Recovery and Readiness Monitoring Using Wearable Technology

in Young Triathlon Athletes

Ine De Bot

, Jasper Gielen

and Jean-Marie Aerts

M3-BIORES, Division Animal and Human Health Engineering, Department of Biosystems, KU Leuven,

Kasteelpark Arenberg 30, 3001 Heverlee, Belgium

Keywords: Readiness, Recovery, Triathlon Athletes, Wearables, Physiological Monitoring, Regression Analysis.

Abstract: This study explored the concept of ‘readiness to perform’ by monitoring twelve youth triathletes (under 23

(U23) and 19 (U19) years old) over three months using the Oura Ring. Physiological data from the wearable

were analyzed for all participants; subjective assessments of training intensity (Rating of Perceived Exertion

(RPE)) and recovery (Total Quality Recovery (TQR) questionnaire) were conducted only in the U23 subgroup.

Stepwise linear regression was used to describe five (Balance) Scores contributing to the Readiness Score

(RS). Subsequently, given the limited transparency of Oura’s algorithm, the RS was modeled using three

approaches: through (1) its real contributors (RMSE = 3.18, R² = 0.71), (2) approximated contributors via

regression and three additional contributors (RMSE = 4.09, R² = 0.52), and (3) directly measured variables

with RPE and TQR scores (RMSE = 4.88, R² = 0.29). Individual-level analysis was prioritized, though a

general model for describing the RS was also developed (RMSE = 3.48, R² = 0.60). Sleep emerged as the

primary contributor to readiness, followed by physical activity and resting heart rate.

1 INTRODUCTION

Technology has become integral to modern sports,

driving advancements in real-time monitoring of

athletes’ physiological data and performance

analysis. However, trust in the reliability and validity

of data provided by innovative technology remains a

key concern among coaches and practitioners,

highlighting the need for critical evaluation and

informed use of technological tools (Aerts et al.,

2025).

Training load quantifies the overall demand and

impact of a training session, both physical and

psychological, on an individual's body (Impellizzeri

et al., 2023). Readiness and recovery are

conceptualized in this study as the readiness to train

well and potentially perform well, and the adaptation

to the (previous) training load, respectively. These

processes are influenced and determined by various

factors, with training load and intensity being key

factors, as reflected by the fundamental concept of

training theory. This concept revolves around the

https://orcid.org/ 0009-0009-8357-6550

https://orcid.org/0000-0002-9459-836X

https://orcid.org/0000-0001-5548-9163

structured and systematic planning of exercise

sessions to improve athletic performance.

Furthermore, rest, sleep, nutrition, and various

physiological markers are critical determinants of

readiness and recovery. Adequate rest and sleep is

essential for comprehensive recovery, as also proper

nutrition, balanced in macro- and micronutrients and

hydration, is fundamental for optimal performance

(Walsh et al., 2021; Watson, 2017; Beck et al., 2015).

Equally relevant are physiological markers, such as

heart rate (variability) (HR(V)) and resting heart rate

(RHR), which reflect the state of the autonomic

nervous system (Schneider et al., 2018). Lastly, also

immunological, biochemical and hormonal markers

can be assessed in monitoring an athlete’s recovery

status. However, no single marker serves as a gold

standard in monitoring readiness and recovery.

Effective recovery monitoring of an athlete

requires a multidimensional approach, incorporating

subjective feedback and social factors to account for

non-training related factors, with objective

(physiological) data. Context is crucial, as

208

De Bot, I., Gielen, J. and Aerts, J.-M.

Recovery and Readiness Monitoring Using Wearable Technology in Young Triathlon Athletes.

DOI: 10.5220/0013733100003988

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 13th International Conference on Sport Sciences Research and Technology Support (icSPORTS 2025), pages 208-216

ISBN: 978-989-758-771-9; ISSN: 2184-3201

physiological measures like HRV are sensitive to

non-training related stressors (Flatt et al., 2018).

This study investigates recovery and readiness of

young triathletes, hypothesizing that data collected by

wearable technology (i.e., Oura Ring), combined with

reference training load and recovery measures, will

demonstrate associations with the overall readiness

level of the athletes. Four main objectives are

formulated, focusing on the individual athlete.

The first objective is to map the athletes' current

sleep, training load, recovery and readiness patterns,

and accordingly investigate associations between the

corresponding collected variables. The second

objective involves modeling and approximating five

different (Balance) Scores that contribute to Oura’s

Readiness Score (RS), using (linear) regression

analyses. The third objective is to identify and

characterize the most significant contributors to the

athlete’s readiness, more specifically Oura’s RS, by

(linear) regression analyses. By this interpretable

method, insights into the compilation of the RS and

the importance of the most significant contributors

are identified. The fourth objective parallels the third,

but focuses on approximating the RS by including

physiologically measured variables and subjective

training load and recovery data in (linear) regression

models, rather than Oura's stated contributors.

2 MATERIALS AND METHODS

2.1 Participants and Study Design

Twelve young triathletes participated in this study.

Six U23 triathletes (3 female, 3 male, aged 19-21

years), with a weekly average amount of training

hours between 13 and 19 hours, of which on average

20% of the time is spent on running, 33% on

swimming, 40% on cycling and the remainder on

strength training. In addition, six U19 triathletes (2

female, 4 male, aged 17-18 years) participated in this

study (13 to 17 weekly training hours).

Over the course of three months, the participants

wore a personal Oura Ring (Gen3, firmware 2.9.32-

2.9.33) throughout both day and night, to monitor key

biometrics (i.e., Activity Scores; High/Low Activity;

Steps; Total Burn; Sleep Scores; Deep Sleep, Light

Sleep, REM Sleep, and Total Sleep Time; Respiratory

Rate; Sleep Efficiency; Sleep Latency; Sleep Timing;

RHR; Average HRV; Body Temperature; Recovery

Index; RS). Oura was selected because of its reported

technical quality, wearing comfort, data reporting and

data availability, and its functionalities within

monitoring readiness. This smart ring provides daily

scores (0-100), with the RS being a key metric in this

study. It is a composite measure, derived from, inter

alia, HRV, and recent sleep and activity levels,

providing an indication of the user's readiness to face

more challenges, or the need for rest and recovery.

Furthermore, participants were instructed to

report an individual rating of perceived exertion

(RPE) score per training session, on a 10-point scale,

ranging from ‘rest’ to ‘maximal exertion’. The

session RPE (sRPE) method, as proposed by Foster et

al. (2001), was used to calculate the total load index

of a session. Additionally, both subjective and

objective data regarding recovery were collected via

the Total Quality Recovery (TQR) questionnaire

(Kenttä & Hassmén, 1998). To maximize compliance

of the athletes and to avoid questionnaire fatigue

(Halson, 2014), the TQR questionnaire was preferred

above longer questionnaires, and no strict daily

completion was adopted; instead, participants were

instructed to complete the questionnaire at least twice

a week, enabling retrospective reporting for multiple

past days. The TQR scale consists of two parts: TQR

perceived (TQRper) – perceived recovery on a 6-20

scale, ranging from ‘no recovery at all’ to ‘maximal

recovery’ – and TQR action (TQRact) – a more

objective score based on an athlete's engagement in

recovery actions across four domains: nutrition, sleep

and rest, relaxation and emotional support, and

stretching and cooling-down. Research has confirmed

the effectiveness of TQR in monitoring training load

effects and individual responses, for both daily and

less frequent (e.g., microcycle) implementation

(Nässi et al., 2017; Debien et al., 2020). No explicit

additional interventions were performed that deviated

from their daily routines and training schedules.

RPE and TQR data were not collected for U19

participants due to their late inclusion and the non-

routine recording of subjective data. In total, Oura

data were recorded over an average of 77.83 (± 6.70)

days, with ~8% missing data due to, inter alia, device

non wear. For the U23 population, an average number

of 76.67 (± 19.55) TQRper, 70.00 (± 14.30) TQRact,

and 168.83 (± 50.23) RPE datapoints were obtained.

The study was approved by the Social and Societal

Ethics Committee of KU Leuven (G-2023-7108-R2).

2.2 Data Processing and Analysis

Data processing comprised three main phases (Figure

1), focusing on (linear) regression analysis. While the

true relationships may be non-linear, linear models –

which additionally allow the inclusion of interaction

terms – were chosen for their interpretability and

alignment with the study’s exploratory objectives.

Recovery and Readiness Monitoring Using Wearable Technology in Young Triathlon Athletes

209

Figure 1: Overview of the implemented three-phase data processing flow for analyzing Oura’s Readiness Score.

RPE = Rating of Perceived Exertion; TQR = Total Quality Recovery.

Model performance for fitting and predicting the

investigated scores per individual athlete was

assessed using the (adjusted) coefficient of

determination (R

(a)

²) and root mean squared error

(RMSE). Overall model significance was assessed

through a global F-test, as were individual estimated

regression coefficients evaluated for significance

(with 𝛼 = 0.05). Model validation for each athlete's

scores was performed using 10-fold cross-validation.

Prior to regression analysis, the assumption of

independence among predictor variables was

assessed by investigating for multicollinearity, with a

condition index tolerance level of 30 and a variance-

decomposition proportion tolerance of 0.5

Phase 1 (Figure 1) analyzed five scores provided

by Oura, each contributing to the RS. For these

scores, individual regression models were

constructed, using non-collinear, actual measured

variables from the Oura dataset (i.e., no composite

measures) that best represent the respective score, as

predictor data. Oura's Recovery Index was not

approached as no measured variables allowed

approaching this score. Following data-fitting,

athlete-specific predictor variable sets were defined.

Subsequently, in a cross-validation loop, a regression

model was reconstructed for each individual score

and athlete, based on the aforementioned selection of

predictor variables. Through iterative variable

subsetting, an average performance of R² > 0.75 and

RMSE < 4.00 for each of the five investigated scores

was pursued. The resulting predictions from this

phase served as inputs for Phase 2.

In Phase 2 (Figure 1), Oura’s RS was analyzed in

more detail, adopting a structure analogous to Phase

1. Regression models were developed distinguishing

between two predictor sets: (1) Oura's real

contributors, and (2) approximations of these

contributing scores derived from Phase 1 combined

with three additional factors (i.e., RHR, Temperature

Deviation, and Recovery Index). These eight

variables collectively represent Oura's official RS

contributors. After the final cross-validation step and

obtaining a model with a proper accuracy, a second

model fitting on the corresponding, this time

normalized, predictor data was performed. This

process was performed for the models based on the

real contributors only, where variable importance was

derived from the fraction of its estimated model

coefficient to the total absolute sum of the different

coefficients, excluding the intercept. Moreover, an

attempt was made to obtain a general model that

might approximate the RS for each of the athletes.

In Phase 3 (Figure 1), again an analogous

structure was adopted. Unlike previous phases, this

phase expanded the potential predictor variables

beyond Oura's stated contributors by including all

measured, non-collinear Oura variables, alongside

RPE and TQR scores. This allowed the construction

and evaluation of regression models representing the

RS by using an alternative subset of variables.

icSPORTS 2025 - 13th International Conference on Sport Sciences Research and Technology Support

210

3 RESULTS AND DISCUSSION

3.1 Population Characteristics and

Variable Interrelationships

3.1.1 Sleep Features

Adequate sleep is vital for comprehensive recovery,

encompassing physical, psychological, immune, and

endocrine functions, whereby insufficient sleep can

impair performance and hinder recovery (Walsh et

al., 2021; Watson, 2017). On average, the U23

athletes slept 7.94 (± 0.58) hours, and the U19 athletes

7.86 (± 0.33) hours, including daytime naps. Our

observations align with previous research suggesting

that athletes often report sleep durations below

recommended levels, which may be associated with

poorer sleep efficiency and quality compared to non-

athletes. Contributing factors include demanding

training and competition schedules, travel, and

increased stress levels (Roberts et al., 2019).

The U23 population averaged a Sleep Score of

80.45 (± 5.72), while the U19 athletes averaged 79.79

(± 3.91). The lowest average Sleep Score (i.e., 71.39

± 8.68) was associated with the least sleep hours,

while the highest average Sleep Score (i.e., 86.53 ±

4.01) was associated with the most sleep hours. This

strong association between sleep duration and Sleep

Score was confirmed by significant positive

correlations in all U23 athletes (p < 0.05), yielding

daily and seven-day average 𝑟 values of 0.74 and

0.65, respectively. Furthermore, Sleep Scores were

significantly correlated with RS for two U23 athletes

(p < 0.05), with daily 𝑟 values of 0.70 and 0.75

(seven-day average 𝑟 values of 0.68 and 0.87).

Due to collinearity among several sleep features,

total sleep duration was selected at the beginning of

the regression analyses without prioritizing collinear

variables with this variable.

3.1.2 Cardiac Features

Two overnight RHR variables – lowest and average –

were analyzed. As anticipated, these variables

showed an average daily and seven-day average

correlation of 𝑟 = 0.85, and 𝑟 = 0.90, respectively. The

average RHR was on average 49.96 (± 4.68) bpm for

the U23 athletes, and 51.35 (± 2.80) bpm for the U19

athletes. These lower values reflect exercise-induced

cardiac adaptations that enhance blood-pumping

efficiency, commonly observed in athletes compared

to non-athletes (Nystoriak and Bhatnagar, 2018).

The average HRV, as measured overnight and

provided as the rMSSD, ranged between 57.45 (±

9.51) ms and 170.73 (± 26.95) ms. Due to significant

inter-individual variability in HRV, focusing on

individual trends is a more meaningful approach than

comparison to others. Furthermore, since inconsistent

findings have been observed in scientific literature

when based on daily measurements, it has been

suggested that weekly or seven-day moving averages

offer a more valid approach (Plews et al., 2013).

The average RHR exhibited a significant

correlation (p < 0.05) with the average HRV for two

U23 athletes (i.e., 𝑟 = −0.71 and 𝑟 = −0.94 for daily

measurements, respectively, 𝑟 = −0.89 and 𝑟 = −0.95

for seven-day averages). One U19 athlete showed a

significant seven-day average correlation of 𝑟 = −0.70

between the average RHR and HRV. Hence, average

RHR and HRV appear to be inversely related.

Moreover, with consistently negative (significant)

correlation results, it was observed that the RS and

average RHR are likewise inversely correlated (i.e.,

daily and seven-day average correlation values up to

𝑟 = −0.68 and 𝑟 = −0.87, respectively).

Due to collinearity between average and lowest

RHR, the average RHR was preferred for further

analysis due to its greater robustness and reliability.

3.1.3 Readiness Score

An average RS of 79.31 (± 2.19) was observed. The

RS exhibited no significant high correlations with

other variables, besides these previously mentioned.

Given the extremely low and mostly non-

significant individual correlations, and daily average

values of 𝑟 = 0.01, 𝑟 = 0.07 and 𝑟 = 0.16 between the

RS and sRPE, TQRact and TQRper, respectively, in

the U23 group, the RS did not correlate with these

variables. This suggests that the RS was not reflected

in the athletes’ training load or recovery ratings. Since

training schedules were pre-determined, without

considering Oura data on the actual training day, the

lack of correlation likely stems from the absence of

RS data integration into the decision-making process.

Notably, it may be assumed that a recovery score is

linked to a readiness score for the athlete. However,

the results indicated no significant link between the

RS and TQR scores. Due to the lack of data, no

conclusions could be drawn for the U19 group.

3.1.4 Recovery Measures

The U23 athletes had an average TQRper score of

14.14 (± 1.42), indicating a ‘reasonable’ to ‘good’

recovery, and an TQRact score of 14.62 (± 1.42).

The two-part TQR questionnaire provides

comprehensive insight into training responses, as a

discrepancy between perceived recovery and

Recovery and Readiness Monitoring Using Wearable Technology in Young Triathlon Athletes

211

recovery actions can indicate maladaptation to

training load. Moreover, a practical guideline

suggests aligning TQRact and TQRper scores with

RPE ratings (reported on or converted to a 0-20 scale)

for adequate recovery (Kenttä & Hassmén, 1998).

However, in this study, consistently low and non-

significant correlations were observed between TQR

scores and the (s)RPE scores on the one hand, and the

two TQR scores among themselves on the other hand.

3.2 Investigation of Oura's (Balance)

Scores Contributing to the

Readiness Score

3.2.1 Sleep Score and Sleep Balance Score

Oura’s Sleep Score claims to reflect how well the

athlete slept each night, and is determined by the total

sleep duration, sleep efficiency, restless sleep, REM

sleep duration, deep sleep duration, sleep latency and

sleep timing. For the U23 population, the cross-

validation of the established models yielded an

RMSE between 2.07 and 3.85, with the exception of

one athlete (RMSE = 8.62). Hence, the model made

an overall error of 2 to 4 units out of 100. Equally, a

large fraction of the variance in the Sleep Score was

explained by all models, represented by an R² of 0.69-

0.86, with the exception of one athlete’s model that

performs worse than a simple mean-based prediction.

For the U19 population, an average RMSE of 2.86

and R² of 0.77 were obtained from cross-validation.

Total and deep sleep duration consistently

emerged as the key determinants of the Sleep Score

across all models, whether or not with an additional

interaction effect. Sleep efficiency was also a crucial

predictor, excluded from the model for only one

athlete. The primarily positive regression coefficients

for these three predictor variables indicate their

positive impact on the Sleep Score. Conversely, sleep

latency and timing were the least incorporated into

the models. Sleep latency consistently exhibited

negative regression coefficients, suggesting that

longer nocturnal wakefulness negatively impacts the

Sleep Score. The effect of sleep timing was

inconsistent. These findings underscore the critical

importance of sleep quantity and the deep sleep stage

for Sleep Score determination, aligning with their

recognized roles in physical recovery.

The Sleep Balance Score was approximated by a

combination of total sleep duration with sleep

efficiency. Given it is a Balance Score, time-shifted

data from the last 14 days, were presented as predictor

data. The validated models for the U23 population

exhibited an average RMSE of 3.98 and an R² of 0.75.

The U19 models demonstrated a considerable lower

performance (i.e., average RMSE of 5.30 and R² of

0.56, excluding one athlete). It is reasonable that the

excluded athlete's model exhibited substantial

overfitting, as indicated by an RMSE > 20 and an R²

< 0. The inclusion of 17 predictor variables in this

athlete’s model likely introduced model complexity

beyond the data's explanatory capacity.

The significantly explanatory variables included

in the descriptive models mainly corresponded to the

athlete's sleep duration from one up to ten days prior,

aligning with Oura's assignment of greater weight to

recent sleep patterns in computing the score.

Regression coefficients for recent sleep duration (1-5

days) were consistently positive, indicating their

positive effect. Furthermore, sleep efficiency was

only included in one U23 athlete’s model, indicating

that the sleep efficiency of the last day, as well as nine

to thirteen days prior were determinant. In contrast,

all U19 models incorporated at least one sleep

efficiency variable, ranging from the athlete’s sleep

efficiency from one to thirteen days before the score.

3.2.2 Previous Day Activity Score and

Activity Balance Score

The Previous Day Activity Score quantifies an

athlete’s (in)activity relative to their long-term

average. Initial model fitting identified burned

calories, inactive, low (HR < 60% of maximum), and

high active time (HR 80-100% of maximum) as the

consistently selected explanatory variables for this

score. The validation results exhibited an average

RMSE of 5.59 and 6.40 for the U23 and U19

population, respectively, indicating deviations of 5 to

6 units from true values. However, an overall average

R² of 0.80 was obtained from the validation process.

Almost all estimated regression coefficients were

negative, but a true interpretation is challenging as the

(regression coefficient of the) intercept for the

majority of athletes started with a value above 100.

Oura’s Activity Balance Score was analyzed

using burned calories, steps and high activity time,

from the preceding 14 days, shifted in time, as

predictor data. The cross-validation process yielded

an average RMSE of 6.56 (R² = 0.68) for the U23

population and 8.41 (R² = 0.49) for the U19

population. Ultimately, no adequate desired

performance was achieved for this score.

3.2.3 HRV Balance Score

The HRV Balance Score compares the athlete’s

average HRV from the past 14 days relative to their

long-term average, thereby giving greater emphasis

icSPORTS 2025 - 13th International Conference on Sport Sciences Research and Technology Support

212

on recent data. Due to limited long-term data, the

athlete’s average RHR over the past 14 days was also

included as predictor data alongside 14-day HRV

data, significantly improving model representation.

Cross-validation of the descriptive models for the

U23 group yielded RMSE values between 1.51 and

4.15 and R² values between 0.46 and 0.95, with the

largest prediction error coinciding with the greatest

explained variation in the score. For the U19 group,

cross-validation yielded an average RMSE of 2.76

and an R² of 0.84. Model analysis revealed that all but

one athlete's model included at least one RHR

variable. Proportionally, more HRV variables were

included, generally exhibiting positive coefficients,

indicating higher HRV correlated with an increased

HRV Balance Score. In contrast, RHR coefficients

exhibited both positive and negative values,

precluding unambiguous interpretation.

3.3 Investigation of the Readiness

Score

3.3.1 RS Approximated by Its Real

Contributors

A first approximation of the RS by regression

analysis was conducted based on its real reported

contributors (i.e., Sleep Score, Sleep Balance Score,

Previous Day Activity Score, Activity Balance Score,

average RHR, HRV Balance Score, Temperature

Deviation and Recovery Index). Validation of the

fitted regression models yielded an RMSE between

2.05 and 4.67, and an R² between 0.28 and 0.90. The

lowest explained variation by the validated regression

models (R² = 0.28) was obtained for the athlete with

the second highest estimated error (RMSE = 4.38).

Stepwise regression consistently selected the

Sleep Score across all athletes. Four U23 athletes

showed a significant positive regression coefficient

for this variable. Among the U19 athletes, five times

a positive regression coefficient, and one non-

significant negative one was observed. Hence, an

overall positive association with the RS was found.

The second contributor assessed was the Sleep

Balance Score. Likewise, this variable exhibited a

positively estimated regression coefficient for the

same four U23 athletes. For the U19 population, this

score was three times included with a positively

estimated coefficient, once with a negative

coefficient. Likewise, an overall positive association

with the RS was assumed. Using subsequently

normalized predictor data, it was shown that the Sleep

Score determined 15-35% of the RS among the U23

population, and 20-30% for the U19 population, with

one exception (<15%). For both populations, the

Sleep Score was one of the strongest contributors to

the RS. For the Sleep Balance Score, a contribution

to the RS of 5-15% for the U23, and 5-20% for the

U19 athletes, respectively, was found.

A third variable examined, and the second

variable included in the descriptive model for all

athletes, was the Previous Day Activity Score,

consistently showing a positive regression

coefficient, except for one athlete, indicating a

positive association with the RS. This may reflect a

confounding effect, where increased activity

correlates with higher readiness. However, Oura

posits that maintaining 5-8 hours of inactivity daily

would have a positive impact on the athlete’s activity

score, and subsequently RS, while both excessive

inactivity and overexertion will reduce the score. The

recurrent positive regression coefficient for the

Previous Day Activity Score in the descriptive

models of the RS is thus a rational finding. The

Activity Balance Score was included in ten athletes’

models, with nine showing a positive and one a non-

significant negative regression coefficient, likewise

suggesting a positive association with the RS. Across

the entire population, these two activity scores

contributed 7-15% (Previous Day Activity Score) and

7-25% (Activity Balance Score) to the RS models.

The athlete’s average RHR is the third variable

included in all regression models, with a consistently

negatively estimated regression coefficient, except

one positive but non-significant one, indicating an

inverse relation with the RS. An overall contribution

to the RS of 8-35% was observed. Furthermore, the

regression coefficient for the HRV Balance Score

showed mixed results. With five positive and three

times a negative regression coefficient, an overall

positive effect predominated. A relative importance

in describing the RS of 5-10%, up to 20%, for the U23

and U19 population, respectively, was found.

Temperature Deviation was the least frequently

included predictor variable, and by this considered as

the least contributing variable in describing the RS.

The variable appeared in only two models, with

positive coefficients, contributing from virtually 0 to

15%. In contrast, the Recovery Index was included in

all U23 models and three U19 models, consistently

showing a positive coefficient, indicating a positive

association with the RS. The relative importance in

determining the RS was 5-15%. Finally, for all

athletes, generally none to three interaction terms

were included in their model, but these terms’

importance did not prevail over the main effects.

In addition to the individual approach, a general

model was constructed to investigate whether also

Recovery and Readiness Monitoring Using Wearable Technology in Young Triathlon Athletes

213

this model could adequately describe the RS of all

athletes. For this general model, the following criteria

were applied: once a variable's main effect was

included for at least four out of the six U23 athletes,

this variable was included in the general model. For

interaction terms, a required occurrence for at least

two out of the six athletes was applied. Accordingly,

it was found that the Temperature Deviation variable

was the only RS contributor variable that did not meet

the selection criteria. However, this variable was still

included in the general model as an interaction term

associated with this variable was selected.

This led to the general regression model ([Eq. (1)]):

Readiness Score = Intercept + A*Sleep Score

+ B*Sleep Balance Score + C*Previous Day

Activity Score + D*Activity Balance Score

+ E*Average RHR + F*HRV Balance Score

+ G*Temperature Deviation + H*Recovery Index

+ I* Sleep Score*Sleep Balance Score

+ J*Sleep Score*Average RHR

+ K*Sleep Score*Temperature Deviation

(1)

With the values of the RS contributors and the

coefficients A to K being individual-specific.

The U23 models were used for variable selection,

whereupon the general model ([Eq. (1)]) was tested

on both the U23 and U19 group for its descriptive

capability. Because of the explicit requirement to

include the aforementioned selected variables in the

general model, not all variables had a significant

regression coefficient for each athlete. Positive

associations with the RS were found for sleep and

activity scores, as well as for the HRV Balance Score

and Recovery Index, while average RHR showed

consistently negative effects. Temperature Deviation

and interaction terms showed no clear pattern.

Model validation yielded RMSE values between

1.40 and 3.98, with the exception of one athlete

(RMSE = 7.10). Overall, the explained variation in

the RS by the general model (R²) ranged between 0.58

and 0.95, excluding one athlete. This athlete was the

only athlete for whom the general model exhibited a

poor performance in the validation process. These

results suggest that the proposed general model has

strong potential for describing and predicting the RS

but individual adjustments should not be excluded.

3.3.2 RS Approximated by Approximations

of Its Real Contributors

This regression analysis aimed to examine whether

models based on predicted (Balance) Scores could

adequately describe the RS. Consequently, the RS,

along with its contributors, would be uncovered and

identified in its totality, as opposed to the black-box

nature of the score that previously dominated.

It was found that sleep-related variables,

particularly the approximated Sleep Score and Sleep

Balance Score, were strong positive predictors of the

athletes' RS, underscoring the importance of sleep for

recovery and readiness. Likewise, activity-related

variables, especially the Previous Day Activity Score,

showed primarily positive associations with the RS.

Average RHR appeared inversely related to the RS,

while the HRV Balance Score and Recovery Index

had positive effects. Consistent with previous

findings, Temperature Deviation was the least

significant predictor of the RS. Model validation

yielded RMSE values between 2.60 and 5.44, with

badly an average predictive accuracy R² of 0.56,

excluding one athlete (RMSE = 7.74 and R² = 0.18).

In a one-to-one comparison of validation statistics,

models based on the true contributors exhibited

superior performance for all but two athletes.

3.3.3 RS Approximated by an Alternative

Subset of Predictor Variables

To meet the fourth objective, regression models using

multiple non-collinear, directly measured Oura

variables, in combination with sRPE and TQR scores,

were developed. Due to missing data, sRPE and TQR

scores were excluded for the U19 group. The models

exhibited poor performance (average validation R² =

0.29), with sleep-related variables, average RHR and

HRV, and respiratory rate most frequently included.

The models failed to adequately describe and predict

the RS, indicating that the subset of contributor

variables to the RS proposed by Oura is needed to

adequately describe the RS using regression models

which only allow the inclusion of main effects and

first-degree interaction terms.

3.4 Limitations

While devices like the Oura Ring show high validity

for directly measured metrics (e.g., RHR, HRV;

Kinnunen et al., 2020), their proprietary scores lack

transparency and gold-standard validation, without

which it remains unclear how well the RS reflects

actual physiological readiness. Therefore, caution is

advised when interpreting wearable-derived – black-

box – metrics for modeling purpose, as compounded

estimation errors may reduce accuracy – especially

for inferred parameters like readiness.

This study’s effort to deconstruct Oura’s scores

into interpretable models offers valuable insights, yet

icSPORTS 2025 - 13th International Conference on Sport Sciences Research and Technology Support

214

also highlights the need for prioritizing directly

measured, well-contextualized data. However, due to

the assumption of linearity, multicollinearity among

physiological variables, the inclusion of interaction

terms that may increase the risk of overfitting, and the

hierarchical structure and limited size of the dataset,

linear regression may not be the most robust

modeling approach for this context. Therefore, future

research should consider alternative modeling

techniques, and pursue external validation using

larger independent datasets and comparisons with

established physiological reference measurements to

assess the validity and generalizability of both the

model and underlying wearable metrics.

While the sRPE method is widely used and

correlates well with HR zones (up to 𝑟 = 0.84 for

endurance athletes (Borresen & Lambert, 2008)), it

lacks precision in time quantification, as it includes

total session duration regardless of pauses (Halson,

2014). Despite this, the simplicity, reliability, and

demonstrated agreement of the (s)RPE method with

more complex metrics support its continued use. In

addition, the TQR questionnaire lacked specificity for

triathlon, with outdated or not clearly defined items

(e.g., cooling down, stretching), limiting its relevance

and score potential. A sport-specific and updated

version, aligned with modern recovery strategies, is

recommended for future research. Noteworthy is the

unavailability of (s)RPE and TQR data for the U19

subgroup which restricts the generalizability of

findings, which are based on only six (U23) athletes.

Lastly, this study focused on twelve youth pre-

elite triathletes monitored over three months, limiting

generalizability to other populations or long-term

trends. Individualized monitoring prevailed over a

generalized approach due to varied physiological

responses among the athletes.

4 CONCLUSION

Readiness and recovery levels of young triathletes

can (potentially) be monitored using wearable

technology in combination with reference training

load and recovery measures. The primary focus

should be on the individual athletes’ responses, rather

than general trends, and their sleep patterns, both in

the short- and long-term. Beside objectively collected

data, the significance of subjective data should not be

underestimated. A novel contribution is presented, as

no prior published work has approximated the RS by

using simple regression analysis based on Oura’s

stated contributing factors, nor based on other

(physiological) wearable data or subjective measures.

REFERENCES

Aerts, J-M., Moons, K., Masci, F., Meeusen, R., Gielen, J.

(2025). Identifying stakeholder perspectives and

priorities for athlete monitoring technology: A mixed-

methods study based on interviewing Belgian high-

performance sports organisations. International Journal

of Sports Science & Coaching, 20(3), 973–987.

Beck, K., Thomson, J. S., Swift, R. J., von Hurst, P. R.

(2015). Role of nutrition in performance enhancement

and postexercise recovery. Open Access Journal of

Sports Medicine, 259.

Borresen, J., Lambert, M. I. (2008). Quantifying Training

Load: A Comparison of Subjective and Objective

Methods. International Journal of Sports Physiology

and Performance, 3(1), 16–30.

Debien, P. B., Miloski, B., Werneck, F. Z., Timoteo, T. F.,

Ferezin, C., Filho, M. G. B., Gabbett, T. J. (2020).

Training Load and Recovery During a Pre-Olympic

Season in Professional Rhythmic Gymnasts. Journal of

Athletic Training, 55(9), 977–983.

Flatt, A., Esco, M., Nakamura, F. (2018). Association

between Subjective Indicators of Recovery Status and

Heart Rate Variability among Divison-1 Sprint-

Swimmers. Sports, 6(3), 93.

Foster, C., Florhaug, J. A., Franklin, J., Gottschall, L.,

Hrovatin, L. A., Parker, S., Doleshal, P., Dodge, C.

(2001). A new approach to monitoring exercise

training. Journal of Strength and Conditioning

Research, 15(1), 109–115.

Halson, S. (2014). Monitoring Training Load to Understand

Fatigue in Athletes. Sports Medicine, Vol. 44, 139–147.

Impellizzeri, F. M., Shrier, I., McLaren, S. J., Coutts, A. J.,

McCall, A., Slattery, K., Jeffries, A. C., Kalkhoven, J.

T. (2023). Understanding Training Load as Exposure

and Dose. Sports Medicine, 53(9), 1667–1679.

Kenttä, G., Hassmén, P. (1998). Overtraining and Recovery

A Conceptual Model. Sports Medicine, 26(1).

Kinnunen, H., Rantanen, A., Kentt, T., Koskim ki, H.

(2020). Feasible assessment of recovery and

cardiovascular health: Accuracy of nocturnal HR and

HRV assessed via ring PPG in comparison to medical

grade ECG. Physiological Measurement, 41(4).

Nässi, A., Ferrauti, A., Meyer, T., Pfeiffer, M., Kellmann,

M. (2017). Psychological tools used for monitoring

training responses of athletes. Performance

Enhancement and Health, 5(4), 125–133.

Nystoriak, M.A., Bhatnagar, A. (2018). Cardiovascular

Effects and Benefits of Exercise. Front. Cardiovasc.

Med. 5:135.

Plews, D. J., Laursen, P. B., Stanley, J., Kilding, A. E.,

Buchheit, M. (2013). Training Adaptation and Heart

Rate Variability in Elite Endurance Athletes: Opening

the Door to Effective Monitoring. Sports Medicine,

43(9), 773–781.

Roberts, S. S. H., Teo, W. P., Aisbett, B., Warmington, S.

A. (2019). Extended sleep maintains endurance

performance better than normal or restricted sleep.

Medicine and Science in Sports and Exercise, 51(12),

2516–2523.

Recovery and Readiness Monitoring Using Wearable Technology in Young Triathlon Athletes

215

Schneider, C., Hanakam, F., Wiewelhove, T., Döweling,

A., Kellmann, M., Meyer, T., Pfeiffer, M., Ferrauti, A.

(2018). Heart rate monitoring in team sports-A

conceptual framework for contextualizing heart rate

measures for training and recovery prescription.

Frontiers in Physiology, 9:639.

Walsh, N. P., Halson, S., Sargent, C., Roach, G. D.,

Nédélec, M., Gupta, L., Leeder, J., Fullagar, H. H.,

Coutts, A. J., Edwards, B. J., Pullinger, S. A.,

Robertson, C. M., Burniston, J. G., Lastella, M., Le

Meur, Y., Hausswirth, C., Bender, A. M., Grandner, M.

A., Samuels, C. H. (2021). Sleep and the athlete:

Narrative review and 2021 expert consensus

recommendations. British Journal of Sports Medicine,

55(7), 356–368.

Watson, A. M. (2017). Sleep and Athletic Performance.

Sports Medicine, 16(6), 413-418.

icSPORTS 2025 - 13th International Conference on Sport Sciences Research and Technology Support

216