Lekkas D, Jacobson N. (2021). Using artificial intelligence and longitudinal location data to differentiate persons who develop posttraumatic stress disorder following childhood trauma. Scientific Reports. 11(1): 10303. doi: 10.1038/s41598-021-89768-2
Researchers examined the efficacy of using time-anchored Global Positioning System (GPS) location data to detect post-traumatic stress disorder (PTSD) diagnostic status among women (ages 18–65 years) who had experienced child abuse (n = 185). Study data came from a previous study that used passive, smartphone-based GPS tracking across seven days to investigate the relationship between PTSD and functional impairment among 228 German women (including individuals diagnosed with PTSD, healthy trauma-exposed controls (HTC), and healthy controls). The current study focused on participants diagnosed with PTSD (n = 150) and HTC participants (n = 35) to determine whether a machine learning framework could differentiate participants with PTSD from HTC participants using GPS data. Researchers developed a novel machine learning pipeline comprising several different types of machine learning models. Using just two GPS variables (maximum daily radius traveled from home and daily minutes spent away from home), the pipeline was able to correctly identify participants diagnosed with PTSD approximately 81% of the time. GPS movement data appears to be a useful digital biomarker for the detection of PTSD. Future research could test the efficacy of using passively collected GPS movement data to identify PTSD in larger, more diverse populations. After further testing, this modeling pipeline could be used to create a smartphone app that collects GPS data, generates movement-based profiles of PTSD patients (highlighting PTSD symptom states and/or deterioration), and sends information to healthcare providers.