Associations of movement behaviors and body mass index: comparison between a report-based and monitor-based method using Compositional Data Analysis

The study addresses how time-use movement behaviors—sleep, sedentary behavior (SED), light-intensity physical activity (LPA), and moderate-to-vigorous physical activity (MVPA)—relate to body mass index (BMI) when accounting for the compositional nature of a 24-hour day. With rising obesity prevalence and substantial daily time spent sedentary among US adults, prior work has linked high SED to adverse health outcomes independent of MVPA. Yet most research has focused narrowly on SED and MVPA, overlooking sleep and LPA, and often relies on self-report measures prone to error. Furthermore, standard analyses typically ignore that time spent in one behavior necessarily displaces time from others. The purpose was to evaluate associations of the full 24-hour composition with BMI using Compositional Data Analysis (CoDA) and to compare associations derived from a report-based method (24-hour Physical Activity Recall; 24PAR) versus a monitor-based method (SenseWear Armband; SWA) in a representative adult sample.

The introduction highlights: (1) high sedentary time in adults (~8 hours/day) and links between SED and chronic disease and mortality; (2) evidence that individuals with obesity are more sedentary and less active than their normal weight peers; (3) ongoing measurement challenges and definitional issues in sedentary behavior research; (4) discrepancies between report-based and monitor-based estimates of activity (e.g., >50% vs <5% meeting PA guidelines via self-report vs accelerometry), and inconsistent associations between SED and obesity across methods; (5) limitations of count-based accelerometer processing for SED, and greater error in long-term recalls compared to short-term recalls; and (6) the advantage of Compositional Data Analysis (CoDA) to properly model time-use data and evaluate effects of reallocating time between behaviors on weight outcomes. These motivate applying CoDA to compare report- versus monitor-based associations with BMI.

Design and sample: Ancillary analysis of the Physical Activity Measurement Survey (PAMS), a cross-sectional study conducted across eight consecutive 3-month quarters to capture seasonal variation. A multi-level stratified sampling approach recruited a representative sample of adults (ages 20–75) from four Iowa counties (two rural, two urban). Inclusion: able to walk and complete English/Spanish surveys; exclusion: critical medical conditions precluding PA. Participants provided consent. The final analytic sample with replicate measures included 1247 adults. Protocol: Each participant completed two independent replicate trials. For each trial, participants wore a SenseWear Armband Mini (SWA) continuously for 24 hours on a randomly selected day, then completed a telephone-administered 24-hour Physical Activity Recall (24PAR) the following day to report activities from the monitoring day. Trials were separated by at least 12 days. Field staff delivered and retrieved devices and provided logs for non-wear activities. Instruments: SWA is a multi-sensor monitor (heat flux, galvanic skin response, skin temperature, near-body temperature, tri-axial accelerometer) providing 1-min metrics (METs, activity time, energy, etc.). Data were processed in Software v8.0 with algorithms v5.2. The 24PAR is a computer-assisted telephone interview capturing previous-day activities in episodes ≥5 minutes, with assigned METs using a reduced list of 270 Compendium codes. Data processing: For SWA, each minute was classified by MET thresholds: ≤1.5 MET = SED; 1.5 < MET < 3.0 = LPA; ≥3.0 MET = MVPA. SWA-derived sleep time was subtracted from total sedentary to yield SED. For 24PAR, activities were mapped to Compendium codes and the same MET thresholds applied; SED was computed as (sum of minutes in 27 specific ≤1.5 MET sedentary activities) minus self-reported sleep; LPA time was computed as 1440 minutes minus SLEEP, SED, and MVPA. For each participant, behavior minutes were averaged across the two days. BMI was calculated as kg/m² and also dichotomized at 30 kg/m² for obesity status. Covariates included sex, age, ethnicity (White, Black, Other), annual income (<$25k, $25–75k, >$75k), employment (full-time, part-time, unemployed/retired/homemaker), education (<high school, some college/post-high school, college/graduate), marital status (married/living as married; divorced/separated/widowed; single/never married), current smoking (yes/no), and measurement day composition (2 weekdays; 2 weekend days; 1 weekday + 1 weekend day). Compositional Data Analysis (CoDA): Minutes/day in SLEEP, SED, LPA, MVPA were converted to proportions of wear-time summing to 100%, with geometric means adjusted to a 1440-min day. Isometric log-ratio (ILR) transformations encoded the 4-part composition into 3 coordinates. Sets of ILRs were created by permuting parts to obtain parameter estimates for each focal part (ilr_SLEEP, ilr_SED, ilr_LPA, ilr_MVPA). Compositional mean bar plots and variation matrices were produced. Total variance computed per standard CoDA methods. Statistical analysis: Multiple linear regression models examined associations between BMI (continuous outcome) and each compositional part via the ILR coordinates. Separate models were run for 24PAR and SWA, with four permutations to obtain gamma coefficients and 95% CIs for each part. Models adjusted for all covariates listed above; model fit assessed with R². Logistic regression with the same permutation approach assessed odds of obesity (BMI ≥30 kg/m²). Compositional isotemporal substitution, following Dumuid et al., estimated predicted BMI changes associated with reallocating time between behaviors while holding the remaining parts constant at their mean relative amounts. Significance set at p<0.05; analyses conducted in STATA v15.0.

- Sample characteristics: Of 1247 adults, 44.4% were classified as obese (BMI ≥30 kg/m²). Obesity status differed significantly by income, education, and smoking status (p<0.05); no significant differences by sex, age, ethnicity, employment, marital status, or measurement day distribution. - Linear models using 24PAR: Greater relative time in sleep was associated with lower BMI (γ = −3.58, p = 0.011); greater SED with higher BMI (γ = 3.70, p = 0.002); greater MVPA with lower BMI (γ = −0.53, p = 0.018). LPA was not significantly associated with BMI in 24PAR models. Model fit: R² = 0.07. - Linear models using SWA: Greater relative time in sleep (γ = −5.10, p < 0.001), LPA (γ = −3.12, p < 0.001), and MVPA (γ = −1.43, p < 0.001) were associated with lower BMI, while greater SED was associated with higher BMI (γ = 8.93, p < 0.001). Model fit: R² = 0.28. - Compositional isotemporal substitution: Reallocating time from SED to MVPA, to sleep, and to LPA (the latter not evident with 24PAR) was associated with predicted reductions in BMI. Effect sizes for substitutions were larger when using SWA-derived compositions than 24PAR.

The findings support that, within the fixed 24-hour composition, higher relative time in sleep, LPA, and MVPA and lower relative time in SED are associated with more favorable BMI levels. Applying CoDA allowed appropriate modeling of the interdependent time-use components and quantification of predicted BMI changes under time reallocations. The stronger associations and greater explained variance observed with the monitor-based SWA compared to the report-based 24PAR suggest that objective monitoring may capture behavior distributions more accurately and relate more strongly to BMI. The results emphasize that both the composition of daily movement behaviors and the choice of assessment method influence observed relationships with adiposity.

Using Compositional Data Analysis on replicate-day data, the study demonstrates that favorable distributions of 24-hour movement behaviors—more sleep, LPA, and MVPA and less sedentary time—are generally associated with lower BMI. Associations were consistently stronger and models explained more variance when behaviors were measured with the SenseWear Armband than with the 24-hour Physical Activity Recall, underscoring the value of monitor-based assessment for studying obesity-related outcomes.