Beyond the 24-Hour Rhythm: Functional Principal Components of Daily Activity Residuals and Mortality in NHANES

\begin{abstract} \noindent\textbf{Background.} Accelerometer-derived 24-hour activity profiles are usually summarised by cosinor rhythm parameters or by functional principal component analysis (FPCA). Applied to the raw profiles, however, the leading functional components largely re-express what the cosinor already captures: activity level, rhythm strength and timing. Whether the shape of the day carries health information beyond these parameters is unclear. \noindent\textbf{Methods.} We analysed adults with seven valid days of accelerometry from two National Health and Nutrition Examination Survey (NHANES) cohorts: hip activity counts (2003--2006; $n = 2{,}615$) and wrist Monitor-Independent Movement Summary units (2011--2014; $n = 3{,}208$). Minute-level activity was transformed with the inverse hyperbolic sine, and a single 24-hour cosinor curve was fitted across each participant's seven days. Each participant's average residual curve, which contains no 24-hour component, was decomposed by FPCA and, to separate stable between-person patterns from day-to-day variation, by multilevel FPCA. In exploratory analyses, we related the three leading components to all-cause mortality and, secondarily, to cardiovascular and cancer mortality. We used Cox proportional hazards models adjusted for age, sex, race/ethnicity, income and the cosinor parameters, with the hip cohort for discovery and the wrist cohort for replication. \noindent\textbf{Results.} The three leading residual components explained 44\% (hip) and 51\% (wrist) of residual variance and consisted mainly of 12-hour and 8-hour patterns. The second component (FPC2) contrasts an early-starting, flat day with a day of distinct mid-morning and evening peaks. Its shape was nearly identical in the two cohorts (similarity 0.96), and multilevel FPCA recovered it as a stable between-person pattern. In unweighted models, FPC2 was the only component associated with all-cause mortality in the discovery cohort (hazard ratio per SD 1.17, 95\% CI 1.06 to 1.29; Benjamini--Hochberg $q = 0.008$), and the association replicated in the wrist cohort (1.18, 1.03 to 1.35; $p = 0.017$). With survey weights that also accounted for selection into the analytic sample, the association was concentrated in cardiovascular deaths (hip 1.53, 1.12 to 2.09, $p = 0.007$; wrist 1.38, 1.10 to 1.72, $p = 0.005$). This association persisted after further adjustment for lifestyle factors and traditional activity metrics. FPC2 was not associated with cancer mortality. Scores of the leading components did not form distinct clusters (average silhouette width $\leq 0.23$), but FPC2 was selected as a continuous predictor of cardiovascular mortality in both cohorts by forward selection on 10-fold cross-validated, survey-weighted AUC. \noindent\textbf{Conclusions.} The within-day structure of activity that a 24-hour cosinor cannot represent varies continuously between people. One replicated aspect of it, the loss of a two-peak daily pattern, was associated with mortality beyond rhythm strength and timing, most clearly cardiovascular mortality. The shape of daily activity therefore carries prognostic information beyond standard rhythm summaries. It can support mortality risk prediction that needs only a few personal details, and may help identify individuals at elevated risk and tailor interventions to their daily activity patterns. \medskip \noindent\textbf{Keywords:} functional principal component analysis, cosinor, rest-activity rhythm, mortality, accelerometry \end{abstract}

Authors

Institutions

Publication Details

Journal
medRxiv
Published
2026-11-12
DOI
https://doi.org/10.64898/2026.11.10.26367785
Primary Topic
Physical Activity and Health
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Beyond the 24-Hour Rhythm: Functional Principal Components of Daily Activity Residuals and Mortality in NHANES

Xin Zheng
medRxiv
Physical Activity and Health
preprint

Beyond the 24-Hour Rhythm: Functional Principal Components of Daily Activity Residuals and Mortality in NHANES

Xin Zheng
preprint en

Abstract

\begin{abstract} \noindent\textbf{Background.} Accelerometer-derived 24-hour activity profiles are usually summarised by cosinor rhythm parameters or by functional principal component analysis (FPCA). Applied to the raw profiles, however, the leading functional components largely re-express what the cosinor already captures: activity level, rhythm strength and timing. Whether the shape of the day carries health information beyond these parameters is unclear. \noindent\textbf{Methods.} We analysed adults with seven valid days of accelerometry from two National Health and Nutrition Examination Survey (NHANES) cohorts: hip activity counts (2003--2006; $n = 2{,}615$) and wrist Monitor-Independent Movement Summary units (2011--2014; $n = 3{,}208$). Minute-level activity was transformed with the inverse hyperbolic sine, and a single 24-hour cosinor curve was fitted across each participant's seven days. Each participant's average residual curve, which contains no 24-hour component, was decomposed by FPCA and, to separate stable between-person patterns from day-to-day variation, by multilevel FPCA. In exploratory analyses, we related the three leading components to all-cause mortality and, secondarily, to cardiovascular and cancer mortality. We used Cox proportional hazards models adjusted for age, sex, race/ethnicity, income and the cosinor parameters, with the hip cohort for discovery and the wrist cohort for replication. \noindent\textbf{Results.} The three leading residual components explained 44\% (hip) and 51\% (wrist) of residual variance and consisted mainly of 12-hour and 8-hour patterns. The second component (FPC2) contrasts an early-starting, flat day with a day of distinct mid-morning and evening peaks. Its shape was nearly identical in the two cohorts (similarity 0.96), and multilevel FPCA recovered it as a stable between-person pattern. In unweighted models, FPC2 was the only component associated with all-cause mortality in the discovery cohort (hazard ratio per SD 1.17, 95\% CI 1.06 to 1.29; Benjamini--Hochberg $q = 0.008$), and the association replicated in the wrist cohort (1.18, 1.03 to 1.35; $p = 0.017$). With survey weights that also accounted for selection into the analytic sample, the association was concentrated in cardiovascular deaths (hip 1.53, 1.12 to 2.09, $p = 0.007$; wrist 1.38, 1.10 to 1.72, $p = 0.005$). This association persisted after further adjustment for lifestyle factors and traditional activity metrics. FPC2 was not associated with cancer mortality. Scores of the leading components did not form distinct clusters (average silhouette width $\leq 0.23$), but FPC2 was selected as a continuous predictor of cardiovascular mortality in both cohorts by forward selection on 10-fold cross-validated, survey-weighted AUC. \noindent\textbf{Conclusions.} The within-day structure of activity that a 24-hour cosinor cannot represent varies continuously between people. One replicated aspect of it, the loss of a two-peak daily pattern, was associated with mortality beyond rhythm strength and timing, most clearly cardiovascular mortality. The shape of daily activity therefore carries prognostic information beyond standard rhythm summaries. It can support mortality risk prediction that needs only a few personal details, and may help identify individuals at elevated risk and tailor interventions to their daily activity patterns. \medskip \noindent\textbf{Keywords:} functional principal component analysis, cosinor, rest-activity rhythm, mortality, accelerometry \end{abstract}

medRxiv
Amsterdam University Medical Centers (NL)
Good health and well-being
Physical Activity and Health
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon