A Turn in the Road: Real-life Driving Differences in Obstructive Sleep Apnoea and Excessive Daytime Sleepiness
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
A driver with untreated obstructive sleep apnoea (OSA) is at an increased relative risk of crashing their vehicle. Nonetheless, the chance of any individual being involved in a collision remains low and it is not clear how we can identify which drivers with OSA are most at risk. Since motion detection is used by insurers to assess risky driving, we sought to characterise how driving changes with OSA and daytime sleepiness in real life.
Methods
Ninety-one recruited participants (mean age 51 [22 to 77] years; 35% female; mean ESS 12.5, mean ODI 15/h) completed 7,834 journeys (mean duration 15.6 minutes), yielding 508,544 driving events. Participants were recruited at referral to our sleep service and installed a smartphone app that recorded driving data, at least until diagnostic outcomes were known. Data were analysed for driving events of braking, accelerating and turning using linear mixed-effects models against participant data including age, sex, BMI, Epworth Sleepiness Scale (ESS), oxygen desaturation index (ODI), time into trip and journey duration.
Findings
ESS and ODI combined showed all clinical groups robustly differed from non-OSA / non-EDS, [ODI<5 / ESS<11] in turning behaviour. Younger age was linked to different, potentially riskier behaviour, while ESS and ODI, when analysed independently, showed no robust association with identifiable differences in driving.
Interpretation
Smartphone telematics can distinguish people with self-reported sleepiness and OSA from non-OSA / non-EDS individuals, via turning behaviour. This supports shifting emphasis from diagnosis and self-report towards objective behavioural indicators of impairment. Combining telematics with cognitive, physiological and/or subjective measures could yield practical criteria to improve crash-risk assessments and provide objective support for decisions around driving cessation or continuance.
Funding
This research was supported by funds from the Royal Papworth Charity and Consciousness and Cognition Lab.
Research in context
Evidence before this study
We searched Google Scholar and PubMed from inception to September 1 st 2025, with no language restrictions, using terms related to ‘obstructive sleep apnoea’, ‘sleepiness’, ‘driving’, ‘crash’, and ‘telematics’. Existing literature consistently links OSA with increased crash risk. Almost all reviewed studies used cohort, database reviews, self-report or driving simulator designs, which lack accurate real-world risk prediction at subgroup and individual level. This prominent issue has led to high profile organisations including the European Respiratory Society, The National Sleep Foundation and The Lancet to call for better research into detection of at-risk sleepy drivers.
Added value of this study
In a prospective clinical cohort at referral, we captured smartphone passive telematics (Insights, Sentiance) across 7,834 journeys, generating 508,544 events from 91 adults. Using linear mixed-effects models, we show that turning behaviour differentiated participants with combined sleepiness and OSA from those with neither (ODI <5 / ESS <11), whereas ESS or ODI alone were not independently associated with identifiable differences in braking/acceleration profiles. This study demonstrates the feasibility and signal of smartphone-derived features to help phenotyping driving risk within routine OSA pathways.
Implications of all the available evidence
Smartphone telematics could improve current risk assessments for drivers with suspected or confirmed OSA, offering scalable, low-cost, passively collected markers that align with real-world behaviour. If externally validated, manoeuvre-specific metrics (e.g., turning) may help triage patients for further evaluation, tailor driving advice or restrictions, and provide objective support for decisions regarding driving cessation or continuance. Next steps include independent replication, integrating these data with cognitive, physiological and subjective dynamics, linkage to hard outcomes (near-misses and crashes), assessment of fairness and privacy, and development of clinically usable thresholds suitable for regulatory and occupational settings.
