Re-creating daily mobility histories for health research from raw GPS tracks: Validation of a kernel-based algorithm using real-life data. (July 2016)

• Record Type: Journal Article
• Authors: Kestens, Yan; Thierry, Benoit; Chaix, Basile
• Is Part Of: Health & place. Volume 40(2016:Jul.)
• Page Start: 29
• Page End: 33
• Publication Date: 2016-07
• Subjects: GPS tracking -- GPS algorithm -- Validation study -- Mobility -- History -- Multiple exposures; Health -- Social aspects -- Periodicals; Health services accessibility -- Periodicals; Public health -- Periodicals; Political planning -- Periodicals; Social medicine -- Periodicals; Epidemiology -- Periodicals; Health Policy -- Periodicals; Health Services Accessibility -- Periodicals; Public Health -- Periodicals; Public Policy -- Periodicals; Sociology, Medical -- Periodicals; Épidémiologie -- Périodiques; Politique sanitaire -- Périodiques; Santé, Services de -- Accessibilité -- Périodiques; Health services accessibility; Health -- Social aspects; Political planning; Public health; Social medicine; Periodicals; 613
• DOI: 10.1016/j.healthplace.2016.04.004
• Languages: English
• ISSNs: 1353-8292

Abstract

Abstract: Background: GPS tracking is increasingly used to document daily mobility, allowing refined analysis of daily exposures and health behaviour. Validation of algorithms processing raw GPS data to identify activity locations and trips are lacking. Objective: Propose novel ways to evaluate GPS processing algorithms data while validating an existing kernel-based algorithm with real-life GPS tracks. Methods: Seven-day GPS tracking and GPS-prompted recall interviews were conducted among 234 adult participants of the RECORD GPS Study. Raw GPS data was transformed using a kernel-based algorithm. Two match and nine mismatch configurations are analysed. Algorithm detection of activity locations and trips were validated. Results: Some 95.8% of available GPS time was correctly classified as an activity location or a trip. The algorithm falsely identified a trip for 2.2% of the tracking time, and falsely identified an activity location 0.7% of time. Missed trips and missed activity locations counted for less than .4% of the time. Conclusion: The tested kernel-based algorithm provides histories of activity locations and trips that are highly concordant with GPS-prompted follow-up interviews.