Abstract
Sleep disturbances are an underrecognized factor associated with reduced workplace productivity ("presenteeism"). Previous studies have largely relied on self-reported sleep data, limiting their scalability and accuracy. We investigated associations between smartphone-based sleep metrics and presenteeism using real-world data from 79,048 working adults in Japan (mean age: 42.1 years [range: 18-66 years]; 47.8% female). Over 2.1 million nights of sleep data were collected over 28 days. Sleep parameters included total sleep time (TST), sleep latency, percent wake after sleep onset (%WASO), chronotype (mid-sleep on free days corrected for sleep debt), and social jetlag. Generalized additive models showed U-shaped associations between TST and presenteeism. Longer sleep latency, higher %WASO, delayed chronotype, and greater social jetlag were each linked to higher presenteeism scores. Unsupervised clustering using UMAP and the Leiden algorithm identified five sleep phenotypes: "Healthy Sleepers," "Long Sleepers," "Fragmented Sleepers," "Poor Sleepers," and "Social Jetlaggers." The latter two clusters showed the worst scores for insomnia, daytime sleepiness, and presenteeism. These findings highlight that not only sleep duration but also quality, timing, and regularity may be associated with workplace functioning. Smartphone-based tracking may offer a scalable means of identifying at-risk individuals and informing future personalized strategies.