Author(s): BACKES, A., MALISOUX L., Institution: LUXEMBOURG INSTITUTE OF HEALTH, Country: LUXEMBOURG, Abstract-ID: 766

INTRODUCTION:
Daily step counts are a central component of physical activity and are increasingly highlighted in public health guidelines. Advances in open-source algorithms now enable the extraction of detailed step‑related indicators from 24‑hour raw accelerometer data. This study aimed to identify distinct movement behaviour profiles based on multiple step-related indicators and examine their associations with cardiometabolic health biomarkers.
METHODS:
This cross-sectional study included 959 Luxembourg residents aged 25-79 years with 1-week of accelerometry data (≥4 valid days). A set of 51 indicators were derived from the step-count time-series. A reversed graph embedding approach was used to reduce this high‑dimensional dataset into a two-dimensional tree structure, enabling the identification of distinct step profiles. Associations between the tree dimensions, individual indicators and cardiometabolic biomarkers, including systolic and diastolic blood pressure, fasting plasma glucose, triglycerides, high‑density lipoprotein cholesterol (HDL‑c), apolipoprotein B/A1 ratio, fasting insulin and a clustered metabolic risk score (CMRS), were examined using adjusted linear regression models.
RESULTS:
Five key indicators emerged as the most influential descriptors of movement behaviours: active-to-inactive transition probability (AC-to-IN TP), inter-daily stability (IS), intra-daily variability (IV), lowest consecutive 5-hour cadence (LC5h cadence) and total time spent at incidental movement cadence (<20 steps/min, TIM cadence). IS, LC5h cadence and TIM cadence were positively associated with tree dimension 1, whereas AC-to-IN TP and IV were negatively associated with tree dimension 1. For tree dimension 2, IS showed a positive association, while IV and LC5h cadence showed negative associations. Four distinct step‑based movement profiles were identified from the tree structure. Regression models showed that individuals located in the upper‑right region of the tree displayed more regular daily patterns and higher levels of very light movement during the most sedentary periods. In contrast, those in the lower‑left region showed greater within‑day variability with a greater amount of fragmented and unsustained activity bouts, alongside higher levels of very light movement during the most sedentary periods. Adjusted regression models showed that tree dimension 1 was positively associated with HDL‑c and negatively associated with triglycerides, apolipoprotein B/A1 ratio, fasting insulin and CMRS. No significant associations were observed for dimension 2 after adjustment.
CONCLUSION:
Movement profiles reflecting more stable daily routines, greater sustained movement and lower activity fragmentation were favourably associated with several cardiometabolic biomarkers. These findings highlight the added value of step‑derived indicators beyond total step counts and demonstrate how data‑driven movement behaviour profiling can reveal key movement dimensions that are linked to cardiometabolic health.