4.2.2.1 ‘Temporal stability’ study
A cross-sectional design was initially used to generate the latent class sleep models for both Waves 1 and 4 separately, first using participants who had participated in both Waves. Thereafter, a longitudinal study design was used to compare the stability of the resulting latent sleep models from each Waves (i.e. at Wave 1 vs. Wave 4) over time.
4.2.2.2 ‘Correlates of latent sleep classes’ study
After assessing the stability of sleep models over time, a third sample of participants (comprising all those offering responses to the survey’s questionnaire sleep items; and based on the very first time each participant answered these items [be that in Wave 1 or in Wave 4]) was analysed to identify latent sleep classes from the largest possible sample of participants available. For assessing the sociodemographic, behavioural and clinical correlates of latent sleep classes, relevant data were extracted from responses given to items in the same wave as that providing the sleep data used. (i.e. the sociodemographic/behavioural/clinical data generated from the questionnaire in which participants first gave responses to the sleep items).
4.2.3 Participants
As sleep data were only available from the questionnaires used in Waves 1 and 4 of the UKHLS, only participants providing responses to from these two UKHLS Waves were eligible for inclusion in the present chapter.
Male participants as well as female participants were included in the present chapter’s analyses to identify the latent sleep patterns amongst the UK population, while analyses of their relationship(s) with pregnancy outcomes necessarily include only (pregnant) women. This two-stage process was considered necessary for the following reasons;
First, the aim was to discover all possible, stable sleep patterns amongst UKHLS participants, and since women might share similar pattern with men we aimed to maximize the number of participants involved.
Second, since the aim was then to describe the distribution of sleep in women as compared to men, subsequent analyses were then conducted to establish which patterns were more/less commonly associated with female participants.
Finally, the aim of these (gender-disaggregated) analyses was to examine whether pregnant women’s sleep patterns differed in comparison to those of the general population (i.e. male and female), since the literature contained claims that pregnant women’s sleep was substantively less ‘favourable’ than the population as a whole (claims that therefore required conducting latent class analyses not only in pregnant women, but also amongst female participants separately and male and female participants combined).
4.2.3.1 Temporal stability study
Male and female participants who participated in both Waves 1 and 4 of the UKHLS and had complete sets of sleep data were included in the the sub-studies that examined the stability of the latent class analysis approach over time when using the UKHLS sleep module.
4.2.3.2 ‘Correlates of latent sleep classes study’
Male and female participants who participated in Wave 1 and/or Wave 4 and had complete sets of sleep data were included in the latent class analysis to generate the UK population sleep clusters .
4.2.4 Measurements
4.2.4.1 Sleep variables
To generate the sleep clusters in the temporal stability and correlates of sleep classes studies, all seven sleep variables generated by items in the UKHLS sleep module were used: sleep duration, latency, sleep disturbances due to snoring or coughing, use of medication to help with sleep, next-day sleepiness, sleep disturbance, and perceived sleep quality.
4.2.4.2 Coding of sleep variables
The sleep variables, which include latency, disturbance, snoring and/or coughing, usage of medication and next-day sleepiness, were re-categorized using fewer categories (3 ordinal categories) to make it easier to visualise more distinguished patterns of variations between the generated clusters when conducting the latent analysis (Table 4-1). The following classification was carried when categorising the
variables; absent, non-habitual (< 3 events a week) and habitual (≥ 3 times a week), as it was the classification most often used in the literature and the classification of insomnia symptoms used by the DSMVI (Amarican Psychiatric Association, 2013). Sleep duration was categorised into four categories, using similar to the Pittsburgh Sleep Quality Index (PSQI; Buysse et al, 1989) scoring categories (Table 4-1; the sleep questionnaire most commonly used by sleep researchers). This categorisation was chosen to facilitate comparison of any sleep patterns identified in the present chapter with those described in the literature.
Table 4-1 The seven items in the Understanding Society Sleep Questionnaire, together with the original response categories and the categories adopted to facilitate comparison with studies using the PSQI.
Sleep questions Original responses Responses after
reducing the numbers of categories
Q1: “How many hours of actual sleep did you usually get at night during the last month? Note: This may be different than the actual number of hours you spent in bed”.
Reported in hours and minutes Reference: ≥7 hours Short: ≥6 and <7 hours Restricted: ≥5 and <6 hours Severely restricted: <5 hours
Q2: During the past month, how often have you had trouble sleeping because you “Cannot get to sleep within 30 minutes”?
1. Not during the past month 2. Less than once a week 3. Once or twice a week 4. Three or more times a week 5. More than once most nights
1. Not during the past month (1) 2. less than three times a week
(2&3)
3. Three or more times a week (4&5)
Q3: During the past month, how often have you had trouble sleeping because you “Wake up in the middle of the night or early in the morning”?
1. Not during the past month 2. Less than once a week 3. Once or twice a week 4. Three or more times a week 5. More than once most nights
1. Not during the past month (1) 2. less than three times a week
(2&3)
3. Three or more times a week (4&5)
Q4: During the past month, how often have you had trouble sleeping because you “Cough or snore loudly”?
1. Not during the past month 2. Less than once a week 3. Once or twice a week 4. Three or more times a week 5. More than once most nights
1. Not during the past month (1) 2. less than three times a week
(2&3)
3. Three or more times a week (4&5)
Q5: During the past month, how often have you taken medicine (prescribed or “over the counter”) to help you sleep? 2. less than three times a week
(2&3)
3. Three or more times a week (4)
Q6: During the past month, how often have you had trouble staying awake while driving, eating meals, or engaging in social activity? 2. less than three times a week
(2&3)
3. Three or more times a week (4)
Q7: During the past month, how would you rate your sleep quality overall?
4.2.4.3 The coding of sleep variables included in the exploratory regression analyses
When running the regression analyses to evaluate the association between each of the seven self-reported sleep characteristics and each of the sociodemographic, and health characteristics, all of the sleep variables were re-categorised as binary variables. The rationale for using binary variables was to permit comparison of the logistic odd ratios generated for analyses of each sleep characteristic with those generated for analyses of any/each LCA-derived sleep cluster.
Sleep duration was categorized as ≥7 hours and < 7 hours, while sleep latency, disturbance, snoring and/or coughing, next-day sleepiness and sleep medication was categorised as ‘ever versus never’. Sleep quality was categorised as very/fairly good vs. fairly/very bad.
Sociodemographic features (i.e. age, gender, current employment status, household composition and highest educational qualification achieved) were used to study the associations between each of the seven sleep characteristics and each/any sleep cluster. Data on each of the sociodemographic variables used were extracted from the same wave as that providing the self-reported sleep data. Most of the sociodemographic variables were re-categorised to simplify those with large numbers of response categories, thereby ensuring that there were likely to be sufficient numbers of participants in each of the reduced categories to permit robust analysis.
A full description of how each of the sociodemographic variables were categorised (before and after re-categorisation) is available in the appendices. A brief summary is provided below:
I. Age was re-categorised into: ≤19 years, 20-39years, 40-59 years and ≥60 years.
II. Education was re-categorised into: ‘A’ level or above vs. GCSE level or
‘other’
III. Occupation was categorised into: employed, unemployed, sick or disabled, in training/education, and retired
IV. Household structure was categorised into: Single without children, Single with children, Couple without children, and Couple with children
4.2.4.4 Health indicators
Three indicators were chosen to provide measures of participant health, including the physical and mental components of the eight items of the SF12v (a shorter version of the SF36v developed by Quality Metric Incorporate; Quality Metric Inc.,
2007). These were used to subjectively evaluate each participant’s well-being and functional health, including the first item of the SF12v (which provides a general subjective evaluation of an individual’s overall health). The physical and mental components of the SF12v were originally rated from 0 to 100, where the higher the score, the ‘healthier’ the participant. The components were re-categorised into three equal categories (i.e. 66, <66 and ≥ 32, <32) using ‘norm scoring’ (i.e. based on the mean [M=50] and standard deviation [SD=16]; see Table 8-12). Likewise, the general health variable was re-categorised into a binary variable: ‘excellent to good’ and ‘fair to poor’. For further detail with regard to the SF12v questionnaire please refer to the appendices (see Chapter 8, Section 8.3.1, page 314).