2.1.

Subjects and Protocol

Sixty adult subjects were recruited, either from the vascular laboratory in the cardiovascular clinic (patients) or using local advertisements (healthy controls). For this clinical trial, 57 subjects contributed data. One subject refused the study post-enrollment, one had corrupted optical data, and one was missing all demographic data.

We report results of two tasks requested of each subject: plantar flexion (pedal) and treadmill exercise. Both tasks were performed on the same day and in the same order, for each subject. For pedal, the first exercise, subjects were in the supine position during the test protocol. The resistance of the exercise device remained constant, and the maximal force of each individual was tested prior to the task. Subjects were then instructed to perform up to 30 plantar flexion exercises within 2 min, at 50% of their maximal force. The treadmill exercise was carried out after a short rest, following the completion of the plantar flexion, according to the Gardner protocol.^25,26 Subjects walked on the treadmill at a constant speed ($2\text{miles}/\mathrm{h}$) for up to 10 min. The platform inclination was raised 2 deg every 2 min. All procedures were approved by the Institutional Review Board at the University of Pennsylvania, where the experiments were carried out. Informed consent was obtained from all subjects prior to enrollment into the study.

The optical probe was placed over the calf flexor to make NIRS and DCS measurements. ${\mathrm{S}}_t{\mathrm{O}}_2$ and relative blood flow change from baseline (rBF) were calculated from these measurements as described in the sections below. ${\mathrm{S}}_t{\mathrm{O}}_2$ was obtained before, during, and after exercise, but rBF was obtained only before and after exercise because DCS measurements were susceptible to motion artifacts caused by muscle motion during exercise.^23,27 In order to investigate potential differences across different levels of PAD, subjects were separated into three groups according to their ABI: healthy population ($\mathrm{ABI}>0.9$; $n=31$); mild PAD patients ($0.9>\mathrm{ABI}>0.7$; $n=11$), and moderate/severe PAD patients ($0.7>\mathrm{ABI}>0.3$; $n=15$).

2.2.

Optical Measurements

A hybrid diffuse optical device was built and employed to assess blood flow and oxygenation in muscle.²⁰ The DCS module uses a continuous-wave, long-coherence length 785-nm laser (CrystaLaser Inc., Reno, Nevada). Four photon-counting avalanche photodiodes (PerkinElmer, Canada) feed a four-channel autocorrelator board (Correlator.com, Bridgewater, New Jersey) that computes the temporal intensity autocorrelation function ($g_2$) of collected light. The NIRS module employs three laser diodes in the near-infrared (685, 785, and 830 nm; Thorlabs, Newton, New Jersey). The laser diodes are amplitude modulated at 70 MHz. Light collected from the tissue was delivered to four photomultiplier tubes (PMTs; Hamamatsu Corp., Japan), and the PMT signals were bandpass filtered, amplified, demodulated, and digitized. A laptop computer controlled both DCS and NIRS modules and recorded data. Two optical switches (Dicon Fiberoptics, Richmond, California) were employed to select among the four lasers (three from NIRS and one from DCS) and the three different source positions on the muscle. Optical data were acquired from 2 min prior to the start of the task (baseline period) until 4 min after the task was finished (recovery period). The whole data acquisition cycle of NIRS and DCS took approximately 2.5 s.

The optical probe was placed over the calf flexor in a region that covered approximately an area equivalent to a circle of 3.5 cm radius, similar to previous studies involving muscle (Fig. 1).^20,21 The source-detector distance ranged from 0.5 to 3.8 cm, so that the depth probed was approximately 1.5 to 2.0 cm into the muscle. Black, rigid, plastic material was used to tightly hold the fibers in place, and elastic straps fixed the probe around the muscle to reduce motion artifacts during exercise.

Fig. 1

Schematics of the optical probe designed for the study, containing three laser sources, four detectors for tissue oxygen saturation (${\mathrm{S}}_t{\mathrm{O}}_2$), and four detectors for relative blood flow (rBF). The probe covers an area of approximately $40{\mathrm{cm}}^2$ on the calf flexor. The different source-detector distances range from 0.5 to 3.8 cm, and are combined to provide a single measure of ${\mathrm{S}}_t{\mathrm{O}}_2$ and an average measure of rBF.

2.3.

Optical Data Analysis

For NIRS, the arrangement of optodes permitted continuous measurement of absolute absorption coefficients by calibrating the source and detector coupling coefficients with a self-calibrating approach.²⁰ Absolute ${\mathrm{HbO}}_2$ and Hb concentrations were then derived from the absorption coefficients.⁹ Total hemoglobin concentration (THC) and ${\mathrm{S}}_t{\mathrm{O}}_2$ were estimated as ${\mathrm{THC}=\mathrm{HbO}}_2+\mathrm{Hb}$ and ${\mathrm{S}}_t{\mathrm{O}}_2{=\mathrm{HbO}}_2/\mathrm{THC}$.

For DCS, rBF changes were estimated from DCS data by extracting a BFI for each source-detector separation at each time point. ($\mathrm{rBF}(t)=\mathrm{BFI}(t)/\mathrm{BFI}(t_0)$, where $t_0$ denotes the baseline period.) The BFI was estimated by fitting the measured intensity autocorrelation function to the solution of the photon correlation diffusion equation in the semi-infinite geometry with extrapolated zero boundary conditions.^9,10 In this work, the diffusive motion model was used to approximate the mean-square particle displacement of the moving red blood cells in tissue and thus derive the BFI. The changes in the absorption coefficient at 785 nm measured by NIRS were used as input in the correlation diffusion equation. The goodness of fit was evaluated for each source-detector separation at each time point, and the decay curves that failed to fit the model (i.e., curves whose fitting residuals were higher than 75% of the mean residual over the entire time-series) were discarded from the analysis. For statistical analysis, changes in rBF were derived from changes just after exercise relative to the period prior to exercise and averaged over all source-detector separations. Figure 2 shows representative time courses of rBF and ${\mathrm{S}}_t{\mathrm{O}}_2$ for two different subjects (one healthy control and one moderate/severe PAD patient) during both tasks. Although ${\mathrm{S}}_t{\mathrm{O}}_2$ was also obtained during exercise, in order to maintain correspondence with DCS data, we only analyzed changes in the ${\mathrm{S}}_t{\mathrm{O}}_2$ data before and after exercise.

Fig. 2

Representative time-courses of ${\mathrm{S}}_t{\mathrm{O}}_2$ (top row) and rBF (bottom row) for a moderate/severe peripheral artery disease (PAD) patient and a healthy control for the treadmill (left column) and pedal (right column) exercises. The gray area indicates the duration of the exercise, while the up (down) arrow indicates the start (end) of the exercise. Note, ${\mathrm{S}}_t{\mathrm{O}}_2$ was obtained before, during and after exercise, but rBF was accurately obtained only before and after exercise (due to motion artifacts that arise during exercise). The statistical analysis was performed for the period after cessation of the exercise, taking the period immediately before each exercise as the baseline.

2.4.

Statistical Analysis

The baseline covariate data for each ABI group were summarized using the median and interquartile range (IQR) for continuous variables or counts (proportions) for categorical data. Kruskal-Wallis tests were used to assess differences in exercise times and the amount of power or work, defined as the product of the power and the length of exercise. Spearman’s correlation coefficient, $\rho$, was used to measure the association between exercise times.

To explore differences between healthy and PAD patients, mixed effects models were used as the basic tool to describe both pre-exercise means as well as post-exercise temporal patterns in rBF and ${\mathrm{S}}_t{\mathrm{O}}_2$ (Ref. 28). The models for ${\mathrm{S}}_t{\mathrm{O}}_2$ incorporated baseline mean ${\mathrm{S}}_t{\mathrm{O}}_2$ levels for each individual as a covariate and allowed nonlinear changes over time; the models for rBF allowed either a constant rate or linear changes over time (see Appendix for details on model selection). While the model fits a continuous curve, we illustrate our results with means and 95% confidence intervals (95% CI) at specific time points (immediately post-exercise, 0.5, 1, and 3 min). We note that while data were potentially collected up until about 4 min post exercise, there was substantial missing data for some subjects toward the end of the collection period. Thus we chose to illustrate results for the first 3 min as the dataset is much more complete across subjects for this time period. Overall significance of the model was determined using a likelihood ratio test; contrasts between specific terms were based on Wald tests. $p$-Values for pair-wise comparisons among ABI groups were adjusted using a Bonferroni correction.

The mixed effects models explored differences among ABI groups in temporal patterns of ${\mathrm{S}}_t{\mathrm{O}}_2$ and rBF, assuming that all individuals within an ABI group follow the same mean pattern over time. To relax this assumption, we analyzed the ${\mathrm{S}}_t{\mathrm{O}}_2$ data using time to return to baseline as the outcome of interest. Because several individuals did not return to baseline by the end of our monitoring period, we estimated the median time to return to baseline levels using Kaplan-Meir methods, and then used a Cox model to test the hypothesis that differences in the relative risk, or equivalently, rate of returning to baseline are associated with ABI group or actual ABI value.

In contrast with ${\mathrm{S}}_t{\mathrm{O}}_2$, individual traces for rBF often did not return to baseline over the measured post-exercise period, and, for some individuals, outlier values were not uncommon. Here, for each exercise, we used a Kruskal-Wallis test to determine whether there were differences in the extremes of the distribution of rBF among ABI groups. Specifically we examined the median proportion of post-exercise interval that the rBF measurements for an individual were either elevated ($>1.5$ fold baseline) or depressed ($<0.5$ baseline). This was done separately for the first 1.5 min (early) or 1.5 to 3.0 min (late) periods post-exercise. If the global test achieved significance ($p<0.05$) Bonferroni-adjusted pair-wise tests of significance were constructed using a Wilcoxon rank-sum test.

Last we developed two models to explore the association between the three key variables measured in this study: ABI, ${\mathrm{S}}_t{\mathrm{O}}_2$, and rBF. The first, a linear model, considers an individual’s ABI as a function of mean ${\mathrm{S}}_t{\mathrm{O}}_2$ and percent elevated or depressed post-exercise rBF as described above for the early or late time periods. The second, a Cox model, considers the relative rate of returning to baseline ${\mathrm{S}}_t{\mathrm{O}}_2$ levels, as a function of ABI and these same percent elevated or percent depressed rBF variables. For purposes of brevity, we report only a single two-variable model for each outcome and exercise, with the rBF variable that achieved the highest level of significance in our analyses. All statistical analyses were carried out using either R 2.10 or R 2.15.²⁹

3.1.

Baseline Characteristics

Table 1 shows baseline characteristics of the subjects. The ABI from healthy subjects ranged from 0.92 to 1.54, while patients with mild impairment presented with $0.7\le \mathrm{ABI}<0.9$. The ABI for the moderate/severe patients ranged from 0.33 to 0.7. The moderate/severe group included 15 subjects whose median ABI of 0.62 corresponds to 56% of the median ABI in the healthy group.

Table 1

Subject characteristics. Values shown are medians [interquartile range (IQR)], except for race and gender, which show number of subjects [proportion of subjects for each ankle brachial index (ABI) group].

While the protocol specified a target length for the duration of each exercise, the healthy group tended to exercise longer because subjects with PAD often stopped when they felt muscle pain. For treadmill, healthy, mild, and moderate/severe PAD patients exercised for a median of 10.0, 4.0, and 4.2 min, respectively. For pedal, healthy controls exercised for a median of 1.9 min compared with a median of 1.4 and 1.5 min for the mild and moderate/severe PAD groups, respectively. Exercise times for the two tasks were significantly correlated ($\rho =0.45$, $p=0.0004$). For the pedal exercise, no significant differences in the distribution of work or power was observed among groups, noting that these variables were missing for five, two, and one subjects in the healthy, mild PAD and moderate/severe PAD groups, respectively ($p=0.39$ for work, $p=0.55$ for power).

Mean ${\mathrm{S}}_t{\mathrm{O}}_2$ differed among groups pre-exercise ($p=0.002$, Table 2). The mean (95% CI) ${\mathrm{S}}_t{\mathrm{O}}_2$ for treadmill was 2.6% (0.4, 5.6%) higher than for pedal ($p=0.089$). Before treadmill, the mean ${\mathrm{S}}_t{\mathrm{O}}_2$ of 66.2% for the healthy group was 10.6% (3.5, 17.7%) and 9.3% (2.8, 15.9%) higher than the mild and moderate/severe PAD groups ($p=0.013$ and 0.017, respectively). By contrast, the mild and the moderate/severe PAD patients differed by only 1%.

Table 2

Hemodynamic parameters (95% CI) pre- and post-exercise. The models for post-exercise tissue oxygen saturation (StO2) were adjusted for the pre-exercise mean for each subject. The values shown here are for a subject with the mean equal to the pre-exercise mean for that group.

3.2.

Treadmill Exercise: Individual Trends in $S_t{O}_2$ and rBF

Pre-exercise mean ${\mathrm{S}}_t{\mathrm{O}}_2$ was strongly associated with post-exercise ${\mathrm{S}}_t{\mathrm{O}}_2$ ($p<0.0001$). Figure 3 shows the estimated mean hemodynamic responses for each group. Strong evidence was found for an increase in ${\mathrm{S}}_t{\mathrm{O}}_2$ over time ($p<0.0001$), but rates of increase did not differ significantly between groups.

Fig. 3

Model-based estimates of mean (95% CI) ${\mathrm{S}}_t{\mathrm{O}}_2$ (top row) and rBF (bottom row) in healthy, mild, and moderate/severe subjects immediately, 0.5, 1.0, and 3.0 min post-treadmill exercise. For the top row, the dashed line is the pre-exercise mean ${\mathrm{S}}_t{\mathrm{O}}_2$ for each group, and dotted lines are the 95% CI (top row, see Table 2 for values) or a value of 1.0 for rBF (bottom row).

Table 2 shows the estimated means at discrete time points, based on the fit of our model immediately following the treadmill exercise. Immediately post-exercise, mean ${\mathrm{S}}_t{\mathrm{O}}_2$ was depressed to 56.9% (53.2, 60.6%) in the healthy group. Mean ${\mathrm{S}}_t{\mathrm{O}}_2$ in the healthy controls exceeded by 11.7% (6.9, 16.5%, $p<0.0001$) that of the mild PAD patients, and 16.6% (12.3, 20.9%, $p<0.0001$) that of the moderate/severe PAD patients. The difference of 4.9% ($-0.4$, 10.1%) between the two PAD groups was not significant. Mean values in all groups gradually returned to the baseline mean for that group.

Mean pre-exercise ${\mathrm{S}}_t{\mathrm{O}}_2$ appears as a dashed line in Fig. 3, with 95% CI as a dotted line. The post-exercise mean for all three groups recovered to either their pre-exercise mean or near it, with the recovery appearing most rapid in healthy controls. Consistent with the patterns in Fig. 3, Table 3 shows that the median time for an individual to return to his/her pre-exercise level, as determined from the Kaplan-Meier method, ranged from 0.72 min for the healthy subjects to 1.68 min for the moderate/severe PAD patients. While the trend was consistent with that seen for the mean models (Fig. 3), differences in the rate of returning to baseline ${\mathrm{S}}_t{\mathrm{O}}_2$ levels for the PAD and healthy subjects did not achieve statistical significance.

Table 3

Median time in minutes (95% CI) to return to pre-exercise mean StO2 along with relative rate (the hazard ratio) for returning to pre-exercise mean StO2 based on a Cox model; UD is undetermined based on the length of time followed. p-Values are shown for the Cox model comparing each peripheral artery disease (PAD) groups to healthy.

In contrast to ${\mathrm{S}}_t{\mathrm{O}}_2$, rBF differed among ABI groups post-exercise ($p=0.0036$), but temporal changes in blood flow during the 3-min post-exercise observation window were not detected. Healthy subjects and mild PAD patients showed pronounced elevation in rBF post-exercise with means (95% CI) of 1.45 (1.14, 1.84) and 1.78 (1.22, 2.60), respectively, while mean rBF in the moderate/severe PAD group of 0.80 (0.51, 1.11) did not differ from pre-exercise levels. Pairwise comparisons indicated that mean rBF in the moderate/severe PAD patients differed from healthy controls ($p=0.017$) and mild PAD patients ($p=0.008$), but that mean rBF did not differ significantly between the healthy controls and the mild group.

Table 4 examines the rBF data in terms of the proportion of the post-exercise period when rBF was reduced to 0.5 of baseline (percent depressed) or exceeded 1.5 of baseline (percent elevated). For treadmill, there were pronounced differences, particularly in percent elevated for both early ($p=0.007$) and late post-exercise periods ($p=0.011$). Both healthy and mild PAD subjects had pronounced elevations in blood flow, with a median percent elevated of at least 45% during early and late post-exercise periods; in contrast, the median percent elevated for the moderate/severe group was 20% in the early post-exercise period and 7% in the late post-exercise period. Differences between groups in percent depressed were also significant in the early post-exercise period ($p=0.010$); median percent depressed for healthy and mild subjects were zero compared to 5.9% for the moderate/severe group.

Table 4

Median (IQR) for percent of time when measurements of rBF for each subject were depressed (below 0.5 of baseline) or elevated (greater than 1.5 of baseline) either early (first 1.5 min) or late (1.5 to 3.0 min post exercise) post-exercise. p-Values are for the overall test of equality among groups; symbols indicate Bonferonni-adjusted pair-wise significance.

3.3.

Treadmill Exercise: Associations Between ABI, $S_t{O}_2$, and rBF

Because ABI is widely used in the diagnosis and monitoring of PAD patients, we first assessed the extent to which characteristics of the post-exercise distribution of rBF added to the prediction of ABI, beyond what could be achieved with ${\mathrm{S}}_t{\mathrm{O}}_2$ alone. Table 5 shows that mean ${\mathrm{S}}_t{\mathrm{O}}_2$ in the treadmill post-exercise period was strongly correlated with ABI ($p=0.001$) but with not percent-depressed rBF. Other characteristics of the rBF distribution (e.g., median rBF or percent-elevated rBF) also showed no association (results not shown). We then modeled the relative rate of returning to baseline for ${\mathrm{S}}_t{\mathrm{O}}_2$ as a function of ABI, and the percent-elevated rBF at the early post-exercise interval. In this two-variable model, the rate of return to baseline ${\mathrm{S}}_t{\mathrm{O}}_2$ was not associated with ABI, and it was only marginally associated with percent elevated rBF ($p=0.089$). Similar trends occurred for the later, 1.5 to 3.0 min, post-exercise period (not shown).

Table 5

Joint modeling of ABI and rate of StO2 return to baseline.

3.4.

Pedal Exercise: Individual Trends in $S_t{O}_2$ and rBF

Pre-exercise mean ${\mathrm{S}}_t{\mathrm{O}}_2$ was strongly associated with post-exercise ${\mathrm{S}}_t{\mathrm{O}}_2$ for pedal exercise ($p<0.0001$). Over time, ${\mathrm{S}}_t{\mathrm{O}}_2$ increased significantly ($p=0.0001$), but the rate of increase did not differ among ABI groups. As with treadmill exercise, a pronounced depression in ${\mathrm{S}}_t{\mathrm{O}}_2$ was observed, followed by recovery (Fig. 4 and Table 2). Immediately post-exercise, mean ${\mathrm{S}}_t{\mathrm{O}}_2$ was 51.8% (47.9, 55.8%) in the healthy group. This value was 18.0%, (14.0, 22.2%, $p<0.0001$) and 15.6% (10.3, 20.9%, $p<0.0001$) higher than for the mild and moderate/severe PAD groups, respectively. The difference of 2.4% ($-3.4$, 8.2%) between mild and moderate/severe PAD groups was not significant.

Fig. 4

Model-based estimates of mean (95% CI) ${\mathrm{S}}_t{\mathrm{O}}_2$ (top row) and rBF (bottom row) in healthy, mild, and moderate/severe subjects immediately 0.5, 1, and 3 min post pedal exercise. For the top row, the dashed line is the mean pre-exercise ${\mathrm{S}}_t{\mathrm{O}}_2$ for each group (see Table 2 for values) and dotted lines are the 95% CI. For the bottom row (rBF), the dashed line is a value of 1.0.

Mean pre-exercise ${\mathrm{S}}_t{\mathrm{O}}_2$ appears as a dashed line in Fig. 4, with 95% CI as a dotted line. The post-exercise mean for all three groups recovered to either their pre-exercise mean or near it, with the recovery appearing most rapid in healthy controls and slower in the two PAD groups. More formally, Table 3 shows that the median time to return to pre-exercise mean ${\mathrm{S}}_t{\mathrm{O}}_2$, as determined from the Kaplan-Meier method, was longer than for treadmill, ranging from 1.14 min for the healthy subjects to 3.32 min for the mild PAD patients. The relative risk of returning to baseline was a factor of 0.29 fold lower for mild patients compared with healthy subjects ($p=0.0007$), and 0.44 fold lower for moderate/severe patients compared with healthy subjects ($p=0.027$).

In contrast with the treadmill exercise, the mean rBF immediately post-pedal exercise did not differ significantly from 1.0 in any group. The healthy group essentially showed no change in rBF, with a mean of 1.08 (0.83, 1.40); mild and moderate/severe PAD groups showed slight, albeit nonsignificant depressions with means of 0.75 (0.49, 1.14) and 0.83 (0.58, 1.19), respectively. There was evidence of increases in post-pedal exercise mean rBF over time ($p=0.003$) and a suggestion of differences in post-pedal exercise mean rBF patterns among groups over time, although these differences did not achieve strict statistical significance ($p=0.078$ overall).

Table 4 shows that, for the pedal exercise, differences in the degree to which subjects displayed elevation or depression in rBF for either the early or late periods were more subtle than for treadmill. There were no differences among groups at either time interval in the percent elevated, with values ranging from 5.3% to 8.9% during the early post-exercise period, and from 7.1% to 17.5% for the later post-exercise period. Differences between groups in the percent depressed differed both in the early ($p=0.025$) and the late ($p=0.017$) time period. Notably, the moderate/severe group had median rates of percent depressed of 13.8% in the early post-exercise phase and 10.7% in the late post-exercise phase, in contrast with median rates of 0% for the healthy group.

3.5.

Pedal Exercise: Associations Between in ABI, $S_t{O}_2$, and rBF

Table 5 shows that, as with the treadmill exercise, mean ${\mathrm{S}}_t{\mathrm{O}}_2$ in the pedal post-exercise period was strongly correlated with ABI ($p=0.001$), but only marginally associated with rBF ($p=0.071$). Other characteristics of the rBF distribution (e.g., median rBF or percent elevated rBF) showed no association with ABI (results not shown). In contrast with the treadmill exercise, the relative rate of return to baseline ${\mathrm{S}}_t{\mathrm{O}}_2$ as the outcome was marginally associated with ABI ($p=0.097$), with a 0.10 increase in ABI predicting a 1.10 fold increase in the rate of return to baseline. In the same model, a 10% decrease in percent-depressed rBF was associated with a 1.23 fold increase in the rate of returning to baseline ($p=0.041$). Similar trends were found for the later, 1.5 to 3.0 min, post-exercise period (not shown).