Combined Effects of Time Spent in Physical Activity, Sedentary Behaviors and Sleep on Obesity and Cardio-Metabolic Health Markers: A Novel Compositional Data Analysis Approach

This was not an experiment but an observational analysis of existing data from the 2005–2006 National Health and Nutrition Examination Survey (NHANES). The researchers tested whether the *composition* of how people spend their 24-hour day—specifically the proportions of time in sleep, sedentary behavior (SB), light-intensity physical activity (LIPA), and moderate-to-vigorous physical activity (MVPA)—is associated with obesity and cardio-metabolic health markers.

The key innovation was using a statistical method called "compositional data analysis" (CoDA). Traditional studies treat each behavior (e.g., "30 minutes of exercise") as independent, but time is finite—if you add 10 minutes of exercise, you must subtract 10 minutes from something else (sleep, sitting, or light activity). CoDA accounts for this codependence.

The outcome measures were:

**Obesity markers:** Body mass index (BMI, kg/m²) and waist circumference (cm)

**Cardio-metabolic markers:** Triglycerides (mg/dL), plasma glucose (mg/dL), plasma insulin (µU/mL), HDL cholesterol (mg/dL), LDL cholesterol (mg/dL), systolic blood pressure (mmHg), diastolic blood pressure (mmHg)

**Sample size:** 1,937 adults (from the NHANES 2005–2006 cycle)

**Population:** Non-institutionalized US civilians, aged 20–79 years

**Inclusion criteria:** Had at least 4 days of valid accelerometry data (≥10 hours/day wear time), plus complete data on sleep duration, BMI, waist circumference, blood pressure, and fasting blood markers

**Exclusion criteria:** Pregnant women, people with missing accelerometry data, and those with fewer than 4 valid days of monitoring

**Setting:** Free-living, community-dwelling adults in the United States

**Demographics:** Mean age ~47 years; approximately 50% female; mixed ethnicities (non-Hispanic white, non-Hispanic black, Mexican American, other)

**Physical activity and sedentary time:** Measured using an ActiGraph AM-7164 accelerometer worn on the right hip for 7 consecutive days. The device recorded activity counts per minute. Cut-points were:

- Sedentary: <100 counts/min

- Light intensity (LIPA): 100–2,019 counts/min

- Moderate-to-vigorous (MVPA): ≥2,020 counts/min

**Sleep duration:** Self-reported via questionnaire: "How much sleep do you usually get at night on weekdays or workdays?" (hours per night)

**Anthropometrics:** Measured by trained health technicians using standardized protocols. BMI calculated from height and weight. Waist circumference measured at the iliac crest.

**Blood markers:** Fasting blood samples (≥8 hours) analyzed for triglycerides, glucose, insulin, HDL, and LDL using standard enzymatic methods.

**Blood pressure:** Measured up to 3 times after 5 minutes seated rest; the average of the last 2 readings was used.

**Study design:** Cross-sectional observational study using secondary analysis of NHANES 2005–2006 data.

**Statistical approach:** Compositional data analysis (CoDA). The researchers converted raw time-use data (minutes/day in sleep, SB, LIPA, MVPA) into "isometric log-ratio" (ILR) coordinates. This allowed them to treat the 24-hour day as a single composition (a set of parts that sum to 1,440 minutes) rather than analyzing each behavior independently. They then used multiple linear regression to test associations between the composition and each health outcome, adjusting for age, sex, ethnicity, smoking, alcohol intake, and total energy intake.

**Key analytical technique:** They performed "reallocation modeling"—mathematically simulating what would happen to health markers if you reallocated a fixed amount of time (e.g., 10 minutes) from one behavior to another, while keeping total time constant. This is the closest you can get to a causal estimate from cross-sectional data.

**What this design can prove:**

It can identify *associations* between time-use compositions and health markers.

It can quantify the *direction and magnitude* of these associations.

It can show *asymmetry* (e.g., the harm of losing MVPA is larger than the benefit of gaining it).

**What this design cannot prove:**

**Causality.** This is cross-sectional data—all measurements were taken at one time point. You cannot determine whether changing your time use *causes* changes in health, or whether people with better health simply *choose* to be more active.

**Direction of effect.** It is equally plausible that lower waist circumference leads people to do more MVPA, rather than MVPA reducing waist circumference.

**Temporal sequence.** No follow-up data; no intervention was applied.

**Sleep measurement.** Sleep was self-reported (not accelerometry), which is known to be inaccurate (people typically overestimate sleep duration by 30–60 minutes). This introduces measurement error that could bias results.

**Confounding.** Despite adjusting for several covariates, residual confounding is possible (e.g., diet quality, socioeconomic status, occupational demands, genetic factors).

**Major methodological weakness:** The sleep data is self-reported while all other behaviors were measured by accelerometry. This creates an inconsistency in measurement precision. If sleep is systematically misreported, the compositional reallocation estimates involving sleep may be unreliable.

**Primary finding:** The overall 24-hour time-use composition (sleep, SB, LIPA, MVPA) was significantly associated with:

BMI (p < 0.001)

Waist circumference (p < 0.001)

Triglycerides (p < 0.001)

Plasma glucose (p < 0.001)

Plasma insulin (p < 0.001)

Systolic blood pressure (p < 0.001)

Diastolic blood pressure (p = 0.003)

**Not significantly associated with:**

HDL cholesterol (p > 0.05)

LDL cholesterol (p > 0.05)

**Within the composition, the strongest positive effect was for MVPA.** The proportion of time spent in MVPA had the largest beneficial association with all significant outcomes.

**Asymmetry of reallocation (the key novel finding):**

Reallocating 10 minutes of SB to MVPA was associated with **0.001% lower** waist circumference (essentially negligible benefit)

But reallocating 10 minutes of MVPA to SB was associated with **0.84% higher** waist circumference (substantial harm)

This asymmetry was consistent across all outcomes: the *loss* of MVPA was 100–1,000 times more harmful than the *gain* of the same amount was beneficial.

**Replacing SB with LIPA:**

For diabetes risk markers (glucose, insulin), replacing SB with LIPA was associated with favorable outcomes (lower glucose and insulin)

For obesity markers (BMI, waist circumference), replacing SB with LIPA showed smaller but still beneficial associations

**Replacing sleep with other behaviors:**

Replacing sleep with SB was associated with worse outcomes (higher BMI, waist circumference, glucose)

Replacing sleep with MVPA was associated with better outcomes

Replacing sleep with LIPA showed mixed results

**Magnitude of associations (example for waist circumference):**

Replacing 30 minutes of SB with 30 minutes of MVPA: associated with ~0.003% lower waist circumference

Replacing 30 minutes of MVPA with 30 minutes of SB: associated with ~2.5% higher waist circumference

To translate these numbers into plain English:

For a person with a waist circumference of 90 cm (about 35.4 inches):

Gaining 10 minutes of MVPA by sitting less would reduce waist circumference by roughly **0.09 mm** (the thickness of a credit card)

Losing 10 minutes of MVPA to sitting more would increase waist circumference by roughly **7.6 mm** (about the width of a finger)

This asymmetry is the headline finding. It suggests that the body's response to physical activity is not linear and symmetric. The protective effect of MVPA may be more about *preventing loss* of activity than about *adding* more activity. In practical terms: if you currently do 30 minutes of brisk walking daily, dropping to 20 minutes may harm your health more than increasing to 40 minutes would help.

For blood markers:

Replacing 10 minutes of SB with MVPA was associated with ~0.5% lower triglycerides

Replacing 10 minutes of MVPA with SB was associated with ~1.2% higher triglycerides

Effects on glucose and insulin followed similar asymmetric patterns

**Acknowledged by authors:**

Cross-sectional design prevents causal inference

Sleep was self-reported, not objectively measured

Accelerometry cannot capture all types of activity (e.g., cycling, swimming, upper-body strength training)

NHANES data is from 2005–2006, and population activity patterns may have changed

The compositional approach is novel and requires validation in other datasets

**Additional critical observations:**

**No blinding or randomization:** This is purely observational; no intervention was applied

**Measurement error in sleep:** Self-reported sleep duration is notoriously unreliable (people typically overestimate by 30–60 minutes). Since sleep is a key component of the composition, this could systematically bias the reallocation estimates involving sleep

**Accelerometry cut-points are arbitrary:** The threshold of 2,020 counts/min for MVPA is based on laboratory calibration studies and may not apply equally to all ages, fitness levels, or body types

**No adjustment for diet quality:** While total energy intake was adjusted, diet composition (e.g., macronutrient ratios, fiber, sugar) was not. Diet is a major confounder for obesity and cardio-metabolic outcomes

**No adjustment for socioeconomic status beyond education:** Income, occupation, and neighborhood environment are strong predictors of both activity patterns and health

**Single time point:** A single week of accelerometry may not represent habitual behavior

**Population limited to US adults:** Results may not generalize to children, older adults (>79), or non-US populations with different activity patterns

**No data on sleep quality:** Only sleep *duration* was measured; sleep quality, fragmentation, and timing are also important for metabolic health

**The asymmetry finding is mathematically expected:** In compositional data analysis, reallocation effects are inherently asymmetric because the baseline composition matters. The authors acknowledge this but the magnitude of asymmetry is still striking

For someone running their own n=1 experiment:

### What to test

Test the **asymmetry hypothesis** directly: Does losing 10 minutes of MVPA per day harm your waist circumference and blood markers more than gaining 10 minutes helps?

**Specific intervention:**

Phase 1 (Baseline, 2 weeks): Maintain your current activity pattern. Measure everything.

Phase 2 (Add MVPA, 3 weeks): Add 10 minutes of brisk walking or jogging per day by replacing 10 minutes of sitting. Do not change anything else.

Phase 3 (Subtract MVPA, 3 weeks): Remove 10 minutes of MVPA per day (replace with sitting). Return to baseline activity otherwise.

Phase 4 (Washout + repeat, optional): Return to baseline for 1 week, then repeat to confirm.

### Minimum meaningful duration

**3 weeks per phase** is the minimum to see changes in waist circumference and fasting glucose/insulin. Blood lipid changes may require 4–6 weeks.

**Total experiment duration:** At least 8–10 weeks (2 weeks baseline + 3 weeks add + 3 weeks subtract + 1 week washout). A more robust design would be 12–14 weeks.

### What to measure (specific metrics)

**Primary outcome:** Waist circumference (measured at the same time each morning, after voiding, before breakfast, using a consistent tape measure at the level of the iliac crest). Take 3 measurements and average.

**Secondary outcomes:**

- Fasting glucose (finger-stick glucometer, morning after ≥8 hour fast)

- Fasting insulin (if you have access to lab testing)

- Triglycerides (finger-stick lipid panel)

- Body weight (same scale, same time, same clothing)

- Resting heart rate (measured upon waking, before getting out of bed)

**Activity tracking:** Wear an accelerometer or use your phone's step counter to verify that you actually added/subtracted 10 minutes of MVPA. Log daily minutes of MVPA, SB, and sleep.

**Sleep tracking:** Use a sleep diary or wearable to track sleep duration and quality. This is critical because changing activity may alter sleep, which is a confounder.

### Key confounds to control for

**Diet:** Keep your diet as constant as possible across all phases. Log all food intake for the first week of each phase to verify consistency. If your diet changes, the results are uninterpretable.

**Sleep duration:** Aim for the same bedtime and wake time across all phases. If adding MVPA makes you sleep better (or worse), this could independently affect outcomes.

**Stress and alcohol:** Track daily stress (1–10 scale) and alcohol intake. Both affect waist circumference and blood markers.

**Menstrual cycle (if applicable):** Waist circumference and insulin sensitivity vary across the menstrual cycle. If possible, start each phase at the same cycle phase (e.g., follicular phase, days 1–14).

**Time of day for measurements:** Measure waist circumference and blood markers at the same time of day (±30 minutes) for every measurement.

**Hydration status:** Dehydration can falsely elevate waist circumference and blood glucose. Maintain consistent water intake.

### What a positive result would look like

**If the asymmetry hypothesis holds:** You would see a larger *increase* in waist circumference (e.g., +0.5–1.0 cm) during the "subtract MVPA" phase than the *decrease* (e.g., –0.1–0.2 cm) during the "add MVPA" phase.

**For blood markers:** Fasting glucose might rise by 3–5 mg/dL when you lose MVPA, but only drop by 1–2 mg/dL when you add it.

**A null result** (symmetric changes or no changes) would also be informative—it might mean the asymmetry only appears over longer timeframes, or that your baseline activity level is already high enough that small changes don't matter.

**Important caveat for n=1:** The asymmetry reported in this paper is a *population average*. Your individual response may differ due to genetics, baseline fitness, age, sex, and body composition. The only way to know is to test it on yourself.