Patients randomized to cycling received 30 minutes per day of in-bed cycling, in addition to usual physiotherapy interventions, 5 days per week, during their ICU stay. Cycling continued for a maximum of 28 days or until the patient could march in place for 2 consecutive days, whichever came first.

Patients randomized to usual physiotherapy received interventions according to current institutional practice. Depending on the patient’s alertness and medical stability, usual physiotherapy included activities to maintain or increase limb range of motion and strength, in- and out-of-bed mobility, ambulation, and assistance with optimizing airway clearance and respiratory function [9,18-20].

The primary outcome for this study was the Physical Function Intensive Care Unit Test-scored (PFIT-s), measured at 3 days post-ICU discharge [21,22]. The PFIT-s includes 4 items (arm strength, leg strength, ability to stand, and step cadence), each scored from 0 to 3, summed to a maximum of 12 points, and then transformed to a total score of 10 (Table 1) [21]. Higher scores represent a better function. The PFIT-s was developed for an ICU population and includes functional items commonly performed during physical rehabilitation sessions. Unlike the 6-minute Walk Test, which is challenging to administer during ICU awakening as few patients can walk, the PFIT-s can be measured serially over time [23]. Psychometric studies of the PFIT-s identified a minimal clinically important difference of 1.0 points [21,24]. We chose the PFIT-s because we anticipated that all ICU patients could complete at least part of the assessment, even if they could not stand (eg, arm or leg strength), thereby reducing the risk of floor effects.

The PFIT-s is reliable and valid in critically ill patients, demonstrating strong psychometric properties (reliability range 0.996-1.00 [22]; convergent validity with the 6-minute Walk Test and muscle strength [21]). We selected 3 days post-ICU discharge because it is close to the intervention period and prior studies have documented variable delivery of rehabilitation post-ICU [25], which could influence later evaluations. Tables 2-4 describe the preplanned primary outcome, subgroup, and sensitivity analyses.

Secondary outcomes include performance-based measures, patient-reported outcomes, and those collected by chart review. Performance-based measures include muscle strength (Medical Research Council Sum Score) [26,27] and function (30-second Sit-to-Stand Test [28,29] and 2-minute Walk Test [30]). The 30-second Sit-to-Stand Test and 2-minute Walk Test are reliable in critically ill or frail older adult populations and also have age- and sex-matched norms [29,30]. Patient-reported measures included the Patient-Reported Functional Scale for ICU [31,32], critical care–related psychological distress using the Intensive Care Psychological Assessment Tool [33,34], health-related quality of life using the EuroQoL (EQ-5D-5L) [35-37], and the Hospital Anxiety and Depression Scale [38]. We also collected data on frailty (Clinical Frailty Scale) [39], Katz activities of daily living scale [40], duration of mechanical ventilation, length of stay (ICU and hospital), mortality at multiple time points (ICU, hospital, and 90-day postrandomization), and changes in living location at hospital discharge from baseline. Because of funding limitations, 90-day postrandomization outcomes were restricted to patients enrolled after March 7, 2018. Table 5 outlines our preplanned secondary outcome analyses and their timing.

We collected data on the following adverse events if they occurred during or immediately after in-bed cycling or usual physiotherapy interventions, were attributed by the clinical team to the randomized intervention, and resulted in clinical deterioration of the patient’s status [8,12,48-50]: concern for myocardial ischemia or suspected new unstable/uncontrolled arrhythmia; sustained symptomatic bradycardia (<40 bpm) or tachycardia (>140 bpm); sustained hypertension (mean arterial pressure >120 mmHg); sustained oxygen desaturation below baseline (typically <90% or 88%); marked ventilator dyssynchrony; bleeding at the femoral catheter site; and new bruising at the femoral catheter site. Serious adverse events included unplanned extubation, cardiac arrest, or falls to the knees during usual physiotherapy activities.

We conducted a blinded interim analysis of the first 180 patients enrolled (half of the sample size) to assess benefits and harms, including serious adverse events. We adhered to conservative statistical guidelines for data monitoring based on the modified Haybittle-Peto rule [53]. The Data Monitoring Committee recommended continuing the trial on September 29, 2020, based on this interim analysis. To maintain the overall type-I error rate (ie, α), we evaluated the primary endpoint using a fixed conservative α of .001 for the interim analyses and plan to use α of .05 for the final analysis.

The first publication of the trial results will focus on comparing the cycling + usual physiotherapy group with the usual physiotherapy group once every patient has reached 90 days postrandomization and data on vital status at hospital discharge have been received. Longer-term endpoints for the economic evaluation will be reported in a separate publication. This document will outline only the analyses included in the primary CYCLE manuscript.

Table S1 in Multimedia Appendix 1 outlines the schedule of study procedures with 5 time points for outcome assessments.The ICU awakening time point was based on the physiotherapist’s assessment of the patient's ability to consistently follow 3 out of 5 verbal commands [42]. A patient’s discharge might be delayed for reasons unrelated to their readiness for discharge (eg, unavailability of hospital beds in the transfer ward). Therefore, ICU discharge measures were recorded either when the patient was discharged from the ICU or when a discharge order was written, whichever occurred first. The 3-day post-ICU time point was scheduled for 3 days after the patient’s physical discharge from the ICU. The hospital discharge time point was recorded when a discharge order was written for the patient for the index admission, including transfer to an alternative level of care. The 90-day time point was scheduled for 90 days after randomization.

All statistical tests will be 2-sided and performed at a 5% significance level. We will report 2-sided 95% CIs and conduct all analyses using SAS version 9.4. Our final RCT report will adhere to (1) the CONSORT (Consolidated Standards of Reporting Trials) 2010 Statement for reporting parallel group randomized trials (Multimedia Appendix 2) [54]; (2) the Extension for Reporting Trials of Nonpharmacologic Treatments [55]; (3) the Guidelines for Reporting Outcomes in Trial Reports, The CONSORT-Outcomes 2022 Extension [56]; and (4) the CONSORT 2024 Statement [57].

We will report the total number of patients screened, including those meeting all inclusion criteria and those with exclusion criteria, based on screening logs from participating sites. For eligible patients, we will provide reasons for non-enrollment. Patient withdrawals and losses to follow-up will be documented in our CONSORT diagram (Figure 1).

The following definitions are used in this study.

“Study days” included all days in the ICU from the day of randomization to 28 days postrandomization.

We did not plan for the randomized intervention to occur under the following circumstances:

The remaining days were designated as “Planned intervention days.” On weekdays (ie, nonholiday Monday through Friday), physiotherapists reviewed study patients for 1 or more of the following “Temporary exemptions” before offering the randomized intervention:

If the patient had no “temporary exemptions,” we proceeded with offering the randomized intervention.

Each “planned intervention day” without a “temporary exemption” was considered an “eligible day.” An eligible day where the patient did not receive the randomized intervention was classified as a “missed opportunity.” Missed opportunities may have occurred due to the following:

We define percent adherence as the ratio of days on which patients received the randomized intervention or had a temporary exemption (numerator) to all planned intervention days, including days with the randomized intervention, temporary exemptions, and missed opportunities (denominator). Descriptive statistics on percent protocol fidelity will be reported for the cohort, broken down by randomization group.

If a patient randomized to usual physiotherapy alone received cycling, this was considered a major protocol deviation.

We will include all eligible randomized patients (ie, excluding postrandomization exclusions representing noneligible patients) according to the treatment they were randomized to receive. Analyses of the primary outcome will be restricted to patients who survived to 3 days post-ICU discharge, as specified in our original protocol and sample size calculation [15]. Analyses of the PFIT-s at other time points (ICU awakening, ICU discharge, and hospital discharge) and all performance-based (strength and function) and patient-reported (eg, quality of life) outcomes will be restricted to patients who survived to the respective time point. We will include only patients discharged alive from the hospital in the analysis of hospital discharge location. The analyses of the duration of mechanical ventilation, ICU and hospital length of stay, and mortality will include all enrolled patients.

To determine if there is a difference in PFIT-s score at 3 days after ICU discharge between the cycling + usual physiotherapy and usual physiotherapy groups, we will conduct a linear regression analysis with randomization group (cycling + usual physiotherapy vs usual physiotherapy) as the independent variable [58]. We will adjust for age (≥65 years vs <65 years) and clinical site, as these were used as randomization stratification variables. We will report the results of the regression analysis as the mean difference in PFIT-s scores with corresponding 95% CIs and P values. Although the goal was to have all outcome assessors remain blinded to treatment allocation, this was not always feasible. To maximize the use of available data, we will include all PFIT-s measures at 3 days postrandomization, regardless of the blinding status of the outcome assessor, and will report the proportion of assessments conducted by blinded assessors.

To account for incomplete component data in the PFIT-s at 3 days post-ICU, we will concurrently consider data from the PFIT-s, 30-second Sit-to-Stand, and 2-minute Walk tests. We will evaluate all PFIT-s data components at 3 days post-ICU discharge. We will identify patients with any incomplete physical function data and review the scores for all 4 components of the PFIT-s (ie, shoulder flexion, knee extension, level of assistance required for sit-to-stand, and step cadence). In the PFIT-s, a score of “0” indicates a lack of physical ability to complete the measure. Therefore, if a patient attempts an item and is unsuccessful, the item receives a score of “0”, which accurately reflects their performance (Tables 1). Further details are provided in Tables S2 and S3 in Multimedia Appendix 1. Table 2 describes the primary outcome analysis.

We will conduct 3 exploratory a priori subgroup analyses to investigate potential treatment effect modification for the primary outcome.

In separate linear regression models for each of the 3 subgroup analyses, we will include randomized treatment allocation, the subgroup variable, and the interaction between the subgroup variable and randomized treatment allocation as independent variables. These analyses will be adjusted for age and center. We hypothesize that the treatment effect will be greater for older patients compared with younger patients [59], greater in patients with frailty compared with those without [59], and greater in males compared with females [60,61]. For statistical significance in the subgroup analyses, we will use an α of .10 for the interaction term. We will assess the credibility of any statistically significant subgroup effect using the method described by Schandelmaier et al [62]. These data will be reported in a forest plot.

To assess the robustness of the findings, we will conduct 5 sensitivity analyses for the primary outcome. All sensitivity analyses will be adjusted for age and center, unless specified:

See Table 4 for further details.

For each continuous secondary outcome, we will conduct a linear regression analysis [58]. We will conduct secondary outcome analyses adjusting for age (≥65 years versus <65 years) only. To avoid the risk of overfitting, we will not adjust for center when analyzing our secondary outcomes. We will report the results of the linear regressions as mean differences with corresponding 95% CIs. If needed to normalize the data, we will perform the linear regression on the log-transformed outcome [58]. If the data are still skewed, we will perform nonparametric analyses. As secondary analyses are underpowered and therefore hypothesis-generating, we will not present P values. In Multimedia Appendix 1, we describe the scoring algorithm to account for incomplete data in the 30-second Sit-to-Stand and 2-minute Walk tests based on a patient’s observed function.

We will analyze time to ICU, hospital, and 90-day mortality using Cox proportional hazards regression analysis [63]. We will report hazard ratios and corresponding 95% CIs [63]. All other binary outcomes will be analyzed using logistic regression analysis, reporting odds ratios with corresponding 95% CIs [58]. We will check the assumptions of the different regression analyses by examining residuals and using other relevant methods. Table 5 describes secondary outcomes analyses.

For the safety analysis, we will only include the days on which the patients received the randomized intervention (ie, days at risk of a safety event associated with rehabilitation activities). We will report the frequency and percentage of patients with severe and serious adverse events, by group. We will also report the frequency and percentage of randomized intervention days with severe and serious adverse events, by group.

We will use multiple imputations to account for missing data in performance-based and patient-reported outcomes [64-66]. In the Cox proportional hazards analyses for ICU and hospital mortality outcomes, we will censor patients with incomplete follow-up at the time of last contact.

We will summarize categorical data as counts and percentages. We will summarize continuous data as means, SDs, or median and IQR, if data are nonnormally distributed. For baseline variables, we will not conduct tests of statistical significance between randomized groups; rather, we will note the clinical importance of any imbalance between groups. We will report subgroup analyses in a forest plot.