Simplifying Daily Cortisol Cycle Analysis: Validation and Benchmarking of the Cortisol Sine Score Against Cosinor and JTK_CYCLE models

This study introduces and validates the Cortisol Sine Score (CSS), a novel, model-free scalar metric that simplifies the analysis of daily cortisol rhythms into a single interpretable value, demonstrating strong agreement with established circadian models while enabling practical application in large-scale epidemiological studies, clinical workflows, and machine learning pipelines with reduced sampling requirements.

The Gist

The Problem: The "Too Complicated" Hormone Report

Imagine your body has a built-in alarm clock and a stress meter called cortisol. In a healthy person, this hormone acts like a sunrise: it spikes high and bright when you wake up, then slowly fades away like a sunset as the day goes on.

However, when people are under chronic stress, depressed, or facing tough life circumstances, this "sunrise" can get messed up. The peak might happen too late, or the whole curve might flatten out so there's no peak at all.

The old way to measure this was like trying to describe a painting by listing every single brushstroke, the exact shade of blue, the angle of the light, and the texture of the canvas. Scientists used complex math models (like Cosinor and JTK_CYCLE) to get these details. While accurate, the results were a massive list of numbers that were hard to plug into simple computer programs, hard to explain to a doctor, and impossible to use for big AI projects that need just one number to make a decision.

The Solution: The "Cortisol Sine Score" (CSS)

The researchers invented a new tool called the Cortisol Sine Score (CSS).

Think of the CSS as a "Rhythm Compass." Instead of giving you a map with every street name, it just tells you:

• Positive Number (+): You are a "Morning Person." Your cortisol peaks early (healthy sunrise).
• Negative Number (-): You are a "Night Owl" (or your rhythm is broken). Your cortisol peaks late or is inverted.
• Zero (0): You are "Flat." Your cortisol is the same all day long (like a flat line on a heart monitor), which usually means high stress or a broken internal clock.

The Magic Trick:
The CSS takes all your cortisol samples from the day and runs them through a simple math filter (a "sine wave"). It asks: "Does this hormone pattern match the natural rhythm of the sun?" If yes, you get a high score. If no, you get a low or negative score. It collapses a whole day of complex data into one single, easy-to-understand number.

The Validation: Does it Work?

The team tested this new "Compass" against the old, complicated "Painting Descriptions" using data from 501 pregnant women who took saliva samples every 4 hours for a full day.

1. It Matches the Experts: The CSS agreed perfectly with the complex models. If the complex model said, "This person has a strong morning peak," the CSS gave a high positive number. If the complex model said, "This person is stressed and flat," the CSS gave a near-zero number.
2. It Predicts Real Life: The CSS was better at spotting the link between Social Disadvantage (like poverty or lack of resources) and stress hormones than any other method. It also did a great job predicting the makeup of the gut microbiome (the bacteria in your stomach), showing that your stress rhythm is deeply connected to your digestion.

The Bonus: Less Work for You!

Usually, to get a perfect reading, you have to wake up in the middle of the night (2:00 AM) to give a sample. That is exhausting and annoying.

The researchers asked: "Can we get a good score with fewer samples?"
They tested every combination of times. They found that you don't need the middle-of-the-night sample to get a great result.

• The Magic Formula: You only need 4 samples (6:00 AM, 10:00 AM, 2:00 PM, 6:00 PM) to get a score that is 95% as accurate as the full 6-sample day.
• The "Lazy" Formula: You can even get away with just 3 samples (2:00 AM, 6:00 AM, 6:00 PM) and still get a very reliable score.

Why This Matters

1. For Doctors: It's a quick "stress check." Instead of looking at a spreadsheet of 20 numbers, they can look at one score to see if a patient's stress system is working.
2. For AI and Big Data: Computers love simple inputs. Because the CSS is just one number, AI can easily use it to find patterns in thousands of people, linking stress rhythms to diseases, gut health, or mental health.
3. For Wearable Tech: Because the math is so simple, future smartwatches or sweat-sensors could calculate this score in real-time, telling you, "Hey, your stress rhythm is off today, try to relax," without needing a supercomputer in the cloud.

The Bottom Line

The Cortisol Sine Score is like turning a complex, 100-page novel about your stress levels into a simple one-sentence summary. It tells you if your internal clock is ticking with the sun, fighting against it, or has stopped altogether. It's simpler, faster, and just as accurate as the old complicated methods, making it easier to study stress in the real world.

Technical Summary

1. Problem Statement

The daily cortisol cycle is a critical indicator of Hypothalamic-Pituitary-Adrenal (HPA) axis function, with alterations linked to chronic stress, psychiatric disorders, and metabolic dysfunction. However, current analytical methods for characterizing this cycle present significant barriers to large-scale application:

• Complexity of Outputs: State-of-the-art methods like Cosinor regression (parametric) and JTK_CYCLE (non-parametric) produce multiple outputs (amplitude, acrophase, mesor, p-values) rather than a single scalar value.
• Integration Challenges: These multi-output metrics are difficult to integrate into standard statistical workflows, mediation analyses, or Machine Learning/AI pipelines that often require single-value inputs.
• Participant Burden: Accurate circadian profiling typically requires intensive sampling (e.g., 6+ time points over 24 hours), which is burdensome for participants and impractical for large epidemiological studies or wearable sensor applications.

There is a critical need for a unified, model-free, scalar metric that captures the timing, alignment, and directionality of the cortisol cycle while remaining interpretable and computationally efficient.

2. Methodology

The Cortisol Sine Score (CSS)

The authors developed the Cortisol Sine Score (CSS), a novel, model-free scalar metric designed to collapse the temporal dynamics of cortisol exposure into a single interpretable index.

• Mathematical Basis: Inspired by Fourier decomposition, the CSS computes a weighted sum of cortisol measurements using sine-transformed time-of-day weights.
• Formula:
  $$CSS = \sum_{i=1}^{n} (Cortisol_i \times \sin(\frac{2\pi t_i}{24}))$$
  Where $Cortisol_i$ is the concentration at time $t_i$ (in hours).
• Interpretation:
  • Positive Values: Indicate morning-dominant patterns (healthy, circadian-aligned).
  • Negative Values: Indicate evening-shifted or inverted rhythms.
  • Near-Zero Values: Indicate flattened rhythms (characteristic of chronic stress or circadian disruption).

Data Source and Validation

• Cohort: Data from 501 pregnant women (March of Dimes cohort) with 3,006 salivary cortisol samples collected at 6 time points daily (02:00, 06:00, 10:00, 14:00, 18:00, 22:00) during the second trimester.
• Benchmarking: The CSS was validated against:
  1. Cosinor Regression: Estimating amplitude and acrophase.
  2. JTK_CYCLE: Estimating rhythmicity, period, and phase.
  3. Other Metrics: Area Under the Curve (AUCg), MESOR, and a custom "Atypicality" metric.
• Sampling Optimization: The authors systematically tested all 63 possible combinations of the 6 time points to determine the minimum sampling protocol required to maintain high fidelity to the full 6-sample CSS.

Use Cases

1. Social Disadvantage (SD): Correlation analysis between CSS and SD (a composite of socioeconomic factors) to test known associations with cortisol dysregulation.
2. Gut Microbiome: A Random Forest machine learning framework was used to predict gut microbiome composition based on cortisol metrics, testing CSS utility in multi-omics contexts.

3. Key Results

Sampling Efficiency

• Minimum Requirement: The CSS can be reliably estimated with as few as 3–4 daily samples without significant loss of accuracy compared to the 6-sample gold standard.
• Optimal Protocols:
  • 3-Sample Protocol (02:00, 06:00, 18:00): $r = 0.952$, MAE = 0.087.
  • 4-Sample Protocol (06:00, 10:00, 14:00, 18:00): $r = 0.953$, MAE = 0.023.
• Critical Time Point: The 06:00 measurement (morning peak) was identified as essential; combinations lacking this time point showed significantly reduced correlation ($r < 0.69$).

Validation Against Gold Standards

• Amplitude Correlation: CSS showed strong positive associations with observed, Cosinor-estimated, and JTK_CYCLE-estimated amplitudes ($\beta \approx 0.73 - 2.08, p < 0.0001$).
• Acrophase Correlation: CSS significantly distinguished between high, mid, and low tertiles of acrophase (peak timing), showing a ~5-hour shift in peak time between low and high CSS groups.
• Classification Performance: CSS successfully classified individuals with high vs. low amplitude with an AUC of 0.89.

Benchmarking and Use Cases

• Social Disadvantage: CSS demonstrated the strongest correlation with Social Disadvantage ($\rho = -0.37, p < 0.0001$), outperforming or matching other metrics like Cosinor/JTK amplitudes.
• Gut Microbiome Prediction: In a Random Forest model, CSS was the top predictor of gut microbiome composition, outperforming traditional amplitude metrics and the custom "Atypicality" score. This suggests CSS captures biologically relevant phase alignment that amplitude alone misses.

4. Key Contributions

1. Novel Scalar Metric: Introduction of the CSS, a single-value metric that encapsulates both the magnitude and timing (directionality) of cortisol rhythms.
2. Reduced Burden Protocol: Demonstration that high-fidelity circadian analysis is possible with 3–4 samples (reducing burden by 33–50%), specifically highlighting the necessity of the 06:00 time point.
3. ML/AI Compatibility: Proof that CSS is superior for integration into machine learning pipelines and multi-omics studies compared to multi-output parametric models.
4. Open-Source Tool: Release of the CortSineScore R package to democratize access to this analysis method.

5. Significance and Implications

• Bridging the Gap: The CSS bridges the divide between detailed circadian biology and large-scale data science, enabling the inclusion of cortisol rhythmicity in complex statistical models and AI frameworks where it was previously too cumbersome to include.
• Clinical and Epidemiological Utility: The metric allows for rapid screening of HPA-axis dysfunction in large cohorts, clinical settings, and resource-limited environments.
• Wearable Technology: The mathematical simplicity of the CSS makes it ideal for implementation in real-time wearable biosensors (e.g., sweat-based sensors) for continuous HPA-axis monitoring without heavy computational overhead.
• Biological Insight: The superior performance of CSS in predicting gut microbiome composition suggests that the timing and directionality of cortisol exposure are more critical for systemic physiological regulation than amplitude alone.

Limitations: The authors note that CSS trades information richness for simplicity. It cannot distinguish between hypocortisolism (Addison's) and hypercortisolism (Cushing's) if both result in flattened rhythms (near-zero CSS). Therefore, it should be used alongside total cortisol output measures (like AUCg) for absolute diagnosis.