How consistently does sleep quality improve at retirement? Prospective analyses with group-based trajectory models

Growing evidence indicates that retiring from paid work is associated, at least in the short-term, with dramatic reductions in sleep difficulties and more restorative sleep. However, much is still not known, in particular how universal these improvements are, how long they last, and whether they relate to the work environment. A methodological challenge concerns how to model time when studying abrupt changes such as retirement. Using data from Swedish Longitudinal Occupational Survey of Health (n = 2,148), we studied difficulties falling asleep, difficulties maintaining sleep, premature awakening, restless sleep, a composite scale of these items, and non-restorative sleep. We compared polynomial and B-spline functions to model time in group-based trajectory modelling. We estimated variations in the individual development of sleep difficulties around retirement, relating these to the pre-retirement work environment. Reductions in sleep difficulties at retirement were sudden for all outcomes and were sustained for up to 11 years for non-restorative sleep, premature awakening, and restless sleep. Average patterns masked distinct patterns of change: groups of retirees experiencing greatest pre-retirement sleep difficulties benefitted most from retiring. Higher job demands, lower work time control, lower job control, and working full-time were work factors that accounted membership in these groups. Compared to polynomials, B-spline models more appropriately estimated time around retirement, providing trajectories that were closer to the observed shapes. The study highlights the need to exercise care in modelling time over a sudden transition because using polynomials can generate artefactual uplifts or omit abrupt changes entirely, findings that would have fallacious implications.