1, age, gender, vascular conditions, number of medications, use of sleep medications, depression, anxiety and stress explained 22.8% of the variance (R2 = 0.23, F8,161 = 5.95, P < 0.001), with better performance associated with younger age (sr = −0.21, = 0.002), female gender (sr = 0.18, = 0.010) and lower anxiety (sr = −0.16, = 0.019). In Step 2, WASOActi, SOLActi and TSTActi explained an additional 4.8% of the variance (∆R2 = 0.05, ∆F3,158 = 3.52, = 0.016), with longer TSTActi and SOLActi being associated with poorer performance (sr = −0.16, = 0.020 and sr = −0.13, = 0.056, respectively; Table 2). In contrast, using subjective sleep measures (Sleep DisturbancePSQI, SOLPSQI, TSTPSQI, Daytime DysfunctionPSQI) in Step 2 did not explain significant additional variance (∆R2 = 0.02, ∆F4,149 = 0.72, = 0.581; Table 3). A further hierarchical regression was performed to determine whether vigilance explained the relationship between actigraphic sleep and HVLT-R delayed recall. In Step 1, PVT lapses were added to the demographics, which in total explained 24.2% of the variance (R2 = 0.24, F9,150 = 5.32, P < 0.001); however, PVT lapses did not contribute significant unique variance to the model (sr = −0.12, = 0.108), and the amount of variance explained by the subsequent actigraphic variables in Step 2 remained much the same, 4.6% of the variance (∆R2 = 0.05, ∆F3,147 = 3.14, = 0.027), with similar unique contributions from TST and SOL (sr = −0.16, = 0.025 and sr = −0.12, = 0.080, respectively).

Table 2. Summary of regression and hierarchical regression analyses for objective measures of sleep, demographic, health and mood variables predicting retrospective memory, working memory and prospective memory
Dependent variablePredictor blockChange statisticsOverall model
ΔR 2 ΔF PredictorsβsrrR2dfF
  1. HVLT-R, Hopkins Verbal Learning Test – Revised; Meds, number of medications; PM, prospective memory; Sleep meds, use of sleep medication over the past month; SOL, sleep-onset latency; TST, total sleep time; Vascular, number of vascular conditions; WASO, wake after sleep onset.

  2. The following transformations were applied to variables used in analysis: O = outlier transformed; R = reflected prior to transformation; SQ = square root transformed; LN = natural log transformed.

  3. To aid interpretation, a negative sign for coefficients has only been used when the relationship is negative, regardless of whether the variable was reflected in transformation.

  4. Note. The degrees of freedom reported in the table vary because of missing data on some variables.

  5. *P < 0.05, **P < 0.01, ***P < 0.001.

  6. For each dependent variable, the first block shows the contribution of demographics, medical factors and mood, and the second block shows the additional contribution of objective sleep.

HVLT-R delayed recallRSQ1. Demographics, health and mood0.235.95***Age−0.23**−0.21−0.300.23(8, 161)5.95
Gender (F)0.19*0.180.23
Vascular−0.17−0.14−0.27
MedsSQ−0.08−0.06−0.28
Sleep meds0.030.030.02
DepressionO.LN0.090.08−0.07
AnxietyO.LN−0.18*−0.16−0.21
StressLN−0.13−0.12−0.13
2. Objective sleep0.053.52*WASOO.SQ−0.07−0.06−0.160.28(3, 158)5.50
SOLO.LN−0.15−0.13−0.19
TSTO−0.17*−0.16−0.12
2-back accuracyO.RLN1. Demographics, health and mood0.091.84Age−0.15−0.14−0.140.09(8, 157)1.84
Gender (F)0.080.080.09
Vascular−0.10−0.08−0.08
MedsSQ0.070.05−0.03
Sleep meds0.130.120.12
DepressionO.LN0.19*0.170.09
AnxietyO.LN−0.14−0.12−0.10
StressLN−0.10−0.09−0.06
2. Objective sleep0.052.71*WASOO.SQ−0.10−0.09−0.120.13(3, 154)2.12
SOLO.LN−0.13−0.12−0.16
TSTO−0.15−0.14−0.09
Rise-time PM accuracy1. Demographics, health and mood0.051.08Age0.060.05< 0.010.05(8, 162)1.08
Gender (F)0.020.020.02
Vascular−0.04−0.03−0.11
MedsSQ−0.17−0.13−0.16
Sleep meds0.050.050.02
DepressionO.LN0.160.140.10
AnxietyO.LN−0.05−0.05−0.03
StressLN−0.06−0.06−0.04
2. Objective sleep0.031.78WASOO.SQ−0.11−0.10−0.160.08(3, 159)1.29
SOLO.LN−0.05−0.04−0.11
TSTO0.090.080.09
Table 3. Summary of regression and hierarchical regression analyses for subjective measures of sleep, demographic, health and mood variables predicting retrospective memory, working memory and prospective memory
Dependent variablePredictor blockChange statisticsOverall model
ΔR 2 ΔF PredictorsβsrrR2dfF
  1. Day-Dysf., daytime dysfunction; Disturbance, sleep disturbance; HVLT-R, Hopkins Verbal Learning Test – Revised; Meds, number of medications; PM, prospective memory; Sleep meds, use of sleep medication over the past month; SOL, sleep-onset latency; TST, total sleep time; Vascular, number of vascular conditions.

  2. The following transformations were applied to variables used in analysis: O = outlier transformed; R = reflected prior to transformation; SQ = square root transformed; LN = natural log transformed.

  3. To aid interpretation, a negative sign for coefficients has only been used when the relationship is negative, regardless of whether the variable was reflected in transformation.

  4. Note. The degrees of freedom reported in the table vary because of missing data on some variables.

  5. *P < 0.05, **P < 0.01, ***P < 0.001.

  6. For each dependent variable, the first block shows the contribution of demographics, medical factors and mood, and the second block shows the additional contribution of subjective sleep.

HVLT-R delayed recallRSQ1. Demographics, health and mood0.235.65***Age−0.23**−0.21−0.300.23(8, 153)5.65
Gender (F)0.19*0.180.23
Vascular−0.17−0.14−0.27
MedsSQ−0.08−0.06−0.28
Sleep meds0.030.030.02
DepressionO.LN0.090.08−0.07
AnxietyO.LN−0.18*−0.16−0.21
StressLN−0.13−0.12−0.13 
2. Subjective sleep0.020.72Disturbance−0.03−0.03−0.050.24(4, 149)3.98
SOLO.SQ−0.08−0.07−0.07
TST−0.02−0.020.03
Day-Dysf.−0.11−0.09−0.14
2-back accuracyO.RLN1. Demographics, health and mood0.091.75Age−0.15−0.14−0.140.09(8, 150)1.75
Gender (F)0.080.080.09 
Vascular−0.10−0.08−0.08
MedsSQ0.070.05−0.03
Sleep meds0.130.120.12
DepressionO.LN0.19*0.170.09
AnxietyO.LN−0.14−0.12−0.10
StressLN−0.10−0.09−0.06
2. Subjective sleep0.021.00Disturbance−0.06−0.05−0.010.11(4, 146)1.50
SOLO.SQ< 0.01< 0.010.02
TST−0.12−0.11−0.11
Day-Dysf.−0.09−0.07−0.03
Rise-time PM accuracy1. Demographics, health and mood0.051.03Age0.060.05< 0.010.05(8, 153)1.03
Gender (F)0.020.020.02
Vascular−0.04−0.03−0.11
MedsSQ−0.17−0.13−0.16
Sleep meds0.050.050.02
DepressionO.LN0.160.140.10
AnxietyO.LN−0.05−0.05−0.03
StressLN−0.06−0.06−0.04
2. Subjective sleep0.031.28Disturbance< 0.01< 0.01−0.020.08(4, 149)1.12
SOLO.SQ−0.18*−0.17−0.19
TST0.030.030.04
Day-Dysf.−0.01−0.010.05

Working memory

For 2-back accuracy, in Step 1, age, gender, vascular conditions, number of medications, use of sleep medications, depression, anxiety and stress did not explain significant variance (R2 = 0.09, F8,157 = 1.84, P = 0.074). In Step 2, WASOActi, SOLActi and TSTActi explained an additional 4.6% of the variance (∆R2 =  0.05, F3,154 = 2.71, = 0.047), although no individual variable contributed significant unique variance (Table 2). In contrast, subjective sleep (Sleep DisturbancePSQI, SOLPSQI, TSTPSQI, Daytime DysfunctionPSQI) only accounted for an additional 2.4% of the variance in Step 2 (∆R2 = 0.02, ∆F4,146 = 1.00, = 0.411; Table 3). Another hierarchical regression was also performed to determine whether vigilance explained the relationship between actigraphic sleep and 2-back accuracy. In Step 1, PVT total lapses was added to the demographics, which still only explained 8.6% of the variance (R2 = 0.09, F9,149 = 1.56, = 0.133), and the actigraphic variables in Step 2 explained the same amount of variance as previously (i.e. 4.6%), although the overall step became marginally non-significant probably due to loss of power from having fewer participants (∆R2 = 0.05, ∆F3,146 = 2.55, = 0.058).

Habitual prospective memory

For prospective memory accuracy, in Step 1, age, gender, vascular conditions, number of medications, use of sleep medications, depression, anxiety and stress did not explain significant variance (R2 = 0.05, F8,161 = 1.08, = 0.378). In Step 2, WASOActi, SOLActi and TSTActi did not explain significant additional variance (∆R2 = 0.03, ∆F3,158 = 1.78, = 0.152; Table 2). Similarly, using Sleep DisturbancePSQI, SOLPSQI, TSTPSQI and Daytime DysfunctionPSQI in Step 2 did not account for significant additional variance (∆R2 = 0.03, ∆F4,149 = 1.28, = 0.280; Table 3). Although neither objective nor subjective sleep predicted prospective memory performance, significant zero-order correlations were observed between poorer prospective memory, and longer objective WASOActi (= −0.16, = 0.018) and longer subjective SOLPSQI (= −0.19, = 0.007; see Table 4 for correlations). Four of the 173 participants made no button presses upon rising from bed throughout the entire study period, which may indicate that they had forgotten or did not understand the instructions. After re-running the analyses without these four participants, prospective memory results were essentially unchanged.

Table 4. Correlations between objective memory, sleep, demographics, mood and health
 HVLTRSQ2-backO.RLNPMWASOActi O.SQSOLActi O.LNTSTActiOAgeGend.Vasc.MedsSQSleep medsDepO.LNAnxO.LNStressLNDist.PSQISOLPSQI O.SQTSTPSQIDysf.PSQIPVTO.LN
  1. 2-back, n-back, 2-back accuracy; Anx, anxiety; Dep, depression; Dist.PSQI, PSQI sleep disturbance subscale; Dysf.PSQI, PSQI daytime dysfunction subscale; HVLT, Hopkins Verbal Learning Test (HVLT)-R delayed recall; Meds, number of medications; PM, prospective memory; PSQI, Pittsburgh Sleep Quality Index; PVT, psychomotor vigilance task, number of PVT lapses > 500 ms; Sleep meds, use of sleep medications over the past month (positive association = association with use of sleep medications, negative association = association with lack of use of medications); SOL, sleep-onset latency; TST, total sleep time; Vasc, number of vascular conditions; WASO, wake after sleep onset.

  2. For gender, positive associations = greater relationship with women, negative associations = greater relationship with men.

  3. The following transformations were applied to variables used in analysis: O = outlier transformed; R = reflected prior to transformation; SQ = square root transformed; LN = natural log transformed.

  4. To aid interpretation, a negative sign for coefficients has only been used when the relationship is negative, regardless of whether the variable was reflected in transformation.

  5. *P < 0.05, **P < 0.01, ***P < 0.001.

HVLTRSQ                  
2-backO.RLN0.27***                 
PM0.22**0.16*                
WASOActi O.SQ−0.16*−0.12−0.16*               
SOLActi O.LN−0.19**−0.16*−0.110.46***              
TSTActi O−0.12−0.090.09−0.19**−0.19**             
Age−0.30***−0.14*< 0.01−0.02−0.040.12            
Gend.(F)0.23**0.090.02−0.05−0.13*0.23**−0.11           
Vasc−0.27***−0.08−0.110.100.070.030.17*−0.16*          
MedsSQ−0.28***−0.03−0.16*0.12−0.010.110.36***−0.110.57***         
Sleep meds0.020.120.020.110.040.20**0.050.20**< 0.010.14*        
DepO.LN−0.070.090.100.130.02−0.13*0.01−0.18**0.090.07−0.04       
AnxO.LN−0.21**−0.10−0.030.20**0.050.030.060.04−0.020.090.070.33***      
StressLN−0.13−0.06−0.040.090.05< 0.01−0.130.04−0.02−0.020.070.28***0.34***     
Dist.PSQI−0.05−0.01−0.02−0.18**0.050.09−0.12−0.010.040.100.25***0.110.13*0.13    
SOLPSQI O.SQ−0.070.02−0.19**0.16*0.24**0.05−0.020.17*0.050.070.21**−0.070.070.22**0.11   
TSTPSQI0.03−0.110.040.020.010.18*−0.15*0.07−0.04−0.04−0.23**0.030.04−0.07−0.03−0.18*  
Dysf.PSQI0.14*−0.030.05−0.010.01−0.050.10−0.090.110.15*0.14*0.45***0.110.16*0.14*−0.100.09 
PVTO.LN−0.23**−0.06−0.100.030.060.040.28***−0.020.20**0.19**0.010.10−0.020.16*−0.030.030.020.15*

Discussion

The primary findings indicate that objective, but not subjective, sleep predicted memory performance. Specifically, after controlling for demographics, health and mood, the collective contribution of actigraphically measured longer WASO, SOL and TST were predictive of poorer retrospective memory (HVLT-R delayed recall) and poorer working memory (2-back accuracy), with greater unique contributions coming from TST and SOL. The inclusion of both objective and subjective sleep measures, the large sample of community-living older adults, and screening for self-reported sleep disorders were the study's major strengths.

In demonstrating that longer objectively measured TST was negatively associated with memory performance, the present study extended previous research in older adults that has relied on self-report to associate sleep quantity with poorer memory (Miller et al., 2014). One possible explanation relating memory decline and age-related changes in sleep–wake control would be if they shared a common underlying aetiology; for example, age-related reductions in wake-promoting neurons, such as the cholinergic neurons of the nucleus basalis of Meynert, are implicated in both memory performance and sleep–wake control (for review, see Stern and Naidoo, 2015). Furthermore, in older adults, self-reported longer sleep duration has been associated with other negative health outcomes, such as higher prevalence of stroke (Fang et al., 2014) and increased brain white-matter hyperintensity volumes (Ramos et al., 2014), which are associated with decreased memory performance (Silbert et al., 2008). The possibility that objectively measured, long TST in the current study indirectly reflected poor sleep quality was not supported by the current data, as the correlation between actigraphic time in bed and sleep efficiency (the proportion of time spent asleep compared with time spent in bed) was low (= −0.07, = 0.361). However, such an association may be more likely to be observed in individuals with insomnia who, in an attempt to compensate for perceived lack of sleep, may stay in bed for longer periods, thus leading to lower sleep efficiency (Friedman et al., 1991). Alternatively, long sleep duration could reflect less restorative sleep in a way that is not detected by actigraphy, or the effect of long sleep may be mediated by sleep stage or other neurophysiological markers that occur during sleep, for example rapid eye movement sleep or sleep spindles (Lafortune et al., 2014). A further theory suggests that in older adults, slow-wave sleep may in fact be deleterious for memory (Scullin, 2013). Synaptic downscaling has been proposed as a process that occurs during slow-wave sleep in which there is a pruning of synaptic weights that accrue over the waking day as a result of experiences (Tononi and Cirelli, 2003). It has been suggested that slow-wave sleep could have a negative impact on memory in older adults through a process of synaptic over-downscaling when the accrual of synaptic weights has decreased as a process of ageing but downscaling during slow-wave sleep is maintained (Scullin, 2013). However, support for this theory is difficult to gauge as sleep stages were not measured in the current study.

The current study also demonstrated a relationship between longer objective SOL and poorer memory, adding to the growing body of evidence suggesting that longer SOL is associated with lower performance in other domains, such as attention allocation (Yaffe et al., 2007). The current finding was not accounted for by mood or age, which has previously been associated with SOL (Naismith et al., 2011; Ohayon et al., 2004). It has been speculated that difficulties with sleep onset and maintenance with age may reflect, as is observed with other aspects of sleep, broader changes in cortical volume and structure (Buysse et al., 2005), and for this reason may be associated with reduced memory performance. Alternatively, preliminary findings have indicated an association between longer subjective SOL and greater beta-amyloid burden in healthy adults [Branger et al., (conference abstract); Spira et al., 2013]. However, whether underlying Alzheimer's disease pathology could explain the relationship between longer objective SOL and poorer memory in the current study is speculative, and further research is required to understand the potential mechanism.

Lack of a strong relationship between prospective memory and sleep may be due to the nature of the prospective memory task, in a naturalistic setting. Because older adults often outperform younger adults in naturalistic settings (Rendell and Craik, 2000), this may indicate that older people with subtle memory decline may be able to modify their behaviour in naturalistic settings, thereby mitigating cognitive impacts that sleep may have on performance. The prospective memory task (i.e. pressing the Actiwatch Event Marker button within 10 min of rising from bed) may, however, have been affected by sleep inertia, which has been shown to occur in older adults over the first 30 min following waking (Silva and Duffy, 2008), and could explain why sleep did not predict prospective memory performance.

The current study found no association between memory and subjective sleep quality or quantity, which is consistent with some (Sutter et al., 2012), but not other research (Miller et al., 2014; Schmutte et al., 2007). Previous studies are likely to have included people with sleep disorders (Miller et al., 2014; Schmutte et al., 2007; Waller et al., 2015), such as sleep apnoea, which is independently associated with memory impairment (Wallace and Bucks, 2013), and may account for some of the associations previously reported between subjective sleep measures and memory performance. Furthermore, a difficulty with interpreting subjective measures of sleep is that they do not always correlate highly with objective measures of sleep, particularly in older adults (Buysse et al., 2008), and this was also observed in the current study.

The current findings are important because of the significant inconsistencies and lack of consensus in previous research about how sleep predicts memory in older adults, particularly in non-clinical populations. Although controlling for a variety of potential confounds was attempted, limitations include the reliance on self-reported presence of sleep disorders, number of medications and vascular medical conditions. It is therefore possible that the sample contained people with undiagnosed sleep disorders. However, sleep disorders are commonly, although not always, associated with daytime dysfunction (as indicated by the PSQI), of which there were not high levels in this sample, nor did daytime dysfunction predict memory performances. Both of these observations would indicate that the association between actigraphic sleep indices and memory performances are less likely to be explained by the presence of sleep disorders. A further limitation is that cognitive testing occurred at different times of day across participants and should be controlled in further studies even though, as noted in the statistical method section, controlling for vigilance, which may change at different times of the day, did not change the study results. Due to the cross-sectional design, the present study cannot confidently address the causal directions on reported associations. This suggests the importance of further longitudinal research to determine trajectories of change over time, as understanding factors that affect memory decline will have implications for the early detection and treatment of people at risk of developing memory impairment with advancing age. The relevance of the findings is highlighted when considering the body of research demonstrating a relationship between sleep disturbance and Alzheimer's disease (Peter-Derex et al., 2015). Further development of behavioural and neurobiological models will be useful in explaining the mechanism behind the current findings. For example, future studies may be able to test whether there is a causal association between longer sleep and poorer memory by restricting sleep (minimally) in older adults with long sleep duration to see whether memory performances improve.

Acknowledgements

The authors would like to acknowledge Fenny Muliadi and Stephen D. Lee for assistance in recruitment and data collection. This work was supported by the Mason Foundation, ANZ Trustees (grant number 13039 to C. L. N.). This research was performed during the tenure of an Award from Alzheimer's Australia Dementia Research Foundation for Ms Cavuoto. Dr Pike is funded by a National Health and Medical Research Council of Australia Clinical Research Training Fellowship (602543).

Author contributions

This study was completed by the first author (M. C.) in partial fulfilment of the requirements for the degree of PhD. M. C. was responsible for data collection and initial manuscript drafting. All authors were involved in the design, analysis and manuscript preparation.

Ancillary