Physical Fitness Training in Patients with Subacute Stroke (PHYS-STROKE): multicentre, randomised controlled, endpoint blinded trial

Abstract Objective To determine the safety and efficacy of aerobic exercise on activities of daily living in the subacute phase after stroke. Design Multicentre, randomised controlled, endpoint blinded trial. Setting Seven inpatient rehabilitation sites in Germany (2013-17). Participants 200 adults with subacute stroke (days 5-45 after stroke) with a median National Institutes of Health stroke scale (NIHSS, range 0-42 points, higher values indicating more severe strokes) score of 8 (interquartile range 5-12) were randomly assigned (1:1) to aerobic physical fitness training (n=105) or relaxation sessions (n=95, control group) in addition to standard care. Intervention Participants received either aerobic, bodyweight supported, treadmill based physical fitness training or relaxation sessions, each for 25 minutes, five times weekly for four weeks, in addition to standard rehabilitation therapy. Investigators and endpoint assessors were masked to treatment assignment. Main outcome measures The primary outcomes were change in maximal walking speed (m/s) in the 10 m walking test and change in Barthel index scores (range 0-100 points, higher scores indicating less disability) three months after stroke compared with baseline. Safety outcomes were recurrent cardiovascular events, including stroke, hospital readmissions, and death within three months after stroke. Efficacy was tested with analysis of covariance for each primary outcome in the full analysis set. Multiple imputation was used to account for missing values. Results Compared with relaxation, aerobic physical fitness training did not result in a significantly higher mean change in maximal walking speed (adjusted treatment effect 0.1 m/s (95% confidence interval 0.0 to 0.2 m/s), P=0.23) or mean change in Barthel index score (0 (−5 to 5), P=0.99) at three months after stroke. A higher rate of serious adverse events was observed in the aerobic group compared with relaxation group (incidence rate ratio 1.81, 95% confidence interval 0.97 to 3.36). Conclusions Among moderately to severely affected adults with subacute stroke, aerobic bodyweight supported, treadmill based physical fitness training was not superior to relaxation sessions for maximal walking speed and Barthel index score but did suggest higher rates of adverse events. These results do not appear to support the use of aerobic bodyweight supported fitness training in people with subacute stroke to improve activities of daily living or maximal walking speed and should be considered in future guidelines. Trial registration ClinicalTrials.gov NCT01953549.


Study Design
The study was designed to test the efficacy of an aerobic physical fitness intervention that can be administered in a clinical setting. Thus, trial interventions were directly applied at neurorehabilitation clinics next to usual care. Several aspects of the trial design follow principles of pragmatic trials. Accordingly, we have evaluated the PHYS-STROKE trial regarding the nine dimensions proposed by the Pragmatic-Explanatory Continuum Indicator Summary 2 (PRECIS-2) tool for assessing the level of pragmatism in a trial (1). As a result, PHYS-STROKE should be regarded as a rather pragmatic trial (see Figure S1). Figure S1: The Pragmatic-Explanatory Continuum Indicator Summary-2 (PRECIS-2) wheel of the PHYS-STROKE trial. Nine dimensions display the level of pragmatism of a trial with scores ranging from one (very explanatory) to five (very pragmatic).

Study protocol
The English version of the study protocol 1 was first published in the 'Trials' Journal:

Study protocol: Amendments
In the course of the study the ethics committee of the Charité-Universitätsmedizin Berlin approved four amendments (02.09.2013; 23.04.2015; 15.11.2016; 21.08.2017). The following data were added: At follow-up 3 months post stroke, most subjects are still in inpatient clinical care where daily variation of physical activity is limited. Variation of physical activity is greater in a home setting. Thus, a seven-day accelerometry is considered a more precise assessment at follow-up 6 months post stroke.
The German Freiburg questionnaire on physical activity is validated in a broad age range in the German population and is capable of estimating daily energy expenditure. It is regarded as a valuable measure in addition to the accelerometry measurement to assess longterm effects of the intervention on physical activity.
Two more trial sites are added as a consequence of difficulties in recruitment.
Amendment and Votum 15.11.2016 Audio recording of the Regensburg semantic and phonemic word fluency test.
To increase the explanatory power of the Regensburg semantic and phonemic word fluency test, audio recordings are necessary to distinguish temporal patterns (temporal cluster analysis) of word production.
Amendment and Votum 12.09.2017 Small administrative changes (no additional assessments).
The local PI changed from Dr. Flöel to Dr. Endres due to a change of the affiliation of Dr. Flöel.

Randomisation and masking
Randomisation used Functional Ambulation Category (FAC; dichotomised FAC≤3), centre, and age (dichotomised age at ≤65 years) as strata and was done in a 1:1 fashion. Each strata was organised in blocks of 10 subjects. The randomisation procedure was done by a clinician of each study site after the baseline visit was performed. The assignment was subsequently communicated to the treating therapists. The trial centre was not informed about patient allocation during the course of the trial.

Procedures: Care providers
Each study site assigned at least two therapists per intervention group to the trial, who were responsible for conducting the intervention. Care providers were supposed to be physiotherapists or sport therapists by training delivering the PHYS intervention and neuropsychologists or psychologists for the RELAX intervention respectively. A team member of the trial centre regularly visited each study site to discuss issues regarding the intervention and to control the adherence to adequate documentation. In addition, the trial monitoring controlled for every patient whether intervention documentation was adequate.

Procedures: Intervention manual
A written intervention manual was distributed to all trial therapists containing extensive description of the intervention and documentation procedures. Additionally, a web page with frequently asked questions was maintained. A regular newsletter informed about specific procedures. A member of the trial centre visited each study site to discuss current issues with the treating therapists. The intervention manuals for both, PHYS and RELAX, are described in the following section: The aerobic physical fitness training (PHYS) intervention was designed to improve cardiorespiratory fitness while being functional in terms of applying locomotion therapy. The goal was to reduce deconditioning and enhance endogenous neuroplasticity following a cerebrovascular event. Cardiorespiratory load was targeted at 50-60 % of the estimated maximal heart rate to ensure an aerobic training in the lower range of recommended cardiorespiratory training intensities endorsed by the American College of Sports Medicine (ACSM) (2).
The target heart rate (THR) was calculated by care providers (therapists) at the beginning of the intervention period based on the formula: '180 -age'. The THR was adjusted with minus 10 beats in case of beta blocker intake. Before the first intervention session patients were allocated to receive training either in a gait trainer (FAC score 0 -2) or on a body-weight supported treadmill (FAC 3 -5). FAC was assessed by the therapists on a daily basis and subsequent training was administered using the respective allocated device.
Before each session, resting (pre-training) heart rate (HR) and blood pressure (RR) was assessed while the patient was still in a resting position. Training was only commenced if HR was below 180 bpm or systolic blood pressure below 200 mmHg, as recommended by the ACSM. Additionally, patients were equipped with a pulse belt for the therapist to constantly monitor the patient's HR.
Gait trainer: The intervention manual refers to the Gait Trainer GT1, Reha-Stim, Berlin, Germany. General recommendations equally applied to other manufacturers. The patient was seated in the gait trainer following the manufacturer's manual. Attention was paid to the individual step length, and ensured that the hemiparetic arm was fixed to the handle bar in front of the patient. Body-weight support (BWS) was adjusted by the therapist, and only given as functionally needed. Training started with a three minute warm-up phase in which the patient started to practice at a low speed. After three minutes, the speed was increased until the THR was reached. After twenty minutes, the speed was reduced for another two minutes to reach a cool-down phase. Training at the THR was possible and constantly achieved due to the following adjustments of treatment modalities: decreasing BWS or increasing speed. The training aimed at delivering a constant cardiorespiratory load of 50 -60 % maximum HR, represented by the THR. In addition patients who recover functionally were able to transition to treadmill based training if FAC was > 2. Treadmill: The intervention manual refers to the Multi-disk treadmill Callis, Model Therapie, SPRINTEX Trainingsgeräte GmbH, Kleines Wiesental, Germany. General recommendations equally applied to other manufacturers. The patient was supplied with the BWS irrespective of need to ensure safety. Body-weight support (BWS) was adjusted by the therapist only as much as functionally needed to allow proper trunk and limb alignment as well as weight shifting on the paretic limb. Patients started at the preferred walking speed over a warm-up period of three minutes. After three minutes, the speed was increased until the THR was reached. After twenty minutes of training, the speed was reduced for another two minutes to the individually preferred walking speed to reach a cool-down phase. Training at the THR was possible and constantly achieved due to the following adjustments of treatment modalities: increase of inclination, reduction of BWS, and increase of speed. The goal was to reduce BWS until 0 % was reached while still allowing for a proper weight bearing of the paretic limb with less than 15° of knee flexion. Patients were able to use handle bars, if needed. Therapists were advised to provide functional support by facilitating hip movement, support weight shifting, stabilizing of the knee, or setting of the paretic limb in case of paresis of the peroneus muscles. Recommendations were given for positioning of staff to ensure ergonomic posture of the therapists during foot placement assistance. Patients were allowed to use orthoses if used also during usual care physiotherapy.
After the training the subject was seated on a chair to assess post-training HR, RR, and perceived rate of exertion during training on a visual analogue scale (0 -10; 10 denotes highest possible exertion). After all equipment had been removed from the patient, he or she was asked if any adverse events occurred during the last 24 hours.
Safety procedures: Throughout the training, HR was monitored by the treating therapist. In case of a HR increase above 180 bpm training was stopped and a physician was contacted. Additionally training was stopped if the patient reported strong pain, constant dizziness, severe fatigue, or a strong urge to urinate. Patients were allowed to take short breaks, but therapists were advised to resume training as quickly as possible. The number and length of breaks were to be decreased over time. To ensure the patient was still practicing below the anaerobic threshold the talk test (3) was administered during each intervention session.

RELAX -Intervention (control)
The relaxation control intervention was designed according to the muscle relaxation after Jacobsen (4) in order to release from overall stress but to restrain from cardiorespiratory load.
A quiet room with either a comfortable chair or bed was used to administer the intervention. Before each session resting heart rate (HR) and blood pressure (RR) was assessed while the patient was still in a resting position. Additionally the patient was equipped with a pulse belt to constantly monitor HR. The patient was positioned to enable relaxation, and soft cushion support was provide to the limbs if needed.
The care provider (therapist) read the commands slowly to the patient (see Table S2). Additionally the patient was encouraged to focus on the feeling of warmth and heaviness in the addressed muscle group. Throughout the intervention session, the therapist ensured that the patient focused on the relaxation, but did not fall asleep.
After 25 minutes of relaxation training, HR and RR were documented and perceived rate of exertion assessed on a visual analogue scale (0 -10; 10 denotes highest possible exertion). After all equipment had been removed from the patient, he or she was asked if any adverse events occurred during the last 24 hours.
The RELAX intervention was stopped if patients reported severe pain, strong fatigue, or the urge to urinate, and continued after the problem had been solved. Tension phase of muscles are supposed to be hold for 5 -10 seconds while relaxation phase is supposed to yield for about 30 -40 seconds. Only exception is for the chest in which the tension phase is much longer. T denotes tension, R denotes relaxation. If in any of the two intervention arms, a session was missed for any reason, study sites were asked to administer missed sessions until twenty sessions were reached. Patients were not allowed to miss training on more than five consecutive days.

Procedures: Protocol adherence
To achieve a standardised intervention regimen, therapists of study centres were trained in the application of the intervention procedures (instructor: Daniela Krohne) and in the documentation of daily intervention diaries following GCP-guidelines (instructor: Regina Schlieder). Quality checks of the intervention diaries were done with every monitoring visit. Additionally, regular visits were conducted at study sites by a member of the trial coordinating centre (Torsten Rackoll) to discuss the enrolment progress and answer inquiries concerning the intervention. Intervention diaries of both groups documented the time of attention of therapists, time spent in the active phase of each session (core intervention), heart rate, and blood pressure before and after the intervention as well as ratings of perceived exertion and adverse events that occurred in the course of the last 24 hours.

Procedures: Standard Care
Standard care was delivered following the German guidelines for neurorehabilitation after stroke (www.barfrankfurt.de). Neurorehabiliation in Germany is categorized into several phases depending on the medical status of the patient and is organized as follows: Recommendations follow the guidelines of the American Heart Association Stroke Council (13). To assure equal administration of therapies we assessed the duration of therapies received per patient between baseline and followup three months after stroke. The length of therapy sessions per patient was recorded by the study site during the period of inpatient rehabilitation therapy. After discharge from the study site, patients documented subsequent outpatient rehabilitation therapies in minutes and presented all documentation at follow-up visits.

Outcomes: Assessments of Secondary Endpoints
All outcome measures were assessed using a standardised manual. Study assessors trained the ratings until they reached an agreement for all ratings in each assessment.

6-min Walking Test (6-MWT):
The distance walked in six minutes (6-minute Walking Test, 6-MWT) was assessed at each centre using hallways not used by other patients or personnel during the test. Thirty-five metres were marked with a clear start and ending mark. The 6-MWT was conducted after the 10-metre gait test, but with a resting period in between. Subjects were asked to walk in a speed in which they would be able to walk safely for six minutes without interruption, but to try to cover as much distance as possible. Patients used the same orthoses and / or walking aids during walking which they used during standard rehabilitative care. An additional person from the assessment team secured every patient's safety and assisted with walking, if needed. The test commenced when subjects started to walk and ended after six minutes. Patients were instructed to turn after 35 metres. The distance walked after six minutes was marked by the assessor. Blood pressure was measured directly after the test and exhaustion was subsequently assessed using a visual analogue scale (zero [0] marking the least possible effort and ten [10] the maximal exhaustion).
Rivermead Mobility Index (RMI) and Rivermead Motor Assessment Subtest Arm: The Rivermead Mobility Index assesses functional mobility with 15 tasks in increasing difficulty. Patients were instructed to perform each task starting with the least challenging one. Each patient was allowed three attempts per task. If a patient was not able to perform one test, he or she was allowed to try the next, more challenging one.
The assessor gave precise instructions and demonstrated the task, if instructions were not understood correctly. The last correctly finished task was counted as the maximum score.
Rivermead Motor Assessment Subtest arm is a subscale ("Upper Limb/Extremity" ('Arm')) of the Rivermead Motor Assessment that determines motor performance of patients after stroke, and consists of 15 arm movements such as pronating/supinating the forearm, bouncing a ball as well as functional items such as cutting putty, grasping and releasing objects, and tying a bow. It consists of test items in three sections that are ordered hierarchically, i.e., first items are easier and become more difficult towards the end of the evaluation. The subtest Arm is assessed as described above for the RMI.

Modified Ranking Scale (mRS):
Trained assessors received information of functional parameters from inpatient care centres and performed all ratings. Actigraphy: GT3x accelerometers were initialized using ActiLife Software, Version 6.8.2 (Actigraph Corp, Pensacola, USA) with 100 Hz sample rate and programmed to record the entire day (24 hrs. starting at 12 a.m.) starting the day after each study visit. The assessors explained the rationale of the devices and asked the patient to wear the device until the morning after the recording started. Patients were requested to wear the device the entire time and only take them off during washing or because of extreme discomfort. In order to guarantee a correct relocation of the device in case it was taken off, patients were shown how to place the device by the assessor. If patients suffered from severe paresis of the upper limb, personnel or relatives were instructed to handle the device. The actigraph was placed on the paretic ankle joint with the device pointing to the lateral side. For return shipping patients were provided with prepaid envelopes. Data were downloaded in 60-second intervals. Wear time validation as well as data scoring for step count assessment were executed using ActiLife software. As a cut-off point, we used more than 60 minutes of continuous zeros, with allowance of 1-2 minutes with counts between 1 and 100. Rated nonwear times were subsequently removed from the analyses. For step count, we applied the company-made low frequency extension filter to discriminate steps of slow walkers from random noise, as suggested by Webber & St.
John. 3 As a cut-off we used the filter for patients exhibiting a walking speed ≤0.4 m/s.

Walking aids
During gait assessments (10m Walk West and 6-minute Walk Test) patients used the same walking aids and/or orthoses as during standard care physio therapy. All walking aids and orthoses were noted by trial assessors. For statistical analysis walking aids were dichotomised as follows.

Box and Block Test:
The Box and Block Test (BBT) (5) assesses gross manual dexterity of the upper limb to determine functional levels of the upper extremity in people with disability compared with those without disability. The test consists of two wooden boxes and a partition in the middle. One of the boxes is filled with 150 squared blocks with a length of the edges of 2.5 cm each. The patient is seated in front of the boxes. After a precise description of the task and verification that the task was understood correctly, the patient is asked to move as many blocks as possible over the partition. He or she is given 60 seconds per hand. After the 60 seconds the examiner counts the number of blocks the patient has moved. If the patient has moved two or more blocks at the same time those blocks are subtracted from the result.
Medical Research Council Scale: The Medical Research Council (MRC) (6) scale is the accepted clinical tool for assessing muscle strength. It is rated on an ordinal scale ranging from 0 (plegic) to 5 (full strength), that has shown high intra-and inter-rater reliability. We evaluated the muscle force of six functional muscle groups (hip extension and flexion, knee extension and flexion as well as ankle extension and flexion) of the lower limb.
Trained study assessors compared the impaired with the non-impaired leg assessed the muscle strength. The force of the impaired leg is noted. Prior to the trial, all assessors validated their ratings against each other and against the trainer.
Resistance to passive movement scale: The Resistance to passive movement scale (REPAS) (7)  The CES-D provides with a lie criteria which rates validity of the answers provided. We used -28 as a cut off and consequently excluded patients with a value below the cut off from analysis.
EuroQol quality of life questionnaire (EQ-5D-5L): The EQ-5D-5L is a questionnaire in which the patients rate (1 = no problem -5 = severe problem) their wellbeing within five dimensions as well as rate the impression of one's health on a vertical visual analogue scale. Questionnaires were filled by each patient at each visit except from those with paresis in their dominant hand.
Calculations of EQ-5D-5L was done using the German validation set recently published. (9) Pittsburgh Sleep Quality Index (PSQI): The PSQI (10) is a measure to rate sleep and asks for reasons for bad sleep. It is filled at each visit. Except for baseline the PSQI asks for sleep behavior during the last four weeks preceding the visit. At baseline the PSQI is filled with respect to the sleep behavior within the four weeks prior to the cerebrovascular event.
Montreal Cognitive Assessment (MOCA): The MOCA (11) is a screening tool for cognitive impairment and has a good validity in the stroke population and it provides with three different versions. It is filled at each visit. Except for the six months follow-up visit a different version is used at each assessment. At the six months follow-up visit the version from the baseline visit is filled.
Trail Making Test (TMT): The trail making test (12) is a neuropsychological test which assesses executive functioning and task switching. The test is administered to each patient at each visit. Cases with an initial neglect symptom were excluded from analysis as the neglect was most likely to overshadow the performance in executive functioning.
Regensburger Wortflüssigkeitstest (RWT; word fluency): The RWT asks for phonemic and semantic word fluency in four tasks. The sum score is equivalent to the words produced within each minute per tasks minus the mistakes. The test is only administered at baseline and at the three months follow-up visit.

Statistical Analysis Plan
A brief description of planned statistical analyses were published in the study protocol (14). The detailed statistical analysis plan (SAP) was published prior to unblinding (SAP published: Sept 1, 2017; last patient's follow-up visit: Nov 1, 2017) and can be found here: https://doi.org/10.6084/m9.figshare.5375026.v1

Statistical Analysis Plan: Changes
Handling of missing values − In the analysis plan from the study protocol published in Trials (14) it was stated to use baseline value imputation in case of missing values.
In the SAP and in the final analyses, we used multiple imputation methods (MICE: multiple imputation by chained equations) for all missing data, except for data missing not at random (MNAR). This approach is regarded as the most appropriate method for missing value imputation in clinical trials (15).

Analyses of safety outcomes
− In the SAP, it was stated to use Fisher's exact test when analysing safety outcomes.
In the primary analyses we used Poisson regression models instead to calculate incidence rates and incidence rate ratios. These models allow accounting for the time at risk for each patient, which differed in case of individual early termination of the trial.

Statistical Analysis Plan: Imputation
Taking into account that our target population was moderately to severely affected, we expected missing values.
Handling of missing values and definitions of missings are described here (15,16): Occurrence was expected for follow-up timepoints. We imputed missing values with multivariate imputation by chained equations (17). Imputation was planned to be conducted using the 'miceadds' package of R Statistical Software. Dependent on the variable with missing values groupwise imputation was conducted using that variable with baseline and follow-up data, and additionally the variables centre, sex, age, baseline FAC, and treatment received. We imputed data on the basis of 10 datasets using the r-command: mice(dat_for_imp, m = 10, seed = 123) With default settings where 'dat_for_imp' denotes the selected variables. The seed was set at '123' to ensure data analysis reproducibility. Missing completely at random (MCAR) Occurrence was expected for logistic reasons or failure of measurement equipment. Thus no observed data was at hand to explain missings. Missings termed MCAR were treated in the same way as described for MAR. Missing not at random (MNAR) Occurrence was expected for patients unable to be assessed for the reason of general or specific impairment. For baseline we imputed missing values with single value imputation (18). If patients were not able to conduct an assessment due to their impairment as was seen in walking related assessments, single value imputation was used using half of the lowest value observed in the entire cohort. We decided to take the half of the lowest value of the group for patients that were not able to perform the desired task (MNAR), so that the mean of imputed values is lower than the mean of the rest of the cohort. This simple approach generates a very small individual maximal walking speed reflecting the moderately to severely impaired cohort without defining zero as a maximum walking speed at baseline, or loosing data points.

Data sharing
All scripts for the primary endpoint analyses as well as the corresponding datasets can be found here: http://doi.org/10.5281/zenodo.3341240

Protocol adherence
At one study site (Charité, Campus Benjamin Franklin) physiotherapists delivered all study interventions (PHYS and RELAX). In all other participating study centres, care providers were available on site delivering the respective study interventions (physiotherapists or sport therapists for PHYS; psychologists and neuropsychologists for RELAX). Heart rate post-session in bpm, mean (SD) 94 (14) 71 (10) Heart rate delta in bpm, mean (SD) 15 (9) -2 (3) Blood pressure pre-session in mmHg, mean (SD) 126/75 (11/9) 126/74 (  We performed a sensitivity analysis to compare adherence to PHYS intervention regimen in severely impaired patients (FAC 0 -1) to the rest of the training cohort (FAC 2 -5): Tables S7: Sensitivity analysis comparing training responses between severely impaired patients to the rest.
FAC 0 -1 (n = 41) FAC 2 -5 (n = 63) Sessions, mean (SD) 16 (7) 16 (6) No. of stopping intervention due to fatigue, mean (SD) 3 (4) 2 (3) Minutes per session, mean (SD) 21 (5) 21 (4) Visual analogue scale, mean (SD) 6 (2) 5 (2) Delta heart rate, mean (SD) 15 (9) 15 (9) Protocol adherence: Target Heart Rate assessment We compared the pragmatic PHYS-STROKE approach to estimate the desired target heart rate (THR) of 50 -60% of a patient's maximum heart rate (maxHR) to a conventional approach from the American College of Sports Medicine (19) which was also used in previous stroke trials (20) calculating 55% of an estimated maxHR (see Table S7). The data presented in Table S8 demonstrate that the PHYS-STROKE approach resulted in a higher desired THR compared to the conventional approach aiming for 55% of an estimated maxHR. This difference was observed throughout all age groups, but was particularly pronounced in younger individuals. The mean age of patients of the PHYS group was 69 (12) years. Therefore, it can be assumed that sufficient cardiovascular stress was induced in patients of PHYS-STROKE allocated to aerobic physical fitness training. The pragmatic PHYS-STROKE approach calculated the target heart rate (THR) with the formula: THR = 180age. The conventional approach calculated 55% of an estimated maximum heart rate (maxHR) with the formula: THR = 0.55*(207 -0.7 * age).

Outcomes: Intervention facilities
Four out of seven study sites had a gait trainer available. We performed a sensitivity analyses on severely affected patients (FAC < 3) who were not trained on a gait trainer as described in the protocol but no difference to the primary outcomes were observed.

Outcomes: Imputation
Data needed to be imputed for missing values due to attrition or calculated for patients too severely affected to complete the assessment of gait. Data imputation displayed is for co-primary and key secondary outcomes. Calculated data for gait speed of patients not able to walk ten metres at baseline by using half of the lowest value of the cohort, no (%). Assuming missing not at random MNAR. We performed a sensitivity analysis to assess the quality of our multiple imputation approach by giving the imputation a hierarchical order as this has been suggested previously (21). We used centre as the highest order, to impute missing data in a groupwise fashion. However, no improved assumption to missing data was found using hierarchical order as outlined in Table S9. Although it seems reasonable to cluster data in a hierarchical order, in our case at least, centre did not improve the outcome, most likely because of the low number of data in each centre. Outcomes: Exploratory analysis of subgroups Figure S2: Subgroup analyses demonstrating the difference in maximal walking speed and Barthel-Index (follow up three months after stroke -baseline) as a function of Functional Ambulation Category (FAC, panel A) and the time from stroke onset to start of intervention (Time to intervention, panel B), respectively, using splines.

Standard care
Times of standard care therapies depend on the degree of functional disability. Data have thus a skewed distribution.