Simultaneous monitoring of heart rate and motion to assess energy expenditure

Sedentary lifestyle and low levels of energy expenditure are associated with a number of chronic illnesses, including obesity, coronary heart disease, noninsulin-dependent diabetes, and hypertension⁽¹⁵⁾. Unfortunately, the accurate measurement of physical activity and energy expenditure is problematic, and all of the existing measurement techniques have significant limitations.

Evidence linking low levels of physical activity to chronic disease states has been derived primarily from epidemiologic studies using questionnaires and exercise logs^{(3,6,8,10,14,21)}. While these instruments may be reliable for the estimation of energy expenditure for population groups, they have repeatedly been shown to be invalid for individuals when compared with expenditure measured by indirect calorimetry or doubly labeled water ^(1,9,16,19). The use of questionnaires and self-recorded exercise logs is particularly questionable for the quantification of changes in an individual's energy expenditure or physical activity level observed during intervention trials. Determination of the amount of physical activity needed to derive specific health benefits requires accurate and reliable methods of measuring energy expenditure in free-living individuals.

Heart rate monitors and electronic motion sensors have also been used to measure daily energy expenditure and physical activity^(4,7,18,20). While each of these directly measures a physical determinant of energy expenditure and thus represents an improvement over subjective measures such as questionnaires and logs, heart rate monitors and motion sensors both have significant limitations. For example, the relationship between heart rate and oxygen consumption (˙VO₂), which reflects energy expenditure directly, varies between persons depending on endurance capacity, necessitating individual calibration curves ^(7,12). The˙VO₂-heart rate relationship is also subject to significant intra-individual variability during different activities⁽¹¹⁾ and is influenced by external factors such as emotion, posture, and environmental conditions^(7,13). The greatest limitation to the use of heart rate for measuring energy expenditure is the almost flat slope of the relationship at low expenditure levels, lending significant error to quantification below a heart rate threshold for each individual^(9,12). Electronic motion sensors are also subject to limitations. Most notably, the relationships between motion and energy expenditure resulting from different types of movements and varying levels of resistance or intensity have not been well characterized^(2,5,7,17). Furthermore, a single motion sensor cannot detect movements of different segments of the body nor can sensors detect the static exercise that may be a substantial component of normal daily activity ⁽⁷⁾.

It has been suggested that the simultaneous recording of heart rate and motion may overcome the limitations of the individual techniques⁽⁷⁾. Thus, once the relationship between heart rate plus movement and oxygen uptake for an individual is defined in the laboratory, these instruments could be used in free-living situations to estimate accurately energy expenditure. The combination of heart rate and motion has been tested in the laboratory on a relatively homogeneous population performing exercise-type activities by Haskell et al. ⁽⁷⁾ and was not found to be a significant improvement over heart rate alone for the prediction of oxygen consumption. In that study, all of the exercises performed to create the calibration curves were at the high end of energy expenditure where the relationship between heart rate and oxygen consumption tends to be linear.

Since it is already known that correlations between energy expenditure and heart rate at high levels of expenditure are good, motion may provide a better measure of the relationship at the low end of expenditure. The addition of motion sensor data to heart rate data at low levels of expenditure may improve the prediction of oxygen consumption from that of heart rate alone. This study was designed to test the accuracy of heart rate in combination with motion for the determination of oxygen consumption during an exercise activity and during activities of daily life. In addition, it was designed to test the practicality of using an instrument developed to record both heart rate and motion with the hopes of applying it to future epidemiologic studies.

Subjects. Ten subjects, eight female and two male, were recruited from the staff of Loyola University's Department of Preventive Medicine and Epidemiology and the Edward Hines, Jr. Veterans Affairs Hospital's Rehabilitation Research and Development Center to participate in this investigation (Table 1). The subjects were all healthy but not involved in any personal or organized physical fitness programs. The protocol was approved by the Institutional Review Board of Loyola University Stritch School of Medicine, and informed consent was obtained from all subjects. All subjects participated in two tests of physical activity: the Activities of Daily Living Circuit (ADLC) and a submaximal treadmill test. Tests were conducted by trained personnel of the Hines Rehabilitation Research And Development Center.

Measurements. During both tests motion was recorded by the Ambulatory Monitoring System 1000 (AMS-1000, Consumer Sensory Products, Palo Alto, CA). The AMS-1000 is a compact and battery-operated instrument and, depending upon the sampling interval chosen, can record motion for up to several weeks. The AMS-1000 was worn at the waist with the mercury switch of the motion sensor positioned at the top of the left calf and held in place by a velcro strap. Movement was quantified by one omnidirectional motion sensor made up of several mercury tilt switches. The switches in the motion sensor were positioned such that a switch either opens or closes when the motion sensor was moved through any plane. The change, i.e., open or close, of any switch was recorded by a counter in the monitor which totaled and stored the number of switch changes over the sampling period. Although designed to also record heart rate, when compared with independently measured rates the AMS-1000 provided highly inaccurate data in 30% of the tests and no data at all in 5%. It was unclear whether the difficulty in accurately recording heart rate using the AMS-1000 was because of the operators or the instrument; thus, the data obtained from the AMS-1000 are neither presented nor used in subsequent analyses.

Heart rate was simultaneously recorded during the ADLC by an electrocardiograph (ECG) telemetry unit (Escort 200T, Medical Data Electronics, Arleta, CA) which provided a digital readout every 30 s. During the treadmill test, heart rate was independently monitored using a standard 12-lead ECG with a standard strip chart recorder (Quinton B3000, Quinton Instruments, Seattle, WA).

Oxygen consumption was measured during both the ADLC and treadmill tests by indirect calorimetry using a portable metabolic cart (MMC Horizon, Sensor Medics, Yorba Linda, CA). The metabolic cart monitored oxygen consumption and carbon dioxide production via a face mask positioned over the nose and mouth. The sampling period for ˙VO₂ (ml·min^-1) was set at 30 s. Prior to testing, the oxygen and carbon dioxide sensors were calibrated with room air and reference gases. Immediately following each test, the calibration was verified. Reproducibility of ˙VO₂ results for the ADLC and submaximal treadmill tests in this laboratory was shown to be quite good (r = 0.907).

Protocol. Each subject completed two activity tests; the ADLC was performed first and then followed by the treadmill test. The ADLC consisted of four different stages, each 3 min in duration. During stage 1, the subject sat quietly in a straight backed chair after being connected to the AMS-1000 telemetry unit and metabolic cart. Stage 2 consisted of typical activities performed while shopping for groceries, e.g., pushing a shopping cart around the periphery of the room, loading cans into the cart, carrying two loaded bags around the room, and unloading the cans onto the shelves. During stage 3, subjects vacuumed a section of rug with an upright electric vacuum. The final stage of the ADLC consisted of the subject walking the periphery of the room several times alternating with climbing eight steps.

The ADLC protocol was designed to test subjects at varying levels of daily activity. The sitting provided baseline data on heart rate and˙VO₂. The shopping was typical of many everyday activities because it combined moderate and low levels of energy expenditure. The vacuuming was primarily an upper body activity with minor lower limb involvement, while the walking and stair climbing was more demanding, combining activities with varying degrees of intensity. We estimated that the final stage of the ADLC would be representative of the maximum heart rate and ˙VO₂ people obtained during daily nonexercise situations. An examination of the ADLC data showed that the average ˙VO₂ for each subject during stage 4 exceeded his or her ˙VO₂ levels during the early stages of the treadmill. Stage 4 data were subsequently omitted from analysis to avoid confounding analyses designed to compare heart rate and ˙VO₂ during low expenditure activities.

The subjects were allowed a brief rest between the ADLC and the treadmill test while the telemetry unit was removed and the standard 12-lead ECG was attached. Before beginning the second test, subjects were allowed to practice walking slowly (2 mph) on the treadmill set at 0% incline for about 1 min. The treadmill test consisted of seven 3-min stages. Stages 1 through 4 were performed at 3% incline with speeds increasing from 2.4 to 3.6 mph. During stages 5 through 7, the speed of the treadmill remained at 3.6 mph while the incline was increased to 6, 9, and 12%, respectively. The treadmill test continued until 1) the subject reached 80% of her/his maximum heart rate, defined as 220 minus age, 2) it appeared the subject would exceed 80% of maximum heart rate early in the next stage, or 3) she/he completed the seven stages. The treadmill test was designed to measure HR at walking rather than running levels, thus the speed of the treadmill did not exceed 3.6 mph. This design ensured that data could be collected on all participants for at least three stages (9 min), regardless of their fitness level.

Data analysis. Mean 30-s motion counts were calculated from the AMS-1000 data for both the ADLC and treadmill tests, converted to an ASCII file, and imported to a SAS (version 6.03, SAS Institute, Cary, NC) data base. A determination of heart rate was made for each test by manually counting the beats per minute from the strip chart recordings of the ECG wave forms from the telemetry unit or the standard 12-lead ECG.

Data analysis was performed using programs available on SAS. Measured oxygen consumption was compared with ˙VO₂ as predicted by heart rate from the two monitors and by activity from the motion sensor for both the ADLC and treadmill tests. Individual and group general linear models were created for the prediction of ˙VO₂ from heart rate and/or motion data. Differences between r² for the prediction of ˙VO₂ from HR alone or in combination with motion were tested by partialF-test.

RESULTS

Table 2 presents the mean values (± SD) for˙VO₂, heart rate, and motion counts for each stage of the ADLC and treadmill tests. The recorded motions per minute exhibited significantly greater variability across subjects than either ˙VO₂ or heart rate(P < 0.001).

The correlation coefficients for the individual calibration equations for the prediction of ˙VO₂ from heart rate or motion alone and in combination are presented in Table 3. The data inTable 3 indicate that motion alone was a poor predictor of ˙VO₂ in both activity tests, whereas heart rate was a better predictor (ADLC r² = 0.81, SEE = 3.25 ml·kg^-1·min^-1; treadmill r² = 0.90, SEE = 2.76 ml·kg^-1·min^-1). Oxygen consumption as predicted by heart rate was greater than measured ˙VO₂ during the ADLC by 4.5 ± 8.3% (range -16 to +12% for individuals) and during the treadmill test by only 1.0 ± 3.7% (range -5 to +7% for individuals). There was no difference in the prediction of ˙VO₂ by heart rate alone or heart rate plus motion during the treadmill test. In contrast, the r² for each individual calibration equation for the ADLC, except Subject 10, was greater with motion included in the regression model, although the increase in r² from 0.81 to 0.86 was not significant for the group as a whole (P = 0.23).

Following the example of Haskell et al. ⁽⁷⁾, we tested whether a single calibration equation for the prediction of ˙VO₂ from heart rate and motion was feasible. We constructed a general linear model for our population by pooling all subjects' heart rate and motion data from both the ADLC and treadmill tests (r² = 0.45). To test whether heart rate above a resting level could be predictive of ˙VO₂ in a single group equation, we calculated the difference between the baseline heart rate for each subject and the heart rate measured during each subsequent sampling period. Baseline heart rate was defined as the lowest heart rate achieved during stage 1 of the ADLC. Thus, when the difference in heart rate, rather than observed heart rate, was used in general linear models for individuals, the mean r² = 0.85 (SEE = 2.95 ml·kg^-1·min^-1), and in a single general linear model for pooled data, r² = 0.74 (SEE = 4.16 ml·kg^-1m·min^-1). While the correlation from the pooled data is reasonably precise for the group as a whole, the˙VO₂ predicted for individuals from this equation differed from measured ˙VO₂ by -25.6% to +14.1% during the ADLC and by -22.5% to+27.9% during the treadmill test.

DISCUSSION

This small study on healthy adult subjects was undertaken to test the relative accuracy of simultaneous heart rate and motion monitoring for the estimation of oxygen consumption during low and moderate intensity activities. Similar to the findings of Haskell et al. ⁽⁷⁾, motion alone was a poor predictor of ˙VO₂, and heart rate plus motion did not enhance the predictions for the moderate intensity treadmill test over heart rate alone. The combination of heart rate and motion, however, did predict oxygen consumption better during the lower intensity ADLC in the present study. On average, the correlation coefficients for ˙VO₂ and heart rate alone were quite high for the treadmill test and thus did not allow for much improvement on these predictions (mean r² = 0.90). This mean correlation coefficient was slightly lower than has been observed previously in moderate-to-high intensity exercise by others where group means for r² are typically around 0.95^(7,13,20). While most previous investigation with moderate-to-high intensity exercise has been conducted among well trained individuals, the subjects in this study were not especially aerobically fit; if more subjects had been able to complete the seven stages of the treadmill test, the overall relationship between heart rate and˙VO₂ may have been improved. Similarily, the mean correlation coefficient (r² = 0.81) for ˙VO₂ and heart rate during the ADLC is comparable to reports from previous investigations involving low intensity activities (r² from 0.75 to 0.91)^(2,11,13).

Although the improvement in prediction of oxygen consumption with the addition of motion to the regression analyses was not significant for the group as a whole, correlations for individuals improved as much as 0.13. Movement during both sets of activities, as measured by the motion sensor, was much more variable among subjects than either heart rate or ˙VO₂. A positive, albeit not significant, relationship existed between the number of motions per minute and the improvement in r² during the ADLC, suggesting that the capacity of movement to enhance prediction of oxygen consumption is influenced by the type of activity and the individual's response. Since it is during low-intensity activity, as well as very high intensity activity, that heart rate fails to have a linear relationship with oxygen consumption or energy expenditure and the vast majority of the average person's physical activity is of low intensity throughout the day, the addition of one or more motion sensors to heart rate monitoring for measurement of daily energy expenditure appears be warranted.

The use of heart rate monitoring for the prediction of oxygen consumption and energy expenditure is predicated on the need for individual calibration curves, and the addition of motion sensors does not appear to affect this requirement. The amount of data management required for this small study demonstrates that the creation of individual calibration curves for subjects in an epidemiologic field setting would be a daunting task. While we did not keep records on how much time was spent on data editing for each subject, it clearly exceeded 2 h of manual editing in addition to the extensive automated data checking. An alternative approach might be the use of a standard equation for all participants. To test for loss of information that might accompany this procedure, we pooled the heart rate and motion data for all subjects and created a single equation for the prediction of ˙VO₂. Unfortunately, as has been observed, the relationship between heart rate and oxygen consumption is highly individual ^(7,11), and the results of this pooling effort were not promising, yielding an r² of only 0.45. When a pooled equation was developed for heart rate alone, a similar result was obtained. This correlation was worse than that reported by Haskell et al. ⁽⁷⁾. However, when we used the calculated difference between resting and measured heart rate, the predictive value increased significantly (r² = 0.74). This equation created from the pooled data, however, produced individual estimates of ˙VO₂ that differed from measured values by as much as 25%, suggesting that the pooled equation would be unusable in an intervention trial where the comparison of oxygen consumption across time in the same individual is crucial. In addition, even if a single prediction equation were possible, the feasibility of collecting information in a noncontrolled study is questionable. In our study, we could use the more reliable laboratory data to infer which data was unreliable. This capability would not be available for a field study as the unreliable data could not be identified using standard editing techniques.

In summary, the use of a motion sensor in conjunction with heart rate monitoring was a better predictor of oxygen consumption in a laboratory setting that simulated activities of daily life than heart rate alone. The addition of motion to heart rate did not improve the prediction of˙VO₂ for the higher intensity exercise of a submaximal treadmill test. The use of a single motion sensor by itself did not adequately predict oxygen consumption for use in epidemiologic studies. The specific instrument tested, the AMS-1000, provided adequate motion data but inconsistent heart rate data in our laboratory. Thus, for further studies we would consider using a separate heart rate monitor and motion sensor(s) with omnidirectional mercury switches. Finally, although it is time consuming and a significant subject burden, it is necessary to develop individual calibration curves for the relationship of heart rate and movement to oxygen consumption. This suggests that heart rate or heart rate plus motion are not viable methods of estimating energy expenditure in large epidemiologic studies but may be useful in small-to-moderate sized intervention trials.