a novel model to predict cutaneous finger blood flow via finger and rectal temperatures
TRANSCRIPT
A Novel Model to Predict Cutaneous Finger Blood Flowvia Finger and Rectal Temperatures
ANDRES E. CARRILLO,* STEPHEN S. CHEUNG,� AND ANDREAS D. FLOURIS*,�
*FAME Laboratory, Institute of Human Performance and Rehabilitation, Centre for Research and Technology Thessaly, Trikala, Greece;�Department of Kinesiology, Brock University, St. Catharines, Ontario, Canada; �Department of Research and Technology Development, Biomnic
Ltd., Trikala, Greece
Address for correspondence: Andreas D. Flouris, FAME Laboratory, Institute of Human Performance and Rehabilitation, Centre for Research and
Technology Thessaly, Karies, Trikala 42100, Greece. E-mail: [email protected]
Received 8 June 2011; accepted 19 September 2011.
ABSTRACT
Objectives: To generate a model that predicts fingertip blood
flow (BFf) and to cross-validate it in another group of subjects.
Methods: We used fingertip temperature (Tf), forearm
temperature minus Tf (TFor-f), rectal temperature (Tre), and their
changes across time (lagT) to estimate BFf. Ten participants (six
male, four female) were randomly divided into ‘‘model’’ and
‘‘validation’’ groups. We employed a passive hot–cold water
immersion protocol during which each participant’s core
temperature increased and decreased by 0.5�C above ⁄ below
baseline during hot ⁄ cold conditions, respectively. A hierarchical
multiple linear regression analysis was introduced to generate
models using temperature indicators and lagT (independent
variables) obtained from the model group to predict BFf
(dependent variable).
Results: Mean BFf (109.5 ± 158.2 PU) and predicted BFf (P-BFf)
(111.4 ± 136.7 PU) in the model group calculated using the
strongest (R2 = 0.766, p < 0.001) prediction model [P-BFf =
Tf · 19.930 + lag4Tf · 74.766 + lag4Tre · 124.255 – 447.474] were
similar (p = 0.6) and correlated (r = 0.880, p < 0.001). Auto-
regressive integrated moving average time-series analyses
demonstrated a significant association between P-BFf and BFf
(R2 = 0.381; Ljung–Box statistic = 8.097; p < 0.001) in the
validation group.
Conclusions: We provide a model that predicts BFf via two
practical temperature indicators that can be implemented in both
clinical and field settings.
Key words: finger perfusion, skin temperature, core temperature,
forearm temperature, cold-induced vasodilation
Abbreviations used : ARIMA, auto-regressive integrative moving
average; BFf, fingertip blood flow; P-BFf, predicted fingertip blood
flow; PU, perfusion units; R2, coefficient of variation; Tf, fingertip
temperature; TF or -f, forearm temperature minus fingertip
temperature; lagT, temperature lag; Tre, rectal temperature.
Please cite this paper as: Carrillo, Cheung, and Flouris (2011). A Novel Model to Predict Cutaneous Finger Blood Flow Via finger and Rectal Temperatures.
Microcirculation 18(8), 670–676.
INTRODUCTION
Thermoregulatory measurements are widely important in
both clinical and field settings [4,12], and are necessary
components for various research and bioengineering appli-
cations [6,7]. However, the high equipment cost and the
extensive requirements from participants necessary for
accurate measurements introduce complications that may
prevent practical use. In this light, measurements from
practical and inexpensive techniques have been used to
estimate data recorded from direct techniques. For exam-
ple, measurements of skin-surface temperature gradients
(forearm temperature minus fingertip temperature [TFor-f])
have been associated with fingertip blood flow (BFf) in
steady-state conditions [17]. To our knowledge, however,
no studies have attempted to predict BFf during dynamic
fluctuations in thermal balance, which may introduce com-
plications with acquiring accurate blood flow predictions.
It is well known that the measured fingertip temperature
(Tf) is a slow indicator of what occurs deeper in the tissue
[2]. Hence, problems arise when Tf measurements are
solely used to represent BFf changes within the microcircu-
latory vessels of the finger. Interpretation of the results may
be inaccurate particularly when a strong thermal stimulus
is introduced that induces an immediate change in BFf,
which is associated with a delayed Tf response. For
example, cold-induced vasodilation is commonly reported
as a cyclical increase in BFf during exposure to cold,
and has been assessed often by our lab [8] and others
[3,13,16] through Tf data. Given the delayed Tf response,
DOI:10.1111/j.1549-8719.2011.00136.x
Original Article
670 ª 2011 John Wiley & Sons Ltd, Microcirculation, 18, 670–676
measurements of Tf that define the occurrence of cold-
induced vasodilation should be complemented by estimated
measurements of BFf when direct measurements are
unavailable or inappropriate, such as during field testing or
athletic competition. It is important therefore to address
the commonly reported delays in Tf that introduce incon-
sistencies regarding the timing of BFf estimations [14], to
extend the link between Tf and BFf beyond the simple posi-
tive relationship previously reported [17].
Our aims in this study were to generate a model that
provides an accurate prediction of BFf via practical and
inexpensive indicators, and to cross-validate it in an inde-
pendent group of subjects. The use of Tf as an indicator of
BFf was deemed vital a priori. In addition, we examined
the potential predictive efficacy of TFor-f due to its reported
association with BFf [17]. While we previously reported
that rectal temperature (Tre) alone was not associated with
reflex alterations in BFf, when used in combination with
peripheral skin temperatures, we have found that it shows
a strong association with BFf [9]. As such, the predictive
efficacy of Tre was also examined. More importantly, to
correct for the response delay of the three chosen indica-
tors (i.e., Tf, TFor-f, and Tre) to reflect fluctuations in BFf,
their changes across time were also evaluated as potential
valuable indicators. Multiplying the three chosen indicators
by their changes across time (e.g., Tf · [Tf at time t minus
Tf at time t ) 1]) converts their expected small and slow
alterations into the large and rapid fluctuations usually
observed in BFf. We speculated that a combination of the
three temperature indicators and their changes across time
would allow for an accurate estimation of BFf.
MATERIALS AND METHODS
Participants and ProceduresThe experimental protocol conformed to the standards set
by the Declaration of Helsinki, and was approved by the
ethical review board at Dalhousie University. Ten partici-
pants (six males, four females; age: 25.8 ± 5.3 years; height:
173.9 ± 6.8 cm; weight: 76.6 ± 15.0 kg; percent body fat:
15.3 ± 9.9%) volunteered for this study. The sample was
randomly divided into ‘‘model’’ (five males, three females)
and ‘‘validation’’ (one male, one female) groups. One-way
ANOVA showed no differences in age, height, weight, or
percent body fat between the model and validation groups
(p > 0.05). All participants were screened for Raynaud’s
phenomenon and high blood pressure. Female subjects
completed testing during the early follicular phase (days 1–
6) of their menstrual cycle, and were screened for preg-
nancy prior to participation. Following a full explanation
of procedures, written informed consent was obtained from
all participants. During a familiarization session conducted
three days prior to testing, participants were given a
detailed verbal description of the protocol, followed by
extensive familiarization with all data collection procedures
and instruments. Anthropometric measurements were also
obtained at this time.
Participants were instructed to abstain from alcohol and
caffeine for the 12 hours prior to data collection, as well as
physical activity and excessive stressors, such as exposure to
extreme hot or cold temperatures for the 48 hours prior to
participation. Immediately prior to data collection, all par-
ticipants changed into appropriate water immersion attire.
Males wore a regular one-piece swim suit and females wore
either a one (full)- or two-piece swim suit. To normalize
testing conditions at baseline all participants were asked to
sit for 15 minutes in a ventilated and air-conditioned ther-
moneutral environment (ambient temperature: 25�C; rela-
tive humidity: 40%). To record a wide range of BFf, Tf,
TFor-f, and Tre values without the application of local stim-
uli, we employed a thermal manipulation protocol exten-
sively used by our group to assess peripheral adaptations to
abrupt changes in thermal homeostasis [7,9,11]. Partici-
pants entered a water tank maintained at 42�C water tem-
perature and passively rested until their Tre increased by
0.5�C above baseline. Thereafter, they exited the water tank
and entered a different water bath maintained at 12�C
water temperature until their Tre was decreased by 0.5�C
below baseline. Ad libitum water intake was permitted dur-
ing data collection. During baseline, participants relaxed in
a semisupine position, while during water immersion they
remained in the same position and were immersed up to
the upper part of their chest, slightly below the level of the
clavicle. Their arms during baseline and water immersion
were supported at the level of the heart and were not
immersed in the water at any time.
Blood Flow and Temperature MeasurementsFinger blood flow (BFf) was measured using laser-Doppler
velocimetry (PeriFlux System 5000, main control unit;
PF5010 LDPM, function unit; Perimed, Stockholm, Swe-
den) at the pulp of the index finger on the right hand. A
standard calibration device (PF 1000; Perimed) was used to
adjust the laser-Doppler flowmeter readings to coincide
with the readings obtained with Perimed’s Motility Stan-
dard. The probe (PR 407 small straight probe; Perimed)
was held in place with a plastic mini holder (diameter:
5 mm; PH 07-5; Perimed) and double-sided adhesive strips
(PF 105-3; Perimed). The BFf was expressed in PU and was
sampled at 32 Hz. These data were then used to provide
one-minute mean BFf values.
For the measurement of finger (Tf) and forearm temper-
ature, a ceramic chip skin thermistor (MA-100; Thermo-
metrics, Edison, NJ, USA) was securely placed next to the
Doppler probe on the pad of the right index finger and on
the posterior aspect of the forearm midway between the
Cutaneous Finger Blood Flow Prediction
ª 2011 John Wiley & Sons Ltd, Microcirculation, 18, 670–676 671
elbow and the wrist joints, respectively. Both skin thermis-
tors were held in place using surgical tape (3M Transpore
Tape, 3M, London, Canada). Rectal temperature (Tre) was
measured using a self-inserted flexible probe (2 mm in
diameter [Mon-A-Therm Core; Mallinkrodt Medical, St.
Louis, MO, USA]) to a depth of 15 cm beyond the anal
sphincter. Values for Tf, forearm temperature, and Tre were
recorded throughout baseline and during water immersion
at eight-second intervals using a data logger (Smartreader 8
Plus; ACR, Vancouver, BC, Canada) interfaced with a com-
puter for continuous monitoring by the investigators.
These values were used to provide one-minute mean values
for each variable.
Statistical AnalysisThe prediction model was created using data from the
model group. As expected, initial assessment of BFf, Tf,
TFor-f, and Tre data revealed a delay in Tf, TFor-f, and Tre
during the hot and cold sessions compared with fluctua-
tions in BFf. To account for the delay, dummy variables
were calculated representing the difference between the cur-
rent Tf, TFor-f, or Tre and the Tf, TFor-f, or Tre collected at
either one, or two, or three, or four, or five minutes before
(Tf example for one-minute change: Tf at time t minus Tf
at time t ) 1). These data are henceforth referred to as
‘‘lag’’ data (lagT). A hierarchical multiple linear regression
analysis was introduced to generate models using Tf, lagTf,
TFor-f, lagTFor-f, Tre, and lagTre data (independent variables)
obtained from the model group to predict BFf (dependent
variable). Similar to a previous study [9], Tre was not
highly associated with BFf (R2 = 0.062, F1,403 = 31.87,
p < 0.001). Thus, Tf and TFor-f were used separately as
independent variables, and the BFf was used as the depen-
dent variable in all models, while lagTf (i.e., lag1T f, lag2Tf,
lag3Tf, lag4Tf, and lag5Tf), lagTFor-f (i.e., lag1TFor-f, lag2TFor-f,
lag3TFor-f, lag4TFor-f, and lag5TFor-f), and lagTre (i.e., lag1Tre,
lag2Tre, lag3Tre, lag4Tre, and lag5Tre) were inserted in each
model iteratively as independent variables. This procedure
allowed for the optimal lagTf, lagTFor-f, and lagTre determina-
tion through the assessment of change in the coefficient of
determination (R2). Paired sample t-tests and Pearson’s
product moment correlation coefficients were employed to
assess differences and associations between the measured
BFf and the predicted BFf (P-BFf).
Validation of the prediction model was performed using
data from the validation group via ARIMA analysis, given
the time series nature of the data. This is because common
techniques are limited by the lack of independence amongst
sequential data points within a time series. This limitation
does not influence ARIMA that can mathematically
describe the association between two variables across time
and has been successfully employed to identify associations
between time series variables in thermal physiology [8,10].
ARIMA combines three statistical processes, autoregression,
integration ⁄ differencing, and moving averages, to mathe-
matically describe the association in a disturbance in one
time series (in this case, P-BFf [independent variable]) and
the possible associated perturbations in a second time series
(BFf [dependent variable]). Using specific model building
procedures, an appropriate ARIMA (a,d,q) model can be
specified based on the calculated autoregression (a), differ-
encing (d), and moving averages (q) integers. Given the
stringent nature of an ARIMA analysis, data must be col-
lected in time series with adequate resolution to construct
a viable model. Therefore, ARIMA was used to determine
whether changes in BFf across time were associated with
similar fluctuations in P-BFf. The overall integrity of the
ARIMA model was determined using the Ljung–Box test in
which a probability level of p > 0.05 implies that the model
is correctly specified, while a probability level of p < 0.05
suggests that there is structure in the observed series, which
is not accounted for by the model. Paired sample t-tests
and Pearson’s product moment correlation coefficients
were employed to assess differences and associations
between BFf and P-BFf, respectively. All statistical analyses
were completed using PASW Statistics (version 18; SPSS,
Inc., Chicago, IL, USA) statistical software package with
the level of significance set at p < 0.05 except in the AR-
IMA Ljung–Box test.
RESULTS
Preanalysis screening procedures were conducted to deter-
mine whether the model-group data satisfied the assump-
tions of hierarchical multiple linear regression analysis.
Assessment of residuals detected no violation of normality,
linearity, and homoscedasticity between P-BFf and errors of
prediction (p > 0.05). Results of the two hierarchical multi-
ple regression analyses are presented in Tables 1 and 2,
demonstrating that the model providing the strongest pre-
diction incorporated Tf, lag4Tf, and lag5Tre, while the second
strongest prediction came from the model incorporating
Tf, lag5Tf, and lag5Tre. However, cross-validation results for
these models (see below) demonstrated that calculated P-
BFf was statistically different from BFf (p = 0.048 and
0.036, respectively; Table 3). Thus, the prediction model
that most appropriately satisfied the selection criteria was
the model incorporating Tf, lag4Tf, and lag4Tre:
P� BFf ¼ Tf � 19:930þlag4 Tf � 74:766þlag4 Tre
� 124:255� 447:474
The above prediction model can generate negative
values. However, given that negative blood flow is not
plausible, all negative values must be recorded as zero.
A.E. Carrillo et al.
672 ª 2011 John Wiley & Sons Ltd, Microcirculation, 18, 670–676
Paired-sample t test results revealed no mean difference
between BFf and P-BFf (109.5 ± 158.2 vs. 111.4 ± 136.7
PU, respectively; p = 0.6). The correlation coefficient
between BFf and P-BFf is shown in Table 1.
Cross-validation of the five strongest models for the
prediction of P-BFf was conducted using data from the
validation group. Table 3 outlines the ARIMA-specific
Ljung–Box statistic and R2 values associated with each pre-
Table 1. Results derived from the hierarchical regression analysis conducted in the model group for the prediction of cutaneous finger blood
flow via Tf, lagTf, and lagTre
IndVariables R2 SEE t P-BFf (mean ± SD) BFf (mean ± SD) r*
Tf, lag1Tf, lag1Tre 0.697 86.1 )0.599 115.9 ± 127.6 113.6 ± 155.9 0.842
Tf, lag1Tf, lag2Tre 0.707 85.1 )0.570 114.5 ± 129.2 112.3 ± 156.6 0.847
Tf, lag1Tf, lag3Tre 0.722 83.5 )0.586 113.7 ± 131.7 111.4 ± 157.8 0.856
Tf, lag1Tf, lag4Tre 0.744 80.4 )0.627 111.8 ± 134.0 109.5 ± 158.2 0.870
Tf, lag1Tf, lag5Tre 0.754 78.6 )0.598 109.0 ± 135.0 106.7 ± 158.0 0.875
Tf, lag2Tf, lag2Tre 0.725 82.4 )0.515 114.3 ± 131.3 112.3 ± 156.6 0.857
Tf, lag2Tf, lag3Tre 0.740 80.8 )0.504 113.3 ± 133.8 111.4 ± 157.8 0.865
Tf, lag2Tf, lag4Tre 0.760 77.8 )0.539 111.4 ± 136.0 109.5 ± 158.2 0.877
Tf, lag2Tf, lag5Tre 0.769 76.2 )0.499 108.6 ± 136.8 106.7 ± 158.0 0.882
Tf, lag3Tf, lag3Tre 0.746 79.8 )0.484 113.2 ± 134.5 111.4 ± 157.8 0.868
fT, lag3Tf, lag4Tre 0.764 77.1 )0.516 111.3 ± 136.5 109.5 ± 158.2 0.879
Tf, lag3Tf, lag5Tre 0.772 75.7 )0.466 108.4 ± 137.3 106.7 ± 158.0 0.883
Tf, lag4Tf, lag4Tre 0.766 76.7 )0.525 111.4 ± 136.7 109.5 ± 158.2 0.880
Tf, lag4Tf, lag5Tre 0.773 75.6 )0.465 108.4 ± 137.4 106.7 ± 158.0 0.883
Tf, lag5Tf, lag5Tre 0.772 75.8 )0.502 108.6 ± 137.1 106.7 ± 158.0 0.883
Tf = finger skin temperature; Tre = rectal temperature; lagT = temperature lag; R2 = regression analysis coefficient of determination; SEE = stan-
dard error of the estimate; t = paired sample t-test statistic; P-BFf = predicted fingertip blood flow (PU); BFf = measured fingertip blood flow (PU);
r = correlation coefficient.
*Pearson’s correlation coefficient significant at p < 0.001.
Table 2. Results derived from the hierarchical regression analysis conducted in the model group for the prediction of cutaneous finger blood
flow via TFor-f, lagTFor-f, and lagTre
IndVariables R2 SEE t P-BFf (mean ± SD) BFf (mean ± SD) r*
TFor-f, lag1TFor-f, lag1Tre 0.516 98.8 )0.466 104.5 ± 99.1 102.4 ± 141.5 0.726
TFor-f, lag1TFor-f, lag2Tre 0.528 97.6 )0.428 102.8 ± 100.5 100.8 ± 141.5 0.733
TFor-f, lag1TFor-f, lag3Tre 0.552 95.4 )0.448 101.5 ± 103.2 99.5 ± 142.0 0.750
TFor-f, lag1TFor-f, lag4Tre 0.572 93.0 )0.515 99.3 ± 104.7 97.0 ± 141.6 0.764
TFor-f, lag1TFor-f, lag5Tre 0.577 91.7 )0.520 96.2 ± 104.2 93.9 ± 140.4 0.767
TFor-f, lag2TFor-f, lag2Tre 0.535 96.9 )0.426 102.8 ± 101.4 100.8 ± 141.7 0.738
TFor-f, lag2TFor-f, lag3Tre 0.561 94.6 )0.432 101.5 ± 104.4 99.5 ± 142.2 0.755
TFor-f, lag2TFor-f, lag4Tre 0.583 91.8 )0.486 99.2 ± 105.9 97.0 ± 141.6 0.770
TFor-f, lag2TFor-f, lag5Tre 0.587 90.6 )0.487 96.0 ± 105.3 93.9 ± 140.4 0.773
TFor-f, lag3TFor-f, lag3Tre 0.565 94.2 )0.451 101.4 ± 104.8 99.4 ± 142.3 0.758
TFor-f, lag3TFor-f, lag4Tre 0.590 91.1 )0.491 99.1 ± 106.7 96.9 ± 141.8 0.775
TFor-f, lag3TFor-f, lag5Tre 0.594 89.8 )0.472 96.0 ± 106.1 93.9 ± 140.4 0.777
TFor-f, lag4TFor-f, lag4Tre 0.595 90.7 )0.522 99.2 ± 107.2 96.9 ± 142.0 0.778
TFor-f, lag4TFor-f, lag5Tre 0.600 89.2 )0.475 96.0 ± 106.8 93.9 ± 140.6 0.781
TFor-f, lag5TFor-f, lag5Tre 0.602 89.0 )0.513 96.1 ± 106.8 93.9 ± 140.6 0.783
Tf = finger skin temperature; Tre = rectal temperature; lagT = temperature lag; R2 = regression analysis coefficient of determination; SEE = stan-
dard error of the estimate; t = paired sample t-test statistic; P-BFf = predicted fingertip blood flow (PU); BFf = measured fingertip blood flow (PU);
r = correlation coefficient.
*Pearson’s correlation coefficient significant at p < 0.001.
Cutaneous Finger Blood Flow Prediction
ª 2011 John Wiley & Sons Ltd, Microcirculation, 18, 670–676 673
diction model for BFf. All models passed the Ljung–Box
test (p > 0.05) confirming that they were correctly specified
in the ARIMA process. The highest R2 value was observed
in the model incorporating Tf, lag4Tf, and lag4Tre. The
increased efficacy of this model was confirmed also through
the more typical paired t test and Pearson’s product–
moment correlation coefficient. These results suggest that
perturbations in BFf were systematically followed by similar
perturbations in the P-BFf calculated through Tf, lag4Tf, and
lag4Tre. This is illustrated in Figure 1, which presents the
fluctuations in BFf and the P-BFf (calculated through Tf,
lag4Tf, and lag4Tre) during the thermal protocol in the model
and the validation groups. Indeed, it becomes apparent that
the prediction model is very responsive following closely
the large and rapid fluctuations observed in BFf.
DISCUSSION
In this study, we generated a model that provides an accu-
rate prediction of fingertip blood flow via practical and
inexpensive indicators, and also cross-validated it in an
independent group of subjects. Our results show that fin-
gertip blood flow can be accurately predicted through fin-
gertip temperature, the difference between current fingertip
temperature and that recorded four minutes before, as well
as the difference between current rectal temperature and
that recorded four minutes before. The resulting prediction
model provides an effective, easy, and cost-efficient alterna-
tive for the estimation of fingertip blood flow that only
requires a two-site temperature measurement.
Skin temperature gradients have been validated for the
estimation of blood flow and vasomotor tone [17]. Our
model provides a complementary approach to predicting
changes in fingertip blood flow with the consideration of
the expected lag using a two-site temperature measurement
from the finger and the core. During certain clinical situa-
tions such as the preoperative period, the evaluation of fin-
ger blood flow is useful for the determination of
thermoregulatory responses that have been reported to be
disrupted with anesthesia [19]. Furthermore, finger blood
flow evaluations may also provide useful information dur-
ing the assessment of thermoregulatory adaptations in
applied settings during which actual measurements are not
plausible. For example, a recent study evaluated the train-
ability of cold-induced vasodilation responses from highly
trained mountaineers prior to and following a three-week
expedition [5]. Thus, estimates for finger blood flow have a
Table 3. Results and comparisons of the five strongest models predicting cutaneous finger blood flow in the validation group
IndVariables P-BFf (mean ± SD) BFf (mean ± SD) r t R2 Ljung–Box statistic
Tf, lag2Tf, lag5Tre 112.3 ± 165.7 97.4 ± 138.7 0.884* )1.750 0.254 13.931
Tf, lag3Tf, lag5Tre 111.5 ± 162.1 97.4 ± 138.7 0.905* )1.868 0.339 8.143
Tf, lag4Tf, lag4Tre 111.7 ± 154.1 99.8 ± 138.0 0.917* )1.801 0.381 8.097
Tf, lag4Tf, lag5Tre 111.0 ± 157.5 97.4 ± 138.7 0.920* )2.004� 0.308 13.944
Tf, lag5Tf, lag5Tre 111.0 ± 153.2 97.4 ± 138.7 0.925* )2.134� 0.380 7.009
Tf = finger skin temperature; Tre = rectal temperature; lagT = temperature lag; t = paired sample t-test statistic; P-BFf = predicted fingertip blood
flow (PU); BFf = measured fingertip blood flow (PU); r = correlation coefficient; R2 = ARIMA coefficient of determination.
*Pearson’s coefficient significant at p < 0.001.�Paired sample t-test significant at p < 0.05.
Figure 1. Fluctuations in measured blood flow and the blood flow
predicted through Tf, lag4Tf, and lag4Tre during the thermal protocol in
the model and the validation groups. The validation group data have
been expanded to improve clarity.
A.E. Carrillo et al.
674 ª 2011 John Wiley & Sons Ltd, Microcirculation, 18, 670–676
wide range of use and should be applied when actual mea-
surements are difficult to implement.
Vasodilation and vasoconstriction of the small arteries in
peripheral body parts are key functions that regulate ther-
mal stability [8,18]. Cold-induced vasodilation has been
reported to occur, at least in part, from the dynamic struc-
tural characteristics of arteriovenous anastomoses that are
capable of complete closure or large amounts of blood
transfer when the vessels are open [1]. A number of meth-
ods are available to quantify the amount of vasodilation,
but because of the invasiveness and difficulty of direct mea-
surement, the majority of used methods are indirect [2].
The most commonly used method to measure cold-
induced vasodilation is finger temperature that represents a
delayed indicator of what is actually occurring in the dee-
per tissues [14]. The quantification of cold-induced vasodi-
lation determined via finger temperatures is well described
[2]. Nevertheless, the difference in timing between finger
blood blow and skin temperature introduces some ambigu-
ity that may hinder interpretation. Thus, additional estima-
tions of blood flow reported with finger temperature data
may provide further insight into thermoregulation research
such as with cold-induced vasodilation.
Despite a previously reported association between TFor-f
and BFf [17], our results suggest that accounting for the
delay in skin and rectal temperatures by incorporating the
change in Tf and Tre provides a stronger prediction of
BFf. Furthermore, our results confirm a previous study
[9] reporting that Tre alone is not associated with reflex
alterations in BFf, but when used in combination with
peripheral skin temperatures, it shows a strong association
with BFf. More importantly, our cross-validation results
confirm that multiplying the skin surface temperature
indicators by their changes across time converts their
expected small and slow alterations into the large and
rapid fluctuations usually observed in BFf. Therefore, the
predictive efficacy of the change across time of different
skin surface temperatures should be evaluated to correct
for the response delay of these thermal indicators to
reflect fluctuations in BFf.
To record a wide range of BFf, Tf, TFor-f, and Tre values
without the application of local stimuli, data collection was
conducted in a stable thermoneutral ambient air tempera-
ture environment. Therefore, our prediction model should
be validated also in a changing air temperature environ-
ment. Furthermore, hot water immersion increases skin
blood flow faster than dry heat [15]. Local effects of heat
transfer, however, were eliminated during water immersion,
given that the arms were supported at the level of the heart
and were not immersed in the water. Nevertheless, the
generated model should be validated in conditions with
varying rates of heat transfer to ensure the usefulness of
this measurement in clinical and other settings. Finally,
although our study was conducted in a relatively small
group of subjects, our recordings were prolonged, resulting
in a vast amount of data to be used to create and validate
our model. Indeed, even after data clustering into one-min-
ute mean values, the total number of cases in our model
group was 403. According to strict criteria [20], the mini-
mum number of observations for a multiple linear regres-
sion model incorporating three independent variables
(predictors) is 74, which represents only 18% of the avail-
able data in our study. Therefore, although generating a
model from 403 cases is not the same as having 403 sub-
jects, the methodology used in the present study adheres to
the appropriate statistical protocols.
In conclusion, we provide a model that accurately pre-
dicts fingertip blood flow via two practical and inexpensive
temperature indicators: current fingertip temperature and
its difference with that recorded four minutes before, as
well as the difference between current rectal temperature
and that recorded four minutes before. The prediction is
accurate and provides a simple cost-efficient alternative
that can be implemented in both clinical and field settings,
as well as for various research and bioengineering applica-
tions where thermoneutral ambient air remains relatively
stable, and when the actual measurement of finger blood
flow is not plausible.
PERSPECTIVE
The high equipment cost and the extensive requirements
from participants necessary for accurate measurement of
fingertip blood flow introduce complications that may pre-
vent practical use. In this paper we provide a model that
accurately predicts fingertip blood flow via two practical
and inexpensive temperature indicators: current fingertip
temperature and its difference with that recorded four min-
utes before, as well as the difference between current rectal
temperature and that recorded four minutes before. The
model provides a simple cost-efficient alternative that can
be implemented in both clinical and field settings.
ACKNOWLEDGMENTS
We are grateful to our participants for their effort and per-
severance. The authors report no conflict of interest. This
work was supported in part by funding from the European
Union 7th Framework Program (FP7-PEOPLE-IRG-2008
Grant No. 239521), and a Discovery Grant (227912-07, S.S.
Cheung) from the Natural Science and Engineering
Research Council of Canada.
Cutaneous Finger Blood Flow Prediction
ª 2011 John Wiley & Sons Ltd, Microcirculation, 18, 670–676 675
REFERENCES
1. Bergersen TK, Hisdal J, Walloe L. Perfusion
of the human finger during cold-induced
vasodilatation. Am J Physiol 276: R731–
737, 1999.
2. Daanen HA. Finger cold-induced vasodila-
tion: a review. Eur J Appl Physiol 89: 411–
426, 2003.
3. Daanen HA, Van de Linde FJ, Romet TT,
Ducharme MB. The effect of body tem-
perature on the hunting response of the
middle finger skin temperature. Eur J
Appl Physiol Occup Physiol 76: 538–543,
1997.
4. Ezri T, Szmuk P, Weisenberg M, Serour F,
Gorenstein A, Sessler DI. The effects of
hydration on core temperature in pediatric
surgical patients. Anesthesiology 98: 838–
841, 2003.
5. Felicijan A, Golja P, Milcinski M, Cheung
SS, Mekjavic IB. Enhancement of cold-
induced vasodilatation following acclimati-
zation to altitude. Eur J Appl Physiol 104:
201–206, 2008.
6. Flouris AD, Cheung SS. Design and con-
trol optimization of microclimate liquid
cooling systems underneath protective
clothing. Ann Biomed Eng 34: 359–372,
2006.
7. Flouris AD, Cheung SS. Influence of ther-
mal balance on cold-induced vasodilation.
J Appl Physiol 106: 1264–1271, 2009.
8. Flouris AD, Cheung SS. On the origins of
cold-induced vasodilation. Eur J Appl
Physiol 108: 1281–1282, 2010.
9. Flouris AD, Cheung SS. Thermal basis of
finger blood flow adaptations during
abrupt perturbations in thermal homeo-
stasis. Microcirculation 18: 56–62, 2011.
10. Flouris AD, Cheung SS. Thermometry and
calorimetry assessment of sweat response
during exercise in the heat. Eur J Appl
Physiol 108: 905–911, 2010.
11. Flouris AD, Cheung SS. The validity of
tympanic and exhaled breath tempera-
tures for core temperature measurement.
Physiol Meas 31: N35–42, 2010.
12. Flouris AD, Cheung SS, Fowles JR, Kruis-
selbrink LD, Westwood DA, Carrillo AE,
Murphy RJ. Influence of body heat con-
tent on hand function during prolonged
cold exposures. J Appl Physiol 101: 802–
808, 2006.
13. Keramidas ME, Musizza B, Kounalakis SN,
Mekjavic IB. Enhancement of the finger
cold-induced vasodilation response with
exercise training. Eur J Appl Physiol 109:
133–140, 2010.
14. Nilsson AL, Eriksson LE, Nilsson GE. Effects
of local convective cooling and rewarming
on skin blood flow. Int J Microcirc Clin
Exp 5: 11–25, 1986.
15. Petrofsky J, Bains G, Prowse M, Gunda S,
Berk L, Raju C, Ethiraju G, Vanarasa D,
Madani P. Does skin moisture influence
the blood flow response to local heat? A
re-evaluation of the Pennes model. J Med
Eng Technol 33: 532–537, 2009.
16. Purkayastha SS, Selvamurthy W, Ilavazha-
gan G. Peripheral vascular response to
local cold stress of tropical men during
sojourn in the Arctic cold region. Jpn J
Physiol 42: 877–889, 1992.
17. Rubinstein EH, Sessler DI. Skin-surface
temperature gradients correlate with fin-
gertip blood flow in humans. Anesthesiol-
ogy 73: 541–545, 1990.
18. Sessler DI. Thermoregulatory defense
mechanisms. Crit Care Med 37: S203–
210, 2009.
19. Sessler DI, Rubinstein EH, Eger EI 2nd.
Core temperature changes during N2O
fentanyl and halothane ⁄ O2 anesthesia.
Anesthesiology 67: 137–139, 1987.
20. Tabachnick BG, Fidell LS. Using Multivari-
ate Statistics, 4th edn. New York, NY:
HarperCollins College Publishers, 2001.
A.E. Carrillo et al.
676 ª 2011 John Wiley & Sons Ltd, Microcirculation, 18, 670–676