SEARCH

SEARCH BY CITATION

Keywords:

  • activity;
  • ageing;
  • exercise;
  • insomnia;
  • sleep

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Method
  5. Sample
  6. Questionnaire assessment
  7. Physical activity
  8. Health and wellbeing
  9. Sleep
  10. Statistical analyses
  11. Results
  12. Discussion
  13. References

Laboratory evidence linking exercise with improved sleep quality raises the possibility that the lower levels of physical activity characteristic of older age groups may contribute to late-life insomnia. While support for this hypothesis appears to come from epidemiological surveys, few such studies have distinguished satisfactorily between social and physical activities which differ widely in terms of energy cost and theoretical significance. The present analyses were, therefore, designed to assess the independent influence of physical and social activity levels on the prevalence and natural history of late-life insomnia. Survivors from a nationally representative UK sample (n = 1042) of elderly people originally interviewed in 1985 were reassessed in 1989 (n = 690) and 1993 (n = 410). Detailed assessments of physical and social activities, mental and physical health status, and sleep quality were made at each survey wave. Logistic regression models, adjusted for age, sex and health status, were used to assess relationships between activity levels and the prevalence, remission/persistence, and incidence of late-life insomnia. Lower physical health, depressed mood and lower physical (but not social) activity levels consistently emerged as significant risk factors for prevalent, persistent and incident insomnia. Age was unrelated to insomnia variables in all the cross-sectional models, but did emerge as a significant risk for cumulative 4–8-year insomnia incidence. These findings suggest that, independent of those activities more closely associated with social engagement, higher levels of customary physical activity per se appear to be protective against incident and chronic late-life insomnia.


Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Method
  5. Sample
  6. Questionnaire assessment
  7. Physical activity
  8. Health and wellbeing
  9. Sleep
  10. Statistical analyses
  11. Results
  12. Discussion
  13. References

Evidence from laboratory and intervention studies linking exercise with improved sleep quality (see Driver and Taylor, 2000) raises the possibility that late-life insomnias may be mediated, at least in part, by the lower levels of physical activity characteristic of older age groups (see Health Education Authority, 1992; Ruchlin and Lachs, 1999; DiPietro, 2001). Support for this hypothesis appears to come from recent epidemiological studies which report significantly elevated levels of insomnia risk among the less active (Habte-Gabr et al., 1991; Janson et al., 2001; Ohayon et al., 2001; Sherrill et al., 1998). However, while laboratory findings suggest the existence of distinct physiological mechanisms connecting exercise with increased sleep need (Driver and Taylor, 2000), interpretation of the epidemiological data is complicated by the range of ‘activity’ definitions employed. Unlike laboratory studies, where ‘activity’ tends to be narrowly defined in terms of the duration and intensity of physical exercise, epidemiological sleep studies have used a variety of ‘activity’ definitions ranging from ‘exercise participation’ (Sherrill et al., 1998) through ‘occupational’ (Ohayon et al., 2001), ‘recreational’ (Janson et al., 2001), ‘social’ (Habte-Gabr et al., 1991) and ‘personal maintenance’ (Newman et al., 1997) activities. As these activities may vary considerably in energy cost and physiological impact, it cannot be assumed that their influence on sleep reflects a common physiological mechanism. Rather, some habitual activities which are emotionally and intellectually satisfying may, irrespective of energy cost, contribute to sleep quality through their contribution to mood and wellbeing. It has been shown, for example, that passive engagement in ‘stimulating’ daytime activities can significantly affect both sleep architecture and the subjective need for sleep (Horne and Minard, 1985). In order, therefore, to clarify relationships between activity and sleep in epidemiological studies of later life, it is important to discriminate between activity as physiologically relevant ‘exercise’, and activity as participation in the social milieu. It is also the case that both physical and social activity levels confound with health status. This is particularly relevant in older populations where disturbed sleep and lower activity (social and physical) are likely to result from mental or physical health problems.

Using a range of activity measures, the present analyses were designed to explore relationships between self-reported activity levels and sleep quality, and assess whether, in models containing relevant indices of social activity and health, physical activity levels are independently prognostic for late-life insomnia. To address this issue, activity–insomnia relationships were examined in three areas: (i) cross-sectional relationships between activity and (prevalent) insomnia; (ii) longitudinal relationships between activity and subsequent incident insomnia; and (iii) longitudinal relationships between activity and the subsequent remission of insomnia symptoms.

Sample

  1. Top of page
  2. Summary
  3. Introduction
  4. Method
  5. Sample
  6. Questionnaire assessment
  7. Physical activity
  8. Health and wellbeing
  9. Sleep
  10. Statistical analyses
  11. Results
  12. Discussion
  13. References

Data were derived from the Nottingham Longitudinal Study of Activity and Ageing (NLSAA), full details of which are presented elsewhere (Morgan, 1998). Briefly, the NLSAA is an ongoing study of activity and health outcomes among people originally aged 65 years and over, living in the UK. The baseline survey for the NLSAA was conducted in 1985, at which time 1042 people, randomly sampled from general practitioners’ lists, were interviewed in their own homes (a response rate of 80%). In order to admit sufficient numbers for subsequent longitudinal analyses, those aged 75 years or older were intentionally over-sampled. Follow-up surveys were conducted in 1989 and 1993, with re-interview rates of 88% (n = 690) and 73% (n = 410), respectively obtained among survivors (Table 1). Information on mortality within the baseline sample provided by the UK National Health Service Central Register allows for ongoing survival analyses of the NLSAA dataset.

Table 1.  Study sample and attrition 1985–1993
1985 (baseline) survey 
 Original random sample1299
 Respondents interviewed in 19851042
 Response rate (1985)  80%
1989 (follow-up) survey 
 No. of survivors (to 1989)781
 Refused interview63
 Untraceable25
 Emigrated3
 Interviews with survivors (1989)690
 Reinterview rate (1989)  88%
1993 (follow-up survey) 
 No. of survivors (to 1993)564
 Refused interview50
 Untraceable11
 Emigrated2
 Interviews with survivors (1993)410
 Reinterview rate (1993)  73%

Physical activity

  1. Top of page
  2. Summary
  3. Introduction
  4. Method
  5. Sample
  6. Questionnaire assessment
  7. Physical activity
  8. Health and wellbeing
  9. Sleep
  10. Statistical analyses
  11. Results
  12. Discussion
  13. References

At each survey wave, levels of customary physical activity (CPA) likely to promote muscle strength, joint flexibility, or stamina were assessed using detailed activity inventories administered in a single face-to-face interview. Activities were divided into seven mutually exclusive categories. Continuous activities (those with a probable minimum energy cost of 2 kcal  min−1, performed for a minimum of 3 min, at least weekly, for at least the previous 6 weeks) provided five categories: outdoor productive activities (e.g. gardening); indoor productive activities (e.g. housework); walking (purposeful walking outside the home); shopping; and leisure activities. In administering the questionnaire on outdoor, indoor, and leisure activities the interviewer first determined whether the respondent's participation in the activity met the criteria for ‘customary’, and then asked in detail about the frequency and duration of participation. Each reported activity was scored as minutes per week. Non-participation was scored as zero.

In the assessment of walking, the interviewer asked in detail about walking carried out on the day prior to interview. If, however, this day had been atypical, then another was selected (up to a maximum of 6 days previously). At analysis, walking per se, and walking as shopping were divided into two separate categories. Both were scored as minutes per typical day. In addition, two further categories of non-continuous activities likely to contribute to muscle strength (e.g. climbing high steps) and joint flexibility (e.g. reaching for high shelves) were also included. Typically, these tasks form discrete units of physical activity and were, therefore, scored in terms of frequency of performance on a 5-point scale (i.e. performed never, occasionally, once or several times a week, daily, or several time a day). For a full descriptive overview of the activity data at each survey wave (see Armstrong and Morgan 1998).

At each of the three survey waves factor analysis (principal components with varimax rotation) of the seven CPA categories was used to extract principal components and associated factor scores. For the baseline (1985) data, the first principal component accounted for 33% of total physical activity variance, with factors scores derived from this component correlating positively and significantly with anthropometric measurements of handgrip strength and shoulder flexibility (Morgan et al., 1991). In subsequent survival analyses controlling for health at the time of activity measurement, the same factor scores were significantly predictive of 10-year (Morgan and Clarke, 1997a) and 12-year (Bath and Morgan, 1998) all-cause mortality in both sexes (with scores indicative of lower activity associated with higher mortality). First principal components from the 1989 and 1993 waves similarly accounted for 32.5 and 36% of total physical activity variance, respectively.

Given that: (i) all continuous activities were required to meet an estimated energy cost threshold; (ii) all first principal component factor scores correlated significantly with anthropometric measurements; (iii) the baseline activity factor structure was replicated over two subsequent survey waves; and (iv) principal component factor scores were significantly associated longer-term mortality, we concluded that the CPA factor scores provided valid and reliable measurements of physiologically relevant physical activity in this population. In the present analyses, therefore, the factor scores derived from these first principal components at each survey wave (and termed CPA1) provide the main indices of physical activity. In addition, as moderate levels of walking (six blocks per day at average pace) have been associated with reduced insomnia risk (Sherrill et al., 1998), and as NLSAA baseline walking scores (minutes/typical day) showed only a modest correlation with CPA1 scores (1985: r = 0.1; 1989: r = 0.1; 1993: r = 0.2), NLSAA walking scores were also included as possible predictors of sleep quality in the present analyses.

Health and wellbeing

  1. Top of page
  2. Summary
  3. Introduction
  4. Method
  5. Sample
  6. Questionnaire assessment
  7. Physical activity
  8. Health and wellbeing
  9. Sleep
  10. Statistical analyses
  11. Results
  12. Discussion
  13. References

General health was assessed using a health index checklist scored from zero (no health problems) to 13 (multiple health problems) covering the presence or absence of: heart, stomach, eyesight, or foot problems; giddiness, headaches, urinary incontinence, arthritis and falls; long-term disabilities, and drug and walking aid use, and contact with (primary and secondary care) medical services. Depression was assessed using the Symptoms of Anxiety and Depression (SAD) Scale with the cutpoint set at 6. At baseline, scores at or above this cutpoint showed high levels of concordance with clinical diagnostic ratings of depression (kappa coefficient = 0.7, P < 0.001; see Morgan, 1998).

Levels of social activity were assessed using the 20-item Brief Assessment of Social Engagement (BASE) Scale developed specifically for this study. BASE includes items on voting, attending religious services, taking holidays, library attendance, letter writing, reading newspapers/magazines, TV and radio use, receiving or making telephone calls, and expressed satisfaction with close social contacts (i.e. family and friends). At baseline BASE scores showed an acceptable level of reliability (α = 0.7), and correlated significantly with CPA1 scores (r = 0.3) and SAD scores (r = −0.3; see Morgan, 1998), with greater engagement associated with fewer symptoms of depression in the latter case.

As obesity has been reported to predict both the onset and course of insomnia (Janson et al., 2001), the present analyses also included the body mass index [weight/(estimated height)2], where height was estimated from the demi-span, a measurement of skeletal size less influenced by kyphotic changes in the older spine (Bassey, 1986). In order to reduce spurious correlations among predictor and dependent variables, two of the assessment scales were modified. An item assessing anxiety-related sleep latency was removed from the SAD, and an item addressing recent sleep problems was removed from the health index. Both scales retained satisfactory levels of reliability and criterion validity following these changes.

Sleep

  1. Top of page
  2. Summary
  3. Introduction
  4. Method
  5. Sample
  6. Questionnaire assessment
  7. Physical activity
  8. Health and wellbeing
  9. Sleep
  10. Statistical analyses
  11. Results
  12. Discussion
  13. References

Interview assessments provided two estimated sleep variables: sleep onset latency and total sleep time. Quality of sleep was assessed using the item ‘do you ever have problems sleeping?’, with five response categories (never, seldom, sometimes, often, all the time). Insomnia was considered to be present if the respondent reported a sleep problem ‘often’ or ‘all the time’, and if that problem had been experienced within the previous week.

Statistical analyses

  1. Top of page
  2. Summary
  3. Introduction
  4. Method
  5. Sample
  6. Questionnaire assessment
  7. Physical activity
  8. Health and wellbeing
  9. Sleep
  10. Statistical analyses
  11. Results
  12. Discussion
  13. References

Data were appropriately weighted to compensate for oversampling of the age group 75+ years at baseline. To assess cross-sectional activity–insomnia relationships, variables were combined in separate logistic regression models with insomnia prevalence (1985, 1989, 1993) as dependent. In each model, the following variables were included as covariates: CPA1 activity scores (quintile ranges); total walking (above and below median value); social engagement (above and below median value); health index (above and below median value); body mass index (above and below underweight/overweight threshold of 25 kg m−2; chronological age (65–74 and 75+ years); depressed mood (present/absent); and sex. Adjusted odds ratios (ORs) with 95% confidence intervals (CIs) were estimated for each covariate in each model. To preserve comparability across models variable codings were based on 1985 threshold or median values, with the exception of CPA1 centiles. As the factor scores at each wave were derived from a unique underlying correlation matrix, these were separately calculated for each model. Finally, because levels of both social and physical activities are gender related (Morgan et al., 1991), interaction terms (CPA1 by sex; total walking by sex; and social engagement by sex) were initially added to each model to test for possible gender by activity interactions. As none of the interaction terms was significant, only main effects are reported here.

To assess the extent to which activity/health variables were predictive of insomnia-related outcomes, two further logistic regression models were applied to sub-groups within the dataset. In the first model 4-year outcomes among the 1985 prevalent cases of insomnia (remission/continued symptoms 1985–1989) were assessed in relation to activity/health factors measured in 1985 (outcome assessment of baseline cases beyond 1989 was not feasible owing to small cell sizes). In the second model, ORs for incident insomnia (i.e. new cases of insomnia arising between 1985 and 1993) were estimated in relation to 1985 activity/health variables. Cumulative incidence was quantified using separate denominators from the 1989 and 1993 surveys (see Table 2), and summing the incident cases identified in these separate analyses. In all the logistic regression models, goodness of fit was assessed by the significance of chi-square resulting from the Hosmer and Lemeshow test.

Table 2.  Nottingham Longitudinal Study of Activity and Ageing: insomnia prevalence, incidence and outcomes 1985–1993
  198519891993
Insomnia prevalenceSample size for prevalence denominator (total sample size)1023 (1042)673 (690)390 (410)
No. of cases at each survey wave (%) 221 (21.6)166 (24.7)85 (21.8)
Symptom persistence among 1985 cases (% of surviving cases)80 (58.0)30 (57.7)
Insomnia incidenceSurviving population at risk for incidence denominator527263
New cases at each survey wave (%)85 (16.1)34 (12.9)

Where appropriate, mean values were compared using independent sample t-tests, and the chi-square test was used to analyse contingency tables. All data were analysed using SPSS for Windows version 10.0.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Method
  5. Sample
  6. Questionnaire assessment
  7. Physical activity
  8. Health and wellbeing
  9. Sleep
  10. Statistical analyses
  11. Results
  12. Discussion
  13. References

From the 1985 survey of 1042 respondents, information on sleep quality was available for 1023 (with omissions due primarily to cognitive impairment at interview). Of these, 221 respondents met the survey criteria for insomnia, giving an overall prevalence of 21.6% (Table 2). Compared with non-cases, insomnia cases reported significantly shorter total sleep times (437.0 min versus 387.0 min, respectively; t = 6.2, d.f. = 807, P < 0.001) and significantly longer sleep latencies (18.5 min versus 52.9 min, respectively; t = −13.1, d.f. = 808, P < 0.001). Levels of insomnia (prevalence) associated with each value of the categorically coded activity variables at baseline (social engagement; CPA1 factor scores; and total walking per day) are shown in Figs 1–3, respectively. In each case a marked activity gradient is present, with lower social engagement levels (χ2 = 11.6; d.f. = 1; P < 0.001), CPA1 scores (χ2 = 39.1; d.f. = 4; P < 0.001) and daily walking levels (χ2 = 8.7; d.f. = 1; P < 0.01), all significantly associated with higher levels of insomnia in these unadjusted bivariate analyses.

image

Figure 1. Prevalence of insomnia according to social activity levels (categories = above/below median social engagement scores; insomnia by social activity χ2 = 11.6; d.f. = 1; P < 0.001).

Download figure to PowerPoint

image

Figure 2. Prevalence of insomnia according to physical activity levels (categories = activity score quintile ranges; insomnia by physical activity χ2 = 39.1; d.f. = 1; P < 0.001).

Download figure to PowerPoint

image

Figure 3. Prevalence of insomnia according to daily walking levels (categories = above/below median minutes/day purposeful walking; insomnia by walking level χ2 = 8.72; d.f. = 1; P < 0.01).

Download figure to PowerPoint

Results from the logistic regression analyses of prevalent insomnia cases are shown in Table 3. Among 221 prevalent cases identified in 1985 both moderately low [OR = 2.2 (CI = 1.0–4.7) P < 0.05] and low [OR = 2.0 (CI = 1.0–4.2) P < 0.05] CPA1 levels, lower physical health status [OR = 2.6 (CI = 1.6–4.0) P < 0.001], and depressed mood [OR = 2.9 (CI = 1.8–4.7) P < 0.001] were significantly and independently associated with an increased risk of insomnia. In the 1989 model, the significant influence of lower physical health status [OR = 2.5 (CI = 1.6–3.9) P < 0.001] and depressed mood [OR = 4.3 (CI = 2.4–7.7) P < 0.001] is maintained. In addition significant ORs were present for being overweight [OR = 1.6 (CI = 1.0–2.6) P < 0.05] and female gender [OR = 2.0 (CI = 1.2–3.2) P < 0.01]. The final (1993) prevalence model more closely resembled that from the baseline data, with both moderately low [OR = 5.0 (CI = 1.6–15.9) P < 0.001] and intermediate [OR = 5.4 (CI = 1.8–16.2) P < 0.001] levels of activity, lower physical health status [OR = 2.9 (CI = 1.5–5.6) P < 0.001] and depressed mood [OR = 5.6 (CI = 2.4–12.9) P < 0.001], all associated with a significantly increased risk of insomnia.

Table 3.  Risk factors associated with prevalent insomnia at three survey waves. Values are odds ratios (OR) with 95% confidence intervals (CI)
Independent variables at each survey wavePrevalent insomnia (1985; n = 1023)Prevalent insomnia (1989; n = 673)Prevalent insomnia (1993; n = 390)
OR95% CIOR95% CIOR95% CI
  1. *P < 0.05; **P < 0.01; ***P < 0.001.

CPA1 activity level
 Low2.21.0–4.7*1. 70.8–3.52.50.7–9.2
 Moderately low2.01.0–4.2*1.230.6–2.65.01.6–15.9**
 Intermediate1.70.8–3.61.40.7–2.75.41.8–16.2**
 Moderately high1.70.8–3.40.90.4–1.82.00.7–6.4
Walking <11 min day−11.10.7–1.70.80.5–1.31.00.5–1.8
Lower social engagement1.00.6–1.61.20.8–1.90.70.3–1.3
Lower physical health status2.61.0–4.7*2.51.6–3.9***2.91.5–5.6**
Depressed mood2.91.8–4.7***4.32.4–7.7***5.62.4–12.9***
Overweight (>25 kg m−2)1.30.9–1.91.61.0–2.6*1.80.9–3.3
Female gender1.40.9–2.12.01.2–3.2**0.70.3–1.3
Age group 75+ years (in 1985)1.10.7–1.61.00.7–1.71.00.5–1.9
Goodness of fit (P)0.090.40.6

From the baseline (1985) survey, 802 (78.8%) individuals did not meet the present criteria for insomnia and were therefore judged to be at risk. Of these, 119 became cases during the period 1985–1993 (see Table 2). Using separate denominators from the 1989 and 1993 surveys, the cumulative incidence was 29% (3.6% year−1). In the logistic regression model low [OR = 5.2 (CI = 2.0–13.6) P < 0.001] and intermediate [OR = 2.4 (CI = 1.1–5.2) P < 0.05] CPA1 levels in 1985, and older age [OR = 1.8 (CI = 1.1–3.1) P < 0.05] were significantly predictive of incident insomnia (Table 4). In this model, depressed mood in 1985 [OR = 2.2 (CI = 1.0–5.2) P = 0.06] just failed to achieve significance.

Table 4.  Baseline (1985) risk factors for persistent (1985–1989) and incident (1985–1993) insomnia. Values are odds ratios (OR) with 95% confidence intervals (CI)
Independent variables (assessed 1985)Insomnia persistence 1985–1989 (n = 102)Incident insomnia 1985–1993 (n = 119)
OR95% CIOR95% CI
  1. *P < 0.05; **P < 0.01; P = 0.06.

CPA1 activity level
 Low1.70.3–8.65.22.0–13.6**
 Moderately low0.50.1–2.42.20.9–5.0
 Intermediate5.01.0–24.7*2.41.1–5.2*
 Moderately high1.20.3–5.41.70.8–3.7
Walking <11 min day−10.50.2–1.41.40.8–2.3
Lower social engagement0.90.3–2.31.20.7–2.0
Lower physical health status5.61.9–16.8**1.40.8–2.4
Depressed mood3.81.1–12.8*2.31.0–5.2
Overweight (>25 kg m−2)2.10.7–6.00.80.5–1.4
Female gender2.40.8–6.90.90.5–1.6
Age group 75+ years0.80.3–2.41.81.1–3.1*
Goodness of fit (P)0.80.8

Of the 221 cases of insomnia identified in 1985, 138 survived to the second wave interview and of these 80 (58%) met the study criteria for insomnia (see Table 2). Full activity/sleep datasets were available for only 102 of these survivors, 59 (58%) of whom met the survey criteria for insomnia. Again, the main reasons for incomplete data were cognitive impairment and ill health at interview. The risk of symptom persistence across the 1985–1989 period was significantly increased by depressed mood [OR = 3.8 (CI = 1.1–12.8) P < 0.05), lower physical health status [OR = 5.6 (CI = 1.9–16.8) P < 0.01], and intermediate levels of CPA1 activity [OR = 5.0 (CI = 1.0–24.7) P < 0.05; see Table 4].

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Method
  5. Sample
  6. Questionnaire assessment
  7. Physical activity
  8. Health and wellbeing
  9. Sleep
  10. Statistical analyses
  11. Results
  12. Discussion
  13. References

Given the representativeness of the original survey sample, the relatively high response rates obtained at each survey wave, and the known reliability and validity of the survey measures used, the present analyses offer generalizable estimates of prevalence, incidence and risk. This conclusion is supported by the descriptive findings. The baseline insomnia prevalence rate (21.6%) falls within the range of values reported for older people in other cross-sectional studies in Europe (e.g. Janson et al., 2001; Jensen et al., 1998; Ohayon, 1996), while the remission and cumulative incidence rates found here accord with those reported in the USA (Foley et al., 1999). In addition, patterns of association in the present analyses are broadly in line with earlier studies. Thus, consistent with the findings of Habte-Gabr et al. (1991), Ohayon et al. (2001), Sherrill et al. (1998) and Janson et al. (2001), all three activity measures (CPA1, daily walking and social engagement) showed significant bivariate associations with insomnia (see Figs 1–3); chronological age did not significantly predict insomnia states in the cross-sectional analyses when health and activity factors were controlled; and, across all multivariate models, mental and/or physical health status consistently emerged as significant risk factors for insomnia. Goodness-of-fit measures, while all appropriately non-significant, ranged from moderate to high, with only the 1985 model (Table 3) showing a relatively low value. It should be emphasized, however, that the principal aim of all analyses was to assess the performance of activity variables as predictive factors, and not to construct the best possible prognostic model.

Broadly, the results support the conclusion that, in multivariate models containing valid and reliable measurements of physical activity, social engagement, and mental and physical health status, low physical activity levels can remain a significant and independent risk for late-life insomnia. With the single exception of the 1989 prevalence data, all models showed significant associations between physical activity variables and insomnia. As the CPA1 scores are derived from measurements designed to capture physiologically relevant levels of activity, this pattern of associations strongly suggests linkage between physical inactivity and sleep quality. However, the relatively low cross-sectional ORs for the 1985 survey wave (Table 3), and the absence of cross-sectional relationships for the 1989 survey wave (see Table 3), also suggest that such relationships may be both modest and fragile. In marked contrast, lower physical health status and depressed mood were consistently associated with a three- to six-fold increase in the risk of insomnia at each survey wave. Differences in the pattern and magnitude of CPA1-insomnia associations between 1985 and 1993 (Table 3) must, however, be interpreted with caution. In the course of this longitudinal study, surviving respondents aged by a further 8 years. It is possible that with increasing age the ‘low’ CPA1 factor scores increasingly confound with ill health, while the ‘moderately low’ and ‘intermediate’ levels of CPA1 may be more influenced by growing frailty (rather than illness episodes). This would help to explain why, in models controlling for health, very low levels of CPA1 failed to achieve significance, while moderate/intermediate CPA1 levels did (Table 3).

Only the 1989 cross-sectional model failed to show significant activity–insomnia associations. It is interesting to note, however, that in this model two variables which remain non-significant in all other models (female gender and overweight) achieve significance (with a BMI of >25 kg m−2 and female gender significantly increasing insomnia risk). Levels of insomnia tend to be higher for women, particularly in older populations (e.g. Habte-Gabr et al., 1991; Newman et al., 1997). Earlier analyses of the present dataset strongly suggest that, in later life, these differences are mediated by health and situational factors rather than by gender per se (see Morgan and Clarke, 1997b). Similarly, in their analyses of sleep survey data from 3097 people aged 65+ years Habte-Gabr et al. (1991) suggest that higher levels of sleep complaints among women reflect higher levels of depressive symptoms and ‘...generally higher levels of health complaints’. Nevertheless, in separate models combining health, age and activity variables, both female gender (Ohayon et al., 2001) and being overweight (Janson et al., 2001) have been reported to significantly increase the risk of insomnia. The significance of these factors in the present models, possibly related to health status, respiratory disorders, or energy balance, is therefore in line with earlier findings.

In the longitudinal models shown in Table 4 lower physical activity levels were associated with a significantly elevated risk of both insomnia persistence and insomnia incidence. Symptom chronicity is a complex variable reflecting not only the possible origins of an insomnia episode (e.g. mental or physical health problems), but also the individual's response to insomnia (e.g. poor sleep hygiene and unsuccessful strategies for dealing with lower sleep quality). The significant association between intermediate CPA1 levels and symptom persistence may reflect the latter element, and indicate an increased opportunity for activity among people experiencing chronic insomnia symptoms. In contrast, ORs associated with activity in the insomnia incidence model (Table 4) are more suggestive of a ‘zone’ of risk which extends from low to intermediate activity levels (the ‘moderately low’ category showed a P-value of 0.07). A reasonable interpretation of these findings is that among older persons at risk (and independent of health status), levels of physical activity above the 60th centile appear to be protective against incident insomnia.

Depressed mood and lower physical health status also emerged as significant risk factors for symptom persistence, although neither factor significantly increased the risk of 4–8-year incident insomnia (Table 4). Again, it may be that health issues and activity increasingly confound over time and that CPA1 estimates in the insomnia incidence model (Table 4) are, to an unknown extent, acting as proxies for frailty. The construct of frailty, characterized by lower activity and vulnerability rather than explicit disease symptoms, may also explain the emergence of older age group as a risk factor for incident insomnia (Table 4). While the likelihood of reporting incident insomnia almost doubled within the older (75+ years) age group (OR = 1.8; 95% CI = 1.1–3.1; P < 0.05; see Table 4), an age by health interaction term added to this model failed to reach significance.

Previous studies of elderly populations have demonstrated excess insomnia risk arising from lower levels of social, recreational or physical activity levels, and compromised physical or mental health status (e.g. Habte-Gabr et al., 1991; Janson et al., 2001; Ohayon et al., 2001; Sherrill et al., 1998). Few studies, however, have analysed three or more of these influential elements in relation to insomnia outcomes. In the present analyses, levels of insomnia prevalence showed significant gradients in relation to each of the activity measures shown in Figs 1–3. It is clear, however, that these various categories of ‘activity’ perform quite differently in multivariate models, with CPA1 scores most associated with insomnia risk (when the other activity measures are controlled). None of the models showed a significant OR either for social engagement scores or for daily walking scores.

The findings reported here are not without practical relevance. While aerobic fitness (Edinger et al. 1993) and moderate-intensity exercise (King et al. 1997) have been associated with higher sleep quality among older people in experimental trials, participation in sports and exercise programmes is generally low within this age group in both Europe and the USA (see Health Education Authority, 1992; Ruchlin and Lachs 1999; DiPietro 2001). Nevertheless, many older people both improve and maintain levels of physical ‘fitness’ through repeated daily activities associated with recreation, and personal and domestic maintenance. The present findings that higher levels of such activities may, through their physiological impact, afford protection against insomnia has important implications for health promotion and treatment in this area.

The limitations of the present analyses should also be recognized. While the NLSAA provides high quality indices of health, anthropometry, night-time sleep quality, and both social and physical activity levels it is, like many longitudinal studies, a product of earlier priorities and emphases. Assessments of daytime sleepiness, daytime napping and snoring were not included in the original design, limiting the extent to which current analyses can investigate what are now recognized as clinically relevant correlates of night-time sleep disturbance. It is also the case that the data demands for CPA1 estimates (recorded scores in each of seven activity categories), together with the wide range of covariates included in the logistic regression models, made the present analyses particularly vulnerable to missing data. Missing values from any of the component or scale scores included in the computation of outcome measures excluded that case from the present models. Most affected was the analysis of insomnia persistence, where 36 cases (of 138) were excluded in this way. As some data were missing because interviews were discontinued through fatigue, ill health or cognitive impairment, these omissions may have biased the remaining sample towards greater health. Examination of the health and activity data used in the present analyses, however, showed that each measure was active across its score range, and that those in poor overall health were well represented.

Features of the NLSAA design, particularly the 4-year inter-wave period, should also be considered when interpreting the present findings. As in other longitudinal studies, the continuity of states between survey waves (insomnia, in the present analyses) has been assumed. Nevertheless, the possibility exists that some of those categorized as ‘persistent’ cases remitted in the inter-wave period. Given, however, the frequency of complaints reported on both occasions (i.e. ‘often’ or ‘all the time’), and the similarly high levels of insomnia continuity among elderly people reported elsewhere (e.g. Foley et al., 1999), it is likely that most respondents who met the survey criteria for insomnia in 1985–1989 and 1989–1993 experienced symptoms throughout the intervening periods. It is also possible that the inter-wave interval may have amplified bias arising from mortality in the present analyses (if new cases or remitting cases died before assessment). As a result, it is likely that incidence levels are underestimated in the present study. Such mortality, however, is less likely to have distorted the insomnia–activity relationships reported here. Across the 8-year study period, overall patterns of survival within this random sample approximated to those in the general population. It is reasonable to assume therefore that attributes associated with survival within the insomnia sub-groups are similar to those which obtained within the general population at the time of follow-up assessment.

Finally, it should also be acknowledged that possible mechanisms mediating activity–insomnia relationships have not been addressed in the present paper. Whether such relationships are mediated by actual changes in sleep structure or by the perception of sleep structure (or by both) remains a question best addressed by a combination of experimental and epidemiological approaches.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Method
  5. Sample
  6. Questionnaire assessment
  7. Physical activity
  8. Health and wellbeing
  9. Sleep
  10. Statistical analyses
  11. Results
  12. Discussion
  13. References
  • Armstrong, G. K. and Morgan, K. Stability and change in levels of habitual physical activity in later life. Age Ageing, 1998, 27-S3: 1723.
  • Bassey, E. J. Demi-span as a measure of skeletal size. Ann. Hum. Biol., 1986, 13: 499502.
  • Bath, P. A. and Morgan, K. Customary physical activity and physical health outcomes in later life. Age Ageing, 1998, 27-S3: 2934.
  • DiPietro, L. Physical activity in ageing: changes in patterns and their relationship to health and function. J. Gerontol. (A), 2001, 56: 1322.
  • Driver, H. S. and Taylor, S. R. Exercise and sleep. Sleep Med. Rev., 2000, 4: 387402.
  • Edinger, J. D., Morey, M. C., Sullivan, R. J., Higginbotham, M. B., Marsh, G. R., Daily, D. S. and McCall, W. V. Aerobic fitness, acute exercise and sleep in older men. Sleep, 1993, 16: 351359.
  • Foley, D. J., Monjan, A., Simonsick, E. M., Wallace, R. B. and Blazer, D. G. Incidence and remission of insomnia among elderly adults: an epidemiologic study of 6,800 persons over 3 years. Sleep, 1999, 22 (suppl. 2): S366S378.
  • Habte-Gabr, E., Wallace, R. B., Colsher, P. L., Hulbert, J. R., White, L. R. and Smith, I. M. Sleep patterns in rural elders: demographic, health, and psychobehavioral correlates. J. Clin. Epidemiol., 1991, 44: 513.
  • Health Education Authority (in association with the and the Sports Council). Allied Dunbar National Fitness Survey: Main Findings. Allied Dunbar, London, 1992.
  • Horne, J. A. and Minard, A. Sleep and sleepiness following a behaviourally ‘‘active’’ day. Ergonomics, 1985, 18: 567575.
  • Janson, C., Lindberg, E., Gislason, T., Elmasry, A. and Boman, G. Insomnia in men – a 10-year prospective population based study. Sleep, 2001, 24: 425430.
  • Jensen, E., Dehlin, O., Hagberg, B. and Samuelsson, G. Insomnia in an 80-year-old population: relationship to medical, psychological and social factors. J. Sleep. Res., 1998, 7: 183189.
  • King, A. C., Oman, R. F., Brassington, G. S., Bliwise, D. L. and Haskell, W. L. Moderate-intensity exercise and self-rated quality of sleep in older adults-A randomized controlled trial. JAMA, 1997, 277: 3237.
  • Morgan, K. The Nottingham Longitudinal Study of Activity and Ageing (NLSAA): a methodological overview. Age Ageing, 1998, 511.
  • Morgan, K. and Clarke, D. Longitudinal trends in late-life insomnia: implications for prescribing. Age Ageing, 1997a, 26: 179184.
  • Morgan, K. and Clarke, D. Customary physical activity and survival in later life: a study in Nottingham UK. J. Epidemiol. Community Health, 1997b,51: 490493.
  • Morgan, K., Dallosso, H., Bassey, E. J., Ebrahim, S., Fentem, P. H. and Arie, T. H. D. Customary physical activity, psychological wellbeing, and successful ageing. Ageing Soc., 1991, 11: 399415.
  • Newman, A. B., Enright, P. L., Manolio, T. A., Haponik, E. F. and Wahl, P. W. Sleep disturbance, psychosocial correlates, and cardiovascular disease in 5201 older adults: the cardiovascular health study. J. Am. Geriatr. Soc., 1997, 45: 17.
  • Ohayon, M. Epidemiologic-study on insomnia in the general-population. Sleep, 1996, 19: S7S15.
  • Ohayon, M. M., Zulley, J., Guilleminault, C., Smirne, S. and Priest, R. G. How age and daytime activities are related to insomnia in the general population: consequences for older people. J. Am. Geriatr. Soc., 2001, 49: 17.
  • Ruchlin, H. S. and Lachs, M. S. Prevalence and correlates of exercise among older adults. J. Appl. Gerontol., 1999, 18: 341357.
  • Sherrill, D. L., Kotchou, K. and Quan, S. F. Association of physical activity and human sleep disorders. Arch. Int. Med., 1998, 158: 18941898.