Abstract 042: Accelerometer-Derived Daily Life Movement Classified by Machine-Learning and Incidence of Cardiovascular Disease in Older Women: The OPACH Study

• Published in: Circulation (New York, N.Y.) N.Y.), 2019-03, Vol.139 (Suppl_1)
• Main Authors: LaCroix, Andrea Z, Bellettiere, John, Di, Chongzhi, Rosenberg, Dori, LaMonte, Michael J
• Format: Article
• Language: English
• ISSN: 0009-7322; 1524-4539
• DOI: 10.1161/circ.139.suppl_1.042

Abstract only Background: OPACH study evidence has shown that both accelerometer-measured light (LPA) and moderate-vigorous physical activity (MVPA) are associated with CVD mortality and morbidity in older women. However, accelerometer-measured PA exposures have been limited to measuring time and intensity of movement, without providing any information about which physical activity behaviors produced the movements. This study examines associations of a validated, machine-learned classification of “daily life movement” (DLM) in relation to incident CVD events. Methods: Women (n=5416, mean age=79±7, 33% Black, 17% Hispanic) without prior myocardial infarction or stroke wore accelerometers for up to 7 days from May 2012-April 2014 and were followed for incident CVD through February 2018 (n=667 events). Machine-learned algorithms were developed in a sample of 39 free-living older women who simultaneously wore small cameras (SenseCam) capturing images of daily activities (used as the ground truth) and hip-worn accelerometers to measure movement. DLM was defined as standing and moving in a confined space such as around the house or garden. The DLM algorithm had 66% sensitivity and 94% specificity compared to SenseCam images and was further validated in 2 other populations. Cox models estimated hazard ratios (HR) and 95% confidence intervals for CVD across quartiles of DLM adjusting for age, race-ethnicity, smoking, alcohol use, education, co-morbidity score, physical function, and self-rated health. We then examined the DLM association with CVD after adjustment for MVPA. Results: Fully-adjusted HRs (95% CIs) for CVD across DLM quartiles were: [Q1-ref 1.00 ; Q2 0.74 (0. 60,0.91); Q3 0.76 (0.61,0.94); Q4 0.56 (0.44,0.72); p-trend