2.1 Recruitment
This cross-sectional study was conducted between 2012 and 2015 at a rehabilitation clinic in the Stockholm area (Sweden). Participants (n = 64) were recruited through various channels, including advertisements on websites and organisations specific to SCI and through word-of-mouth and referrals from the regional SCI unit. Interested individuals contacted the research group via email, and eligibility was confirmed through a phone call with the researcher. Medication and injury level information was obtained, and data collection procedures were explained. If there were any uncertainties about inclusion or exclusion criteria, the researcher requested permission to consult the attending physician or sought the participant's help. All necessary information was provided to the participants by email or regular mail, and informed consent was obtained from each participant on the first day of data collection. Inclusion criteria for the study required motor-complete tetraplegia with a level of injury between C5-C8 and American Spinal Injury Association impairment scale (AIS)21, A + B, or motor-complete paraplegia with a level of injury between T7-T12 and AIS A + B22. Participants had to be 18 years old and at least one year post-injury, with absent or minimal self-reported spasticity on the PENN spasm frequency scale. Exclusion criteria included a known history of coronary artery disease, angina, chronic congestive heart failure, resting hypertension, chronic obstructive pulmonary disease, shoulder pain, or other known conditions that could limit exercise ability.
2.2 Ethics
The present study was carried out according to the ethical standards in the 1964 Declaration of Helsinki. The Stockholm regional ethics committee also approved this study, reference number 2011/1989-31/1. All the participants gave their informed written consent to participate. All methods were performed following the relevant guidelines and regulations.
2.3 Oxygen uptake measurements
The participants were asked to refrain from smoking and vigorous activity for 12 h before testing. A Jaeger Oxycon mobile system (Jaeger Oxycon mobile system, Hoechberg, Germany) was used to collect data on oxygen consumption (VO2 and VCO2) mean VO2 during the arm cranking test. Participants wore a face mask with a ventilation turbine and a sampling tube to collect breath-by-breath data, which was sent to a portable housing unit to analyse oxygen and carbon dioxide content, which was verified with reference gases and room air before the start of each test. A variation of 3% for each steady-state measurement was accepted 23. Heart rate data were collected using Polar chest straps. Each test lasted at least six minutes, and the last three minutes of oxygen consumption data were used to calculate absolute and relative oxygen consumption and heart rate. The mobile system was validated for lower oxygen consumption levels before testing to compare the accuracy of the mobile system with a traditional Douglas Bag method. Ten individuals sat quietly and breathed in the facemask connected to both devices for 10 minutes each. The results showed no significant difference between the two methods, with a mean error of -0.27 range of 0.22–0.32 L·min− 1 and a mean variation of 3%.23
The submaximal cardiorespiratory capacity was tested using the same arm ergometer (Ergommedic 89E Monark, Sweden) placed on a height-adjustable table to achieve the best shoulder position. Participants were allowed to use upper-body straps if they had poor balance or gloves if they had poor hand function. The test began with 6 minutes at a low workload of 18–24 Watt for paraplegia and 5–10 Watt Borg RPE 9–12. The workload was then increased to 36–42 Watt for paraplegia and 10–25 Watt for tetraplegia, with Borg RPE 13–14, for at least 6 minutes. Steady-state was achieved within 4 minutes at the higher work rates, and since continuous work was performed, steady-state was reached even faster at lower work rates.
The VO2peak test protocol was designed based on international laboratory protocols24 in previous research25 and was individualised for each participant. The participants positioned their wheelchairs by the arm ergometer (Ergommedic 89E Monark, Sweden) on a height-adjustable table to achieve optimal shoulder position. Participants with poor hand function were allowed to use an upper-body strap or gloves. The test protocol was (eller The Holmlund-Grooten test was) individualised based on the participant's workload and results during the submaximal arm-cranking test. The test began with a 3-minute warm-up with starting workload based on submaximal testing (Table 1). The cadence was individually chosen between 70 and 110 RPM for persons with motor-complete paraplegia and 50 and 90 RPM for those with tetraplegia. The resistance was increased every minute for at least three and up to six minutes. During the last 1–3 minutes of testing, individualisation included increasing the cadence instead of resistance and was based on Borg RPE and visual assessment. Oxygen consumption (VO2 L·min− 1) and HR were continuously measured and analysed in ten-second averages. VO2peak was determined based on the highest mean oxygen consumption value for 30 seconds. For the test to be considered eligible as VO2peak, participants had to achieve a test time of at least 6 minutes, levelling-off despite increased resistance or cadence (RPM), and a Borg RPE score > 16 supported by respiratory exchange ratio (RER) or respiratory quotient (RQ) above 1.1.
Table 1
Starting workload and the individual workload
Submaximal workload (Watt) | Starting workload (Watt) | Increased workload (Watt) |
10–15 | 10 | 0.25 |
20–25 | 10 | 0.50 |
36 | 24 | 0.50 |
42 | 36 | 0.75 |
2.4 Data analysis
Due to missing data for HR (11%), Watt (11%) and Borg RPE (40%) during the submaximal testing, we imputed these data using automatic analysis. The analysis included five imputations and used linear regression models for scale variables. The models included no interactions, and the maximum percentage of missing values was set to 100%, while the maximum number of parameters in the imputation model was specified as 100. The model was based on descriptive statistics for participant characteristics and the physiological variables (HR, oxygen uptake, RER) collected during the submaximal and maximal arm crank tests. After imputation, the HR data showed a small, non-statistical increase, with the overall mean change from 112.6 to 111.9 beats per minute (bpm). The standard deviation (SD) decreased from 24.7 to 23.7, leaving the data's minimum and maximum values constant. The imputation for Watt changed the mean and SD from 32.6 (10.8) to 31.6 (10.8). Moreover, the imputation of data for Borg RPE changed the overall men from 13.4 to 13.3, and the SD of the original data increased from 1.7 to 2.5. This indicates that imputation has successfully improved the overall accuracy, reduced the variation in the data for heart rate and Watt and brought the mean of the data closer to the actual mean without altering the range of the data.
2.5 Submaximal model development
The physiological responses for exercise for individuals with a mcSCI are related to the level of injury; all cervical and thoracic injuries down to T6 have compromised physiological responses, which need to be considered in the model. Correlation analyses were performed, which included sex, age, years since injury, level of injury (continues,) weight, height, and BMI. All variables were introduced in a multiple linear regression analysis with VO2peak as the dependent variable. Independent variables were added with forward selection in order of significance. Moreover, injury level was introduced as a continuous variable in automatic linear modelling, which merged the variable to three different dichotomous levels for best fit by introducing 0 = C5-C8, 1 = T7-T11 and 2 = T12 in the regression model. A significant difference in relative VO2peak mL·kg·min− 1 (95%CI -6.5;-1.9) and no significant difference in body weight and BMI further strengthened the dichotomisation of the injury level for paraplegia. Moreover, T1-T2 could be classified as a cervical injury since the muscles innervated by nerves from these areas do not contribute to VO2peak measures. At the same time, individuals with T3-T6 could be merged into the T7-T11 group since breathing muscles and heart rate regulation are involved.21
2.6 Test procedures for the Holmlund-Grooten submaximal testing
The table for the test person was adjusted to align with the shoulder. For individuals with poor hand function, gloves were used, and if necessary, a strap for better balance in the wheelchair. The test began with three minutes of warm-up with a low workload in the correct cadence, 60 revolutions per minute (rpm) for paraplegia and 50 rpm for tetraplegia. After five minutes, the test workload was adapted based on Borg RPE for central exhaustion and heart rate. The higher workload was then used for another 5–7 min, and the last 3 minutes were used for steady state. A test protocol for the Holmlund-Grooten test for wheelchair users is available in Supplementary file S1.
2.7 Statistics
Test and demographic data were described in means and standard deviations (SD), min/max, and categorical variables in proportions for continuous variables. A multiple linear regression model based on test data and the abovementioned demographic variables was used and constructed to predict VO2peak. The model added these independent variables in order of significance into the VO2peak estimation model using a probability of F = 0.05 for entry and 0.10 for removal. The model was tested for the assumption that there is a difference in VO2peak and heart rate between tetraplegics and paraplegics and that VO2peak is normally distributed. The variance inflation factor (VIF) was used to measure the degree of multi-collinearity in the regression model, where higher VIF values indicate a higher degree of multi-collinearity. Beta-coefficients were interpreted as small (0-0.29), medium (0.30–0.49) and large (> 0.50) associations.26 Scatter plots of the included variables were used to inspect the data visually and to detect outliers and influential data points that could compromise the results.
We plotted a simple linear regression line to test the validity of the correlation between estimated and measured VO2peak for the different injury levels. Moreover, we calculated the absolute difference between predicted and measured VO2peak for each participant. A one-sample t-test was used to test if the mean of the total sample differed from zero. Intraclass correlation coefficients (ICC[3,1]) with a 95% confidence interval (95% CI) were used to establish the concurrent validity between the predicted and measured VO2peak, i.e. a two-way model using random effects with absolute agreement. Random effects were chosen since we assumed that raters were a random sample from a larger population. Measures of the absolute agreement were chosen since we considered that the systematic differences were relevant. Based on the 95% confidence interval of the ICC estimate, values less than 0.5, between 0.5 and 0.75, 0.75 and 0.9, and greater than 0.90 indicate poor, moderate, good, and excellent reliability, respectively.27 We also calculated the SEM as a measure of validity by multiplying the standard deviation (SD) of the difference between the predicted and measured VO2peak with the square root of (1-ICC). The SEM was then used to calculate the Minimal Detectable Change (MDC) of the prediction model with the following formula: MDC = 1,96·√2·SEM. The MDC indicates the minimal change that can be interpreted as a fundamental change in VO2peak for each individual; a smaller MDC indicates a more sensitive measure.28 As a final step, a visual analysis of the Bland-Altman plot29 revealed a possible systematic error between the predicted and measured VO2peak. In a Bland-Altman plot, the difference between predicted and measured VO2peak are plotted against the mean of the predicted and measured VO2peak and the mean difference and the limits of agreement (LoA); mean ± 2SD. All statistical analyses used a significance level of p ≤ 0.05 (two-tailed) and were performed using SPSS 24.0 software and MS Excel.