RAISE: Early Detection of RED-S Risk Using Machine Learning and Wearable Data

(click to the right in the frame above for the video)

Hi, I’m Sitka, from Whitehorse, Yukon. I’m a nationally competitive skier and cyclist. I’m also into sport science, and using data to improve performance.

So when my older brother suffered a third arm fracture in three years, I did some research and learned about RED-S, or relative energy deficiency, in athletes whose training exceeds their fueling, causing serious short and long-term health problems.

RED-S affects up to half of endurance athletes but is hard to detect, since testing is expensive and specialized, and only possible after symptoms appear and the damage has set in.

So I created RAISE, an early-warning tool that detects signs of overtraining before symptoms appear.

RAISE analyses the training and health data that athletes already collect on our smartwatches every day, to provide our coaches with a real-time RED-S risk score. Not to diagnose, but to support smarter training and recovery.

I’m now testing RAISE with the Yukon Ski and Cycling Teams, so we can help more athletes train safer, prevent injuries, and protect their health.

Raise performance, reduce risk.

Purpose

I am a nationally competitive cross-country skier and cyclist with an interest in sport science and using data to boost performance.

When my older brother (and teammate) had his third arm fracture in three years, I started researching bone injuries, and I discovered Relative Energy Deficiency in Sport (RED-S). It’s a medical condition where an athlete’s energy intake is insufficient relative to energy expenditure, affecting multiple body systems.

Estimated to affect 22–58% of athletes depending on the sport, RED-S increases injury risk while reducing training consistency. In early stages, reduced energy availability can lead to weight loss, which may temporarily improve performance and mask the underlying physiological damage. Diagnosis typically requires specialized testing after symptoms develop and long-term health problems have already set in.

Meanwhile, athletes collect all sorts of training and health data on wearable devices like smartwatches. My novel innovation integrates those existing technologies to analyze the data for earlier detection of the warning signs for RED-S, supporting a preventative approach to athlete health rather than waiting to diagnose after symptoms appear.

Hypothesis

If health and training data from wearable devices are analyzed over time, then patterns associated with RED-S will be detectable before clinical diagnosis, allowing for earlier intervention and prevention of long-term health consequences.

Research

I read peer-reviewed studies and the International Olympic Committee (IOC) RED-S Consensus Statement and Cat2 Clinical Assessment Tool, to understand how RED-S happens, its impacts on athletes and the diagnosis process. I found that RED-S is common, has serious health impacts, and that current diagnosis methods are not accessible or practical for athletes or coaches to see the warning signs before it’s too late.

Consulting experts

I met with high-performance coaches and a sports-medicine doctor to understand how athletes’ training is currently monitored, and how RAISE could be designed to be convenient and effective as part of a RED-S prevention strategy. I identified existing technologies to integrate with my algorithm, and what data would be most helpful to coaches for adapting training plans in real-time, and to doctors for developing treatment plans.

Coding

I built a machine-learning algorithm based on coding courses I took through the University of Michigan and Stanford University, and with the help of Claude AI. I trained RAISE using health indicators associated with RED-S that athletes collect everyday on our wearable devices, then I developed the interface that would be convenient and informative for coaches.

Partnership

I partnered with TrainingPeaks, a popular online platform where coaches build and share training plans with their athletes, so I could integrate the algorithm with their app and allow coaches to see RAISE’s RED-S risk score and warning flags as part of their existing training planning routines.

Testing

I beta-tested RAISE with athletes from the Yukon Ski and Cycling Teams, in a real-life training environment. Athletes uploaded their wearable data and filled out daily surveys, while coaches had access to the full RAISE platform, graphs and risk scores for each participating athlete. My algorithm achieved high accuracy (approx. 99%) in identifying RED-S risk patterns.

RAISE (Recovery Analytics and Intensity Surveillance Engine) is a machine-learning system that identifies early warning signs of Relative Energy Deficiency in Sport (RED-S) using training and wellness data already collected by athletes. It is a screening tool to support early detection, not a clinical diagnosis.

RAISE combines Long Short-Term Memory (LSTM) neural networks and XGBoost decision trees to analyze both trends over time and complex relationships between variables. It analyzes health indicators associated to RED-S risk: objective metrics such as heart rate variability (HRV), sleep quality, and training load, plus subjective athlete input from a short daily wellness survey (fatigue, mood, stress, sleep quality, muscle soreness and menstruation).

The system is integrated with TrainingPeaks, a widely used coaching and training-plan platform, where athletes upload their smartwatch data and complete their survey, which are then securely transferred to RAISE for analysis.

RAISE generates a real-time RED-S risk score, displayed on a coach-facing dashboard with chronological trend graphs and automated alerts when training plan adaptations or more intensive medical interventions are needed. Risk is categorized using a colour-coded system:

0.0–0.3 (Low, green): Maintain training

0.3–0.5 (Moderate, yellow): Monitor and increase recovery

0.5–0.7 (High, orange): Adjust training and address risk factors

0.7+ (Critical, red): Stop training and refer for medical evaluation

During beta testing, RAISE successfully distinguished between adaptive and maladaptive training responses. Athletes with intense training loads but stable health and recovery metrics showed low risk scores, meaning that they were adapting well and getting stronger. Conversely, athletes with poorer health and recovery metrics showed higher risk scores, meaning they are at higher risk of RED-S and may need intervention from their coach or doctor.

Testing so far demonstrates that RAISE can identify patterns in athlete data and provide early-warning signals before clinical symptoms or injury occur. Athlete feedback showed that the system was easy to use and integrate into daily training routines, while coach feedback showed its value in adapting training plans to athlete needs.

While my initial results are promising, further validation across a larger group of athletes is required.

RAISE addresses a critical gap in athlete health monitoring by allowing early detection of RED-S risk before symptoms appear. Current detection methods rely on specialized clinical assessment after physical damage has already begun, which limits opportunities for prevention.

To my knowledge, this is the first system that provides real-time detection of RED-S risk signs using wearable and training data, which allows individual assessments to determine whether an athlete is having an adaptive or maladaptive training response.

RAISE would allow athletes to trust in their training plans - to optimize their performance in their sport, while reducing the risk of burn-out, injuries and long-term health damage caused by RED-S.

By integrating existing technology and practices - wearable devices and training-plan platforms that athletes and coaches use everyday - RAISE is highly scalable across various sports and levels of competition, including young athletes who have less access to specialized testing.

Overall, my project shows that we can feasibly detect early warning signs of RED-S using existing data, allowing a preventative approach to athlete health instead of reacting when symptoms appear and the damage has set in.

Over the summer, I will continue testing athletes on the Yukon Ski and Cycling Teams, then reach out to ski and cycling teams in different regions, and teams in different sports, through my connections as an athlete. My goal is to make RAISE an easy and valuable part of athletes’ and coaches’ training plan routines.

An exciting future step would be to fully integrate RAISE into the TrainingPeaks platform for more seamless access for coaches .

Long-term, I hope to expand RAISE to national sporting organizations, and to the medical community to contribute to the RED-S referral process.

I am grateful for expert mentors who showed genuine interest, encouraged me and guided me along my research and project development. These include:

Training Peaks API Team

Graham Nishikawa, former Paralympic Nordic Ski Guide, and current Yukon Ski Team coach

Hudson Lucier and Stephanie Bester, Cycling Association Yukon coaches

Yukon Ski Team and Yukon Cycling Team athletes

BC Science Fair Foundation

Dr. Anne-Marie Leblanc, Sport medicine physician

Ingenious+ and the Rideau Hall Foundation

References

Cabre, H., Moore, S., Smith-Ryan, A., & Hackney, A. (2022). Relative energy deficiency in sport (RED-S): scientific, clinical, and practical implications for the female athlete. German Journal of Sports Medicine, 73:7, 225–234. https://doi.org/10.5960/dzsm.2022.546

Dave, S., & Fischer, M. (2022). Relative Energy Deficiency in Sport (RED - S). Current problems in Pediatric and Adolescent Health Care, 52:8, 101242. https://www.sciencedirect.com/science/article/abs/pii/S1538544222001110

Dvořáková, K., Paludo, A. C., Wagner, A. et al (2024). A literature review of biomarkers used for diagnosis of relative energy deficiency in sport. Frontiers in Sports and Active Living, 6, 1375740. https://doi.org/10.3389/fspor.2024.1375740

International Olympic Committee (IOC). (n.d.). IOC REDs CAT2. https://stillmed.olympics.com/media/Documents/Athletes/Medical-Scientific/Consensus-Statements/REDs/IOC-REDs-CAT-V2.pdf

Mountjoy, M., Ackerman, K., Bailey, D., et al (2023). 2023 International Olympic Committee’s (IOC) consensus statement on Relative Energy Deficiency in Sport (REDs). British Journal of Sports Medicine 2023. 57:17. 1073-1098. https://bjsm.bmj.com/content/57/17/1073

Peklaj, E., Reščič, N., Seljak, B. et al (2022). Is RED-S in athletes just another face of malnutrition?, Clinical Nutrition ESPEN, 48. 2405-4577. https://www.sciencedirect.com/science/article/abs/pii/S2405457722000420

Stanislas, C. (n.d.). Relative Energy Deficiency in Sport (RED-S), Physiopedia. https://www.physio-pedia.com/Relative_Energy_Deficiency_in_Sport_%28RED-S%29

All photos, unless otherwise noted, are owned by the author, Sitka Land-Gillis or his parents.