Effect of body temperature on cold induced vasodilation

Abstract

Cold-induced vasodilation (CIVD) is an acute increase in peripheral blood flow observed during cold exposures. It is hypothesized to protect against cold injuries, yet despite continuous research it remains an unexplained phenomenon. Contrary to the traditionally held view, we propose that CIVD is a thermoregulatory reflex mechanism contributing to heat loss. Ten adults (4 females; 23.8 ± 2.0 years) randomly underwent three 130-min exposures to −20°C incorporating a 10-min moderate exercise period at the 65th min, while wearing a liquid conditioning garment (LCG) and military arctic clothing. In the pre-warming condition, rectal temperature was increased by 0.5°C via the LCG before the cold exposure. In the warming condition, participants regulated the LCG throughout the cold exposure to subjective comfort. In the control condition, the LCG was worn but was not operated either before or during the cold exposure. Results demonstrated that the majority of CIVD occurred during the warming condition when the thermometrically-estimated mean body temperature (T _b) was at its highest. A thermoregulatory pattern was identified whereby CIVD occurred soon after T _b increased past a threshold (~36.65°C in warming and pre-warming; ~36.4°C in control). When CIVD occurred, T _b was reduced and CIVD ceased when T _b fell below the threshold. These findings were independent of extremity temperature since CIVD episodes occurred at a large range of finger temperatures (7.2–33.5°C). These observations were statistically confirmed by auto-regressive integrated moving average analysis (t = 9.602, P < 0.001). We conclude that CIVD is triggered by increased T _b supporting the hypothesis that CIVD is a thermoregulatory mechanism contributing to heat loss.

Appendix

Plotting the T _f time series against the T _b time series for each participant and condition separately (Figs. 2–4) yielded a systematic pattern (relationship between T _f and T _b described above) but no periodicity [i.e., regular repetition in time (lack of periodicity is an essential ARIMA assumption)] (Box and Jenkins 1976). Further analyses were conducted using data from all participants and conditions simultaneously. To statistically address the possibility that the two time series (i.e., T _f and T _b) were not inherently unpredictable [random-walk/white noise (another ARIMA assumption)], the autocorrelation of the residuals from exponential smoothing (Gardner 1985) was calculated for each time series for a maximum of 38 lags/data points (i.e., ~5 min, given that data were collected every 8 s), confirming that neither of the two was random (Box-Ljung statistic P < 0.001). The exponential smoothing technique is very useful in time series analysis as it emphasizes the regularity of a series by removing some of the random variation and by giving extra weight to more recent observations. This allows the researcher to capitalize on any pattern that is evident in the observed series and to use that pattern to draw relevant conclusions. In the current analysis, the changes in T _f were detected approximately 160 s (or 20 lags/data points) following changes in T _b. Thus, the T _f series was recalculated to incorporate 20 lags and further analyses were conducted using this new variable.

ARIMA models combine as many as three types of processes based on the concept of random disturbances or shocks: autoregression (i.e., AR), integration/differencing (i.e., I) and moving averages (i.e., MA). Between two observations in a time series, a disturbance (in this case cold exposure/LCG heating) occurs that somehow affects the level of the series (in this case T _b) as well as the level of another series (in this case T _f). These disturbances as well as the association between the two time series can be mathematically described by ARIMA models. Using the appropriate model-building procedure for the best possible series model (Box and Jenkins 1976) the identification of the processes underlying the time series was conducted to determine the integers for autoregression, differencing and moving average. Given that the autoregression and moving average components require stationary series (Box and Jenkins 1976), the autocorrelations and partial autocorrelations of T _f and T _b time series were calculated showing that neither was stationary. This was also evident in all preliminary plots of T _f and T _b across time (Figs. 2–4). Following differencing transformation at lag 2, stationarity was achieved with no spikes evident to the 38th lag (~5 min of data collection given that data were collected every 8 s). As the series were differenced twice to achieve stationarity, the ARIMA integration parameter was set to ‘2’ (Box and Jenkins 1976). Further, given that both series had one spike in the first value of the autocorrelation and exponentially declining values of the partial autocorrelation, the appropriate order for autoregression and moving average was ‘0’ and ‘1’, respectively (Box and Jenkins 1976). The resulting ARIMA [0, 2, 1 (i.e., 0 order of autoregression; 2 degree of differencing; 1 order of moving average)] was found to be parsimonious and adequate model with no spikes evident to the 38th lag (Box-Ljung statistic P < 0.001) and T _b significantly associated with T _f (t = 9.602, P < 0.001). These results demonstrate that, across time, fluctuations in T _b were systematically followed (approximately 160 s later) by similar fluctuations in T _f.