Validation Of An Integrated Heart Rate/Physical Activity Monitor

Abstract

PURPOSE: To validate a novel physical activity monitor (PAM-III) with the ability to collect inertial sensor and heart rate measurement data simultaneously in a single, discreet, low-profile device. METHODS: 19 healthy males (age 26.9 ± 6.4yr.; ht: 179.7 ± 7.2 cm; wt: 79.5 ± 10.4 kg; % fat: 16.0 ± 7.4) and 25 healthy females (age 24.8 ± 4.8 yr.; ht: 166.1 ± 5.2 cm; wt: 62.2 ± 8.9 kg; % fat: 25.9 ± 11.4) underwent two separate exercise sessions on different days that consisted of treadmill walking on a level grade at 2.0, 3.0, and 4.0 mph, and running at 6.0 mph. Subjects exercised for 5 min at each speed, with 3 min rest periods between each bout. Pulmonary ventilation and gas exchange were measured breath-to-breath with a Sensormedics VmaxST portable metabolic measurement system. Peak aerobic capacity was assessed during an inclined walking treadmill test to volitional fatigue. During each exercise session subjects wore the PAM-III device (waistband; heart rate monitored via surface electrodes) and a Polar heart rate monitor. RESULTS: An n-fold cross-validation technique was used to establish the coefficients of a (fixed-structure) linear predictive model for each individual and then to evaluate model performance on unseen data. In particular, the training dataset for each subject excluded a single 30-sec exercise epoch, which was then used to evaluate the model; this process was repeated for all 30-sec exercise epochs for that subject and results accumulated over all subjects. Model inputs were heart rate, rotational kinetic energy, and acceleration vector magnitude and activity counts; model output was total energy expenditure (kcal/ min). Data from the VmaxST were 30-second averages. This model produced a mean r2across all individuals of 0.94 (range = 0.82 to 0.98) and a pooled r2 of 0.95. The mean RMS error across all subjects was 0.76 kcal/min. The second model was a “generic” linear model synthesized across all subjects using an n-fold cross-validation technique. The model was trained using all exercise data from (N – 1) subjects and then evaluated using the exercise data from the excluded subject; the process was repeated for all subjects. Model inputs included heart rate, rotational kinetic energy, acceleration vector magnitude activity counts, height, and weight; the output was total energy expenditure (kcal/min). This model produced a mean r2 across all individuals of 0.91 (range = 0.76 to 0.98) and a pooled r2 of 0.76. The mean RMS error =1.81 kcal/min. CONCLUSIONS: The PAM-III integrated heart rate/physical activity monitor provides very good estimates of energy expenditure during walking and jogging. Supported by R21 CA112323-01A1 and an NIH grant to the GCRC RR00847