Validity of the CORE Wearable Sensor During Internal Cooling Induced by Hyperhydration with Cold Water at Rest

1. Introduction

Exercise performed in hot and humid conditions, particularly when interspersed with periods of high-intensity efforts, can predispose athletes to marked elevations in core body temperature (T_C), which could potentially compromise endurance performance [1]. Various strategies can be used by athletes to attenuate the increases in T_C and better sustain performance under such scenarios; pre-exercise internal cooling via the ingestion of large volumes of cold fluid (hyperhydration) or ice slurry is one of them [2]. Indeed, hyperhydration and ice slurry ingestion have been shown to lower T_C before exercise [3,4,5,6]. Moreover, the volume of cold substances entering the body creates a heat sink that enables increasing body heat storage capacity. In turn, this can potentially allow greater physical work to be done before reaching a T_C that compromises the maintenance of an exercise intensity judged satisfactory to sustain performance [2].

Monitoring the impact of a pre-exercise cooling strategy on the reduction in T_C is important and may have important implications from a performance perspective [7,8]. First, it can help athletes choose the most effective pre-exercise internal cooling strategy possible. Next, understanding the kinetics of the reduction in T_C associated with a given internal cooling strategy may help optimize its ingestion timing so that exercise begins in a state of maximal cooling. Finally, it may help athletes plan an optimal pacing strategy for the upcoming exercise. Indeed, the magnitude of the reduction in T_C induced by an internal cooling strategy could be used by athletes as an indicator to estimate how long they may be able to sustain an exercise intensity that they would not normally dare to maintain under conditions in which the body had not been cooled before exercise.

Because of either their invasive nature, need for trained personnel, lack of practicality, the discomfort they may cause, or their high cost, traditionally used and accepted methods to monitor changes in T_C during clinical and exercise settings [9,10,11], i.e., rectal, gastrointestinal or oesopheagal sensors, cannot be used routinely and on a large scale in field settings. Nevertheless, wearable sensors are now available to provide estimates of T_C under real world context [12,13], and these devices could be used by athletes to monitor the impact of pre-exercise cooling. Over the last 5 yrs, the CORE (greenteg, Zürich, Switzerland), a wearable sensor designed to estimate T_C, has received a great deal of scientific attention. Indeed, it has been the object of several validity and reliability studies conducted under various settings such as during intensive care, sleeping, free living, resting and exercise conditions [14,15,16,17,18,19,20,21,22,23,24,25,26,27,28].

To the best of our knowledge, no study has yet examined the validity of the CORE sensor under resting conditions involving cold water-induced hyperhydration, which produces a substantial internal heat sink. This seems relevant, as athletes represent part of the target market of the manufacturer of the CORE sensor, thereby rendering them susceptible to use the device under conditions of internal cooling. The CORE derives estimates of T_C based on an algorithm relying on heart rate, the single-heat-flux method and skin temperature [12,25]. Internal cooling may reduce T_C while producing limited changes in skin heat flux, skin temperature and heart rate [5,29]. Consequently, the CORE sensor may have difficulty detecting changes in T_C under these specific conditions.

Interestingly, our laboratory has recently demonstrated that the CORE was insensitive to a 0.26 °C decline in T_C induced by the ingestion of 7.5 mL · kg fat-free mass (FFM)⁻¹ of cold water (4 °C) [19]. However, since the amount of water ingested was small and the decline in T_C remained within the superior range of the normal daily variation in gastrointestinal or rectal temperatures, i.e., ± 0.25 °C [30], and the mean bias between instruments was lower than the ± 0.27 °C value often regarded as worthwhile in the literature [9,10], it would be premature and inadequate to infer from these results regarding the ability of the CORE to detect decreases in T_C induced by potent internal cooling interventions.

Therefore, the purpose of this study was to compare the changes in T_C measured by a gastrointestinal temperature telemetric sensor (G_TS) and the CORE during a hyperhydration period at rest designed to induce a substantial decline in T_C. It was hypothesized that T_C measured by the CORE and the G_TS would differ by more than the established threshold for a worthwhile practical difference between instruments, i.e., ± 0.27 °C.

2. Materials and Methods

Participants

Eleven individuals (2 women) participated in this study. Their mean age, height, body mass, and FFM were respectively 24 ± 4 yrs, 175 ± 9 cm, 67 ± 8 kg and 58 ± 8 kg. Prior to participating in the study volunteers were required to read a consent form and subsequently a verbal and written consent was obtained. The study was approved by the CIUSSS Estrie-CHUS Ethics Committee (2020-3606). The women completed the experiments in the follicular phase of their menstrual cycle.

Procedures

Context of the study

The current findings represent a secondary analysis of results from a study which compared the effect of pre-exercise hyperhydration and euhydration during a subsequent 5 km running time-trial performance [31]. Herein, we report results pertaining only to the hyperhydration phase and describe the technical procedures only relevant to the current study.

Preliminary Visit

Volunteers completed a preliminary visit where height (standard stadiometer), body mass (BX-300+, Atron Systems, West Caldwell, NJ, USA) and FFM (Lunar Prodigy, GE Healthcare, Chicago, IL, USA) were measured. At the end of the visit volunteers were required to return to the laboratory at 7-day intervals and always at the same time of day to complete two experimental trials administered in a randomized, counterbalanced and crossover fashion: one in which they began a 5 km running time-trial in a euhydrated state, and another in which they began the running time-trial in a hyperhydrated state.

Experiments

Prior to each experiment, diet, training, sleeping and hydration were controlled for, as previously explained [31]. Immediately after arriving at the laboratory, volunteers voided their bladder, collected a urine sample, body mass was measured, a heart rate chest strap was attached to the torso and the temperature and humidity level of the laboratory were measured (Extech H30, Flir, Goleta, CA, USA). Then, participants were seated for a 10 min stabilisation period, after which measures of heart rate and T_C derived from the G_TS and CORE were taken. Following these procedures, volunteers began a 120 min seated period during which they consumed, every 20 min over the first 60 min, four boluses of 4 °C water, each provided at a dose of 7.5 mL · kg FFM⁻¹ with 0.35 g · kg FFM⁻¹ of glycerol [32]. Every 20 min thereafter, measures of heart rate, T_C, urine volume and body mass (post-void) were taken.

Measurements

Core body temperature was measured with G_TS (CorTemp, HQ Inc., Palmetto, FL, USA). When compared to rectal temperature, the G_TS has been demonstrated to represent a valid index of T_C [33]. To avoid any interaction with fluid intake, participants ingested the G_TS 10 h prior to their arrival at the laboratory [34,35]. Gant, Atkinson [35] reported no apparent differences between rectal and gastrointestinal temperatures after the repeated ingestion of 4 °C water before and within bouts of exercise when the G_TS were ingested 10 h prior to experiment commencement. Similarly, when the G_TS were ingested 8 h before the onset of experiment, O’Brien, Goosey-Tolfrey [36] demonstrated that the consumption of 6.8 g · kg⁻¹ of ice slurry during recovery following passive heating did not produce any relevant differences between the fall in rectal and gastrointestinal temperatures. Telemetric gastrointestinal temperature signals were recorded with a single CorTemp Data Recorder (HQ Inc, Palmetto, FL, USA). Each G_TS was calibrated before the beginning of the experiment. Calibration was performed at four different temperatures (37, 38, 39, and 40 °C) in a heated bath (Precision 281, Thermo Scientific, MA, USA) using a high precision, partial immersion, non-mercury glass thermometer (Thermo Scientific Ertco, USA). A 4-point regression line was used to predict gastrointestinal temperature [37]. A single CORE wearable sensor, bought in April 2021, was used for the current study. As recommended by the manufacturer, the CORE was 1) clipped to the chest strap, ~ 20 cm below the left armpit, 2) applied to the skin 10 min prior to the first data collection and; 3) operated paired with signals from the heart rate monitor. Heart rate was measured using a Garmin Premium chest electrode (Garmin, Olathe, KS, USA). Urine specific gravity was assessed using a digital refractometer (PAL-10S, Atago, Bellevue, WA, USA). Urine volume was measured with a graduate urinal. Body mass was evaluated with a high precision digital platform scale (BX-300+, Atron Systems, West Caldwell, NJ, USA). Accumulated fluid retention at each measurement time point was computed by subtracting the accumulated urine volume from the accumulated fluid intake.

Statistical Analysis

A linear mixed-effects model with random intercepts for volunteers was used to analyze the effect of time and sensors as well as their interaction on T_C. Time, sensors and their interaction were treated as fixed effects. A linear mixed-effects model with random intercepts for volunteers was used to analyze the effect of time on accumulated fluid retention, urine volume, and heart rate. Time was treated as a fixed effect. For the above analysis, we used a scaled identity covariance structure to model the repeated measurements. Moreover, a variance components matrix was selected for the covariance structure of the random intercepts. The distribution of residuals was verified with a Shapiro-Wilk test as well as visualization of the Q-Q plot. If relevant, post-hoc analyses were performed using the false discovery rate procedure [38]. Data for these analyses are presented as mean ± standard deviation (SD). Statistical significance was considered as p ≤ 0.05. For describing the relative validity of the CORE, a repeated measures correlation [39] as well as an intraclass correlation coefficient (two-way mixed, absolute agreement, single measure) were computed. The correlation coefficient was interpreted as specified by Akoglu [40] and the intraclass correlation coefficient as specified by Koo and Li [41]. For description of absolute validity, the bias in measurement between sensors, the typical within-subjects variation in measurement and Bland and Altman 95% limits of agreement were calculated. The bias between sensors represents the difference between two comparisons. The typical within-subjects variation in measurement was taken as the square root of the residual covariance obtained from the linear mixed-effects model comparing sensors [42]. For the Bland and Altman-related data, we first examined the relationship over time between the differences in T_C observed between sensors and their averages. A marked heteroscedastic pattern was identified, indicating that the bias between sensors was dependent upon the changes in T_C. Therefore, as recommended by Bland and Altman, a regression approach for non-uniform differences was applied to determine the biases and associated standard deviations [43]. Because repeated measurements were performed within individuals over time with both sensors, a linear mixed-effects model with random intercepts for volunteers was used to derive the regression equations and associated standard deviations between the sensor’s differences and their averages, as specified by Myles and Cui [44]. The average T_C between sensors was taken as the covariable. A variance components matrix was selected for the covariance structure of the random intercepts. No additional covariance structure was specified for the repeated measurements, thereby allowing estimation of a single residual variance for the differences around the predicted bias. Two Bland-Altman plots were generated. The first examined the relationship between the average T_C values of the G_TS and the CORE and the differences in T_C between the CORE and the G_TS over the entire 120 min sitting period. The maximal decline in T_C measured by the G_TS occurred at min 60. Hence, agreement between the CORE sensor and the G_TS was also examined for changes between min 60 and baseline. For each time point (i.e., 20, 40 and 60 min), changes in T_C were calculated relative to baseline for each device, and Bland and Altman analyses were performed on the differences between changes. This approach allowed us to determine whether the CORE sensor accurately tracked the magnitude of the reduction in T_C induced by cold-water hyperhydration. A T_C difference of ± 0.27 °C or 95% limits of agreement of ± 0.27 °C between sensors was considered practically meaningful [9,10,45,46]. Analyses were conducted using Microsoft Excel (version 2502, Redmond, WA, USA), R studio (version 2026.01.2, Posit Team) and IBM SPSS Statistics (version 21, New York, NY, USA).

3. Results

Ambient conditions, hydration state at arrival at the laboratory, glycerol ingestion, fluid intake, urine production, accumulated fluid retention, and heart rate

Mean temperature and relative humidity during the sitting period were 21.8 ± 0.3 °C and 58.0 ± 6.7%, respectively. Volunteers were apparently well hydrated at their arrival at the laboratory. Indeed, the mean urine specific gravity value (1.013 ± 0.007) suggested that the kidneys were not in a state of significant water conservation. Volunteers ingested a total of 82.6 ± 14.0 g of glycerol (1.2 ± 0.1 g · kg body mass⁻¹) with 1762 ± 292 mL (26.4 ± 3.0 mL · kg body mass⁻¹) of cold water. There was a significant time effect for urine production, accumulated fluid retention, and heart rate (all p < 0.01). Urine production peaked at min 60 with a volume of 292 ± 116 mL, which was significantly different from times 0, 20, 40, 100 and 120 (all p < 0.01).

Accumulated fluid retention increased continuously until min 80 where it culminated at 1014 ± 541 mL. This volume was significantly greater than those observed at 0, 20, 40, 100 and 120 min (all p ≤ 0.03). After 120 min, accumulated fluid retention reached 645 ± 546 mL, with a total volume of urine produced of 1143 ± 418 mL. Heart rate changes during the sitting period are shown in Figure 1. Heart rate decreased from 74 ± 7 to 62 ± 7 beats · min⁻¹ during the first 40 min of the sitting period after which it remained relatively stable fluctuating between 58 and 64 beats · min⁻¹. Post-hoc comparisons revealed a significant difference between min 0 and all other timepoints (all p < 0.01), and min 100 with timepoints 20 and 120 min (both p < 0.02).

Comparison between thegastrointestinal telemetric sensor and CORE data

Figure 2 shows the changes in T_C monitored with the G_TS and CORE during the 120 min sitting period. Statistically significant effects of time (p < 0.01), sensors (p = 0.046) and interaction (p < 0.01) were observed.

Hyperhydration was successful in achieving internal body cooling. In fact, the G_TS demonstrated a change in T_C of − 0.76 ± 0.31 °C between min 0 and 60 (p < 0.01). On the other hand, the change in T_C from min 0 to 60 with the CORE was only of − 0.09 °C ± 0.20 °C (p = 0.35). The peak decline in T_C captured by the CORE was of − 0.11 ± 0.19 °C, which occurred at min 120. The CORE demonstrated no significant change in estimated T_C throughout the hyperhydration period, compared with baseline (all p ≥ 0.19). On the other hand, with the G_TS, all timepoints differed significantly from the baseline value (all p < 0.01). Mean T_C during the entire 120 min sitting period was 37.04 ± 0.39 °C with the G_TS, and 36.98 ± 0.12 °C with the CORE (p = 0.046). Post-hoc analyses revealed a significant difference between sensors at min 0, 20, 60, 80 and 120 (all p ≤ 0.02).

For all comparisons combined, i.e., 0-120 min, the typical within-subjects variation in measurement was of the order of 0.19 °C, the repeated measures correlation between sensors was not statistically significant (p = 0.07) and poor with a r_rm = 0.22, and the intraclass correlation coefficient was also poor at 0.12 (p = 0.15). Figure 3 shows the individual responses of all 11 participants to the hyperhydration periods, as measured by the G_TS and the CORE. Only two participants, 4 and 6, showed a response with the CORE that appeared consistent with the initial pattern of change in T_C expected following the use of an internal cooling strategy, namely a relatively sharp decline in T_C. In the remaining 9 participants, fluctuations in CORE-derived T_C relative to baseline were minimal throughout the hyperhydration period. Panels 6 and 7 represent the data for women.

Figure 4 illustrates a Bland and Altman plot showing the relationship between the averages T_C measured with the G_TS and CORE and the magnitude of the differences in T_C observed between the CORE and the G_TS.

These results indicate that the bias between the CORE and the G_TS was inversely proportional to the average value of T_C measured by the CORE and the G_TS. More specifically, based on the line of no difference between sensors, the CORE started overestimating T_C below an average T_C of ~36.96 °C, whereas the opposite was observed at T_C above 36.96 °C. Nevertheless, between average T_C values of ~36.78 and 37.14 °C, the CORE showed no measurement difference compared with the G_TS that fell within the range considered unacceptable from a physiological standpoint, i.e., ± 0.27 °C. In the context of Figure 3, 55% of the differences between the G_TS and the CORE were within ± 0.27 °C.

The Bland and Altman plot shown in Figure 5 indicates that the CORE had difficulty capturing the substantial declines in T_C that occurred during the first 60 min of the seated period. Indeed, for average T_C declines of − 0.1, − 0.2, − 0.3, − 0.4, − 0.5 and − 0.6 °C from baseline, the T_C values estimated by the CORE were respectively + 0.29 (95% limits of agreements, − 0.36-0.94 °C), + 0.41 (− 0.24-1.06 °C), + 0.57 (− 0.08-1.22 °C), + 0.65 (0.00-1.30 °C), + 0.77 (0.12-1.42 °C), and + 0.89 °C (0.24-1.54 °C) higher than the G_TS. Finally, during the first 60 min of the sitting period, only 30% of the T_C differences between the Δ G_TS and the Δ CORE fell within limits considered irrelevant from a practical perspective.

4. Discussion

This study aimed to determine whether the CORE wearable sensor could detect the decline in T_C induced by a potent internal cooling strategy, specifically cold water-induced hyperhydration. The main finding was that, despite a marked reduction in G_TS-derived T_C during the first 60 min of the sitting period, the CORE showed difficulties in detecting the declines in T_C. More specifically, while the G_TS detected a substantial decline in T_C of 0.76 °C from min 0 to 60, the corresponding change measured by the CORE was only of − 0.11 °C. These results raise doubts about the ability of the CORE to detect rapid reductions in T_C induced by hyperhydration with cold water at rest.

The present findings extend previous observations from our laboratory, which suggested that the CORE showed limited ability to detect a modest 0.26 °C reduction in T_C induced by the ingestion of a smaller dose of cold water [19], i.e., 7.5 mL · kg FFM⁻¹ at 4 °C. In that earlier context, any conclusion that the CORE was likely unable to detect this decrease in T_C remained largely speculative and could not reasonably be extended to potent internal cooling strategies, given that the magnitude of the observed change was at the lower limit of the normal daily variation in T_C (i.e., ± 0.25 °C, [30]) and lower than the ± 0.27 °C value often regarded as worthwhile in the literature [9,10,30].

In the present study, however, a peak hyperhydration level of 1014 mL of 4 °C water combined with glycerol induced a much larger decline in T_C, i.e., − 0.76 °C, thereby providing a stronger test of the device’s ability to detect internal cooling. Despite this pronounced decline in T_C, the CORE still showed problems to reflect the magnitude and kinetics of the reduction in T_C demonstrated by the G_TS. To provide support to this observation, the mixed-effects linear model revealed a clear statistically significant effect for interaction (i.e., p < 0.01) between sensors despite the small sample size of the current study. The repeated measures correlation assessing the association between the T_C of the two sensors was poor, thereby suggesting that the responses of the sensors did not move strongly together. The intraclass correlation coefficient was also poor, indicating that agreement between sensors was not strong. Finally, among the eleven individual responses reported, only two suggest that the CORE may detect declines in T_C following hyperhydration. Taken together, these findings suggest that the limitations of the CORE are potentially not only restricted to detecting very small changes in T_C induced by internal cooling strategies, but also extend to larger declines that are clearly meaningful from both a physiological and practical perspective. Studies with larger sample sizes will need to be conducted to confirm the current findings.

One important observation from the current study is that the bias between the CORE and the G_TS was not constant across the range of the average T_C measured. This heteroscedastic pattern is particularly problematic because it indicates that the disagreement between sensors is not random, but rather systematically related to the underlying T_C. For instance, close examination of the Bland and Altman illustrated in Figure 4 reveals that the CORE began to overestimate T_C below a mean T_C of ~36.98 °C and to underestimate it above this threshold. Moreover, data deriving from the Bland and Altman depicted in Figure 5 indicate that average T_C declines of between − 0.1 to − 0.6 °C from baseline over the first 60 min of hyperhydration were associated with T_C values estimated by the CORE that were all above changes suggesting a decline in T_C, as well as being above the acceptable agreement between sensors of ± 0.27 °C. These observations are contrary to what should be expected of temperature sensors from an accuracy standpoint following the use of a potent internal cooling strategy, i.e., be able to detect a decrease in T_C. From an applied standpoint, these results suggest that the CORE could provide the false impression that T_C is stable during cold water-induced hyperhydration when it is in fact decreasing meaningfully. This limitation may be especially relevant in the context where athletes or practitioners seek to monitor the effectiveness of a pre-exercise cooling strategy and optimize its timing before competition.

The lack of sensitivity of the CORE to internal body cooling likely relates, at least in part, to the nature of the physiological signals on which its algorithm is based. Indeed, the device estimates T_C from heart rate, the single-heat-flux technique, and skin temperature. The role of heart rate in the algorithm used by the CORE to estimate the changes in T_C during exercise is likely substantial. Indeed, during exercise, increases in metabolic heat production, which dictate the magnitude of T_C elevation [47], are closely associated with rises in heart rate [48]. Accordingly, algorithms have previously been developed to estimate T_C during exercise using changes in heart rate [49,50]. In the present study, however, substantial changes in T_C occurred concomitantly with only relatively trivial changes in heart rate, as shown in Figure 1. Thus, the contribution of heart rate to the prediction of T_C changes could only be limited, as supported by our findings. The effect of cold fluid or ice slurry ingestion on mean skin temperature under resting conditions in thermoneutral conditions appears to be relatively trivial [4,5,51]. Therefore, this observation suggests that this variable could not have played an important role in the estimation of the changes in T_C in the current study, although we recognize that we did not monitor skin temperature during the hyperhydration period. Finally, considering our results, it must be acknowledged that skin heat flux likely played a negligible role in the ability of the CORE to estimate the changes in T_C in the context of the present study. This is not surprising, as internal cooling acts to extract heat internally and, as a result, heat flux from the core to the skin decreases. Accordingly, our results suggest that the CORE may have difficulties dealing with decreases in skin heat flux, at least with those of the magnitude produced by hyperhydration with cold water. Altogether, these results suggest that the current algorithm used by the CORE may not be fully suited to predicting changes in T_C occurring during internal cooling at rest.

Another noteworthy finding of the current study is that only 33% of T_C differences between sensors from baseline values during the first 60 min of cooling were within ± 0.27 °C. This observation reinforces the conclusion that the CORE lacked acceptable agreement with the G_TS, and this precisely during the period when accurate monitoring was most needed. In addition, and importantly, although the difference in average T_C over the 120 min sitting period was trivial between sensors, this lack of difference should not be construed as evidence of validity, as the CORE showed difficulties in adequately reproducing the true temporal changes in T_C measured by the G_TS.

This study has some limitations that must be acknowledged. The sample size was relatively small and included only 11 participants. Therefore, all findings of the current study must be interpreted with this in mind. Nevertheless, the individual results reported in Figure 3 suggest a tendency supporting the possibility that the CORE may have difficulty capturing declines in T_C during a period of internal cooling. Results of the current study apply specifically to an internal cooling strategy using hyperhydration at rest in temperate laboratory conditions. It remains to be determined whether the CORE would behave differently with other internal cooling strategies used under similar or different surrounding temperatures. Finally, despite that the use of a G_TS provided a strong reference against which to assess the validity of the CORE, it is not impossible that different results would have been observed had the reference method been rectal temperature. However, it has been demonstrated that gastrointestinal temperature represents a valid index of T_C compared with rectal temperature, and that both body areas may respond similarly to internal body cooling [35,36,52].

5. Conclusions

In conclusion, the present study suggests that the CORE wearable sensor may have a limited ability to detect the decline in T_C induced by pre-exercise hyperhydration with cold water. Therefore, caution may be warranted when using the CORE to assess the effectiveness of cold water-induced hyperhydration or to inform performance-related decisions following the use of this internal cooling strategy. Future studies with larger sample sizes are needed to confirm our findings. Moreover, it would be interesting to test the capability of the CORE under scenarios of external body cooling.