The main data used in the thesis are from the Stroke Oxygen Study (SO2S). The aims of the
thesis were designed with these data in mind. Although each chapter will give a more detailed
explanation of the data used from the trial for each analysis, an overview is given below of the trial
Chapter 2 26
2.4.1 Design and setting
This trial was a multi-centre randomised controlled trial of oxygen supplementation in
patients with acute stroke. The main trial hypothesis was that a fixed dose of oxygen-treatment
during the first 3 days after an acute stroke improves outcome, with a primary outcome of the mRS
90 days post-stroke. Further hypotheses extended the times at which the outcome was recorded
to 6 months and 12 months post-stroke (Roffe et al. 2014).
This thesis considers the values of the mRS recorded during the whole follow-up period. In
the trial there are values of the mRS, BI and NEADL scores recorded at 90 days (3 months), 6 months
and 12 months post-stroke.
Patients in the study were randomised, based on covariates from the baseline assessment,
using minimisation to one of three groups in a ratio of 1:1:1:
1. Continuous oxygen supplementation for 72 hours
2. Nocturnal oxygen supplementation for 3 nights
3. No routine oxygen supplementation
Individuals who were randomised to receive oxygen were given it at a rate of 2 to 3 l/min,
depending on baseline oxygen saturation.
2.4.2 Data collection
A pilot study for the SO2S was conducted between July 2004 and April 2008, recruiting 300
patients (Roffe et al. 2011, Ali et al. 2013). Following the implementation of minor amendments,
the main study began recruitment in April 2008. Patients were recruited from multiple (>30) centres
throughout the UK and worldwide. Medical centres were eligible for participation in the study if
Chapter 2 27 oxygen saturation, and if there was a local researcher who was willing to act as the principal
investigator for the locality. All adult patients who were admitted to one of the centres with
symptoms of an acute stroke within the preceding 24 hours were eligible to be considered for
participation in the study, as long as there was no indication for, or contraindication to oxygen
treatment in the doctor’s opinion (Roffe et al. 2014).
Patients were not eligible for inclusion in the trial if the responsible doctor considered the
patient to have definite indications or contraindications to oxygen treatment. This decision was left
to the responsible doctor and used to ensure best clinical practice. Patients were also excluded if
the stroke was not the main health problem, or they had another serious life threatening illness
that was likely to lead to death. These patients were excluded as they were unlikely to receive any
benefit from the study intervention. The eligibility criteria for inclusion in the trial reflected this
uncertainty about who should receive oxygen treatment and for how long. This allowed as many
patients as possible to be recruited into the study.
The initial assessment was conducted at baseline by the researcher randomising the
patient. It included the baseline demographics along with date and time of event, the Glasgow
Coma Scale, predictors of outcome, and the NIHSS. An assessment one week after the stroke was
conducted in order to confirm diagnosis. It was performed by a trained assessor seven days (± one
day to allow for weekends and holidays) after enrolment. The assessment documented deaths and
neurological status (NIHSS), compliance with the intervention, and complications arising during the
first week.
Follow-up at 3 months, 6 months and 12 months was conducted centrally in order to ensure
blinding of the assessors. The assessment was a questionnaire sent out to patients at each specific
time point, after checking that the patient was still alive. Non-responders were contacted and were
able to answer the questions by telephone where possible in order to reduce the amount of missing
data and loss to follow-up. The follow-up questionnaire the patient received contained the
Chapter 2 28 using the EuroQol five dimensions questionnaire (EQ-5D),which provides a single index value for
health status as well as being asked questions regarding memory, sleep, speech and discharge
status.
2.4.3 Response during the follow-up
The original sample size for the study was 6,000 participants. With allowance for a
maximum of 10% loss to follow-up, the trial had a recruitment target of 6,669 (Roffe et al. 2014).
The recruitment target was subsequently revised in October 2012 to 8,000 patients, to provide
greater power to detect an interaction between stroke subgroups (defined by severity) and the
effect of oxygen versus control.
In total there were 8,003 individuals that were recruited to the study between April 2008
and June 2013. All patients had full information recorded at baseline as this information was used
in the minimisation process. There were 129 patients that had died within the first seven days of
the study.
Follow-up questionnaires were sent to individuals with a series of questions that allowed
the mRS, BI and NEADL to be calculated. At 3 months, 7,370 (92%) patients in the trial were still
alive, of which 6,936 (94%) of these patients responded to the 3 month questionnaire. At 6 months
7,167 (89%) patients were still alive and 6,594 (92%) of these patients responded to the 6 month
questionnaire. At 12 months, 6,957 (86%) recruited patients were still alive and 6,020 (87%) of
these patients provided responses to the questionnaires. Unlike the other data set included in the
thesis, where multiple imputations had previously been conducted before the data was obtained,
no values of the SO2S data set were imputed. This is because all methods that are used for analysis
throughout the thesis are able to deal with missing data, and there are some concerns about the
validity of imputing the outcome variable. One if the issues with the missing data for the outcome
Chapter 2 29 introduction of bias within the imputation strategy (Sterne et al. 2009). Also in order to compute
the outcome, it is recommended that there is complete data for all covariates that affect the
outcome, therefore caution should be taken imputing the outcome variable, and it was decided not
to do this for the SO2S data.