A systematic scoping review of latent class analysis applied to accelerometry-assessed physical activity and sedentary behavior
Michael Kebede,Annie Green Howard,Yumeng Ren,Blake Anuskiewicz,Chongzhi Di,Melissa A. Troester,Kelly R. Evenson
DOI: https://doi.org/10.1371/journal.pone.0283884
IF: 3.7
2024-01-22
PLoS ONE
Abstract:Background Latent class analysis (LCA) identifies distinct groups within a heterogeneous population, but its application to accelerometry-assessed physical activity and sedentary behavior has not been systematically explored. We conducted a systematic scoping review to describe the application of LCA to accelerometry. Methods Comprehensive searches in PubMed, Web of Science, CINHAL, SPORTDiscus, and Embase identified studies published through December 31, 2021. Using Covidence, two researchers independently evaluated inclusion criteria and discrepancies were resolved by consensus. Studies with LCA applied to accelerometry or combined accelerometry/self-reported measures were selected. Data extracted included study characteristics and both accelerometry and LCA methods. Results Of 2555 papers found, 66 full-text papers were screened, and 12 papers (11 cross-sectional, 1 cohort) from 8 unique studies were included. Study sample sizes ranged from 217–7931 (mean 2249, standard deviation 2780). Across 8 unique studies, latent class variables included measures of physical activity (100%) and sedentary behavior (75%). About two-thirds (63%) of the studies used accelerometry only and 38% combined accelerometry and self-report to derive latent classes. The accelerometer-based variables in the LCA model included measures by day of the week (38%), weekday vs. weekend (13%), weekly average (13%), dichotomized minutes/day (13%), sex specific z-scores (13%), and hour-by-hour (13%). The criteria to guide the selection of the final number of classes and model fit varied across studies, including Bayesian Information Criterion (63%), substantive knowledge (63%), entropy (50%), Akaike information criterion (50%), sample size (50%), Bootstrap likelihood ratio test (38%), and visual inspection (38%). The studies explored up to 5 (25%), 6 (38%), or 7+ (38%) classes, ending with 3 (50%), 4 (13%), or 5 (38%) final classes. Conclusions This review explored the application of LCA to physical activity and sedentary behavior and identified areas of improvement for future studies leveraging LCA. LCA was used to identify unique groupings as a data reduction tool, to combine self-report and accelerometry, and to combine different physical activity intensities and sedentary behavior in one LCA model or separate models.
multidisciplinary sciences