🧬 biology

From Splits to Heat Strain: A Ten-Year Analysis of Performance Determinants and a Digital Twin Prototype for Olympic-Distance Triathlon Developed Using a Generative AI Workflow

This ten-year analysis of over 7,000 Olympic-distance triathlon records utilizes a human-in-the-loop generative AI workflow to identify the run leg as the primary performance discriminator, assess heat resilience, and develop a digital twin prototype for optimizing race planning and heat-strain management.

Original authors: Marco Cardinale

Published 2026-07-08

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Marco Cardinale

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

The Big Picture: A Decade of Triathlon Data

Imagine trying to understand what makes a triathlete win an Olympic-distance race (1.5km swim, 40km bike, 10km run). For a long time, experts had to guess based on small snapshots of data.

This paper is like a massive time-traveling detective story. The author, Marco Cardinale, used Artificial Intelligence (AI) to dig through ten years (2015–2025) of race results from 339 major championships. He looked at 7,415 race performances by over 3,000 different athletes.

He didn't just look at the clock; he also reconstructed the weather on every single race day to see how heat and humidity affected the runners. The goal was to build a "Digital Twin"—a virtual clone of a race scenario—to help predict who might win and how the heat might change the outcome.

The Three Legs of the Race: Who Really Matters?

Think of a triathlon as a relay race where the same person runs all three legs. The paper asked: Which leg is the most important for winning?

The Swim: This is like the starting line of a sprint. It's important because it decides who gets to stand in the front row for the next part. However, the data shows that once the race gets going, the swim time doesn't actually predict the final winner very well.
The Bike: This is the middle leg. In elite races, the cyclists often ride in a tight pack (like a school of fish), which helps them save energy. Because everyone rides so close together, the bike leg tends to "flatten out" the differences between athletes. It's a necessary part of the race, but it rarely decides the winner on its own.
The Run: This is the boss. The study found that the running leg is the single biggest factor in deciding who stands on the podium (the top 3). Whether it's men or women, if you want to win, you have to be the best runner at the end of the day. The AI confirmed that the run leg explains the most about who wins.

The Analogy: Imagine a three-legged race where the first two legs are just about staying in the pack. The third leg is where everyone has to sprint alone. The paper says the winner is almost always the person who is fastest on that final solo sprint.

The Heat Factor: The Great Equalizer?

The researchers wanted to know: Does the heat make everyone slow down, or do some people handle it better?

They used a special weather metric called WBGT (Wet Bulb Globe Temperature), which measures how hot it feels when you consider the sun, humidity, and wind. The races they studied ranged from a cool 7°C to a scorching 28°C.

The Finding: The heat didn't affect everyone the same way. It's not a simple rule like "hotter = slower."
The "Heat Tolerant" Athletes: Some athletes actually seemed to perform better (or at least not worse) as the temperature rose. About half of the male athletes and about a third of the female athletes showed this "heat tolerance."
The "Heat Sensitive" Athletes: Others struggled significantly as the temperature climbed.
The Conclusion: There is no single "heat rule" for the whole group. Instead, heat acts like a filter that separates the athletes who are built (or trained) for the heat from those who aren't.

The "Digital Twin": A Virtual Crystal Ball

The most futuristic part of the paper is the Digital Twin.

Imagine you have a video game character that is a perfect digital copy of a real athlete. This "Digital Twin" has memorized every race the athlete has ever run.

How it works: The AI looks at an athlete's past swim, bike, and run times, plus the weather forecast for an upcoming race.
What it predicts: It tries to guess how fast the athlete will go in the next race.
How accurate is it? It's like a weather forecast: it's not perfect, but it's useful. The model can predict a race time within a reasonable margin of error (about 0.5 "z-units," which translates to roughly 1.5 to 2 minutes over the whole race).
The Goal: This isn't a magic wand that guarantees a win. Instead, it's a planning tool. It helps coaches and athletes say, "If the weather is going to be this hot, and the athlete has this history, here is a realistic target time to aim for."

The "Did Not Finish" (DNF) Mystery

The paper also looked at how many people quit the race (DNF).

In very hot conditions (the "Blue Flag" category, which is the hottest allowed for racing), the quit rate went up.
However, the data didn't tell the researchers why people quit. Did they get heatstroke? Did they crash their bike? Did they get a cramp? The data just says "they didn't finish."
The study found that in the hottest races, more people quit, especially in women, but it couldn't pinpoint the exact medical cause because the public data doesn't record that level of detail.

Summary

In simple terms, this paper used AI to read a decade of triathlon history and found three main things:

The Run is King: If you want to win an Olympic triathlon, your running speed at the end is the most important thing.
Heat is Personal: Some athletes thrive in the heat, while others crumble. There is no one-size-fits-all reaction to hot weather.
The Virtual Coach: We can now build a "Digital Twin" that uses past data and weather forecasts to give a realistic prediction of how an athlete might perform, helping them plan their race strategy better.

The paper stops there; it doesn't claim to have a cure for heatstroke or a guaranteed training program, but it provides the data and the tools to help experts figure those things out.

Technical Summary: From Splits to Heat Strain in Olympic-Distance Triathlon

Problem Statement
Despite the availability of extensive race-result archives, there has been no large-scale longitudinal analysis that simultaneously links split-level kinematics to meteorological data across a full decade of international Olympic-distance triathlon championships. While existing literature identifies running as a primary performance determinant and acknowledges the impact of thermal loads, gaps remain in understanding the interplay between discipline-specific pacing, transition efficiency, race-field quality, and ambient thermal stress. Furthermore, there is a lack of data-driven tools to support individualized race planning, heat acclimatization strategies, and safety interventions for elite athletes competing in increasingly hot conditions.

Methodology
This study employed a "human-in-the-loop" generative AI workflow to construct a comprehensive analytical pipeline. The research utilized a dataset of 7,415 athlete-race records from 339 championship events (2015–2025) involving 3,060 unique athletes across six continents.

Data Acquisition & Engineering: Race results were retrieved via the World Triathlon REST API. Split times (swim, T1, bike, T2, run) were standardized into z-scores within each race and sex stratum to control for variations in field strength and course conditions.
Meteorological Reconstruction: Race-day Wet Bulb Globe Temperature (WBGT) was reconstructed for 301 events (88% coverage) using the Open-Meteo historical archive and the Liljegren et al. physical energy-balance model. Water temperature data were also integrated where available.
Statistical & Machine Learning Modeling:
- Performance Determinants: Ordinary Least Squares (OLS) regression and XGBoost classifiers were used to quantify the unique variance contributions of each discipline to total race time and podium outcomes.
- Heat Stress Analysis: Linear mixed-effects models with random athlete-level slopes were used to assess individual heat tolerance. Athletes were stratified by their response to WBGT changes (slope < 0 indicating heat tolerance).
- DNF Analysis: Logistic regression and Wilson score confidence intervals were applied to analyze Did-Not-Finish (DNF) rates against WBGT flag categories (Green, Blue, Orange).
- Digital Twin Prototype: An XGBoost regression model was developed to predict total race z-scores based on historical split performance and race-day WBGT. The model utilized a temporal hold-out validation strategy (training on 2015–2022, testing on 2023–2025) to prevent data leakage.

Key Results

Performance Determinants:
- Variance Contribution: In terms of total race time, the bike and run legs contributed near-equal unique variance (semi-partial $R^2 \approx 0.13$ each), while the swim leg contributed minimally (~0.02).
- Podium Prediction: The run leg was the dominant discriminator for podium finishes. XGBoost feature importance ranked the run highest (Men: 45.5%, Women: 48.9%), followed by the bike and swim. This confirms that pack cycling compresses variance, shifting the decisive factor to the run.
- Swim Impact: While the swim contributes little to total time variance, it remains a significant secondary predictor for podiums (XGBoost importance ~25–27%) because it dictates draft-group composition for the bike leg.
Heat Stress & Tolerance:
- Population Level: Mean WBGT slopes were not statistically significant for either sex (Men: $p=0.065$ ; Women: $p=0.104$ ), suggesting heterogeneous individual responses rather than a uniform population effect.
- Individual Variability: Significant individual variation in heat tolerance was observed. Approximately 52.1% of qualifying male athletes and 36.9% of female athletes demonstrated "heat-favorable" slopes (performance improved or remained stable as WBGT increased).
- DNF Rates: DNF rates ranged from 11.7% to 18.8% across conditions. While no significant association was found between WBGT and DNF probability for men, a small but significant positive relationship was observed for women ( $p=0.004$ ). Venues like Singapore and Hangzhou showed notably elevated DNF rates relative to their WBGT.
Digital Twin Performance:
- The prototype achieved a Mean Absolute Error (MAE) of 0.490 z-units (men) and 0.468 z-units (women) on temporal hold-out.
- Pearson correlation coefficients were 0.383 (men) and 0.295 (women).
- WBGT was the single most important feature in the prediction model (~23% importance), followed by historical performance metrics.

Significance and Claims
The paper claims to provide the first decade-long analysis integrating split-level data with reconstructed meteorological conditions to characterize podium prediction, heat-tolerance profiling, and outcome projection at scale.

Scientific Contribution: The study challenges the notion of uniform heat effects, demonstrating that elite triathletes exhibit heterogeneous thermal responses. It reinforces the primacy of the run leg for podium success while highlighting the strategic importance of the swim for bike-leg positioning.
Practical Application: The authors present a Digital Twin prototype that couples performance and thermo-physiological models. While the authors modestly acknowledge the model's current moderate predictive accuracy (MAE > 0.45 z-units, corresponding to ~90–120 seconds), they posit that the tool offers a foundation for evidence-based race selection, heat acclimatization prioritization, and tactical planning.
Methodological Innovation: The study demonstrates the efficacy of a generative AI workflow (using tools like Claude Sonnet/Opus) in scaffolding complex data ingestion, feature engineering, and statistical modeling pipelines for sports science research.

The authors conclude that these findings support the development of AI-driven tools to enhance athlete safety and performance planning, particularly as major races increasingly occur in thermally challenging environments.