Patients scheduled for major elective intra-abdominal surgery at the University Hospital of Wales between June 2017 and February 2018 were included in the study. Inclusion criteria included: aged 18 years or older, capacity to consent and a clinical indication for planned CPET before elective major surgery. Exclusion criteria included: atrial fibrillation, nickel allergy, unable to wear a watch, unable to undergo CPET and pregnancy. Baseline information on age, gender and body mass index (BMI) were recorded. Information about the study was sent by post along with the CPET appointment. Patients had a discussion about the study on the day of attending their CPET appointment and consent taken prior to their test.
Our original sample size was chosen pragmatically to be 100. This would have allowed the estimation of any feasibility proportion to within at least plus or minus 9.8 percentage points using a 95% confidence interval
Preoperative CPET was conducted and interpreted by a consultant anaesthetist experienced in CPET in accordance with national guidelines ii using an electromagnetically braked cycle ergometer (Lode, Gronigen, The Netherlands) and a Medgraphics Ultima metabolic cart (MedGraphics, Gloucester, UK). Calibration was undertaken in accordance with manufacturer's guidelines using a 3 litre syringe (Hans Rudolph, Kansas City, KS, USA) and reference calibration gases. During data collection, the middle five of seven breaths were averaged. An exercise protocol was used whereby patients cycled at 60 r.p.m. for 3 min in an unloaded freewheeling state, followed by a progressively ramped period of exercise (from 5 to 15 W min−1 based on mass, stature, age and sex) to volitional or symptom-limited termination, followed by 3 min recovery. Medgraphics Breeze software automatically determined peak oxygen uptake (VO2peak; defined as the highest O2 uptake during the final 30 s of exercise reported). The AT was manually interpreted using the V-slope method (Beaver, Wassermen, & Whipp, 1986) and supported by comparison of end-tidal oxygen tension (ETO2) and ventilatory equivalent for oxygen (VE/VO2) plots. The ventilatory equivalent for carbon dioxide (VE/VCO2 ) was identified at the AT or was recorded as the gradient of the linear VE/VCO2 relationship if the AT could not be identified.
AT was not recorded if AT was not reached during the test or the Respiratory Exchange Ratio (RER) was persistently >1.0 during the exercise test precluding the determination of AT.
All participants underwent CPET as part of their routine pre-operative workup. Four key CPET measures of activity were recorded: peak oxygen consumption (peak VO2), the ventilatory equivalent for carbon dioxide (VE/VCO2 slope), the anaerobic threshold (AT) and peak work (also known as peak power output). The following thresholds were used: 14ml/kg/min was used for the peak VO2 threshold, 34ml/min for the VE/VCO2 slope and 11ml/kg/min for the anaerobic threshold.[v] A median split was used to divide recorded peak work values.
In addition to this, all participants wore a wearable device continuously for 7 days prior to their surgery. The wearable device used was the Garmin Vivosmart HR+ smart activity tracker. A participant information leaflet was sent to all eligible patients one week prior to attending a routine CPET clinic. The devices were then issued to participants after they had completed their CPET assessment. The devices were worn continuously for the 7 day study duration and were then removed and stored until return during the next routine clinic appointment. Patients were given clear advice regarding wearing the device and issued with a charger in case of power failure, although the battery of the device was sufficient for 7 days of continuous use. The device recorded resting heart rate, average heart rate, maximum heart rate, total steps, floors climbed, number of intense minutes of exercise, total calories and total distance travelled. These variables were averaged across the 7-day period prior to analysis.
After the 7 days, participants completed the International Physical Activity Questionnaire (IPAQ) which identified the metabolic equivalent task (MET) minutes achieved in different domains (work, transportation, domestic, garden, leisure time) by self-report.
The total MET minutes of physical activity per week was computed by summing the MET minutes from each category (walking, moderate or vigorous activity).
The total time per week in each category was then multiplied by a constant depending on the level of intensity of the exercise: 3.3 for walking, 4.0 for moderate-intensity activity and 8.0 for vigorous-intensity activity. Once the MET minutes per week for each category were computed, these were summed to produce the total physical activity MET-minutes/week and the IPAQ global score; the patients were then categorised as having a high, moderate or low level of physical activity.
Descriptive data is presented as mean ± standard deviation. Statistical modelling was used to explore the relationship between parameters provided by CPET, self-reported activities via the IPAQ and activity measurements recorded by the wearable device.
Graphical exploration of the wearable data was performed, alongside range checks.
Incomplete data capture was possibly due to sub-optimal fit of the device or movement of the device on the wrist during capture, with the device documentation suggesting that a issues such as sweat, lotion or sunscreen on the wrist, fit of device (location on the wrist and tightness of the strap) and intensity of the activity can contribute to limitations of data capture. In future these issues could be address by repeated checks to ensure correct fit and care of the device at all times and continued patient education on optimal use of the device.
Correlations between variables were calculated using Pearson correlation coefficients. With Pearson correlation assumptions met, via scatter plots demonstrating linear covariation and visual inspection of Q-Q plots. Linear regression was used to explore whether the eight wearable variables could be used to estimate values for the four CPET measures of activity. In addition, further analysis explored whether the addition of IPAQ global scores could improve this estimation. Three models were fitted for each of the four measures of CPET activity, with all including basic demographic variables (age, gender, BMI). The first model used only the eight wearable values as predictors, the second used only the global IPAQ score and the third model used both of these components together.
The models were compared using the Akaike Information Criteria (AIC). Standard model diagnostics were explored to ensure adequate model fit, including fitted versus residual plots. Models were assessed and compared using appropriate statistics, including R2 and adjusted R2 values, the percentage of correct predictions and Pearson correlation coefficients (with associated 95% confidence intervals) between the fitted and observed values. Receiver Operator Characteristic (ROC) curves for these models were also compared.