Data Science
Sleep disruption and depression outcomes
The challenge
It is relatively well-known that sleep and circadian disruption-related factors are risk factors for poorer mental health broadly, but the precise aspects which underlie this are unclear. Self-report of some variables can be inconsistent and not necessarily always reliable at the population level, where objective assessment using accelerometers/actigraphs, can provide complimentary information.
The research
UK Biobank includes approximately 502,000 participants from the general population, where participants have self-report, electronic health record linkage, and (in a sub-sample) detailed actigraphy data. This data gives measures of daily activity including aspects of sleep duration, timing and quality (among other factors), amount of physical activity, and when. Machine learning can be used to identify key predictors of depression, ‘decomposing’ large numbers of variables into more informative, key risk factors.
The results
Across several different machine learning approach, the results were relatively consistent. Common predictors of depression included: difficulty getting up, insomnia symptoms, snoring, actigraphy-measured daytime inactivity, and lower morning activity (~8 am). The number of these risk factors that a person had, on average, was associated linearly with more risk of depression.
The impact
The clinical utility of sleep assessment could be improved by focusing on specific characteristics of sleep and circadian health rather than overall, broad phenotypes. Aspects of common questionnaires pertaining to sleep health, could be amended to include things like daytime inactivity.
Lead
Read the paper
The study, Subjective and objective sleep and circadian parameters as predictors of depression-related outcomes: A machine learning approach in UK Biobank, is published in Journal of Affective Disorders.
