Background Sleep varies markedly within and between individuals and, as a multifactorial trait, sleep can be challenging to operationalise. The aim of the present study was to assess whether latent class analysis (LCA) might identify distinct sleeping patterns that not only capture the complexity of sleep but are meaningfully associated with key sociodemographic characteristics.

Methods Latent class analysis was performed using Latent Gold software on data from seven sleep variables (duration, latency, disturbance, coughing/snoring, medication, quality and daytime sleepiness) collected by questionnaires in Waves 1 and 4 of the UK Household Longitudinal Study. To avoid the possibility that responses to these questions change with familiarisation, only data provided by the n = 45,144 adults answering these questions for the first time were used. The best fitting latent model was selected on the basis that it had the lowest Bayesian Information Criterion, and potential associations between any latent sleep clusters identified and five key sociodemographic characteristics (age, gender, educational attainment, household composition and subjective health) were explored by conducting multinomial logistic regression analyses using STATA IE-14.

Results In both Waves the best fitting LCA models comprised six latent sleep classes, each containing 6.5–31.6% of respondents. The distribution of the seven sleep variables across these six latent sleep classes suggested that respondents reporting sleep characteristics relevant to each might be described as: ‘long good sleepers’; ‘long moderate sleepers’; ‘snoring good sleepers’; ‘disturbed bad sleepers’; ‘short bad sleepers’; and ‘struggle to sleepers’. Each of the five sociodemographic characteristics displayed statistically significant relationships with all six of the latent sleep classes; and these relationships remained significant (though modestly attenuated) after adjustment for preceding characteristics identified as likely confounders using a directed acyclic graph. Compared to participants classified as ‘long good sleepers’, respondents with self-reported sleep characteristics that led them to be classified in one of the remaining five (suboptimal) sleep classes were generally more likely to be: older; female; poorly educated; employed; living alone without children; and in less-than-excellent health.

Conclusion Latent class analysis revealed six distinct sleeping patterns amongst the UK adult population of the UK. All six of these sleeping patterns were associated with a range of sociodemographic characteristics, suggesting that the patterns observed offer a potentially meaningful classification that captures the complexity of sleep and provides an opportunity to explore the role that such patterns (rather than individual sleep characteristics) play in the aetiology of disease.