S H O R T C O M M U N I C A T I O N

The Validity of Temperature-Sensitive Ingestible Capsules for MeasuringCore Body Temperature in Laboratory Protocols

David Darwent,1 Xuan Zhou,1 Cameron van den Heuvel,2 Charli Sargent,1 and Greg D. Roach1

1Centre for Sleep Research, University of South Australia, Adelaide, South Australia, Australia, 2Children’s Research Centre,University of Adelaide, Adelaide, South Australia, Australia

The human core body temperature (CBT) rhythm is tightly coupled to an endogenous circadian pacemaker located inthe suprachiasmatic nucleus of the anterior hypothalamus. The standard method for assessing the status of thispacemaker is by continuous sampling of CBT using rectal thermometry. This research sought to validate the use ofingestible, temperature-sensitive capsules to measure CBT as an alternative to rectal thermometry. Participants were11 young adult males who had volunteered to complete a laboratory protocol that extended across 12 consecutivedays. A total of 87 functional capsules were ingested and eliminated by participants during the laboratoryinternment. Core body temperature samples were collected in 1-min epochs and compared to paired samplescollected concurrently via rectal thermistors. Agreement between samples that were collected using ingestiblesensors and rectal thermistors was assessed using the gold-standard limits of agreement method. Across all validpaired samples collected during the study (n = 120,126), the mean difference was 0.06°C, whereas the 95% CI(confidence interval) for differences was less than ±0.35°C. Despite the overall acceptable limits of agreement,systematic measurement bias was noted across the initial 5 h of sensor-transit periods and attributed totemperature gradations across the alimentary canal. (Author correspondence: david.darwent@unisa.edu.au)

Keywords: Circadian rhythm, Core body temperature, Ingestible sensors, Limits of agreement, Validation

INTRODUCTION

The status of the endogenous circadian pacemakercannot be assayed directly; its properties can only be in-ferred by continuous observation of overt rhythms thatare tightly coupled to it. The circadian rhythms of corebody temperature (CBT), plasma cortisol, and plasmamelatonin are the most reliable markers for thispurpose (Brown & Czeisler, 1992; Czeisler & Klerman,1999). Of these, CBT is the most commonly cited,because measurements are sensitive to circadian vari-ation across the entire cycle and are relatively easy toobtain (Duffy & Dijk, 2002; Minors et al., 1996).

Continuous CBT sampling has conventionally beencarried out using an indwelling rectal thermistor con-nected to a portable data logger with a length of wire(Cain et al., 2010; Ishibashi et al., 2010; James et al.,2007; Kubo et al., 2010). The requirement to insert andpreserve placement of a rectal thermistor is a source ofdiscomfort and/or embarrassment to some participants.Ingestible sensors that sample temperature while transit-ing through the alimentary canal offer a less offensivealternative. The sensors, which transmit measurementsto a remote data logger via radio, are not susceptible to

the most common faults associated with hard-wiredrectal thermometry, i.e., improperly situated rectalprobes, and transmission failures due to disconnectedwires. So long as participants keep the data logger that re-ceives the transmission within range (∼1.5 m), the scopefor unintentional misuse is greatly reduced.

Estimates of CBT sampled using ingestible sensorshave been validated via comparison with concurrentsamples collected using rectal thermistors. The observedlevels of agreement are reported to be acceptable (Duch-arme et al., 2001; Edwards et al., 2002; Gant et al., 2006;McKenzie & Osgood 2004), but not universally so(Kolka et al., 1993; Sparling et al., 1993). The disparitymay be attributable to differences in the metrics andthresholds used to determine “acceptable agreement.”Thus, Byrne and Lim (2006) undertook a meta-analyticreview of data pooled from five studies using the gold-standard “limits of agreement” method (Bland &Altman, 1986, 2010). Agreement was a priori deemed tobe acceptable if the mean difference (bias) betweenmeasures was less than ±0.10°C, and if the 95%confidence interval (CI) for differences was less than±0.40°C. The analysis proved equivocal. The magnitude

Address correspondence to David Darwent, Centre for Sleep Research, University of South Australia, Level 7, Playford Building, City EastCampus, Frome Road, Adelaide SA 5000, Australia. Tel.: +618 8302 6624; Fax: +618 8302 6623; Email: david.darwent@unisa.edu.au

Submitted January 25, 2011, Returned for revision February 20, 2011, Accepted May 21, 2011

Chronobiology International, 28(8): 719–726, (2011)Copyright © Informa Healthcare USA, Inc.ISSN 0742-0528 print/1525-6073 onlineDOI: 10.3109/07420528.2011.597530

Chr

onob

iol I

nt D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y T

echn

isch

e U

nive

rsite

it E

indh

oven

on

11/2

2/14

For

pers

onal

use

onl

y.

of the mean difference exceeded the ±0.10°C threshold,but the 95% CI was within the specified range.

The general purpose of this study was to assesswhether CBT as measured in the alimentary canal(using ingestible sensors) could be used as a substitutefor CBT as measured in the rectum (using rectal thermo-metry) in laboratory-based, circadian rhythm researchprotocols. Validation of ingestible sensors has primarilybeen concerned with applications in exercise physiology(reviewed in Byrne & Lim, 2006). These may not general-ize to circadian rhythm research, because exercise elicitsextremes of body temperature beyond that generated bythe endogenous pacemaker. Only one study has soughtto validate the use of ingestible sensors for monitoringthe circadian CBT rhythm (i.e., Edwards et al., 2002).An acceptable level of agreement was found, but thestudy was conducted in field-based settings, and partici-pants only ingested a single sensor during trials. Studiesthat assess circadian phase typically have protocols thatare substantially longer than the transit period of asingle sensor. The novel aim of the present study, there-fore, was to establish the validity (or otherwise) of inges-tible sensors for continuous CBT sampling via ingestionof sequential sensors across multiple days in a laboratorysetting.

METHODS

EthicsThe research methods conformed to the standards ofgood practice outlined in Portaluppi et al. (2010).

ParticipantsA total of 12 healthy young adult males participated in thestudy. Data from one individual were excluded becausehe reported discomfort with the rectal thermistor and de-clined to comply with the protocol beyond the secondday of the study. The 11 remaining participants had amean (± standard deviation) age of 22.4 (± 2.4) yr and aBMI (body mass index) of 22.1 (± 2.3) kg/m2. Femaleswere excluded from participation, because hormonalchanges associated with the menstrual cycle can inducevariation in measures that were included in the study(e.g., mood, hormones, and interstitial glucose).

Protocol

Experimental ProtocolData for the analysis were collected during a forced-desynchrony protocol conducted in the sleep monitoringfacility at the University of South Australia. The study wasinitially conducted to investigate the independent contri-bution of circadian and sleep homeostatic processes tosleep and performance. A comprehensive descriptionof this protocol and the measures used here have beenreported elsewhere (see Darwent et al., 2010; Sargentet al., 2010). Only relevant details are recounted here.

The protocol extended across 12 consecutive calen-dar days and was administered to four groups ofthree participants on separate occasions. The scheduleincluded two training days (T1, T2), a baseline day(BL), and 7 × 28-h periods (FD1 to FD7). Rest periodswere imposed from 00:00 to 08:00 h on both trainingdays, and from 22:40 to 08:00 h on the baseline day.Thereafter, each 28-h period comprised an 18.7-hwake period followed by a 9.3-h rest period (seeupper and lower horizontal axes in Figure 1 for studytimeline). Participants consumed four meals (breakfast,lunch, dinner, and supper) and had two snack oppor-tunities (one between lunch and dinner, and anotherbetween dinner and supper) during each wakeperiod. Refrigerated milk and fruit juice were availableonly at these times. Fresh water, kept at room tempera-ture, was available to participants throughout theprotocol.

Ambient room temperature was set to 21.0°C (± 1.0°C)at all times. Lighting levels were dimmed during wakeperiods (10–15 lux) and near-extinguished (<0.03 lux)during rest periods. Participants were prohibited fromsleeping or engaging in any vigorous physical activity(e.g., exercise, indoor sports) during scheduled wakeperiods. With the exception of using the bathroom, par-ticipants were required to spend the entirety of all restperiods recumbent in bed, attempting to sleep. Closed-circuit infrared television cameras were used to ensurecompliance with the protocol.

FIGURE 1. Schematic diagram representing the transit period(successive bold and thin lines) of all functional capsules superim-posed on the fitted core body temperature rhythm. Dotted linesindicate the transit period for capsules that were still in transit atthe end of the protocol. Grey-shaded areas show periods of sched-uled time in bed. Data for participant, P10, are not illustratedbecause the participant withdrew part-way through the protocol,and circadian phase and period estimates could not be derived.The fitted core body temperature rhythms illustrated were basedexclusively on temperature data collected using the mini-loggersystem. The algorithms used to de-mask and determine the circa-dian amplitude and phase parameters are reported elsewhere (seeDarwent et al., 2010).

D. Darwent et al.

Chronobiology International

Chr

onob

iol I

nt D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y T

echn

isch

e U

nive

rsite

it E

indh

oven

on

11/2

2/14

For

pers

onal

use

onl

y.

Temperature Monitoring and Cleaning ProtocolCore body temperature was measured using two systemsinparallel: (1) amini-logger (ML) rectal thermistor system(Phillips Respironics, Bend,Oregon, USA); and (2) a Vital-Sense (VS) ingestible-capsule system (Phillips Respiro-nics). The ML system consisted of a data logger (120 ×65 × 22 mm; 125 g) connected to a disposable Steri-Probe 491B rectal thermistor (Cincinatti Sub-Zero Pro-ducts, Cincinnati, Ohio, USA). This brand of thermistorhas accuracy to within ±0.20°C in the 34°C to 41°C rangeand, in conjunction with the data logger, a resolution of0.12°C. The VS system consisted of a Philips Respironicstelemetric data logger (120 × 90 × 25 mm; 200 g) that re-ceives radio signals transmitted by Jonah temperature-sensitive ingestible capsules (23 mm length × 8.7 mmdiameter; 1.6 g) (Phillips Respironics). This brand of in-gestible sensor has a transmission range of 1 m, accuracyto within ±0.10°C in the 32°C to 42°C range, and a resol-ution of 0.01°C. The internal chronometers of the MLand VS systemswere synchronized on a common compu-ter and configured to sample CBT in 1-min epochs.

Data collection began on the first training day immedi-ately prior to the scheduled rest time, but this wasdelayed by 24 h for the first group of participants (seeP01 and P02 in Figure 1). Each participant self-inserteda rectal thermistor 10 cm into their rectum and con-nected the cable to a data logger. Capsules were activatedby research staff and then ingested by participants with aglass of water. The data loggers for each system werestored in pouches fastened around each participant’swaist. The waist pouch was worn continuously, exceptwhile showering. To minimize the potential for extendedperiods of data loss, each system was checked every 2.5 hduring wake periods. An additional check was performedafter each instance of voiding. When contact between acapsule and data logger was lost, a replacement capsulewas activated by research staff and then ingested by theparticipant. Rectal probes were readjusted by the partici-pant whenever researchers (or participants themselves)suspected slippage.

A process was developed to remove temperaturesamples that were obviously erroneous. The ML datawere processed 6 mo in advance of the VS data by thesame individual. Samples were considered for removalif temperature measurements were (1) outside of thenormal biological range1; (2) increasing or decreasingat an improbable rate (i.e., >0.50°C/min, given lowactivity and quiescent laboratory conditions); or (3) evi-dently associated with probe slippage or capsule inges-tion times.

Assessing AgreementAgreement between the VS and ML samples was deter-mined using the gold-standard “limits of agreement”

(LOA) method (Bland & Altman, 1986, 2010). Themethod requires two intermediary derivatives to be cal-culated for each paired sample. These are the difference(d) between each paired sample i.e., VS −ML, and themidpoint of each paired sample i.e. (ML + VS)/2. Theagreement between two measures is then estimatedusing the mean difference (d̄) across all paired samples,termed the bias, and the standard deviation (SD) of thedifference values. The SD is used to calculate the LOA—which is equivalent to the 95% CI for differences as-suming a gaussian distribution (i.e., LOA = 95% CI =±1.96SD). Absent any heterogeneity of differencesacross the midpoints, agreement is considered accepta-ble if the bias and LOA are within predefined (seebelow) thresholds.

Assumption of homogeneity in the distribution ofdifferences is evaluated visually using a Scatterplot ofthe midpoint (x-axis) and difference ( y-axis) distri-butions (called a Bland-Altman plot). This is identicalto a Tukey mean-difference plot, but is superimposedwith linear horizontal markers that denote the bias ±LOA. If the distribution of differences is uniform acrossthe range of midpoint values, then bias is homogenous.In this event, any measurement difference between anovel and baselinemeasurement device can be correctedusing simple addition. Otherwise, bias is heterogeneous,and samples from the novel measure must be trans-formed using complex mathematics before parity canbe realized.

Data AnalysesEffect sizes were considered significant at the alpha levelof 0.01. Means are presented as mean (± SD), whereappropriate.

Bias and the Limits of AgreementAgreement between the VS and ML measurementdevices was calculated for (1) individual participants,treating all valid paired samples as a continuoustime series; (2) discrete capsules, irrespective of par-ticipant ID; and (3) all paired samples collectively, ir-respective of participant or capsule ID. Agreement wasdeemed to be acceptable if the bias was less than<0.10°C (i.e., if<± 0.10°C), and if the LOA was lessthan 0.40°C (i.e., if LOA <0.40°C). Agreement wasdeemed to be acceptable if the bias was less than0.10°C, and if the LOA was less than 0.40°C. Inrespect to cross-study compatibility, the thresholdsare consistent with those stipulated by Byrne andLim (2006) in their meta-analytic review of analogousstudies. In respect to technical considerations, the in-strument accuracy specifications for the ingestiblecapsule and rectal thermistor devices (±0.20°C and±0.10°C, respectively) together yield a limit for

1The range (mean ± 2SD) of normal human body temperature measured via rectal thermistor is reported to be 34.4°C to 37.8°C, based onmeta-analytic review of 20 studies that included 347 participants (Sund-Levander et al., 2002).

Sampling Core Temperature Using Ingestible Sensors

© Informa Healthcare USA, Inc.

Chr

onob

iol I

nt D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y T

echn

isch

e U

nive

rsite

it E

indh

oven

on

11/2

2/14

For

pers

onal

use

onl

y.

discriminating a real difference of ±0.30°C (i.e., 0.20°C+ 0.10°C = 0.30°C). Thus, differences <± 0.30°C arepotentially attributable to specified instrumentationerror, whereas differences >± 0.30°C must, to someextent, reflect actual temperature differences betweenrectal and alimentary tissues/cavities. The 0.40°Cthreshold thus represents the real difference that devi-ates from rectal temperature by no more than onemeasurement step, accurate to 0.1°C (i.e., accuracylimit of the ingestible sensors).

Paired sample t tests were calculated to determine thesignificance of any observedmeasurement bias. Product-moment correlations between the difference and mid-point distributions were calculated to determine if anyobserved heterogeneity of differences was significantacross temperature midpoints.

Systemic Bias Owing to Elapsed Transit TimeCapsules must transit through the alimentary canal inorder of the mouth and throat, esophagus, stomach,small intestine, large intestine (cecum, colon, rectum),and, finally, the anus. Differences in actual body tempera-ture between sites along the alimentary canal and rectumwould yield systematic bias across the transit period. Thebias would persist for the period taken for capsules totransit through to the rectum, since capsules wouldtherein be situated near the rectal thermistor. The effectof elapsed capsule transit time on measurement biaswas evaluated by collapsing transit-time data intoseven categories (0–0.2 h, 0.2–1 h, 1–4 h, 4–12 h, 12–24 h,24–48 h, and 48–72 h). An exponential timescalewas chosen, because the time span of greatest interest(i.e., the earlier time spans) comprised only a relativelysmall proportion of the expected range of capsuletransit times. The mean temperature differences (d̄)between the time-interval categories were tested for sig-nificance using mixed-model analysis of variance(ANOVA), wherein participant ID was entered into themodel as a random effect.

RESULTS

Sample Size and Missing DataThe transit period of all capsules ingested by participantsacross the 12-day protocol are superimposed on fittedCBT rhythms in Figure 1. Participants ingested a collec-tive total of 100 VS capsules during the data collectionperiod. Two of these capsules failed to transmit dataand were excluded. The final capsule ingested by eachparticipant was also excluded because elimination oc-curred after the protocol had ended. Thus, onlysamples collected between the ingestion time of theinitial capsule and the elimination time of the penulti-mate capsule were considered for analysis. The time in-terval between these instances had a mean duration of220.55 (± 17.38) h across participants.

Analyses were based on the 87 functional capsules thatwere ingested and eliminated by participants during theprotocol. The mean transit period was 27.16 (± 13.69) h,but substantial inter- and intraindividual differenceswere evident in the number of capsules ingested by indi-vidual participants (range = 4–13) and the mean transitperiod of these capsules (range = 16.6–51.7 h) (seeTable 1 for summary). The lapses in data collection thatoccurred between the elimination of one functionalcapsule and the ingestion of another had a mean dur-ation of 42.79 (± 44.19) min. Apart from the time requiredto identify nonfunctional capsules, long lapse times (>2h) occurred occasionally (n = 7), because instances ofelimination were either not disclosed or otherwise en-croached on prioritized test-battery sessions. Outside ofthese lapses, missing data accounted for 8.37% of all VSsamples and 8.87% of all corresponding ML samples.

Level of Agreement

Individual ParticipantsBland-Altman plots showing the distribution of paired VSand ML samples for two participants (P10 and P05) areillustrated in Figure 2. Data for all participants are sum-marized in Table 2. With just a single exception (P07),

TABLE 1. The number and mean transit period and the sum total of missing data and concurrently valid epochs for the capsules ingestedand excreted by each participant during the laboratory internment

ID N capsules Mean transit period (h) Sum transit periods (h) Missing VS (%) Missing ML (%)

P02 9 21.4 (±7.6) 192.3 23% 11%P03 4 51.7 (±25.8) 206.9 15% 8%P04 7 32.8 (±9.7) 229.4 7% 4%P05 8 29.3 (±4.7) 234.2 3% 4%P06 13 16.6 (±7.5) 216.8 6% 9%P07 13 17.0 (±6.3) 221.6 10% 13%P08 5 46.4 (±10.7) 231.8 7% 19%P09 8 27.2 (±7.6) 217.6 6% 11%P10a 7 24.9 (±10) 174.1 6% 3%P11 6 34.3 (±16.6) 205.8 7% 10%P12 7 33.3 (±11.3) 233.0 3% 5%

aParticipant withdrew from the study prior to completing the study protocol.

D. Darwent et al.

Chronobiology International

Chr

onob

iol I

nt D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y T

echn

isch

e U

nive

rsite

it E

indh

oven

on

11/2

2/14

For

pers

onal

use

onl

y.

the VS system yielded mean CBT estimates that werewarmer than those yielded by the ML system. The extentof measurement bias was marginal, grand d̄=0.06°C, andranged from just −0.01°C to 0.16°C. The mean differencesbetween paired samples were significant for all partici-pants, but were in excess of the ±0.10°C bias threshold inonly two instances (P03, P05). The LOA were <0.40°Cthreshold for all but one participant (P08); mean LOA =0.33°C. p values indicated significant heterogeneityof differences across temperature midpoints for almostall participants, but the slope of regression lines (i.e., rvalues) was bidirectional, generally marginal, andtended toward zero, with mean r = −0.04.

Discrete CapsulesAgreement between the VS and ML samples is summar-ized for all discrete capsules in Figure 3. The black dots(n = 87) indicate intersection of the bias (d̄) (y-axis) andmidpoint of the mean (μ) VS and ML values (x-axis) ob-tained for each capsule.

Significant measurement bias was observed for 76 dis-crete capsules (87.36%), but the magnitude was small in

most cases (grand d̄ = 0.05°C, range = −0.10–0.26°C). Bias>± 0.10°C threshold occurred for 27 capsules (31.03%).The LOA (not shown) were also marginal across capsules(mean LOA = 0.31°C, range = 0.17–0.64°C). Only 15 cap-sules had LOA >0.40°C threshold. Significant heterogen-eity of differences was found for 76 capsules (87.36%)(range r = −0.78–0.75), but the central tendency for thesample was close to zero (mean r = −0.05). In aggregate,51 of the 87 capsules satisfied both the bias and LOA cri-teria. Of these, only 10 had difference distributions thatwere homogeneous across midpoint temperatures.

All Samples InclusiveThe superimposed linear markers shown in Figure 3represent bias ± 95% CI for all paired samples (n =120,126), irrespective of participant ID. The meanglobal difference (=0.06°C) and LOA (= 0.35°C) satisfiedthe thresholds for acceptability. The product-momentcorrelation indicated minor, but significant, heterogen-eity of differences across temperature midpoints (r =−0.10, p < .001).

FIGURE 2. Bland-Altman plots illustrating the level of agreement and homogeneity of differences for valid paired samples collected for twoparticipants (P10, P05). Plots include the total number of valid paired samples (n) and depict the bias (dotted horizontal line) and LOA(continuous horizontal lines) for the distribution of differences.

TABLE 2. Overall agreement for all the ML and VS samples collected for each participant

Paired samplesMeasurement bias, °C Heterogeneity of bias

ID n bias t p LOA r p

P02 8,319 0.02 9.95 <0.001 0.39 0.01 =0.192P03 9,787 0.12 76.24 <0.001 0.30 −0.11 <0.001P04 12,519 0.03 21.14 <0.001 0.27 −0.06 <0.001P05 13,063 0.16 114.37 <0.001 0.32 0.15 <0.001P06 11,281 0.09 47.86 <0.001 0.39 −0.05 <0.001P07 10,414 −0.01 −6.71 <0.001 0.32 0.12 <0.001P08 10,687 0.10 47.31 <0.001 0.42 −0.22 <0.001P09 11,082 0.03 16.16 <0.001 0.34 −0.08 <0.001P10 9,649 0.01 4.35 <0.001 0.25 0.05 <0.001P11 10,333 0.08 41.47 <0.001 0.39 −0.32 <0.001P12 12,992 0.02 16.07 <0.001 0.30 0.10 <0.001

Underlined values signify results that were outside a priori defined thresholds for classifying acceptable bias (bias < ±0.10°C) and LOA (LOA< 0.40°C).

Sampling Core Temperature Using Ingestible Sensors

© Informa Healthcare USA, Inc.

Chr

onob

iol I

nt D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y T

echn

isch

e U

nive

rsite

it E

indh

oven

on

11/2

2/14

For

pers

onal

use

onl

y.

Systemic Bias Owing to Elapsed Transit TimeFigure 4 presents the differences between paired samples(y-axis) as a function of elapsed capsule transit time(x-axis). The left-side panel illustrates the functionalrelationship on an exponential timescale. The tempera-ture bias for VS samples is greatest in the initial 0.2 h(12 min) after capsule ingestion (d̄>− 0.40°C), butshrinks to parity between 1 and 4 h into the transitperiod. A steady-state bias of ≈0.07°C above the MLtemperature is reached after 4 to 12 h has elapsed.Mixed-model ANOVA indicated that bias was signifi-cantly reduced as the transit period lengthened,F(6, 120,119) = 1603.50, p < .001. The critical time pointsfor parity and the steady-state differential are representedin greater resolution on the near-linear timescale (0.2–12h) shown in the right-side panel. Parity between themean VS and ML temperature is evident after ∼3 h, but

the steady plateau is not reached until nearly 5 h follow-ing capsule ingestion.

DISCUSSION

Ingestible sensors that sample temperature during transitthrough the alimentary canal yield CBT estimates with anacceptable level of agreement to paired samples of CBTcollected using rectal thermistors. This is substantiatedon the basis that the aggregate bias and LOA observedbetween paired samples were within a priori determinedthresholds. The assessment is given on the proviso ofexistent methods to minimize systemic bias that occursin the initial sensor-transit period. Absent this provision,bias in this initial period has the potential to systemati-cally modify CBT estimates.

Previous investigations of agreement between CBT inrectal and alimentary tissues have yielded equivocalresults (see Byrne & Lim, 2006). In most of thesestudies, however, variations in CBT were beyond thenormal resting range and induced via exercise (e.g.,Gant et al., 2006; Kolka et al., 1993; Lee et al., 2000) or ar-tificially via peripheral heating/cooling (e.g., O’Brienet al., 1998). The robust levels of agreement observed inour study were found despite the minimal variation inCBT associated with the quiescent conditions. CBTagreement between sites should be harder to detect(not easier) when the range of temperatures sampled islesser. Thus, our findings suggest that intervening vari-ables are more likely to differentially affect CBT in rectaland alimentary tissues under nonquiescent conditions.Unfortunately, the absence of an exercise condition is alimitation of our study protocol, and we were, thus,unable to confirm this suggestion in fact.

Metabolic and/or homeostatic processes generated inresponse to exogenous stressors may explain the incon-sistent levels of agreement seen under quiescent andnonquiescent conditions. Physiological mechanismsunderlying differential heat distribution in alimentaryand rectal tissues are complex and pervasive (Afrin

FIGURE 3. Bland-Altman plots illustrating the level of agreementand homogeneity of differences for the mean VS (μVS) and ML(μML) temperatures of each of the 87 capsules ingested and elimi-nated by participants. The dashed horizontal line represents themean difference for the totality of all valid paired samples (n =120,126) collected across the transit period of these capsules.The continuous horizontal lines represent the LOA for the totalityof all singular differences.

FIGURE 4. Bias as a function of capsule transit time plotted on an exponential (left panel) and a near-linear (right panel) time scale. Errorbars indicate standard deviations.

D. Darwent et al.

Chronobiology International

Chr

onob

iol I

nt D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y T

echn

isch

e U

nive

rsite

it E

indh

oven

on

11/2

2/14

For

pers

onal

use

onl

y.

et al., 2011). Heat generated by elevated metabolicactivity in heart and skeletal muscle during exercisewould be directly conducted to proximal organs beforedistal organs, whereas heat convection via blood (inassociation with increased cardiac output) would befaster for organs with greater vascular perfusion (Afrinet al., 2011). The rectum is anatomically and functionallydistal and, thus, less likely to be affected by commonmediators of metabolic and homeostatic processes(Sessler, 2000). Nonuniform tissue thermoregulationwould also have contributed to the bias observed inrespect to sensor-transit periods, since transit throughthe alimentary canal presents tissues in a definitiveorder. Of particular note, hormonal changes associatedwith the female menstrual cycle may have analogouseffects on body heat distribution in females. Theabsence of female participants in our study thus limitsthe generalizability of the findings.

The general level of agreement we observed was notuniformly applicable to the data sampled by individualparticipants. The sequence of discrete sensors ingestedby individuals yielded a continuous time series ofsamples that manifest as a raw CBT rhythm. The agree-ment between this rhythm and the one manifested bythe concurrent samples of rectal temperature was accep-table only for the majority (not the entirety) of partici-pants. The disparity observed between participants isdirectly attributable to measurement bias that occurredacross the transit period of discrete sensors. Of thesensors ingested by participants, only roughly 60% hadagreement levels below the stated thresholds for accept-ability. That the measurement bias was sometimes nega-tive, i.e., mean CBT estimates for the ingestible sensorswere cooler than for rectal thermistors, but most oftenpositive, i.e., CBT estimates were warmer for the ingesti-ble sensors, raises the possibility that bias had multiplecontrary sources.

The tacit assumption underlying rectal thermometryis that probe slippage can be detected, and the offendingsamples removed, in the absence of independent confir-matory evidence. This assumption may lack verisimili-tude when probe slippage induces incremental, ratherthan sudden, temperature declines and/or if reductionsin temperature coincide with CBT maxima, i.e., wheremeasured samples are least likely to be displaced belowthe normal biological range. Thus, it is speculated thatperiods of undetected rectal probe slippage could haveartificially contributed to the systemic positive measure-ment bias that was observed. Systematic negative bias oc-curred as a function of sensor-transit time. This wasgreatest immediately following sensor ingestion andthen gradually decreased until reaching and then goingbeyond parity to settle at a marginal steady-state biasafter 5 h. Bias associated with capsule transit periodscould be minimized by asking participants to ingest cap-sules at periodic intervals, such that CBT would besampled by multiple capsules across some timeperiods. An ingestion interval that was 5 h less than the

mean transit period (i.e., 27 h − 5 h = 22 h) would elimin-ate measurement bias associated with transit times forroughly 50% of sensors and substantially reduce biasfor other sensors.

The results reported in this study were based on alarge data set collected from multiple participants whoingested temperature sensors across a 12-day period.They reveal that CBT measured in the alimentary canalis an appropriate substitute for CBT measured in therectum in laboratory protocols. This measurement meth-odology is preferential in protocols where participantsare relatively quiescent and in which sensors can be in-gested at regular intervals in order to minimize systema-tic biases associated with transit periods. By implication,this approach may be less well suited to protocols whereparticipants are very active, since the reliability ofmeasurements may be reduced under these conditions.

ACKNOWLEDGMENTS

This study was financially supported by the AustralianResearch Council under the Discovery Project grantscheme.

Declaration of Interest: The authors report no conflictsof interest. The authors alone are responsible for thecontent and writing of the paper.

REFERENCES

Afrin N, Zhang Y, Chen JK. (2011). Thermal lagging in living biologicaltissue based on nonequilibrium heat transfer between tissue, arter-ial and venous bloods. Int. J. Heat Mass Tran. 54:2419–2426.

Bland JM, Altman DG. (1986). Statistical methods for assessing agree-ment between two methods of clinical measurement. Lancet1:307–310.

Bland JM, Altman DG. (2010). Statistical methods for assessing agree-ment between two methods of clinical measurement. Int. J. Nurs.Stud. 47:931–936.

Brown EN, Czeisler CA. (1992). The statistical analysis of circadianphase and amplitude in constant-routine core-temperature data.J. Biol. Rhythms 7:177–202.

Byrne C, Lim CL. (2006). The ingestible telemetric body core tempera-ture sensor: a review of validity and exercise applications.Br. J. Sports. Med. 41:126–133.

Cain SW, Dennison CF, Zeitzer JM, Guzik AM, Khalsa SB, Santhi N,Schoen MW, Czeisler CA, Duffy JF. (2010). Sex differences inphase angle of entrainment and melatonin amplitude inhumans. J. Biol. Rhythms 25:288–296.

Czeisler CA, Klerman EB. (1999). Circadian and sleep-dependent regu-lation of hormone release in humans. Recent Prog. Horm. Res.54:97–132.

Darwent D, Ferguson SA, Sargent C, Paech GM, Williams L, Zhou X,Matthews RW, Dawson D, Kennaway DJ, Roach GD. (2010). Contri-bution of core body temperature, prior wake time, and sleep stagesto cognitive throughput performance during forced desynchrony.Chronobiol. Int. 27:898–910.

Ducharme MB, McLellan TM, Moroz D, Buguet A, Radomski MW.(2001). A 36-hour comparison of core temperature at rest andduring exercise using rectal probe and pill telemetry [abstract].Proc. Aust. Physiol. Pharmacol. Soc. 32:28.

Duffy JF, Dijk DJ. (2002). Getting through to circadian oscillators: whyuse constant routines? J. Biol. Rhythms 17:4–13.

Sampling Core Temperature Using Ingestible Sensors

© Informa Healthcare USA, Inc.

Chr

onob

iol I

nt D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y T

echn

isch

e U

nive

rsite

it E

indh

oven

on

11/2

2/14

For

pers

onal

use

onl

y.

Edwards B,Waterhouse J, Reilly T, Atkinson G. (2002). A comparison ofthe suitabilities of rectal, gut, and insulated axilla temperatures formeasurement of the circadian rhythm of core temperature in fieldstudies. Chronobiol. Int. 19:579–597.

Gant N, Atkinson G, Williams C. (2006). The validity and reliability ofintestinal temperature during intermittent running. Med. Sci.Sports Exerc. 11:1926–1931.

Ishibashi K, Arikura S, Kozaki T, Higuchi S, Yasukouchi A. (2010). Ther-moregulatory effect in humans of suppressed endogenous melato-nin by pre-sleep bright-light exposure in a cold environment.Chronobiol. Int. 27:782–806.

James FO, Boivin DB, Charbonneau S, Belanger V, Cermakian N.(2007). Expression of clock genes in human peripheral bloodmononuclear cells throughout the sleep/wake and circadiancycles. Chronobiol. Int. 24:1009–1034.

Kolka M, Quigley M, Blanchard L, Toyota D, Stephenson L. (1993).Validation of a temperature telemetry system during moderateand strenuous exercise. J. Therm. Biol. 18:203–210.

Kubo T, Takahashi M, Takeyama H, Matsumoto S, Ebara T, Murata K,Tachi N, Itani T. (2010). How do the timing and length of a night-shift nap affect sleep inertia? Chronobiol. Int. 27:1031–1044.

Lee SM, Williams WJ, Fortney Schneider SM. (2000). Core temperaturemeasurement during supine exercise: esophageal, rectal, andintestinal temperatures. Aviat. Space Environ. Med. 71:939–945.

McKenzie JE, Osgood DW. (2004). Validation of a new telemetric coretemperature monitor. J. Therm. Biol. 29:605–611.

Minors D, Folkard S, Waterhouse J. (1996). The shape of endogenouscircadian rhythm of rectal temperature in humans. Chronobiol.Int. 13:261–271.

O’Brien C, Hoyt RW, Buller MJ, Castellani JW, Young AJ. (1998). Tele-

metry pill measurement of core temperature in humans during

active heating and cooling. Med. Sci. Sports Exerc. 30:468–472.

Portaluppi F, Smolensky MH, Touitou Y. (2010). Ethics and methodsfor biological rhythm research on animals and human beings.Chronobiol. Int. 27:1911–1929.

Sargent C, Ferguson SA, Darwent D, Kennaway DJ, Roach GD. (2010).The influence of circadian phase and prior wake on neuromuscularfunction. Chronobiol. Int. 27:911–921.

Sessler D. (2000). Perioperative heat balance. Anesthesiology92:578–596.

Sparling PB, Snow BS. Millard-Stafford ML. (1993). Monitoring coretemperature during exercise: ingestible sensor vs. rectal thermistor.Aviat. Space Environ. Med. 64:760–763.

Sund-Levander M, Forsberg C, Wahren LK. (2002). Normal oral,rectal, tympanic and axillary body temperature in adult menand women: a systematic literature review. Scand. J. CaringSci. 16:122–128.

D. Darwent et al.

Chronobiology International

Chr

onob

iol I

nt D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y T

echn

isch

e U

nive

rsite

it E

indh

oven

on

11/2

2/14

For

pers

onal

use

onl

y.