Re-evaluating sleep difficulty as a risk factor for dementia in the UK Biobank cohort: addressing survival bias using a semi-competing risks approach

Abstract Introduction Sleep difficulty, a prevalent sleep disturbance in ageing populations, has shown paradoxical associations with dementia risk in prior epidemiological studies. Emerging evidence suggests that survival bias—where premature mortality in individuals with sleep difficulty obscures dementia risk—may explain these inconsistencies. Methods We analysed data from 457,367 UK Biobank participants aged 40–69 years who were enrolled at baseline between 2006 and 2010 and followed until 2022. Sleep difficulty was assessed via self-reported questionnaires, and dementia was obtained from electronic health records. To address survival bias, we employed a semi-competing risks framework that jointly models dementia incidence and mortality, contrasting results with conventional Cox proportional hazard models. Results Semi-competing risk analyses suggested that usual sleep difficulty modestly increased risks of vascular dementia (fully adjusted HR 1.14, 95% CI 1.02–1.28) and slightly increased all-cause dementia (HR 1.03, 95% CI 0.98–1.08) in the total sample. The association is particularly strong in younger adults (below 55 years old) and low APOE risk (APOE ε4 non-carriers) when adjusted for age, sex and education. Conversely, Cox models suggested a protective association between usual sleep difficulty and all-cause dementia in the total sample, aligning with prior UK Biobank studies. This discrepancy highlights how survival bias distorts risk estimates. Discussion Our findings resolve conflicting evidence by demonstrating sleep difficulty’s direct dementia risk when accounting for competing mortality. The semi-competing risks approach provides a robust framework for ageing research, when survival bias is present. Clinically, these results underscore sleep difficulty management as a modifiable dementia prevention target in mid-to-late life.