Accelerometer-derived circadian rhythm and colorectal cancer risk in UK Biobank: a prospective cohort study

This prospective cohort study of 95,050 UK Biobank participants found that while specific accelerometer-derived physical activity patterns were nominally associated with reduced colorectal cancer risk, these associations largely disappeared after adjusting for lifestyle and metabolic factors, suggesting they are not independent predictors.

The Gist

The Big Picture: Checking Your Body's "Internal Clock"

Imagine your body has a master conductor, like the leader of a symphony orchestra. This conductor is your circadian rhythm (your internal body clock). It tells your cells when to work, when to rest, and when to repair themselves.

For a long time, scientists thought that to keep this orchestra healthy, you just needed to play a lot of music (do a lot of exercise). But this new study asked a different question: Does how you play the music matter more than just the total volume?

The researchers used a massive database (UK Biobank) and gave 95,000 people smartwatches. These watches didn't just count steps; they acted like high-tech spies, recording exactly how people moved, slept, and were exposed to light over a whole week. They then watched to see who developed colorectal cancer (cancer of the colon or rectum) over the next 8.5 years.

The Investigation: 224 Clues

The researchers looked at 224 different clues (metrics) from the watches. They grouped these clues into five categories:

1. Step Counts: How many steps you took.
2. Sleep: How well you slept and how long you stayed asleep.
3. Light: How much natural light you saw during the day and night.
4. Activity Bouts: Did you move in short bursts or long, steady streams?
5. Rhythm: How steady was your daily pattern?

The Findings: What Worked and What Didn't

1. The "Steady Stream" vs. The "Sprint"

The study found that the structure of your movement matters more than just the total number of steps.

• The Good News: People who moved in steady, moderate-to-vigorous bursts (like a brisk 5-to-10-minute walk) had a lower risk of colon cancer.
• The Analogy: Think of your body like a garden hose. If you just let a tiny trickle of water drip (very light, sporadic movement), it doesn't clean the hose well. But if you turn the water on to a steady, strong stream for a few minutes, it flushes out the gunk. The study suggests that these "steady streams" of activity help keep the colon healthy.

2. The "Broken Rhythm" Warning

The study found that people whose daily patterns were fragmented (jumping between activity and sitting down constantly, or having a weak difference between day and night) had a higher risk of colon cancer.

• The Analogy: Imagine a train schedule. If the train runs on time, stops at the station, and leaves on time, the system works smoothly. If the train is constantly stopping and starting, or if the schedule is chaotic, the system breaks down. A "broken rhythm" in your body seems to confuse the cells, making them more vulnerable to cancer.

3. The Surprising Silence: Sleep and Light

You might expect that sleeping poorly or seeing too much light at night would be the biggest red flags. Surprisingly, they weren't.

• In this specific group of people, the way they moved during the day was a much stronger predictor of cancer risk than how they slept or how much light they saw.
• The Analogy: It's like trying to fix a car engine. You might think the problem is the oil (sleep) or the fuel (light), but the mechanic (the data) says, "No, the problem is actually how the driver is shifting gears (movement patterns)."

The Catch: It's Complicated

The study had a major "plot twist." When the researchers adjusted their math to account for other factors like body weight, diabetes, and overall health, the strong signals got much weaker.

• What this means: The "good movement patterns" and "healthy rhythms" are likely just symptoms of a generally healthy body. If you are lean, don't have diabetes, and eat well, you probably also move in a steady, rhythmic way. It's hard to tell if the movement caused the health, or if being healthy allowed for that movement.
• The E-Value: The researchers calculated a "confounding score" (E-value). It suggested that an unmeasured factor (like diet or stress) could easily explain away the link.

The Bottom Line

This study is like a hypothesis generator, not a final verdict. It didn't prove that "walking in 10-minute bursts prevents cancer."

Instead, it gave us a map:

1. Colon cancer seems to care a lot about how we move (steady bursts, good rhythm).
2. Rectal cancer didn't show the same clear patterns (it's a different beast).
3. Sleep and light didn't show up as the main villains in this specific dataset.

The Takeaway for You:
While we wait for more research to confirm the cause-and-effect, the study reinforces a simple, old-fashioned idea: Keep your body moving in steady, rhythmic chunks. Don't just sit all day and then sprint once. Break up your day with consistent, moderate activity. It might just be the "steady stream" your internal orchestra needs to stay in tune.

Technical Summary

1. Problem Statement

Colorectal cancer (CRC) is a leading cause of cancer mortality. While physical activity is a known protective factor, traditional epidemiological studies rely on self-reported data that capture only total volume, missing the temporal structure and fine-grained behavioral patterns of movement.

• The Gap: It is unclear whether multidimensional circadian patterns—comprising step counts, sleep architecture, physical activity bouts, circadian rhythmicity, and light exposure—independently influence CRC risk.
• The Challenge: Accelerometer data generates a massive, highly correlated hypothesis space (hundreds of metrics), making it difficult to distinguish independent causal signals from confounding factors (e.g., total activity volume vs. activity fragmentation).
• Subsite Specificity: Colon and rectal cancers differ anatomically and molecularly; it is unknown if they share the same behavioral risk profiles.

2. Methodology

The study employed a hypothesis-generating Exposure-Width Association Study (ExWAS) design using data from the UK Biobank.

• Study Population:

  • Cohort: 95,050 participants (aged 40–69) with valid 7-day wrist-worn accelerometer data and no prevalent CRC at baseline.
  • Follow-up: Median of 8.5 years.
  • Outcomes: 775 incident CRC cases (503 colon, 269 rectal), identified via ICD-10 codes (C18–C20) linked to NHS records.

• Data Processing & Phenotyping:

  • Device: Axivity AX3 triaxial accelerometers (100 Hz sampling).
  • Software: Processed using GGIR (v3.3-4) in R.
  • Metric Generation: 224 distinct accelerometer-derived metrics were categorized into five core domains:
    1. Objective Light Exposure: 24-hour rhythmic profiles (night, morning, afternoon, evening, late-night).
    2. Step Counts: Total volume, peak cadence (1–30 min windows), and functional categorization (incidental vs. purposeful).
    3. Sleep Architecture: Sleep-period-time (SPT) derived metrics including duration, efficiency, fragmentation (NNA, WASO), and timing.
    4. Physical Activity (PA): Volume (LPA, MVPA, VPA), bout structure (80% criterion for 1, 5, 10 min), and fragmentation (transition probabilities, Gini index).
    5. Circadian Rhythmicity: Non-parametric (Relative Amplitude, Interdaily Stability, Intradaily Variability) and parametric (Cosinor models: Mesor, Amplitude, Acrophase) metrics, plus Detrended Fluctuation Analysis (DFA).

• Statistical Analysis:

  • Model: Cox proportional hazards regression with age as the timescale.
  • Adjustment Tiers:
    • Minimal Discovery Model: Adjusted for sex, race, Townsend Deprivation Index, smoking, alcohol, season, and family history.
    • Sensitivity Tiers: Added adiposity/metabolic factors (BMI, diabetes) and total acceleration volume to test independence.
  • Robustness Checks: Lag-time analyses (excluding 1–2 years of follow-up) to mitigate reverse causation; E-value calculation for unmeasured confounding; subgroup analyses by sex and age.
  • Clustering: Integrated behavioral profiles were generated via cluster analysis to synthesize the five domains.

3. Key Results

• Signal Burden: In the minimal model, 121 metrics showed nominal significance ($p < 0.05$):
  • 31 for overall CRC.
  • 89 for colon cancer.
  • 1 for rectal cancer.
• Dominant Domains: Significant associations were almost exclusively found in Physical Activity Architecture and Step Counts.
  • Colon Cancer: Strongest signals included frequent short light-activity bouts (HR 0.85) and higher intradaily variability (IV), indicating rhythm fragmentation (HR 1.14). Lower activity during the least-active 5-hour window (L5) was protective.
  • Overall CRC: Higher mean acceleration during 5–10 minute MVPA bouts was protective (HR 0.88 per SD).
  • Null Domains: No metrics in Sleep Architecture or Light Exposure reached nominal significance in the primary model.
• Rectal Cancer: Exhibited a sparse signal profile, with only one significant metric (Self-Similarity Parameter from DFA, HR 1.14).
• Impact of Adjustment:
  • Signals attenuated substantially after adjusting for adiposity/metabolic factors and total acceleration volume.
  • For colon cancer, significant associations dropped from 89 (Minimal) to 9 (Independent Pattern model), suggesting that structural behavioral signals are intrinsically linked to total activity volume and metabolic health.
• Temporal Stability: Lag analyses showed high stability; 99 of the 121 primary associations remained significant after excluding the first 2 years of follow-up, arguing against reverse causation.
• Cluster Analysis: A "low-activity/fragmented-rhythm" profile was associated with a 33% higher risk of colon cancer (HR 1.33) compared to an intermediate profile, though this signal was sensitive to feature selection.

4. Key Contributions

• High-Resolution Circadian Mapping: This is one of the first studies to systematically map 224 multidimensional circadian metrics against CRC risk, moving beyond simple "total activity" or "sleep duration" metrics.
• Subsite Differentiation: The study highlights a stark divergence between colon and rectal cancer risk profiles, suggesting they should not be treated as a monolithic entity in behavioral epidemiology.
• Structural vs. Volume: It demonstrates that the structure of activity (bout intensity, fragmentation, rhythm robustness) carries distinct signals, even if they are highly correlated with total volume.
• Digital Biomarker Landscape: Provides a systematic "map" of candidate behavioral phenotypes (specifically MVPA bout structure and rhythm fragmentation) for future mechanistic research.

5. Significance and Limitations

• Significance:

  • Biological Plausibility: The findings align with preclinical evidence that circadian disruption and sedentary behavior drive inflammation and insulin resistance, key drivers of colorectal carcinogenesis.
  • Public Health: Suggests that interventions focusing on breaking up sedentary time and maintaining robust day-night activity contrasts may be more critical than just increasing total exercise volume.
  • Methodological: Establishes a framework for ExWAS in digital epidemiology, balancing discovery with rigorous sensitivity testing.

• Limitations:

  • Hypothesis Generating: As an ExWAS with 224 correlated tests, findings are nominal and require independent replication.
  • Confounding: The substantial attenuation of signals after metabolic adjustment suggests these behavioral markers may be proxies for metabolic health rather than independent causal drivers.
  • Single Time Point: Accelerometry was measured once; long-term changes in behavior were not captured.
  • Selection Bias: The "healthy volunteer" bias of UK Biobank may limit generalizability to higher-risk populations.
  • Sensor Limitations: Wrist-worn devices may lack sensitivity to detect subtle light-oncogenesis pathways compared to chest-worn or environmental light sensors.

Conclusion: The study identifies locomotor step counts, physical activity intensity, and circadian rhythmicity as the most promising digital biomarkers for colon cancer risk, while finding no independent signal for sleep or light exposure in this cohort. The results underscore the complexity of circadian digital biomarkers and the necessity of independent replication to clarify their utility in cancer risk stratification.