The ingestible capsule temperature sensors were validated for accuracy against two Hg thermometers (range 34–42°C; VWR, Dietikon, Switzerland). These two thermometers were compared against a third Hg thermometer with a larger temperature range (0–100°C), which was calibrated using ice (0°C) before reading. Each ingestible capsule was fixed in place in a digital water bath (2.6 L; VWR, Dietikon, Switzerland) directly adjacent to the two Hg thermometers, as well as two electronic EBI310 thermometers with TPX220 probes (EBRO, Ingolstadt, Germany). All equipment were placed in the water bath prior to stepwise heating and during the gradual increase in temperature from room temperature to a starting temperature of 35°C. Based on previous papers reporting submersion times in the range of 4–5 min with multiple submersion temperatures (19, 20), a 0.5°C stepwise increase in temperature was performed from 35 to 40°C with a period of stabilization of 8 min after each 1°C increase in temperature. Recordings were taken during the stabilization period, every minute for 8 min. Ingestible capsules were deemed to be stable after 5 min based on a difference between measurements of ≤0.1°C. Once a stabilization period had been identified, capsules were then compared against the Hg and electronic thermometers using the same stepwise increase in water bath temperature from 35 to 40°C; this test for accuracy being performed in the morning and afternoon for 2 or 3 days. Investigations were carried out using only two capsules at a time with one recorder per capsule. The CorTemp^® recording device was placed within 20 cm of the capsule. Signal interference from other equipment was minimized by turning off all computers and other devices in the room where the water bath was located. The stepwise increase in temperature was performed in two sets of capsules: the first set was measured at three time points (0, 17, and 24 h) and the second set measured at six time points (twice per day, morning, and afternoon) for three consecutive days. All investigations were carried out by the same investigator.

Temperature values for both studies were downloaded from the CorTemp^® recording device. For the ex vivo validation study, in addition to the computer download, data were also manually recorded from the device in real-time. Outlier data points were excluded manually if they were greater than 2SD from the mean of the nearest six values (three preceding and three proceeding) not including the outlier value. During the in vivo study, the mean temperature and SD were calculated over the following time intervals for each subject: in bed before and during sleep, soon after arousal, as well as under laboratory conditions, namely during resting EE measurement (30 min), ergometry cycling (30 min), intermittent isometric leg press (90 min), and postprandial period (60 min). Valid Tc data were available for a subset of 21 subjects during EE measurement and during sleep for 24 subjects.

Data on changes in EE or Tc across time in response to exercise or meal were assessed by ANOVA with repeated measures. For the ex vivo validation studies, difference scores and the level of agreement were calculated for each comparison of temperatures: (i) capsule vs. Hg, (ii) electronic vs. Hg, and (iii) capsule vs. electronic. The mean difference in temperature was calculated to determine systematic bias between the devices. This was done for all days of testing for each capsule. Bland–Altman plots were generated to compare the difference score to the mean of the two of the three devices (capsule, Hg, and electronic). Linear regression was used to determine the relationship between the temperature readings of the capsules and the Hg or electronic thermometers. Data were analyzed using Statistix software (version 8; Tallahassee, FL, USA) and are presented as mean ± SD. Statistical significance was set at the level of p < 0.05.

The time for the capsules to pass through the GI tract was found to be 31 h on average (range 13–82 h), with evacuation time of ≤15 h in five subjects (24%; Figure 2). Of these five subjects, three were female; mean age was 24 years and mean BMI 23 kg/m² (three normal weight, one overweight, and one obese). Based on gender, mean ± SD expulsion times of 27 ± 15 and 36 ± 24 h were observed for men and women, respectively; no significant between-gender differences were detected.

A decline in night time Tc was observed for all subjects in the current study; a typical sleep pattern for one subject is shown in Figure 3. Furthermore, the large range in Tc obtained over the awake or sleep periods, and corresponding to temperature oscillations of 0.4°C, may in part be explained by a change in total heat production not necessarily in phase with the change in total heat loss. In further analysis of individual data while lying in bed, Tc was assessed at the following time points: (A) the highest 30-min average value before sleep onset, (B) the lowest 30-min average value during sleep, and (C) the first 30-min average value after arousal. These three values and the differences between them (B − A; B − C; C − A), computed for each subject, are presented in Table 1. The average drop in Tc during sleep (B − A) was −0.6 ± 0.1°C. The average difference between the maximum drop in Tc and postarousal (B − C) was −0.4 ± 0.1°C. Overall, the change in Tc from before sleep onset to postarousal (C − A) was −0.2 ± 0.1°C.

Mean changes in Tc and EE in response to the exercise tests and to ingestion of a mixed meal (600 kcal) are shown in Table 2. Dynamic exercise consisting of low-power cycling resulted in a small increase in Tc but only as from 40 W (+0.1°C) reaching statistical significance at 50 W (+0.2°C, p < 0.05) compared to resting precycle values; the EE at 40 and 50 W being ~3–4 times higher than at rest (i.e., ~3–4 METs). No significant differences were observed for Tc in response to intermittent leg press exercise resulting in EE < 2 METs or to a 600 kcal mixed meal where the peak increases in postprandial EE was <25% relative to premeal (fasting) values. Furthermore, no significant differences in Tc in response to the light exercise or meal were observed due to differences in BMI. An example of Tc profile showing changes in response to dynamic and isometric exercise is presented in Figure 4.

Homeotherms conserve energy by lowering body temperature; an example of this can be seen at night when body temperature declines during sleep, part of which is explained by the progressive disappearance of the residual postprandial thermogenesis of the evening meal which often lasts 5–7 h. As expected, a decrease in Tc was observed among all subjects in the current study, which confirms previous reports of the general population (6, 21, 22). Greater nocturnal declines in Tc (up to 2°C) have been reported in nomadic populations such as the Australian aborigines (23) and nomadic Lapps (24) and have been suggested to represent an energy-saving adaptation. In the current study, the average decline in Tc from sleep onset to the lowest night time temperature was −0.6°C (range −1.49 to −0.1). This signature drop in Tc provides an interesting point of comparison in Tc studies; however, measurement of body movement during sleep should be undertaken to allow researchers to account for individual variations in movement during sleep, which may affect Tc readings.

Internal Tc is measured as the telemetric capsule moves through the GI tract. Therefore, the speed of progression of the capsule through the gut will have a significant influence on the time taken for evacuation of the capsule from the body (i.e., expulsion time) and hence on the duration of Tc recording. The average expulsion time in the current study was 31 h, with a large variation among subjects (range: 13–82 h). In those in whom the capsule was evacuated in less than 15 h (n = 5), there were no apparent differences in gender (three females and two males) or in BMI category (three normal and two overweight/obese). Similar expulsion time is reported in the literature, with a range of between 8 h (25) and 5.6 days (26), and is likely to be influenced by the transit time of food through the gut. Such food transit time has been shown to vary between individuals according to the type and volume of food/fluid ingested. Additional factors reported to affect transit time in the general population include alcohol (26), heavy exercise (27), age (older adults have greater transit time) (28), and gender (females have a greater transit time than males) (29). In the current study, subjects were advised to avoid exercise and alcohol for 24 h prior to the test to reduce the effects on transit time and EE. Dietary fiber, particularly wheat bran, has also been shown to decrease colonic and whole gut transit time (30)—an effect that is likely to be exacerbated if it is not associated with an increase in water drinking. In the study by Song et al. (29), females had a significantly longer transit time than males and the authors suggest that female hormones, and in particular progesterone, may be responsible for this difference. In our study, mean expulsion time of the capsule was greater in women than in men (36 vs. 27 h), despite the fact that among those who passed the capsule in ≤15 h, three out of five were women. Although Lee et al. (25) originally recommended ingesting the capsule 6 h prior to data collection to ensure the capsule has reached the intestine, it has since been shown that the location of the capsule in the gut does not substantially affect temperature readings (31). In the current study, capsules were ingested before the evening meal, which has been shown in previous studies to promote transit time (32) and may be one explanation for the rapid expulsion time of the capsule observed in some individuals in the current study. Expulsion time of the capsule is an important factor to consider when conducting studies of this nature to optimize measurement time and minimize data loss.

We observed large erratic temperature excursions outside of the physiological range for all subjects, which occurred to a greater frequency in subjects with greater abdominal adiposity (waist circumference > 100 cm). This may relate to noise due to the proximity of the recording device (worn on the waist) to the capsule within the intestine. Greater processing was required for data from these subjects, which reduced the number of data points available. Previous studies have excluded outliers greater than 2SD from the mean (of a particular time period) when the outlier was included in the mean value (6). However, in the current study, such erratic temperature oscillations were often >80°C and if included in the mean would have invalidated the results for that time period. Therefore, manually checking the data for outliers, although time consuming, ensured more representative Tc values. Computer filters can also be used to eliminate the outliers automatically.

Following these technical difficulties (number and uncertainty of outliers), a validation study was carried out to investigate the accuracy of the capsules measured in the physiological range of 35–40°C, twice per day for 3 days. We identified a substantial bias exceeding ±0.5°C compared to Hg thermometers in half of the capsules, with the bias in some capsules exceeding ±1°C (Figure 8, time point 0). In previously studies, bias of 0.27°C (19) and ±0.4°C (33) have been found, but these values are averages for all capsules studied. Furthermore, the latter study tested capsules at non-physiological temperatures between 40 and 55°C, which may account for some of this difference. Hunt and Stewart (20) have previously shown that the measurement error increased above 41°C. Additionally, Mittal et al. (33) used a Bowman probe instead of an Hg thermometer for comparison, which may also explain some of these differences.

While considering the source of bias, it is important to note that there is an inherent investigator error in reading the temperature value (parallax error). In the current study, the degree of investigator-associated bias was assessed before comparing the Tc capsules with Hg thermometers. An investigator-associated bias of <0.05°C was identified, and hence of small magnitude. These inherent errors of random nature were taken into account for all subsequent comparisons between capsule and thermometers and constitute a window of uncertainty that should be an important consideration in studies of Tc where minute differences can have a considerable impact on the final result.

In the current study, repeated accuracy tests of each capsule recording were taken in order to determine whether or not the accuracy of the capsules changed over time (up to 3 days). This information is vital given that Tc studies generally measure subjects for 24 h or more. We were interested in exploring whether the initial bias observed between the capsules and thermometer, as reported previously (8, 20, 25), changed over time. In 16 additional capsules, which underwent several comparisons against an Hg thermometer and an electronic thermometer over three consecutive days, the bias observed was not consistent over time. Of greater concern was the observation that a drift in bias occurred over time for some capsules (see Figure 9). For example, an initial bias measured at time zero disappeared by 17 h, but in some instances, a new bias of different magnitude was identified at 17 h, which was not present at the initial measurement. This drift in accuracy over time has implications not only for the in vivo Tc measurement but also for any calibration procedures undertaken to improve accuracy of the capsules prior to ingestion.

Researchers have sought to remove the bias associated with the Tc capsules by employing an additional calibration step prior to ingestion in vivo. However, the literature on capsule calibration prior to Tc measurement is inconsistent. Moreover, the manufacturer of the Tc capsules used in the current study states that additional calibration is not required prior to use. Despite this, a number of validation studies carried out with the same capsule suggest that initial calibration is required as the capsules show a small positive systematic bias compared to Hg thermometers (19, 20). It is recommended that the initial raw values obtained during calibration should be used in regression equations and then corrected for any bias arising between the capsule and Hg thermometer (19, 20). Challis and Kolb (19) found that applying a 5-point correction equation to raw data reduced to 2.2% the number of readings beyond the standard and was superior to using only three data points. More reasonably, the manufacturing company could improve the product in order to eliminate the bias. However, our study highlights a drift in bias over time for some capsules (Figure 9), whereby the initial bias disappeared by 17 h, or new bias occurred over time, which was not present initially. In fact, in our validation studies comparing capsules from three batches of capsules (bought and tested at different times), one in every four capsules showed substantial drift exceeding ±0.5°C in each of the three batches, thereby implying that the initial recommended calibration step might not completely remove bias in the system. Due to inconsistencies or lack of consistent reporting of calibration procedures in the literature, it is difficult to determine the impact of calibration on the results of existing studies. Greater consistency and harmonization with respect to calibration procedures and improved reporting practices are required in future studies.

The current study has a number of caveats. First, physical activity was not measured in subjects who underwent Tc measurement prior to the standardized laboratory measurements. The use of accelerometry in future studies would provide activity measurements during the non-standardized component of the experiment, as well as during sleep, which could be incorporated into the analysis and determine the extent to which day-to-day movement influences Tc. Second, the meal subjects consumed the night before the experiment was not strictly standardized. This may have accounted for the differences in expulsion time between subjects and the evolution of Tc during the night. Future studies should consider providing a standardized meal, taking into consideration nutritional factors likely to affect GI transit time, e.g., meal size, dietary fiber content, etc., and which in turn may influence the capsule’s expulsion time. Third, in the measurement of Tc with the ingestible capsule, it is assumed that (i) the anatomical location of the capsule in the gut has very little influence on Tc (e.g., heat production as a result of bacterial fermentation by colon microbiota) and (ii) the presence or absence of food in the gut does not influence Tc significantly. In order to offset this effect of fluctuating Tc over time (Figure 3), the average Tc over the total gastrointestinal transit time should be used. Note that the study was not designed to detect differences in Tc based on gender or BMI. Future studies with larger, equal-sized groups of lean and obese would be useful in addressing some of the inconsistencies in the existing literature on excess body weight and Tc. Strengths of the current study include the fact that Tc and EE were measured simultaneously under standardized conditions among male and female subjects across a BMI range of 20–31 kg/m². This study demonstrates the feasibility of measuring Tc change under free-living and well-standardized laboratory conditions. Temperature changes associated with sleep and physical activity above 3 METs were particularly well identified; for most people, physical activity below 3 METs represents most of the activities performed during the day. Nonetheless, comparison of the ingestible capsule with concomitant rectal temperature monitoring would be desirable in future validation studies, which aim to investigate the sensitivity of the capsules to altered Tc in response to light physical activity or meals. Furthermore, future studies need to address the issue of whether the use of a fixed room temperature might also influence the Tc measurement, particularly when comparing people of different morphology, body fat distribution, and indeed when comparing men and women who may differ in thermal demand (34). Additionally, useful information has been garnered on transit time, which will be important in future studies to ensure a satisfactory measurement period. In this context, the ingestion of the capsule following the 12 h fast may constitute a more standardized approach.