Between-visit Reproducibility of Cerebrovascular and Cardiovascular Reactivity During the Cold Pressor Test

Abstract

The cold pressor test (CPT) is a painful test used to assess cardiovascular and cerebrovascular reactivity. Past studies examined between-visit reproducibility of cardiovascular reactivity during the CPT but not cerebrovascular reactivity. Therefore, we performed a retrospective analysis to test the hypothesis that cerebrovascular responses during the CPT would not differ and would have moderate relative agreement between visits. Fifty young adults free from overt disease, not taking pain-modifying medications, completed two trials ≥48 hours apart. We measured middle cerebral artery blood velocity (MCAv) via transcranial Doppler ultrasound, beat-to-beat blood pressure (BP) via photoplethysmography, and end-tidal CO₂ (ET-CO₂) via capnography during a five-minute baseline and two-minute CPT. We calculated reactivity as the change (Δ) from baseline to the first and second minute of the CPT. There were no between-visit differences in baseline mean BP, mean MCAv, or ET-CO₂ (Ps≥0.24). There were no between-visit differences for Δmean MCAv (Ps>0.75) or Δmean BP (Ps>0.07). Relative agreement between visits was moderate for Δmean MCAv (intraclass correlation coefficient (ICC): Δmin-1=0.68, Δmin-2=0.72) and moderate/good Δmean BP (ICC: Δmin-1=0.65, Δmin-2=0.80). Our data suggest that Δmean MCAv to the CPT is not different between visits while the relative agreement is similar to BP reactivity. Further, relative agreement was higher during minute two than minute one for key variables. These data extend our knowledge of the reproducibility of CPT reactivity. These findings will inform future research aimed at comparing cardiovascular and cerebrovascular reactivity as health markers between groups.

New & Noteworthy

We present novel data on between-visit cerebrovascular and cardiovascular reactivity reproducibility during the CPT. Relative agreement was stronger for BP and mean MCAv reactivity during the second minute of the CPT, suggesting that future work should use protocols >2 minutes. Cerebrovascular and cardiovascular variables had similar reproducibility; changes in ET-CO₂ and perceived pain significantly predicted CVCi reactivity. Controlling ET-CO₂ could improve reproducibility; future work could consider pain perception while assessing cerebrovascular reactivity to the CPT.

Introduction

The cold pressor test (CPT) is a painful stimulus that involves immersing a participant’s hand, foot/feet, or forehead into an ice-water mixture (~0–4°C). The CPT elicits sympathetically-mediated blood pressure (BP) and heart rate responses (i.e., cardiovascular reactivity) (1–5). Cardiovascular reactivity during the CPT is clinically relevant because greater BP reactivity is predictive of a future risk of hypertension (6–8), and coronary artery vasomotor reactivity during this procedure is predictive of myocardial infarction and stroke risk (9). The CPT is also used to assess cerebrovascular reactivity during an increase in BP and sympathetic nervous system activity (10).

During the CPT, thermoreceptors and nociceptors send afferent signals that are integrated and lead to increased sympathetic nervous system activity, subsequently increasing peripheral vascular resistance, BP, and heart rate (2, 6, 11). It is debated whether the sympathetic nervous system contributes to cerebral blood flow. Nevertheless, cerebral blood velocity must be regulated as higher BP increases perfusion pressure and higher cortical activity requires greater blood flow and nutrients (i.e., neurovascular coupling) (9, 12). Thus, others have investigated the effect of various participant and research design factors on cerebrovascular responses to the CPT. For example, previous work investigated the effect of biological sex (13), hydration status (1), the limb(s) (e.g., hand vs. foot) submerged in the ice-water mixture (14), the cerebral vessel assessed (15), age (9, 16, 17), and changes in the concentration of end-tidal CO₂ (ET-CO₂) (10, 15) on cerebrovascular responses to the CPT. Considering this previous body of work, it is important to understand the reproducibility of CPT responses.

Although others have investigated the reproducibility of BP and sympathetic responses to the CPT (2, 5, 18–20), to our knowledge, no one has investigated the between-visit reproducibility of cerebrovascular responses during the CPT. Therefore, we retrospectively analyzed the between-visit reproducibility of cerebrovascular reactivity during the CPT among 50 participants. Specifically, we assessed the between-visit reproducibility of cerebrovascular, cardiovascular, and pain responses during the CPT. Informed by past reports from others on the reproducibility of BP responses during the CPT (2, 5, 18–20), we tested the hypothesis that cerebrovascular reactivity would exhibit no systematic bias and at least moderate relative agreement for between-visit reproducibility during the CPT.

Methods

Ethical Approval.

The United States Army Medical Research & Material Command Human Research Protection Office and Institutional Review Board for Human Subjects Research at the University of Texas Southwestern Medical Center (IRB# 092017-068, 092017-069, 092017-070) and Texas Health Presbyterian Hospital Dallas approved the protocols and associated informed consent forms. These studies conformed to the standards set by the latest revision of the Declaration of Helsinki. Each participant provided verbal and written consent before enrollment.

Experimental overview.

We retrospectively analyzed data collected between October 2019 and January 2022 from three registered clinical trials (ClinicalTrials.gov Identifiers: NCT04138615, NCT04136548, NCT03621085). Each protocol was similar, except each protocol examined a different pain medication (ketamine, fentanyl, or morphine). Each protocol included two experimental visits separated by at ≥48 hours (range: 2–48 days; median: 7 days). Visit start time was matched within 1 hour. Each visit included a resting baseline period and pain assessments before the infusion of either the drug or the placebo. Our previous publications focused on comparing post-drug versus post-placebo responses to a lower-body negative pressure challenge (21–24) or the CPT (25–27). Further, we compared sex differences in sympathetic responses during lower body negative pressure during placebo trials (28). Notably, the questions addressed in the present manuscript are distinct from our previous work. In this manuscript, we investigated the between-visit reproducibility of CPT responses before any drug or placebo infusions or lower-body negative pressure testing. We have not previously reported any cerebrovascular responses during the pre-drug/placebo CPTs.

Participants.

We recruited participants from the Dallas-Fort Worth metroplex. Participants were free of any known cardiovascular, metabolic, or neurological diseases. Inclusion criteria were age 18–45 years, body mass ≥60 kg, and body mass index <35 kg/m² at screening. Exclusion criteria were current or recent (i.e., within the past 3 years) nicotine use or current use of pain-modifying or antihypertensive medications.

There are different viewpoints on matching menstrual cycle status between and across participants for cardiovascular-related research (29). We did not design this project to examine menstrual cycle-related effects among the female participants in this study (n = 22). Nonetheless, we matched the menstrual cycle phase between visit 1 and visit 2 for 10 participants (luteal phase n = 5; follicular phase n = 5). Twelve female participants did not have matched menstrual cycle phases: 6 were not matched due to logistical challenges related to scheduling during the SARS-CoV-2 pandemic, 4 had intrauterine devices (i.e., IUDs), 1 previously had a hysterectomy, and 1 had irregular menstrual cycle phase length.

Experimental Protocol.

Before the experimental visits, we instructed participants to limit their food and water intake for 2 hours, caffeine intake for 12 hours, and strenuous exercise, alcohol, and non-steroidal anti-inflammatory drugs for 24 hours. Additionally, we advised them to refrain from using over-the-counter cold or allergy medication and aspirin for 36 hours before each trial. Participants arrived at the laboratory and provided a spot urine sample to confirm 1) a negative urine drug screen, 2) no hypohydration (spot urine specific gravity <1.025) (1, 30) (Atago Inc., Bellevue, WA, USA), and 3) a negative pregnancy screen for female participants. After placing an intravenous catheter, we instrumented participants to measure brachial BP, beat-to-beat BP, heart rate, middle cerebral artery blood velocity (MCAv), cerebral oxygen saturation (ScO₂), respiration rate, and partial pressure of ET-CO₂. After we acquired these signals, participants relaxed quietly in the supine position in our temperature-controlled laboratory. Following an epoch of at least 5 minutes for baseline data collection, participants completed a pressure pain threshold test that was followed by a CPT.

Pressure Pain Assessments & CPT.

We assessed pressure pain threshold in a subset of 43 participants by applying pressure with the tip of a handheld digital algometer (model FPX, Wagner Instruments, Greenwich, CT, USA) just proximal to the webbing between the second and third digits. We instructed participants to “report the first feeling of discomfort, not the most pain that can be tolerated.” We gradually applied force until participants reported the first feelings of discomfort. We repeated this test 3–5 times, ~10 seconds apart, until we obtained 3 readings within 10% of one another (31).

The CPT was completed by having a laboratory team member submerge the participant’s hand in a stirred ice bath (0.35 [0.07]°C) for 2 minutes. We instructed participants to avoid holding their breath or hyperventilating during the CPT; we monitored compliance via capnography. Immediately after removing their hand from the ice bath, we asked participants to rate their pain during the CPT using a 100 mm visual analog scale. The left anchor, at 0 mm, was ‘no pain or discomfort’, and the right anchor, at 100 mm, was ‘worst imaginable pain or discomfort’. The distance from the left anchor, 0 mm, is reported in the results as perceived pain during the CPT.

Cardiovascular, cerebrovascular, and respiratory measures.

We used a single-lead ECG to continuously assess heart rate (GE Medical Systems, WI, USA). We used photoplethysmography to assess beat-to-beat BP (Finometer; Finapres Medical Systems, Netherlands) (32). We confirmed finometer-measured BP with an automated brachial cuff that integrates Korotkoff sounds detected with a microphone that were triggered with ECG-signals (Tango M2 Stress Test Monitor, SunTech Medical, SC, USA). We measured MCAv at the right transtemporal window using a 2 MHz transcranial Doppler probe (Spencer Technologies, Redmond, WA). We measured ET-CO₂ using a nasal cannula connected to a capnograph (9004 Capnocheck Plus; Smiths Medical International Ltd, Watford, UK). We measured ScO₂ using near-infrared spectroscopy (Moor Instruments Inc., Wilmington, DE, USA). All measures were collected with the participants at rest and during the CPT.

Data analysis.

Continuous data were collected at a sampling rate of 625 Hz using Biopac (MP150, Biopac, Santa Barbara, CA, USA). We defined BP and MCAv waveform peaks as systolic BP or MCAv, nadirs as diastolic BP or MCAv, and the average value of the integrated waveform as mean BP or mean MCAv on a beat-to-beat basis using Ensemble (Elucimed, Wellington, New Zealand). We calculated cardiovascular and cerebrovascular reactivity, expressed as a change (Δ), during the CPTs by subtracting the values obtained during the first and second minutes of the CPT from those obtained during the 5-minute pre-CPT baseline period. We calculated the cerebrovascular resistance index as mean BP/mean MCAv (CVRi), cerebrovascular conductance index as mean MCAv/mean BP (CVCi), and MCAv pulsatility index as ([maximum MCAv - minimum MCAv]/mean MCAv).

Statistical analysis.

The results in this manuscript address a secondary research question by leveraging an existing dataset. Thus, there was no a priori power analysis for the presented work. We report normally distributed data as mean ± standard deviation (SD) and non-normally distributed data (P ≤ 0.05 for Shapiro-Wilk normality test) as median [interquartile range]. Reproducibility is considered at least partially subjective, so we report multiple reproducibility indices to be comprehensive and transparent (2, 33). We used two-way repeated-measures analysis of variance (i.e., RM-ANOVAs) to test for main effects of time and visit, and for time-by-visit interactions. We used Bonferroni-corrected pairwise comparison post-hoc tests (e.g., paired t-test) to assess significant effects. We report partial eta squared (η_p²) effect sizes and interpreted η_p² of <0.01 as trivial, 0.01 to <0.06 as small, 0.06 to <0.14 as medium, and ≥0.14 as large effects.

We compared visits to detect fixed bias (constant bias). For variables with normally distributed (parametric) differences (P > 0.05 for Shapiro-Wilk normality test), we tested for fixed bias using paired two-tailed t-tests and reported Cohen’s d effect size. For variables with non-parametric differences (P ≥ 0.05 for Shapiro-Wilk normality test), we tested for fixed bias with Wilcoxon Signed-rank tests and reported rank biserial (r_bc) effect sizes (34). A positive test for fixed bias was a statistically significant difference (P < 0.05) between visits. We interpreted a Cohen’s d of <0.20 as trivial, 0.20 to <0.50 as small, 0.50 to <0.80 as medium, and ≥0.80 as large (35). We interpreted r_bc of 0 to <0.20 as small, 0.20 to <0.30 as medium, 0.30 to <0.40 as large, and 0.40 to ≥1 as very large (36). We calculated absolute agreement as mean absolute error, mean absolute percent error, and coefficient of variation (CV). We only calculated mean absolute percent error and CV for baseline because the mean or median value for most reactivity variables was 1–1.5 standard deviations/interquartile ranges from 0, so mean absolute percent error and CV were inflated and not easily interpreted. We used a mean absolute percent error of ≤10% as a metric of good absolute agreement (34, 37, 38).

We assessed between-visit relative agreement with Lin’s concordance correlation coefficient (CCC) (39) and intraclass correlation coefficients (ICC; 3,k) (40). We interpreted CCC as <0.2 as poor, 0.2 to <0.6 as fair, 0.6 to <0.7 as moderate, 0.7 to <0.9 as strong, and 0.9 to 1.0 as very strong (41). We interpreted ICCs as <0.50 as poor, 0.50–0.75 as moderate, 0.75–0.90 as good, and >0.90 as excellent (40).

We created Bland-Altman plots to display between-visit reproducibility of baseline and Δ values during minutes 1 and 2 of the CPT. We tested for Bland-Altman assumptions of 1) between-visit difference (residual) normality, 2) no proportional bias (no relation between the average of and difference between the visits), and 3) homoscedasticity (even spread of differences across the value range) (33, 34, 42). We assessed between-visit difference normality (parametric) with Shapiro-Wilk tests. For parametric variables, we tested for proportional bias with ordinary least squares regression of the differences and the average of the two visits. For non-parametric variables, we tested for proportional bias with quantile (median) regression. Proportional bias was present if the regression slope was significantly different from 0 (P < 0.05) (34). We tested for homoscedasticity using the same regression type used for proportional bias testing. In this model, we regressed the absolute differences between visits with the proportional bias assessment model on method averages. Heteroscedasticity was present if the regression slope was significantly different from 0 (P < 0.05) (34). Our calculations for the limits of agreement (LoA) were based on the three Bland-Altman assumptions. If all assumptions were met, we calculated LoA as the mean difference ± the standard deviation multiplied by 1.96 (34, 42). When proportional bias was present, LoA followed the slope of the proportional bias line (proportional bias regression model standard error of the estimate x 1.96) (34, 42). If heteroscedasticity was present, regardless of proportional bias, we calculated V-shaped LoAs by multiplying the regression coefficient from the absolute residuals of the ordinary least squares or median regression models by 2.46 (33, 34, 42).

We also explored the factors that determined changes in CVCi (ΔCVCi) during the CPT with a linear mixed model. We used ΔCVCi as our measure of cerebrovascular reactivity because BP is a key determinant of ΔMCAv, and ΔCVCi includes mean BP. Previous work highlights that multiple variables affect cerebrovascular reactivity to the CPT: age (9, 16), sex (13), hydration status (i.e., urine specific gravity) (1), and ET-CO₂ (10, 15). Therefore, we fitted a linear mixed-effects model with fixed effects including visit number, age, sex, urine specific gravity, and end-tidal CO₂ (ET-CO₂). We added perceived pain to the model as a fixed effect because previous work has not investigated the effect of pain perception on cerebrovascular reactivity to the CPT. Finally, participant ID was included as a random intercept to account for within-subject correlation across repeated measures. Because we included ET-CO₂ in the model, we only included participants who had ET-CO₂ measured during both visits (n = 38).

We completed data analysis in R (Version 4.6.1, R Core Team, 2026) (43, 44). We considered α < 0.05 statistically significant, but also took into account the effect size and absolute value of each physiological variable when interpreting these data (45, 46).

Results

We provide participant screening characteristics in Table 1. There was not a significant between-visit difference in visit time (visit 1: 11:30 AM [4 hours] vs. visit 2: 12:00 PM [4 hours]; mean difference = 9 minutes; P = 0.81, r_bc=0.09), participant urine specific gravity (visit 1: 1.013 ± 0.007 vs. visit 2: 1.014 ± 0.007; P = 0.51, d = 0.09), laboratory ambient temperature (visit 1: 23.1 [1.5] vs. visit 2: 23.1 [1.9]°C; P = 0.56, r_bc = 0.09), relative humidity (visit 1: 43 [11] vs. visit 2: 42 [15] %; P = 0.94, r_bc = 0.01), or barometric pressure (visit 1: 750 ± 7 vs. visit 2: 749 ± 6 mmHg %; P = 0.50, r_bc = 0.10).

Table 1.

Participant Characteristics (n = 50)

Sex	22 (44%) Female
	28 (56%) Male
Age, yrs	27 [9] (19 – 45)
Body mass index, kg • m−2	26 ± 5 (20 – 34)
Race	37 (74%) White 6 (12%) Asian 1 (2%) Black 1 (2%) American Indian/Alaskan Native 3 (6%) More than one race 2 (4%) Unknown
Ethnicity	42 (84%) Not Hispanic/Latine 6 (12%) Hispanic/Latine 2 (4%) Unknown

We present categorical variables as n (%). We present normally distributed continuous variables as mean ± standard deviation or non-normal as median [interquartile range] with ranges (minimum – maximum) for age and body mass index.

Reproducibility of pressure pain tolerance.

Pressure pain tolerance was not different (P=0.24, d=0.19) between visit 1 (1.2[0.5] kg/cm²) and visit 2 (1.1[0.5] kg/cm²). The mean absolute difference between visits was 0.25, and the mean absolute percent error was not good (i.e., poor, 20.0%). Pressure pain tolerance had moderate/good relative agreement between the two visits (CCC=0.686, P<0.001; ICC=0.819, P<0.001).

Reproducibility of water temperature and perceived pain during the CPT.

Water temperature was not different between visits (visit 1: 0.36[0.05] vs. visit 2: 0.34[0.08] °C; P=0.18, r_bc=0.23). Perceived pain during the CPT was significantly greater (P=0.04, d=0.30) during visit 2 (65[23]) compared with visit 1 (59[32]). Perceived pain mean absolute error between visits for perceived pain was 9, mean absolute percent error between visits was not good (20.1%), and relative agreement between visits 1 and 2 was strong/excellent (CCC=0.831, P<0.001; ICC=0.915, P<0.001).

Reproducibility of physiological responses to the CPT.

We removed two participants from the reactivity analysis because they had excessive reductions in ET-CO₂ throughout the CPT during one of their visits (Participant 1: −23 mmHg; Participant 2: −33 mmHg), suggesting measurement error rather than physiological responses. In these cases, we removed participant data for both visits.

Figure 1 shows the change (Δ) from baseline for systolic, diastolic, and mean BP, and heart rate, and mean MCAv, CVCi, MCAv pulsatility, and ET-CO₂. There was a significant effect of time for all variables (Ps<0.03; F=3.5–104.9; η_p²=0.07–0.68). The η_p² for Δmean MCAv was medium (η_p²=0.07), for the other variables, the η_p² was large (η_p²=0.26–0.68). There were no significant main effects of visit (Ps>0.14; F=0.18–2.52; η_p²=0.03–0.05) nor were there any significant time-by-visit interactions (Ps>0.06; F=0.06–0.65; η_p²=0.01–0.06). Based on post-hoc testing for systolic, mean, and diastolic BP, and heart rate, there was a significant increase during minutes 1 and 2 compared to baseline (Ps<0.001), and the Δ during minute 2 was significantly greater than minute 1 (Ps<0.001). For CVCi and ET-CO₂, there was a significant decrease during minutes 1 and 2 (Ps<0.009), and the Δ during minute 2 was significantly greater than minute 1 (Ps<0.001). The mean MCAv Δ during minutes 1 (P=0.17) and 2 (P=1.0) was not significant, but the Δ during minute 2 was greater than minute 1 (P=0.03). There was a significant decrease in MCAv pulsatility during minute 1 (P<0.001) and 2 (P<0.001), but the minute 2 Δ was not significantly different from minute 1 (P=1.0).

Figure 1. Cerebrovascular and cardiovascular responses to the cold pressor test (CPT).

The figure shows changes from baseline (Δ; CPT value minus baseline value) during minute 1 and minute 2 of the cold pressor test (CPT) for (A) systolic blood pressure (BP), (B) mean BP, (C) diastolic BP, (D) heart rate, (E) mean middle cerebral artery blood velocity (MCAv), (F) cerebrovascular conductance index, (G) MCAv pulsatility, and (H) end-tidal CO₂ (ET-CO₂). The large colored dots and error bars represent group mean±standard deviation. The grey dots represent individual data. There was a significant effect of time for all variables. There were no significant main effects of visit, nor were there any significant time-by-visit interactions.

Between-visit reproducibility of resting baseline values

Table 2 shows baseline absolute and relative between-visit reliability for the resting baseline period. Based on paired tests, there were no significant between-visit differences. Mean absolute percent error was <10% for systolic, mean, and diastolic BP, heart rate, ET-CO₂, and MCAv pulsatility, but ≥10% for all other variables. Proportional bias was only present for heart rate. There was no heteroscedasticity for any of the variables. Relative agreement based on ICCs was fair to moderate, with heart rate having the strongest ICC. Similarly, relative agreement based on CCC was fair-to-strong, with heart rate having excellent reliability.

Table 2.

Resting baseline absolute and relative between-visit reproducibility.

Variable	n	Visit 1	Visit 2	Fixed Bias	P	Effect size	MAE	MAPE (%)	CV (%)	Proportional bias P-value	Scedasticity P-value	ICC	ICC 95% CI	ICC classification	ICC P-value	CCC	CCC P-value	CCC classification
Systolic BP (mmHg)	49	127 [12]	127 [13]	−1±12	0.71	d=0.05	9	6.9	9.5	0.57	0.45	0.61	0.31–0.78	moderate	<0.001	0.436	0.002	fair
Mean BP (mmHg)	49	94 [10]	92 [12]	−0 [7]	0.89	rbc=0.02	6	6.5	9.7	0.44	0.77	0.67	0.42–0.82	moderate	<0.001	0.506	<0.001	fair
Diastolic BP (mmHg)	49	70 [9]	71 [9]	1 [6]	0.66	rbc=0.07	5	7.4	11.2	0.47	0.41	0.60	0.29–0.77	moderate	0.001	0.423	0.002	fair
Heart rate (beats/minute)	49	58 [14]	59 [13]	−1 [6]	0.29	rbc=0.18	4	6.5	9.2	0.02	0.11	0.94	0.89–0.96	excellent	<0.001	0.878	<0.001	strong
ET-CO2 (mmHg)	44	49 [5]	49 [6]	1 [3]	0.25	d=0.18	2	5.3	6.9	0.08	0.045	0.87	0.76–0.93	good	<0.001	0.764	<0.001	strong
Peak MCAv (cm/second)	46	86 [25]	85 [23]	2 [21]	0.26	rbc=0.19	13	15.8	21.6	0.76	0.55	0.61	0.30–0.79	moderate	0.001	0.440	0.002	fair
Mean MCAv (cm/second)	46	60±13	60±13	1 [14]	0.28	rbc=0.18	9	16.3	21.7	0.79	0.59	0.65	0.36–0.80	moderate	<0.001	0.477	0.001	fair
Minimum MCAv (cm/second)	46	42±9	41±9	1.81 [11]	0.22	rbc=0.21	7	17.1	22.3	0.82	0.47	0.64	0.35–0.80	moderate	<0.001	0.467	0.001	fair
MCAv pulsatility (au)	46	0.77±0.10	0.77±0.12	−0.01 [0.10]	0.94	rbc=0.01	0.07	9.0	13.5	0.15	0.70	0.73	0.51–0.85	moderate	<0.001	0.571	<0.001	fair
CVCi (cm/s/mmHg)	46	0.65±0.155	0.64±0.139	0.006±0.147	0.79	d=0.04	0.11	17.4	22.8	0.41	0.69	0.67	0.40–0.82	moderate	<0.001	0.503	<0.001	fair
CVRi (mmHg/cm/s)	41	1.5 [0.579]	1.6 [0.494]	−0.03 [0.42]	0.46	rbc=0.14	0.3	16.5	25.0	0.46	<0.01	0.62	0.28–0.80	moderate	0.002	0.446	0.003	fair
ScO2 (%)	42	64.0±7.0	62.6±7.2	1.32±5.6	0.13	d=0.24	4.4	7.1	8.8	0.82	0.92	0.81	0.66–0.90	good	<0.001	0.675	<0.001	moderate

We present normally distributed variables as mean ± standard deviation with Cohen’s d sizes. We present non-normally distributed variables as median [IQR] and rank biserial correlation effect (rbc) sizes. MAE = mean absolute error, MAPE = mean absolute percent error, CV = coefficient of variation, ICC = intraclass correlation coefficient, CI = confidence interval, CCC = Lin’s concordance correlation coefficient, BP = blood pressure, ET-CO2 = end-tidal CO2, MCAv = middle cerebral artery blood velocity, au = arbitrary units, CVCi = cerebrovascular conductance index (mean MCAv/mean BP), CVRi = cerebrovascular conductance index (mean BP/meanMCAv), ScO2 = cerebral oxygenation.

Between-visit reproducibility of minute 1 and minute 2 responses during the CPT

Table 3 shows the between-visit reproducibility for the changes (Δ) observed during the first and second minute of the CPT. There was not a significant between-visit difference for the Δ during minute 1 or minute 2 of the CPT for any of the variables listed in Table 3. The effect size for the between-visit difference ranged from trivial to medium with Δheart rate and ΔCVRi being the only variables with a medium effect. Proportional bias was present for Δheart rate during minute 2, ΔET-CO₂ during minute 1, Δpeak-, Δmean-, and Δminimum MCAv during minute 1, and ΔCVCi during minute 1. Heteroscedasticity was present for Δheart rate during both minutes, for ΔET-CO₂ during minute 2, Δpeak MCAv during minute 1, and Δminimum MCAv during minute 1. Relative agreement was poor to moderate based on ICCs, and poor to strong based on CCCs. Notably, relative agreement, based on ICCs and CCCs, was numerically higher during the second minute for 10 of the 12 variables listed in Table 3.

Table 3.

CPT reactivity absolute and relative between-visit reproducibility

Variable	CPT minute	n	Visit 1	Visit 2	Bias	Paired test P-value	Effect size	Effect classification	MAE	Proportional bias P-value	Scedasticity P-value	ICC	ICC 95% CI	ICC P-value	ICC P-value	CCC	CCC P-value	CCC classification
ΔSystolic BP (mmhg)	1	49	7 [13]	6 [9]	2 [13]	0.14	d=0.21	small	9	0.72	0.34	0.63	0.35–0.79	moderate	<0.001	0.45	0.001	fair
	2	49	16 [16]	14 [14]	1 [14]	0.81	d=0.03	trivial	9	0.76	0.85	0.79	0.63–0.88	good	<0.001	0.65	<0.001	moderate
ΔMean BP (mmHg)	1	49	8±7	6±8	2±8	0.07	d=0.27	small	6	0.51	0.98	0.65	0.37–0.80	moderate	<0.001	0.46	0.001	fair
	2	49	15 [13]	14 [10]	1 [10]	0.60	d=0.08	trivial	6	0.12	0.99	0.80	0.64–0.89	good	<0.001	0.64	<0.001	moderate
ΔDiastolic BP (mmHg)	1	49	6±6	4±6	2±6	0.05	d=0.29	small	4	0.32	0.78	0.66	0.41–0.81	moderate	<0.001	0.48	<0.001	fair
	2	49	11 [9]	10 [7]	0 [6]	0.41	d=0.12	trivial	4	0.12	1.00	0.79	0.62–0.88	good	<0.001	0.65	<0.001	moderate
ΔHeart rate (beats/min)	1	49	10 [15]	11 [12]	1 [6]	0.17	rbc=0.23	medium	6	0.23	<0.01	0.83	0.70–0.91	good	<0.001	0.71	<0.001	strong
	2	49	6 [8]	5 [10]	1 [7]	0.17	rbc=0.23	medium	5	0.03	0.02	0.83	0.69–0.90	good	<0.001	0.69	<0.001	moderate
ΔET-CO2 (mmHg)	1	43	−1 [3]	−2 [3]	0.2 [3]	0.49	d=0.11	trivial	2	0.05	0.06	0.44	−0.03–0.70	poor	0.03	0.28	0.065	fair
	2	42	−2 [4]	−3 [4]	1 [4]	0.11	d=0.25	small	3	0.35	0.01	0.66	0.37–0.82	moderate	<0.001	0.48	0.001	fair
ΔMCAv peak (cm/s)	1	46	−2 [6]	−1 [7]	0 [8]	0.91	rbc=0.02	small	5	<0.01	0.04	0.64	0.35–0.80	moderate	<0.001	0.47	0.001	fair
	2	46	−2 [9]	−4 [12]	1 [7]	0.49	rbc=0.12	small	6	0.19	0.16	0.73	0.51–0.85	moderate	<0.001	0.57	<0.001	fair
ΔMCAv mean (cm/s)	1	46	1 [6]	2 [7]	0 [5]	0.75	rbc=0.05	small	4	<0.01	0.13	0.68	0.41–0.82	moderate	<0.001	0.51	<0.001	fair
	2	46	−1±8	−1±8	0±7	0.99	d=0.00	trivial	5	0.76	0.75	0.72	0.49–0.84	moderate	<0.001	0.56	<0.001	fair
ΔMCAv min (cm/s)	1	46	1 [5]	2 [6]	0 [5]	0.52	rbc=0.11	small	4	0.01	0.01	0.62	0.32–0.79	moderate	0.001	0.44	0.002	fair
	2	46	1 [7]	0 [8]	0 [7]	0.85	rbc=0.03	small	5	0.45	0.80	0.61	0.29–0.78	moderate	0.001	0.44	0.002	fair
ΔMCAv pulsatility (au)	1	46	−0.05 [0.09]	−0.06 [0.08]	0.01 [0.10]	0.43	rbc=0.14	small	0.07	0.91	0.17	0.40	−0.08–0.67	poor	0.045	0.24	0.102	fair
	2	46	−0.06 [0.09]	−0.04 [0.11]	0.00 [0.13]	0.93	rbc=0.02	small	0.09	0.68	0.81	0.50	0.09–0.72	moderate	0.012	0.33	0.026	fair
ΔCVCi (cm/s/mmHg)	1	46	−0.040±0.059	−0.017±0.087	−0.023±0.081	0.06	d=0.28	small	0.06	<0.01	0.15	0.58	0.23–0.76	moderate	0.002	0.39	0.008	fair
	2	46	−0.095±0.085	−0.095±0.095	0.000±0.080	0.99	d=0.00	trivial	0.06	0.36	0.61	0.75	0.55–0.86	good	<0.001	0.60	<0.001	moderate
ΔCVRi (mmHg/cm/s)	1	41	0.117 [0.240]	0.092 [0.261]	0.033 [0.298]	0.21	rbc=0.23	medium	0.25	0.10	0.13	0.46	−0.01–0.71	poor	0.027	0.28	0.074	fair
	2	41	0.236 [0.352]	0.285 [0.335]	0.000 [0.343]	0.83	rbc=0.04	small	0.25	0.15	0.05	0.72	0.47–0.71	moderate	<0.001	0.56	<0.001	fair
ΔScO2 (%)	1	42	0.1±2.1	0.1±1.9	0.1±2.7	0.88	d=0.02	trivial	1.81	0.58	0.89	0.19	−0.52–0.56	poor	0.26	0.10	0.523	poor
	2	42	−0.2 [2.8]	−0.2 [3.5]	0.3 [3.9]	0.58	d=0.10	trivial	2.83	0.54	0.31	0.36	−0.19–0.66	poor	0.08	0.22	0.163	poor

We present normally distributed variables as mean ± standard deviation with Cohen’s d sizes. We present non-normally distributed variables as median [IQR] and rank biserial correlation effect (rbc) sizes. MAE = mean absolute error, ICC = intraclass correlation coefficient, CI = confidence interval, CCC = Lin’s concordance correlation coefficient, Δ = change (minute 1 or 2 minute baseline), BP = blood pressure, ET-CO₂ = end-tidal CO₂, MCAv = middle cerebral artery blood velocity, au = arbitrary units, CVCi = cerebrovascular conductance index (mean MCAv/mean BP), CVRi = cerebrovascular conductance index (mean BP/meanMCAv), ScO2 = cerebral oxygenation.

Bland-Altman Analysis for Cardiovascular and Cerebrovascular Measures

Figure 2 shows Bland-Altman plots for (A–C) systolic, (D–F) mean, and (G–I) diastolic BP, and (J–L) heart rate baseline values and Δ values during minutes 1 and 2 of the CPT. There were no significant fixed biases. As noted in Tables 2 and 3, there is proportional bias for baseline and Δ during minute 2 for heart rate reflected by the sloping bias line, and there is heteroscedasticity during minutes 1 and 2, reflected by the V-shaped limits of agreement.

Figure 2. Bland-Altman plots for cardiovascular variables at baseline and changes during the cold pressor test (CPT).

The figure shows agreement between visits 1 (V1) and 2 (V2) for (A–C) systolic blood pressure (BP), (D–F) mean BP, (G–I) diastolic BP, and (J–L) heart rate at baseline and for changes (Δ; CPT values minus baseline values) during minute 1 and minute 2. There is no significant fixed bias for any of the variables at each time point. Limits of agreement are similar for systolic and mean BP at baseline and Δ during minutes 1 and 2. The limits of agreement narrow for diastolic BP. As noted in Tables 2 and 3 there is proportional bias for baseline and minute 2 heart rate reflected by the sloping bias line, and there is heteroscedasticity at minute 1 and 2, reflected by the V-shaped limits of agreement.

Figure 3 shows Bland-Altman plots for (A–C) mean MCAv, (D–F) CVCi, (G–I) MCAv pulsatility, and (J–L) ET-CO₂ baseline values and Δ values during minutes 1 and 2 of the CPT. There were no significant fixed biases. There was proportional bias for Δmean MCAv, ΔCVCi, and ΔET-CO₂ during minute 1 of the CPT. Baseline and minute 2 ΔET-CO₂ were heteroscedastic. Supplemental Figure 1 (https://doi.org/10.6084/m9.figshare.33437203) shows Bland-Altman plots for peak and minimum MCAv, CVRi, and ScO₂.

Figure 3. Bland-Altman plots for cerebrovascular variables at baseline and changes during the cold pressor test (CPT).

The figure shows agreement between visits 1 (V1) and 2 (V2) for (A–C) mean middle cerebral artery blood velocity (MCAv), (D–F) cerebrovascular conductance index (CVCi), (G–I) MCAv pulsatility, and (J–L) end-tidal CO₂ (ET-CO₂) at baseline and for changes (Δ; CPT values minus baseline values) during minute 1 and minute 2. There was no significant fixed bias for these variables for baseline or Δ values during the CPT. Limits of agreement for Δmean MCAv and ΔCVCi were narrower at minute 2 than at baseline. There was proportional bias for Δmean MCAv, ΔCVCi, and ΔET-CO₂ during minute 1 of the CPT; the sloped limits of agreement represent proportional bias. Baseline and minute ΔET-CO₂ were heteroscedastic.; the V-shaped limits of agreement represent heteroscedasticity.

Relative agreement summary and exploratory analysis

Figure 4 summarizes the between-visit relative reliability via (A) ICC with 95% CIs and (B) CCC at baseline, minute 1, and minute 2 for systolic, mean, and diastolic BP, and heart rate, and mean MCAv, CVCi, MCAv pulsatility, and ET-CO₂. Supplemental Figure 2 (https://doi.org/10.6084/m9.figshare.33437203) shows the same data with additional ICC and CCC results for peak MCAv, minimum MCAv, CVRi, and ScO₂. As noted in Figure 3, heart rate has the strongest between-visit relative agreement, and between-visit relative agreement is generally greater during minute 2 of the CPT compared to minute 1 for this variable. Notably, based on ICCs, relative agreement for Δsystolic, Δmean, and Δdiastolic BP, and ΔCVCi changes from ‘moderate’ to ‘good’ (based on ICCs) at minute 2; we observed the same change for CCC with a change from ‘fair’ to ‘moderate’.

Figure 4. Relative agreement summary.

A summary of the between-visit relative reliability via (A) intraclass correlation coefficients (ICC) with 95% CIs and (B) Lin’s concordance correlation coefficient (CCC) for baseline values and change values (Δ) during minute 1 (ΔMin 1) and minute 2 (ΔMin 2) of the CPT for systolic, mean, and diastolic BP, and heart rate, and mean MCAv, CVCi, MCAv pulsatility, and ET-CO₂. As noted in Figure 3, heart rate has the strongest between-visit relative agreement, and between-visit relative agreement is generally greater during minute 2 of the CPT compared to minute 1. Notably, based on ICCs, relative agreement for Δsystolic, Δmean, and Δdiastolic BP, and ΔCVCi changes from ‘moderate’ to ‘good’ (based on ICCs) at minute 2; we observed the same change for CCC, with a change from ‘fair’ to ‘moderate’.

Variables associated with ΔCVCi during the CPT

Age, sex, and hydration (urine specific gravity) were not significant predictors of ΔCVCi, so we removed these variables from the model. Visit number was not significant (P = 0.281), but we retained this variable because of the repeated measures design. Model results are shown in Table 4. Based on our final model, there was a significant negative relation between ΔCVCi and ET-CO₂, and a significant positive relation between ΔCVCi and perceived pain. Variance inflation factor was <1.03 for all variables (ΔET-CO₂ = 1.04, perceived pain = 1.03, and visit number = 1.01). The model fixed effects (perceived pain, ET-CO₂, and visit number) explained 36.7% of the variance in ΔCVCi (Marginal R² = 0.367). The full model—including the random effect of participant ID—explained 64.5% of the variance (Conditional R² = 0.645). The intraclass correlation for the model was 0.439. In a subset of 37 participants, we recorded which hand was submerged during the CPT. The MCAv measurement was always on the right side, so we were able to assess if hand side was associated with ΔCVCi. Hand side (i.e., right vs. left) was not a significant predictor in this new model (marginal R² = 0.38; conditional R² = 0.57; likelihood ratio test vs the original model: X² = 0.11, P=0.75).

Table 4.

Linear mixed model results for cerebrovascular reactivity

Predictor	Unstandardized Estimate (b)	Standardized Estimate (β)	Standard error	t-value	P
Intercept	−0.002	-	0.008	−12.03	<0.001
Δend-tidal CO2	0.0111	0.47	0.002	5.37	<0.001
Perceived Pain	−0.0014	−0.29	<0.001	−3.20	0.001
Visit Number	0.013	-	0.012	1.08	0.28

Discussion

We recognize the following main findings. We confirmed previous findings that resting cardiovascular variables and cardiovascular reactivity to the CPT are reproducible, but we found that relative agreement for BP reactivity was stronger during the second minute of the CPT. Second, we confirmed previous work that resting MCAv and CVCi were reproducible. Third, we found that cerebrovascular reactivity to the CPT was generally reproducible, with that response aligning with some, but not all, previous work. Lastly, we confirmed previous work that ΔET-CO₂ is a key determinant of ΔCVCi during the CPT, and added a new finding that perceived pain is significantly associated with ΔCVCi.

Reproducibility of resting cardiovascular measures and cardiovascular reactivity.

Our findings align with previous work, which indicates good reproducibility for resting heart rate and BP (2, 5). We observed a significant effect of time (increase) for systolic, mean, and diastolic BP, and HR during the CPT with no visit effect or time-by-visit interactions. Previous studies on the reproducibility of cardiovascular reactivity during the CPT have mostly reported good reproducibility, based on ICCs (2, 19, 20), but some have reported less favorable reproducibility while highlighting large individual differences (47). Some of this prior work aligns with our findings that there can be large individual differences in cardiovascular reactivity to the CPT (2, 47). Despite the individual differences, both we and Dillon et al. found that Δmean BP is almost always positive, such that mean BP increases during the CPT (2). Based on previous work and our findings, cardiovascular reactivity to the CPT is typically not different between visits, but individual response heterogeneity persists, and this heterogeneity should be considered in future work (18). Interestingly, we observed that systolic, mean, and diastolic BP relative agreement—based on ICC and CCC—was greater during the second minute of the CPT. To our knowledge, no other findings have indicated whether test time affects reproducibility. Based on this observation, future work using the CPT across visits may want to use at least a two-minute CPT. More work is needed to compare CPT test length to determine whether reproducibility is stronger during longer CPTs (e.g., 3 minutes).

Reproducibility of resting TCD-measured cerebrovascular measures.

Resting MCAv, CVCi, pulsatility, and other cerebrovascular variables were not significantly different between visits, and relative agreement was fair for CCC and moderate for ICCs. We did not observe any proportional bias for these variables. Others have investigated between-visit reproducibility of resting TCD-measured middle cerebral artery blood velocity (48–51), and most have focused on mean MCAv or MCA velocities, but not CVCi, MCAv pulsatility, or CVRi. Past work, and our present findings, suggest an absence of group average differences in resting MCAv between visits (49). Moreover, our findings support prior work, which found that there is no proportional bias for resting MCAv between visits (51). The lack of proportional bias and fair relative agreement for CVCi and CVRi was not surprising because both were calculated using mean BP, which showed no proportional bias and fair/moderate relative agreement. The good/strong reproducibility of ET-CO₂ is an important finding because CO₂ is the most important modulator of MCAv. Specifically, changes in the partial pressure of arterial CO₂ (PaCO₂) is a potent vasodilator (increased PaCO₂) or vasoconstrictor (decreased PaCO₂) of the cerebrovasculature. The cerebrovascular system is especially sensitive to changes in PaCO₂ (52) because of the need for tight regulation of acid-base balance which can further influence the pH of the cerebrospinal fluid near the ventrolateral medulla, a key nucleus for ventilation control and acid-base homeostasis (53–55). Based on past work and our findings, resting mean MCAv values differ on an individual level but reproducibility is moderate/fair.

Reproducibility of cerebrovascular reactivity to the CPT.

We found a significant effect of time for Δmean MCAv, ΔCVCi, ΔMCAv pulsatility, and ΔET-CO₂ to the CPT. Post-hoc comparisons for Δmean MCAv revealed that there was only a significant difference between the Δ during minute 1 and minute 2, and numerically, the values for Δmean MCAv were near 0. This finding indicate that MCAv changes minimally during the CPT. Our results align with work that found no change in mean MCAv during the CPT (10, 14, 15, 56–58), but not with others who observed increases in mean MCAv during the CPT (1, 10, 15, 59–61). Previous findings indicate that changes in ET-CO₂ during the CPT alter the MCAv response (10, 15), so it is noteworthy that we evaluated the reproducibility of cerebrovascular responses to the CPT without fixing ET-CO₂ (i.e., isocapnic vs. poikilocapnic). Our results generally suggest that cerebrovascular reactivity is reproducible between visits because there are no fixed biases, a trivial to small effect sizes, and moderate/fair relative agreement.

There is a notable difference between cardiovascular (i.e., ΔMean BP) and (i.e., Δmean MCAv) cerebrovascular reactivity to the CPT, at least for this TCD-measured MCA response. For almost all trials (94%; 279/296 trials), the Δsystolic, Δmean, and Δdiastolic BP were positive during the CPT, so BP increased from baseline to minute 2. Conversely, the direction of the mean MCAv response during the CPT was not as consistent. Specifically, across the 96 trials for the 48 participants, 48% of ΔMCAv values were positive, 52% of Δmean MCAv values were negative, and 18 participants (38%) had positive Δmean MCAv during one visit and a negative Δmean MCAv during the other visit. These differences highlight that the directional change in ΔMCAv during the CPT is not as consistent with ΔBP.

Considering the difference between the ‘typical’ BP response and ‘typical’ cerebrovascular response, it seems that a ‘typical’ MCAv or CVCi response to the CPT could not be easily established. However, Ogoh and colleagues (57) recently investigated the role of dynamic cerebral autoregulation on mean MCAv responses during the CPT. They found a significant association between low-frequency phase of mean MCAv and the relative decrease in CVCi to the CPT. These data indicate that stronger dynamic cerebroautoregulation was associated with greater decreases in CVCi during the CPT (57). However, the CPT was not isocapnic, and there was a significant association between the ΔCVCi and ΔET-CO₂. Thus, future work is needed to determine whether dynamic cerebral autoregulation is associated with isocapnic MCAv responses to the CPT. If there is a strong relation, cerebrovascular reactivity to the CPT could be used as a surrogate for dynamic cerebral autoregulation. More work is needed to determine if the relation is reproducible.

Associations between ΔET-CO₂, as well as perceived pain, with ΔCVCi during the CPT.

We found that ΔET-CO₂ (positively) and perceived pain (negatively) were associated with ΔCVCi during the CPT, findings that align with previous work. As mentioned, Tymko and colleagues highlighted that fixing ET-CO₂ significantly alters cerebrovascular reactivity to the CPT, which is due to the cerebrovascular system’s unique sensitivity to changes in CO₂ concentrations and ultimately pH (10, 52). Our finding that ΔET-CO₂ was the largest contributor to the linear mixed model supports the foundational principle that cerebro-arterial CO₂—and ultimately pH—has the largest impact on CVCi. Our finding that perceived pain is also associated with ΔCVCi aligns with previous work. For example, Perry et al. found that when participants were mildly dehydrated, pain perception was higher, which was associated with greater reductions in ET-CO₂, and a subsequent attenuation of ΔMCAv during the CPT (1). Their findings suggest that changes in perceived pain alter cerebrovascular reactivity via ET-CO₂ changes, whereas our findings indicate that both ET-CO₂ and pain affect cerebrovascular reactivity during the CPT. Previous work using painful heat stimuli on the forearm proposed that increases in MCAv during such painful stimuli could be related to neural activation in the neuromatrix of nociception (62). More work is needed to understand the effect of pain on cerebrovascular reactivity to the CPT.

Methodological Considerations.

Although all measurements of CPT reactivity were conducted pre-placebo or -drug infusion and visits were well standardized for participant preparation (e.g., fasting, hydration, etc.), ambient conditions, and visit time, this study was not prospectively designed to assess the test-retest reproducibility of CPT responses. Also, the CPT was not matched for the same arm between visits in this retrospective analysis; however, based on our regression analysis, this may have had little effect on results. This study employed radial microneurography, which required us to alternate the hand used for the CPT between visits to avoid placing the microneurography needle in the same arm during relatively close repeated visits. Second, this study was not prospectively designed to examine any differences associated with biological sex differences or the menstrual cycle phase. However, the obtained data provide the foundation for prospective studies to address these important questions and to replicate the exploratory subgroup results reported in this article.

Conclusion

We confirmed that resting BP and heart rate, and ΔBP and Δheart rate during the CPT are reproducible. Interestingly, we found that between-visit relative reliability for BP and heart rate reactivity was stronger during the second minute of the CPT. Thus, future work should consider using at least 2-minute protocols when studying BP and heart rate reactivity to the CPT. Our results suggest that MCAv and CVCi reactivity to the CPT are reproducible. Based on our regression model, the ΔCVCi is strongly related to the ΔET-CO₂. Thus, it is plausible that performing isocapnic CPTs could improve reproducibility, but more work is needed to confirm this hypothesis. We further observed that perceived pain was a significant predictor of ΔCVCi. Future work could explore pain perception on cerebrovascular reactivity to the CPT. More work is needed to understand the role of cerebrovascular reactivity to the CPT for assessing cerebrovascular regulation in health and disease.

Data Availability Statement

Data are available upon reasonable request to the principal investigator after institutional data transfer agreement approvals.