Training Activities

The methods in this paper were first developed in the context of fMRI data analysis, and our examples will come from this domain. A simple way to apply the analyses to fMRI data is to use as activity estimates ( ${\widehat{\mathbf{U}}}^{\left(m\right)}$ ) the regression coefficients, or “beta”-weights, from a first-level time series analysis [36 (link), 37 (link)]. The time-series model accounts for the hemodynamic lag and the temporal autocorrelation of the noise. The activity estimates usually express the difference in activity during a condition relative to rest. Activity estimates commonly co-vary together across fMRI imaging runs, because all activity estimates within a partition are measured relative to the same resting baseline. This positive correlation can be reduced by subtracting, within each partition, the mean activity pattern (across conditions) from each activity pattern. This makes the mean of each measurement channel (across condition) zero and thus centers the ensemble of points in activity-pattern space that is centered on the origin.
Rather than using the concatenated activity estimates from different partitions, encoding analysis and PCM can also be applied directly to time series data. As a universal notation that encompasses both situations, we can use a standard linear mixed model [38 (link)]:
$\mathbf{Y}=\mathbf{\text{ZU}}+\mathbf{\text{XB}}+ϵ,$
where Y is an N × P matrix of all activity measurements, Z the N × K design matrix, which relates the activity measurements to the K experimental conditions, and X is a second design matrix for nuisance variables. U is the K × P matrix of activity patterns (the random effects), B are the regression coefficients for these nuisance variables (the fixed effects), and E is the matrix of measurement errors. If the data Y are the concatenated activity estimates, the nuisance variables typically only model the mean pattern for each run. If Y consists of time-series data, the nuisance variables typically capture additional effects such as time-series drifts and residual head-motion-related artifacts.

We assessed costs from a health system perspective. The unit prices of medical equipment and general consumables were taken from market prices at the time of purchase [18 ]. Annual overhead and building costs of the CCG operations were taken from financial records kept at the supporting clinics. Salaries of personnel involved in the program were collated from a publicly available salary portal [19 ]. Implementation costs related to preparation for launching the CCG intervention were tracked by the central study team and included activities like training CCG teams and developing training materials.

Subjects were randomized into two training groups, applying minimization method (Hu and Hu, 2012 (link)) after the first three visits to laboratory. Training zones were determined based on lactate thresholds. The subjects were instructed to continue their daily physical activities (commuting, non-physical hobbies, etc.), but all strenuous exercise in addition to LIT or HIT was not allowed. The exact training programs are given in Supplementary Material 1.
LIT consisted of cycling under LT₁-power. The weekly 5–6 training days included long (1.5–4 h), medium (1–1.5 h), and short (45–60 min) exercises. Weekly training hours progressed individually (see Progression-paragraph below) based on perceived exertion from 4.5 up to maximum 12.5 h. Subjects did their training mostly outdoors with their own bicycles with Rally RK200 dual-sensing power meters (Garmin Ltd., Taiwan). A possibility for indoors cycling with trainer or Wattbike Trainer (Wattbike Ltd., Nottingham, UK) was given. Three (out from 16) subjects did their training completely indoors, and the others did 3% of their training indoors.
HIT consisted of 2—3 weekly indoor training sessions with Wattbike Trainers, or indoor trainer with their own bicycles with Rally RK200 dual-sensing power meters. Training consisted of high-intensity work intervals 3–7 min long with recovery periods ¾ of the work interval duration. In the first training week, there were 15 min of cumulative high-intensity time in a training session, and it progressed individually (see Progression-paragraph below) up to 30 min per session. Each session included 10 min warm up and cool down. These, as well as recovery periods, were done with power <60 W, and high-intensity segments were initially 110% LT₂ power ( $\pm 15$ W).
In both groups, RPE (0–10) was reported from each training session. HR from each session was recorded with the Garmin HRM-Pro heart belt. In the LIT group, power data were recorded with power meters (Rally RK200), and in the HIT group, power data were collected from the Wattbike Trainer or power meters (Rally RK200). All data was transferred after the session to AthleteMonitoring app (AthleteMonitoring, FITSTATS Technologies, Inc., Moncton, Canada), from which training realization was actively monitored weekly by the first author. In the HIT group, if all weekly training had RPE $\le$ 6 and HR did not rise above LT₂-threshold, the power was increased by 10%. In the LIT group, the subjects were actively given feedback whether training was at the prescribed level. Apart from the first HIT-session, all sessions were unsupervised. Training power was modified in both groups according to VO_2max -test (visit 4) at training week 6.
Daily physical activity was gathered by measuring heart rate from wrist continuously by Garmin Forerunner 945 for two to 4 weeks before the start of the intervention. This was done to estimate the baseline endurance training before training intervention.
Training zones and load. Training was monitored by distributing cycling power output to five zones (Cejuela-Anta and Esteve-Lanao, 2011 (link)): Z1 (below LT₁—10 W); Z2 (LT₁—10 W to LT₁ + 10 W); Z3 (LT₁ + 10 W to LT₂—10 W); Z4 (LT₂—10 W to LT₂ + 10 W); Z5 (above LT₂ + 10 W). For each zone, a weighting factor was linked (in an ascending order: 1, 2, 3, 4, 7.5) and training load was calculated by multiplying the factor by the time spent in the zone (Cejuela-Anta and Esteve-Lanao, 2011 (link)).
Targeted power of LIT was designed to be at Z1 and Z2, and HIT at Z4 and Z5. Heart rate was divided into three zones (HR Zones 1—3) separated by LT₁ and LT₂. In addition, for the LIT group, it was calculated how many times their 30 s -average power exceeded Z2 -zone. For baseline physical activity before the start of the training intervention, Z0 was defined to be a zone below halfway between resting HR and LT₁, and Z1b above Z0 and below LT₁. During physical activity heart rate measurement part of the data was dismissed, 12 $\pm$ 15% of the gathered data, because of heart rate was not adequately recorded from wrist.
Training progression. Both groups had 10 training weeks. Weeks 3 and 7 were load reduction weeks for enhancing recovery. Subjects’ training progression was linked to the perceived exertion. After each training week (excluding load reduction weeks), subjects were asked ‘How much the training has strained your week on a scale of 0–10?’, and the training time was increased more with lower exertion (see Supplementary Digital Content 1). For those finishing the study, adherence rate in training sessions was 98.4%. The weekly training intensity distribution is shown in Figure 2 and training realization in Table 2.