Modelling Sleep Time Series Profiles of Australian Railway Drivers

The focus of this study is modelling of railway drivers (RDs) multivariate sleep duration time series profiles, of in excess of > 1000 sleep episodes, with respect to SOM-derived clusters. Generalized Additive Models for Location, Scale and Shape (GAMLSS) of the RDs sleep duration were tested with respect to RD-specific predictors of duration of next break, hours to next work break, hours to next duty, hours since break onset, hours to next break, sleep onset time, next duty onset times, work break onset times and current break duration. SOMs found 6 clusters of RDs whose patterns of sleep duration differ over time and across whom a subset of predictors operated differently on sleep attainment. GAMLSS confirmed that sleep onset time, next duty onset time, hours since current work break onset, duration of next break, hours to next break and hours to next duty are factors operating differently across the clusters, as do current break onset time and break duration. Sleep onset time interacts differently across the groups according to the RD's next break duration, or on the RD's hours to next break or hours to next duty. In addition next duty onset time interacts differently across the groups according to the RD's hours to next duty, or on the anticipated duration of their next break. Keywords- Generalized Additive Models for Location, Scaleand Shape (GAMLSS); self-organising maps (SOMs); sleep duration time series; hours to next duty, hours since break onset, duration of next break, hours to next break