Fluoxetine to improve functional outcomes in patients after acute stroke: the FOCUS RCT

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 13/04/30. The contractual start date was in October 2014. The draft report began editorial review in May 2019 and was accepted for publication in November 2019. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Martin Dennis, Maree Hackett, Graeme J Hankey, Gillian Mead and Erik Lundström report grants from the National Health and Medical Research Council (Australia) and funding from the Swedish Research Council Framework grant in clinical therapy research during the conduct of the study. Maree Hackett also reports grants from The Stroke Association (London, UK), grants from the National Institute for Health Research (NIHR) Stroke Research Network and a grant in clinical therapy research during the conduct of the study, and grants from the National Heart Foundation of Australia outside the submitted work. She also held a National Health and Medical Research Council (Australia) Career Development Fellowship, level 2 (reference APP1141328) (2018–21). Stephanie Lewis reports being a member of the NIHR Health Technology Assessment General Committee (2016 to present). Peter Sandercock reports lecture fees from Bayer AG (Leverkusen, Germany) paid to his department, outside the submitted work.

Copyright statement

© Queen’s Printer and Controller of HMSO 2020. This work was produced by Dennis et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2020 Queen’s Printer and Controller of HMSO

The burden of stroke

Approximately 130,000 people have a stroke each year in the UK and, even with acute treatments, about 50% of survivors will have long-term residual disability. 1 This places a huge burden on health and social services and informal carers. Although more can be done to implement treatments that we know are effective (e.g. the more widespread provision of thrombolysis and thrombectomy and more rapid access to stroke units), there is still an urgent need to identify new treatments that might reduce neurological impairments, disability and dependency after stroke. One promising intervention that needs to be tested is a widely used antidepressant drug, fluoxetine, which is a selective serotonin reuptake inhibitor (SSRI).

Serotonin reuptake inhibitors in animal models

In animals, SSRIs have several potentially beneficial effects on both normal and diseased brains. First, they have a neurotrophic effect. Neurotrophins are involved in embryogenesis and organogenesis, control neural plasticity in adults, regulate synaptic activity and neurotransmitter synthesis and are essential for the regeneration of nerves. 2 Adult neurogenesis is generally restricted to the subependymal cells of the ventricular system and the subgranular zone of the dentate gyrus in the hippocampus. 3 SSRI antidepressants increase neurogenesis and expression of neurotrophic/growth factors in the adult hippocampus,4 which is likely to account for the behavioural benefits of antidepressants in animals. 5 Importantly, several studies have shown that migration of new neurones to damaged areas of brain may occur,6 and that neurogenesis may also occur in areas of damaged brain in patients who have had ischaemic stroke. 7 Second, fluoxetine may have a neuroprotective effect associated with its anti-inflammatory effect (e.g. repression of microglia activation)8 and enhancement of specific protein expression (e.g. hypoxia-inducible factor-1 alpha and heme oxygenase-1). 9 Third, SSRIs can indirectly affect the adrenergic system through the upregulation of beta-1 receptors. 10

Selective serotonin reuptake inhibitors and motor function in humans

In healthy humans, functional magnetic resonance imaging studies have demonstrated that fluoxetine can modulate cerebral motor activity. 11 In eight patients who had a pure motor stroke who were given fluoxetine, there was hyperactivation in the ipsilesional primary motor cortex during a motor task; moreover, fluoxetine significantly improved motor skills in the affected side. 12 In a small-scale randomised trial of patients who had a unilateral stroke, the administration of citalopram, another SSRI, was associated with a significant improvement in neurological status, as measured with the National Institutes of Health Stroke Scale (NIHSS),13 and a decrease of motor excitability over the unaffected hemisphere, as measured by transmagnetic stimulation. 14 Zittel et al. 15 investigated the effects of a single dose of 40 mg of citalopram in eight chronic stroke patients; dexterity was significantly improved. In a trial of 52 hemiplegic patients who were randomly allocated to receive one of three treatments (20 mg/day of fluoxetine vs. 150 mg/day of maprotiline vs. placebo) for 3 months against a background of physical therapy, those allocated to receive fluoxetine demonstrated the greatest recovery from disability. 16

The FLuoxetine for motor recovery After acute ischaeMic strokE (FLAME) trial17 evaluated the effects of SSRIs on motor recovery after stroke. This double-blind, placebo-controlled, multicentre trial randomised 118 patients who had an ischaemic stroke and unilateral motor weakness to receive either 20 mg of fluoxetine daily or placebo for 3 months. At day 90, the improvement in the Fugl-Meyer Assessment Motor Score from baseline was significantly greater in the fluoxetine group [57 patients, adjusted mean 34.0, 95% confidence interval (CI) 29.7 to 38.4] than in the placebo group (56 patients, adjusted mean 24.3, 95% CI 19.9 to 28.7) (p = 0.003). In a post hoc analysis, the frequency of independent patients [modified Rankin Scale (mRS) of 0–2]18 was significantly higher in the fluoxetine group than in the placebo group (26.3% vs. 8.9%; p = 0.015), although there were no significant differences at other cut-off points. The small sample size limits the study’s generalisability. All patients also received physiotherapy (of unknown intensity), so we do not know whether or not fluoxetine on its own, or with less intense physiotherapy, would also be effective. Importantly, we also do not know whether or not any benefits of fluoxetine persist beyond the treatment period and whether or not fluoxetine might improve outcome in stroke patients without motor deficits. Nevertheless, these promising but inconclusive results clearly justify further larger trials in patients who have motor deficits.

Might selective serotonin reuptake inhibitors be of benefit in recovery of non-motor aspects of stroke?

Several small studies have suggested that fluoxetine might have other neurological benefits (e.g. increased activation of agonist and antagonist muscles in paretic arms after stroke,19 and improvements in executive function after stroke20). We do not know whether or not these beneficial effects of antidepressants are independent of their antidepressant effect. 21

In people with depression, SSRIs modulate the hyperactivity of the hypothalamic pituitary axis (HPA). 22 After stroke, activation of the HPA occurs, resulting in hypercortisolism. Hypercortisolism is associated with the development of delirium after stroke and also predicts worse long-term outcome. 23 Thus, SSRIs might, by attenuating the hypercortisolism that is present after stroke, improve outcomes, including cognition.

Systematic review of effects of fluoxetine on post-stroke outcomes

In 2011, when the Fluoxetine Or Control Under Supervision (FOCUS) trial was being planned, a recent systematic review of randomised controlled trials (RCTs) testing whether or not a course of treatment with fluoxetine started shortly after stroke onset might improve function and prevent post stroke depression identified six RCTs published before December 2009, which together randomised 385 patients. 24 Meta-analysis demonstrated that fluoxetine helped recovery in neurological function (weighted mean difference –4.72, 95% CI –8.31 to –1.13), improved independence in activities of daily living (weighted mean difference –8.04, 95% CI –13.40 to –2.68) and reduced the incidence of post-stroke depression [odds ratio (OR) 0.25, 95% CI 0.11 to 0.56]. A Cochrane review of selective serotonin receptor antagonists in stroke25 subsequently identified 56 trials comparing SSRIs with a control intervention (e.g. usual care or placebo), which were given in the first year after stroke. Fifty-two trials (4059 participants) reported data that could be included in the meta-analyses. Of these 52 trials, 28 used fluoxetine and 31 recruited patients within 3 months of stroke onset. The meta-analyses demonstrated beneficial effects of SSRIs on dependency, disability, neurological deficit, depression and anxiety at the end of treatment. There were benefits even in patients without depression at recruitment. However, there was substantial heterogeneity in the estimates of effect sizes; sensitivity analyses suggested that methodological limitations of many of the included trials may have led to overestimation of effect sizes and there was an excess of gastrointestinal side effects in patients receiving a SSRI. 25 Furthermore, most trials excluded people with cognitive impairment and aphasia, and only eight trials followed patients up after treatment had been discontinued.

Why choose fluoxetine?

There are many SSRI antidepressant medications available. We chose to evaluate fluoxetine because it is one of the most widely studied. Its safety profile is very well established, and the drug is well tolerated in long-term use, even in older patients. There was more evidence for its effectiveness in stroke than for that of alternatives, such as citalopram. 25 A number of manufacturers produce the drug and the price was low, which makes it particularly attractive to health services that are under severe cost pressures. Finally, of all the SSRIs, it has the longest half-life; therefore, gradual reduction in dose is not required when withdrawing the drug (which is inevitable in a trial), which is typically carried out to avoid the possibility of a SSRI-withdrawal syndrome. 26

Potential concerns of using fluoxetine in stroke patients

There are potential risks associated with giving fluoxetine to a wide range of stroke patients. Its reported interaction with antiplatelet and anticoagulant medication might increase bleeding risk, although this is usually minor and limited to bruising. Like other antidepressants, fluoxetine may lower the seizure threshold and, therefore, could increase the frequency of post-stroke seizures. In our Cochrane review, there was a non-significant excess of seizures in patients who were allocated SSRIs. 25 Therefore, we excluded from the FOCUS trial patients who had a history of epileptic seizures. An adverse effect on glycaemic control in diabetic patients has been recorded. Hyponatraemia is a recognised adverse effect and may prove to be more common among stroke patients who may be taking concomitant angiotensin-converting enzyme (ACE) inhibitors, diuretics and proton pump inhibitors. Observational studies have suggested that bone fractures are more common in those taking SSRIs, and this has variably been attributed to an increased risk of fractures in depression, increased risk of falling while taking SSRIs (possibly owing to drowsiness, increased activity or motor effects) and direct effects of SSRIs on bone strength. 27,28 Furthermore, there are already concerns that stroke patients have a greater risk of falls owing to their neurological and functional deficits or concurrent medications (e.g. antihypertensive medication), and greater risk of fractures owing to osteoporosis affecting hemiplegic limbs. 29,30

Nevertheless, fluoxetine has been very commonly prescribed for several years for selected patients who have had a stroke to treat depression and emotionalism without major problems emerging.

Patients who are commenced on psychotropic drugs, including fluoxetine, are encouraged to monitor the effects on their psychomotor function before resuming driving. However, stroke patients in the UK are advised not to drive for at least 1 month after a stroke, which should provide ample time in the trial for any potentially important adverse effects that would affect their driving ability to become apparent.

Rationale for the study

Design overview

The FOCUS trial was a pragmatic, investigator-led, multicentre, parallel-group, double-blind, placebo-controlled trial with broad entry criteria and follow-up to ascertain the primary and secondary outcomes at 6 and 12 months.

Setting

The FOCUS trial was carried out in hospital-based stroke services in the UK.

Participant inclusion/exclusion criteria

Consent

The investigator was responsible for ensuring that informed consent was obtained and the consent form was completed, signed and dated by all parties before any protocol-specific procedures were carried out.

Participant information booklets (PIBs) and informed consent forms (ICFs) were provided (see the project web page: www.journalslibrary.nihr.ac.uk/programmes/hta/130430/#; accessed 7 May 2020). Separate versions were available for patients with capacity; proxies were used for patients without capacity. We developed easy-access versions for patients or proxies with cognitive or communication difficulties (see Appendix 2). The verbal explanation to the participant was provided by the investigator or designated person, and aimed to cover all the elements specified in the PIB/ICF. The participants were given every opportunity to clarify any points that they did not understand and, if necessary, ask for more information. Participants could withdraw their consent to participate at any time without loss of benefits to which they would otherwise be entitled.

The participants consented to their medical records being inspected by regulatory authorities and representatives of the sponsor(s) and agreed that the information held and maintained by NHS Digital and other central UK NHS bodies could be shared with us and may be used to help contact them or provide information about their health status.

Written informed consent from the patient was always sought where possible. If this was not possible because the patient could not write, the randomising clinician or nurse could gain witnessed verbal consent. Laws governing consent procedures, and in particular those governing incapacitated adults and their involvement in research, were followed.

The patient or personal legal representative received a folder including a copy of the relevant version of the PIB, a copy of the completed ICF and a patient diary that contained contact details for the trial co-ordinating centre and prompted the recording and reporting of safety outcomes and adverse events, etc. The original ICF and PIB were filed in the site file with the randomisation form (see the project web page: www.journalslibrary.nihr.ac.uk/programmes/hta/130430/#; accessed 7 May 2020). The completed ICF was also scanned and uploaded onto the secure trial website or e-mailed, or faxed, to the trial office, before randomisation. The trial management system prompted the research team to do so via e-mail and/or fax until the consent form had been received.

Randomisation

Having obtained consent, the randomising person collected the baseline data necessary to complete a randomisation form (see the project web page: www.journalslibrary.nihr.ac.uk/programmes/hta/130430/#; accessed 7 May 2020) and entered the patient’s baseline data into our computerised central randomisation service by means of a secure 24 hours per day/7 days per week (24/7) web interface. After the computer program checked these baseline data for completeness and consistency, it allocated that patient a unique study identification number and a treatment pack number that corresponded to either fluoxetine or placebo. The system applied a minimisation program to achieve balance between the treatment groups for four factors:

1. delay since stroke onset (2–8 vs. 9–15 days)

2. predicted 6-month outcome (based on the six simple variable model)31

3. presence of a motor deficit (based on NIHSS)13

4. presence of aphasia (based on NIHSS).

The six simple variable model is a statistical model that predicts survival and functional outcome after stroke. 31 The variables are (1) the patient’s age, (2) whether or not the patient was independent prior to the stroke, (3) whether or not they lived alone, (4) whether or not after the stroke the patient could lift both arms off the bed, (5) walking without help of another person and (6) talking without being confused (i.e. normal on the verbal component of the Glasgow Coma Scale32).

The minimisation algorithm randomly allocated the first patient to a treatment, but allocated each subsequent patient to the treatment that minimised the imbalance between the treatment groups with respect to the prognostic factors. 33 It was designed to allocate equal numbers to each of the two treatment groups (i.e. a 1 : 1 ratio). To ensure that we retained a random element to treatment allocation, patients were allocated to the group that minimised differences between groups with a probability of 0.8. The system contained a list of treatment codes for each centre and that matched the stocks held at that centre. At the end of the session, each patient was allocated a treatment code that corresponded to an active (20 mg of fluoxetine o.d.) or placebo treatment pack that contained a 6-month supply of capsules held at that centre.

The randomisation system took account of the drug stocks that were held locally to (1) ensure that the allocated treatment was available and (2) minimise wastage. The randomisation system automatically generated an e-mail/fax to the centre co-ordinator and the local research pharmacist to ensure that the allocated treatment was prescribed. The pharmacist or co-ordinator could access treatment codes to replace lost study medication through a secure website by entering the patient’s study ID number and date of birth.

To facilitate drug reconciliation and stock control, the pharmacist or local co-ordinator removed an adhesive treatment number label (flag) from the medication bottle, stuck it onto the confirmation of allocation fax and faxed it back to the trial co-ordinating centre. The trial management system prompted them to do so via e-mail and/or fax until the fax was received.

Following randomisation, the trial co-ordinating centre sent a letter to inform the general practitioner (GP) of the patient’s enrolment in the trial, including a copy of the consent form and the follow-up schedule (see the project web page: www.journalslibrary.nihr.ac.uk/programmes/hta/130430/#; accessed 7 May 2020).

The interventions

The interventions were 20 mg of fluoxetine o.d. or placebo for 6 months. The study medication (active and placebo) was manufactured by Unichem (Mumbai, India), imported by Niche Generics Ltd (Hitchin, UK), purchased from Discovery Pharmaceuticals Ltd (Castle Donington, UK) and quality assured, packaged, labelled and distributed by Sharp Clinical Services (Tredegar, UK). Patients were supplied with 186 capsules and were prescribed the study medication (20-mg capsules of fluoxetine or placebo) to be taken daily. If the patient was unable to swallow capsules and had an enteral feeding tube in place, the capsules were broken open and the contents put down the tube.

We measured adherence to the study medication in several ways: recording the date of first and last dose taken, number of missed doses while in hospital, capsule counts when unused capsules were returned and estimated adherence at 6-month follow-up. We recorded the reasons for stopping the study medication early. Our primary measure of adherence was the best estimate of the interval between the first and the last dose based on all of the information available. Therefore, for a particular participant, a capsule count might lead us to modify the estimate of the timing of the last dose (see Chapter 3 for more detail).

Blinding

The patient, their families, the health-care team including the pharmacist, the staff in the co-ordinating centre and anyone involved in outcome assessments were blinded to the treatment allocation by using a placebo capsule that was visually identical to the fluoxetine capsules, even when broken open to allow the administration of the trial medication down an enteral feeding tube.

An emergency unblinding system was available. If a clinician thought that they needed to know the allocated treatment for a patient, they were asked to telephone a 24/7 helpline that was manned by staff from a co-ordinating centre, and provided access, directly or indirectly, to one of our chief investigators. The case for unblinding was discussed and, if agreed, the clinician was given a unique code (based on a simple arithmetic manipulation of the date) to unlock the web-based unblinding system. The clinician could then enter the patient’s details, along with the reason(s) for unblinding, and they were provided with the treatment allocation. This was designed so that those in the co-ordinating centre and those conducting follow-up remained blind to the treatment allocation. Our information technology system logged any attempts to unblind.

Primary outcome

The primary outcome was the mRS (based on ordinal analysis to maximise power and to avoid the problem of including patients with a mRS of > 2 prior to their stroke) at 6 months after randomisation. 18,34 We also collected data on mRS at 12 months (one of our secondary objectives). Patients who died were attributed a score of 6 for this analysis.

The mRS is a simple, time-efficient measure with well-studied reliability that is used to categorise levels of functional outcome (see Table 10). It has been used extensively in large, multicentre stroke trials.

Any misclassification of patients into an inappropriate mRS category may reduce the power of the trial. To minimise misclassification and intermodality differences, we used the simplified modified Rankin Scale questionnaire (smRSq) described by Bruno et al. 18,35,36 This can be delivered via telephone and postal questionnaires and has been completed by patients and proxies. 36,37

Secondary outcomes

To answer our secondary objectives, we collected the following outcome measures:

• Deaths from all causes by 6 and 12 months.

• The EuroQol-5 Dimensions, five-level version (EQ-5D-5L) to provide an overall measure of HRQoL and to allow a health economic analysis based on quality-adjusted life-years (QALYs). 38

• The Mental Health Inventory – 5 questions (MHI-5), which is derived from the Short Form questionnaire-36 items (SF-36) and provided a measure of depression and anxiety symptoms. This brief measure performs well, compared with longer questionnaires (e.g. Mental Health Inventory – 18 questions, General Health Questionnaire –12 questions and General Health Questionnaire –30 questions), in the detection of depression and anxiety symptoms. 39–41

• The vitality subscale of the SF-36 was used to assess patients’ levels of fatigue. 42,43

• The Stroke Impact Scale (SIS) provided an overall assessment of patient outcome as well as allowing us to assess the effect of treatment on specific outcomes of importance to the patients. The SIS is a stroke-specific, comprehensive health status measure. The scale was developed with input from patients and caregivers and comprises eight domains (strength, hand function, activities of daily living/instrumental activities of daily living, mobility, communication, emotion, memory and thinking, and participation) from across the full impairment–participation continuum. 44–46 It also provided an overall assessment of recovery with a visual analogue scale. The scale has been validated for use by proxy respondents and has been delivered via telephone and postal questionnaires. 45,47,48

Safety outcomes

• New diagnosis of depression since randomisation. This was collected at 6- and 12-month follow-up with GPs and participants. We recorded who had made the diagnosis, whether or not any treatment, and specifically any treatment with antidepressant medication, was initiated and whether or not there was any attempt at suicide or self-harm. Sometimes patients were started on an antidepressant without a clear prior diagnosis; in these cases, we made an individual judgement based on all available information and whether the antidepressant had been started for new depression (or low mood) or for another indication, such as neuropathic pain, anxiety or emotionalism without depression.

• Other adverse events, including further strokes (ischaemic or haemorrhagic), acute coronary events, upper gastrointestinal haemorrhage, falls resulting in injury, new bone fractures, epileptic seizures, symptomatic hypoglycaemia (< 3 mmol/l), hyperglycaemia (> 22 mmol/l) and hyponatraemia (Na⁺ < 125 mmol/l). Retrospectively, we also categorised patients with other serious bleeds (e.g. lower gastrointestinal, renal tract and subdural) and thrombotic events (deep-vein thrombosis, pulmonary embolism, mesenteric thrombosis, ischaemic limbs) that led to hospital admission. Information on these events were collected via centres, GPs and participants at discharge and 6 months, although we became aware of some events occurring later because they led to hospital admissions that we recorded at the 12-month follow-up.

Follow-up

The principal investigator (PI) and researchers at each site collected the local data listed in the schedule of study assessments below. The chief investigators and the research team in the central co-ordinating office collected the central data (Table 1).

Study safety assessments

Our monitoring system was primarily aimed at identifying suspected unexpected serious adverse reactions (SUSARs), but also at identifying whether or not the frequency of serious adverse reactions was greater than in other populations given fluoxetine and sufficiently common to offset any benefits. We did not aim to detect the occurrence of the very many adverse events that occur in stroke patients and that were very unlikely to be related to participation in the trial or the medication.

The trial materials given to the patient and/or their carer contained details of the known adverse reactions to fluoxetine (based on the SmPC) and the adverse events that commonly occur after stroke. They received a diary in which they were encouraged to record the date and nature of any adverse events.

Patients who were enrolled while they were an inpatient had a hospital discharge form completed by the local co-ordinator at the time of discharge from the recruiting centre or shortly after. The data collected were entered on a secure web-based form or faxed to the co-ordinating centre to ensure that we were alerted to any important adverse reactions. We regularly prompted centres to complete discharge forms for patients with incomplete data.

Patients who were enrolled while they were an outpatient had a central follow-up at 1 month after recruitment to detect safety outcomes and adverse reactions.

At 12 weeks after randomisation, the trial co-ordinating centre staff posted a reminder to the patients to report any adverse events or difficulties with the trial medication, but this was not followed up unless a response was received.

All surviving patients who had not withdrawn consent or indicated that they did not want to be contacted directly were followed up at 6 and 12 months after randomisation, whether or not they adhered to their allocated treatment. At each follow-up, the GP was asked about safety outcomes and other adverse events. In order to detect adverse reactions between the scheduled follow-ups, patients, their carers or their GPs could report any adverse reactions to us via:

• post – a Freepost envelope and adverse events form to return to us with details of any adverse reactions that the patient had experienced

• a helpline – a telephone number that allowed the patients or their doctors to leave a message (if non-urgent) or to access a trial doctor (if urgent).

About 2 weeks before any central follow-up was due, the trial co-ordinating centre staff contacted the GPs (or hospital co-ordinators if no discharge form had been received) to check that the patient was alive and that they may be approached for follow-up. The GP was asked (and paid a fee of £56.00) to provide a list of non-IMPs and to complete a questionnaire including information regarding the patient’s adherence to the IMP, details of any safety outcomes or adverse events, hospital admissions and up-to-date contact details for the patient.

If appropriate, the trial co-ordinating centre then posted a questionnaire to the patient at 4 weeks (only for those recruited as outpatients), 26 weeks and 52 weeks. If the patient did not respond to the postal questionnaire, they were telephoned by the co-chief investigators at 6 months and by a trained member of the team at 12 months. The questionnaire at 26 and 52 weeks aimed to capture the primary and secondary outcomes and included the outcome of any adverse events that have been reported earlier in the follow-up. If the patient had incapacity, the next of kin (proxy) was asked to complete and return the forms. If the patient was unable to speak English, we asked that their carer supported them in filling out the forms. If the follow-up information could not be obtained by the postal or telephone questionnaire, we asked the local research team to arrange a face-to-face follow-up at a clinic or home visit.

Data linkage and extract to determine outcome and long-term survival

We collected data from our participating hospitals, the patients and their GPs about hospital admissions during the first 12 months. However, we also planned to obtain information about the health status and resource use of participants to determine outcomes beyond the end of the trial from the Health and Social Care Information Centre. This function has been devolved to NHS Digital in England and Wales and the eData Research and Innovation Service (eDRIS) in Scotland. No centrally held data are available for Northern Ireland.

Management of depression in the trial

Our hypothesis was that new episodes of depression would be less commonly diagnosed and treated in the group allocated to fluoxetine. We ascertained cases of depression by:

• asking about a diagnosis or initiation of an antidepressant during hospital admission or during the first month – this was recorded on the locally completed discharge form or the 1-month central follow-up form

• asking the GP at 6 months and 12 months

• asking the patient (or their proxy) at 6 months and 12 months.

Because the primary question addressed by the FOCUS trial was whether or not a SSRI (20 mg of fluoxetine o.d.) enhanced recovery from stroke, it would be an advantage if the control group were kept free from any SSRIs, including fluoxetine. However, it would be unethical to deny patients in the trial access to effective antidepressant treatment. We therefore asked collaborating clinicians and the patients’ GPs to adhere to the following treatment guideline.

If a patient in the FOCUS trial was diagnosed as having depression (or pathological emotionalism) that the responsible clinician judged to be severe enough to justify treatment with antidepressant drugs, we recommended that, if possible, they should avoid any SSRIs and prescribe either mirtazapine or trazodone. Both drugs are compatible with fluoxetine (there are no common or important interactions), although because mirtazapine has some serotonergic activity there is likely to be a slightly greater risk of precipitating a serotonergic syndrome. Both drugs were recommended by the National Institute for Health and Care Excellence (NICE) for treatment of depression in patients with physical illness. 26 The clinician might alternatively use a tricyclic antidepressant of their choice. We advised that patients taking the trial drug and another antidepressant should be monitored carefully (e.g. check plasma sodium levels to exclude hyponatraemia) to identify any potential interactions.

Sample size

We planned to enrol at least 3000 patients in the main phase of the FOCUS trial. This aimed to provide 90% power with a two-sided 5% level of significance to detect a 5.6% absolute increase in percentage with mRS 0–2 from 27.0% to 32.6% based on an ordinal analysis, which is statistically more efficient than an analysis that dichotomises the mRS. 34

In arriving at this sample size, we took account of the effect sizes seen in the FLAME trial17 alongside the effects that we judged clinicians and their patients would find interesting. Because fluoxetine is safe and inexpensive, the FOCUS trial sought to reliably detect a moderate, but nonetheless clinically important, benefit that might be associated with widespread use of fluoxetine in this population. However, we also took account of the feasibility of enrolling large numbers of patients into the FOCUS trial.

We based our expected outcomes for our placebo group on the distribution of the mRS score measured at 6 months after randomisation in the CLOTS trials,49,50 which evaluated graduated compression stockings.

We used the ordered categorical data method described by Machin et al. 51

The Trial Steering Committee (TSC) reviewed the target sample size and could adjust this based on:

• advice from the Data Monitoring Committee (DMC)

• accruing data on –

  • the enrolment into specific prespecified subgroups

  • completeness of follow-up

  • distribution of mRS categories in the population of enrolled subjects (i.e. both treatment groups combined), overall and in specific patient categories (e.g. those with motor deficits or aphasia).

For example, if the distribution of mRS was different from that anticipated, then the sample size could be increased. This approach had the advantage that such sample size adjustments could be made without reference to the accumulating unblinded data, and avoided the need for conditional power calculations, which could be unreliable.

Statistical analyses

Our statistical analysis plan was published prior to completion of data collection. 52 We summarise the plan here but, rather than reproduce the whole plan in detail, we have indicated in Chapters 3–7 which analyses were not specified in the published plan. For all analyses, unless otherwise specified, we retained participants in the treatment group to which they were originally assigned, irrespective of the treatment they actually received (i.e. an intention-to-treat analysis). A statistical significance level of p < 0.05 (two tailed) was applied to all analyses. The final prespecified analyses were performed on the data set after any ‘cleaning’ that was required had been completed and the database was locked. The treatment allocation was only then unblinded. Certain post hoc analyses presented in this report included data that had been further cleaned after the main analyses had been carried out and published to correct minor anomalies detected during the analyses.

Missing data

Our randomisation systems did not allow investigators to proceed to treatment allocation without entering complete baseline data. The mRS, our primary outcome, includes death; therefore, the number of participants with missing mRS at follow-up was small. Anyone with a missing mRS was not included in any analysis requiring mRS (complete-case analysis).

For secondary outcomes [e.g. SIS, MHI-5, vitality subscale of the SF-36 and EQ-5D-5L] for which missing data were expected because data were not available for patients who did not survive, we presented results for those who were alive at follow-up and any discrepancies in death rates between groups were taken into account in the interpretation. Missing data for single questions within scores were handled as detailed by each scoring method. Where responses to all questions within a scale or subscale were missing, that patient was not included in that part of the analysis.

Protocol deviations, adherence and blinding

Inclusion/exclusion violations: we reported the number and percentage of participants randomised who did not meet the entry criteria (e.g. non-strokes), with exclusion criteria. However, they were included in the primary analysis. A secondary analysis excluded ineligible patients (see below).

Unblinding: we reported the number of patients who required unbinding of study medication during the trial by treatment group and, where available, present the reasons for unblinding.

Adherence: each participant was issued with a 6-month supply of trial medication (186 capsules). At 6 months, they were asked if they had completed the course and taken all of the capsules and how often they took capsules on average. They were asked the reasons for stopping, as well as the date of stopping. Where possible, we retrieved and counted the unused trial medication. Before unblinding, we derived an estimated date on which the patient was thought to have taken their last dose of trial medication and used the interval (days) from first dose to that date as our main measure of adherence. This was based on all of the available information. We used a combination of the following to define several types of non-adherence to the protocol (see 1–8 below):

• inclusion/exclusion violations

• the answers to the adherence questions (see above)

• number, percentage and duration of any open-label SSRI intake before the 6-month follow-up

• the reasons for stopping trial medication.

A so-called intention-to-treat analysis, in which patients’ outcomes are analysed in the groups that they were randomised to regardless of treatment received, provides the least biased and most robust evidence of the effect of treatment. However, the observed treatment effect may be reduced if a large number of patients are included who are unlikely to benefit because they did not have a stroke or more likely where a large proportion of patients do not receive the allocated treatment or actually received the alternative treatment (i.e. cross-overs).

Where the primary analysis does not demonstrate an improvement of functional outcome (mRS) at the 6-month follow-up, the question arises of whether or not this is this likely to be a result of poor adherence to the protocol and/or trial medication? This is important because we would not wish to abandon a potentially useful treatment simply because of poor adherence to trial protocols or trial medication. These might be improved in any future trials. We undertook further per-protocol analyses to reassure the clinical community that the trials have not underestimated any treatment effect to an extent that would alter future clinical practice, or more likely the need for further randomised trials of SSRI in stroke.

Inevitably, analyses that try to take account of adherence introduce a degree of patient selection and, thus, are likely to introduce bias.

These prespecified exploratory sensitivity analyses to account for non-adherence included all of the analyses of the primary outcome and selected secondary outcomes:

• living at home or with relative versus care home, hospital or long-term hospital care

• mRS at 6 months and 12 months (mRS 0–2 vs. mRS 3–6)

• new diagnosis of depression between randomisation and 6 months and 12 months

• SIS domain scores

• averaged score over all SIS domains

• SF-36 vitality subscale score

• utility based on EQ-5D-5L and population preferences.

These analyses do not include any analysis of subgroups defined on the basis of baseline variables. The following groups were sequentially added to the group excluded from the analyses:

1. Patients who did not meet the entry criteria for the trial.

2. Patients who did not receive any trial medication.

3. Patients who received < 90 days of trial medication because of failures in trial procedures, for example failures to transfer trial medication with patients during moves between hospitals, care homes and home. The 90-day cut-off point was chosen because previous trials have tested this duration of treatment with apparent benefit. 17,25

4. Patients who received < 90 days of trial medication because of patient or relative concerns but not because of suspected adverse reactions.

5. Patients who received < 90 days of trial medication because they experienced symptoms that were attributed to the trial medication.

6. Patients who had been allocated to placebo who received a SSRI (fluoxetine or other) within the first 90 days and the SSRI was not known to have been stopped within 10 days of starting.

7. Patients who had been allocated to fluoxetine who received a SSRI (fluoxetine or other) within the first 90 days and the SSRI was not known to have been stopped within 10 days of starting.

8. Patients who did not complete at least 150 days of treatment. We chose this cut-off point because patients sometimes received the questionnaires shortly before 6 months, and some stopped the trial medication at that point, whereas others finished the 186 capsules. We regarded both as fully adherent.

Research governance

The trial was co-ordinated by a Project Management Group: Professor Martin Dennis, Professor Gillian Mead (the joint chief investigators and PIs for two participating sites), Karen Innes (trial manager) and Catriona Graham (trial statistician).

Trial co-ordinating centre

The trial co-ordinating centre was responsible for all aspects of the management of the FOCUS trial and was based at the Centre for Clinical Brain Sciences at the University of Edinburgh. Responsibilities included regulatory submissions and compliance; financial management; monitoring of sites; training; patient information and communication; end-point assessment; data collection systems and data management; IMP management; statistical analysis; reports and publications; and archiving of the trial master file in accordance with funder and sponsor requirements.

Trial Steering Committee

A TSC was established, including a stroke survivor and a carer, to oversee the conduct and progress of the trial. The terms of reference of the TSC, the draft template for reporting and the names and contact details were agreed at its first meeting (see Report Supplementary Material 1).

Data Monitoring Committee

An independent DMC was established to oversee the safety of participants in the trial. During the period of recruitment into the study, interim analyses of the baseline and follow-up data were supplied, in strict confidence, to the chairperson of the DMC, along with any other analyses that the committee requested. In the light of these analyses, the DMC could advise the chairperson of the TSC if, in their view, the randomised comparisons had provided (1) ‘proof beyond reasonable doubt’ that for all, or some, the treatment is clearly indicated or clearly contraindicated and (2) evidence that might reasonably be expected to materially influence future patient management. Appropriate criteria of proof beyond reasonable doubt were not specified precisely, but the DMC worked on the principle that a difference of at least three standard errors in an interim analysis of a major outcome event (e.g. death from all causes or independent survival at 6 months) would be needed to justify halting, or modifying, the study before the planned completed recruitment. This criterion has the practical advantage that the exact number of interim analyses would be of little importance, and so no fixed schedule was proposed. Following a report from the DMC, the TSC decided whether to modify entry to the study (or seek extra data).

The terms of reference of the DMC, the DMC charter and the names and contact details were agreed at the first meeting of the DMC (see Report Supplementary Material 2).

Patient and public involvement

In December 2010, service users overseeing a study of post-stroke fatigue commented on our plans for the FOCUS trial. We met with the group in May 2011. We included their suggestions about patient information booklets (i.e. explaining the rationale for using fluoxetine in people without depression, listing all side effects), and agreed that all trial participants would be sent a summary of the trial results if they wished. A group of stroke survivors with aphasia developed the easy-access version of the patient information booklet, guided by Professor Marian Brady (see Appendix 2). A stroke survivor, Judith Williamson, from the National Institute for Health Research (NIHR) Stroke Research Network (SRN), and Zena Jones (manager of the NIHR SRN patient and public involvement group) attended our first investigator meeting (in June 2011). In March 2013, the SRN patient and public involvement group advised us how to enhance the proportion of eligible patients consenting, and how to facilitate follow-up. This advice featured in our Autumn 2013 newsletter to sites. In November 2013, Ms Jones and Ms Williamson endorsed our plans to telephone patients immediately after hospital discharge. They edited the script that would guide the telephone call. They commented on a draft of the funding application to NIHR. Ms Williamson and a carer were on our TSC, offering advice and comments throughout the trial. They were particularly influential in discussion of how we would disseminate the results of the trial to participants and their families. They commented on the final newsletter, which was sent out on the day the results were presented at conference, and published in The Lancet. 55 They have also had input into the drafting of this report, specifically the Plain English summary.

Recruitment

Between 10 September 2012 and 31 March 2017, 103 UK hospitals consented 3152 patients and enrolled 3127 patients. Recruitment was stopped after we had exceeded our minimum target of 3000 patients to account for recruitment of ineligible patients and withdrawals (Figure 1).

FIGURE 1.

Recruitment graph showing planned vs. actual recruitment.

The network of centres was built on the networks that we had established to carry out previous trials, including the CLOTS49,50 and IST343 trials. We enrolled new centres throughout the trial period. The trial co-ordinating centre worked closely with the research teams at prospective centres, and the majority of site initiation visits were carried out remotely using telephone and video conferencing. The trial was facilitated by the NIHR-funded research networks, which provided funding for local research nurses. Centres also received a £300 pharmacy start-up fee, and approximately £46 per patient recruited.

Thirty-one patients were identified as ineligible between obtaining consent and randomisation; in other cases, the patients, their proxy or their treating clinician changed their mind about participation in the trial. Of the 3127 patients who were enrolled, 1564 were allocated to the fluoxetine group and 1563 were allocated to the placebo group. Eleven of these patients did not meet our eligibility criteria: two in each group had a final diagnosis other than stroke and seven others were identified as meeting exclusion criteria after randomisation (e.g. a history of epilepsy, self-harm or some other contraindication to fluoxetine). The ineligible patients were retained in our intention-to-treat analyses. The patients’ progress through the trial is shown in Figure 2.

FIGURE 2.

Participant flow. a, 1544 inpatients with discharge form; 20 recruited as outpatients; b, 1536 inpatients with discharge form; 27 recruited as outpatients. Reproduced from the FOCUS trial collaboration. 55 © 2018 The Author(s). Published by Elsevier Ltd. This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY-NC-ND 4.0) license. See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

Reproduced from the FOCUS trial collaboration. 55 © 2018 The Author(s). Published by Elsevier Ltd. This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY-NC-ND 4.0) license. See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

Baseline characteristics of recruited patients

The baseline characteristics of the two treatment groups were well balanced with respect to all measured variables (Tables 2–6). 55

Withdrawal

The term withdrawal is used widely in RCTs but it is important to be precise about what is meant. In the FOCUS trial, we separately categorised patients into the following groups:

• patients who had stopped their trial medication, but were content to be followed up

• patients who may or may not have stopped their medication but no longer wanted to be contacted directly for follow-up information, but were content for us to obtain follow-up information from family, friends, GPs or routine data sources

• patients or their proxies who withdrew consent and wanted the patient and their data to be taken out of the study from that point on, so that we used data collected up to that point only.

When we were informed that the patients wished to withdraw, we clarified which of the above applied. Only 31 patients withdrew consent; the timing and treatment allocations are shown in the participant flow diagram (see Figure 2) and treatment allocations were fairly well balanced.

Discharge forms

A discharge form (see the project web page: www.journalslibrary.nihr.ac.uk/programmes/hta/130430/#) was completed for all 3040 patients enrolled as inpatients.

The 6- and 12-month follow-ups

We completed the 12-month follow-up in the trial in June 2018. The participant flow diagram (see Figure 2) shows the completeness of follow-up with respect to our primary outcome and vital status. Table 7 shows the methods of follow-up to obtain these data. Forty-nine per cent of 6-month follow-up assessments were obtained by postal questionnaire (see the project web page: www.journalslibrary.nihr.ac.uk/programmes/hta/130430/#). The remainder required a telephone reminder or were completed by telephone interview. The telephone follow-ups for participants not returning postal 6-month questionnaires were carried out by the two co-chief investigators (MD and GM) who were trained and certified in the use of the mRS and had conducted an independent validation of the smRSq. 18,35,36 Those conducting the 12-month telephone follow-ups had received training in their application.

Based on our previous trials, we had expected 80% of follow-up to be returned by postal questionnaire, rather than 50%. This placed a much greater burden on the central co-ordinating team than expected. In addition, in the FOCUS trial, patients were followed up at 6 and 12 months, doubling the number of follow-ups. Because missing outcome data would reduce the power of the trial, and can more importantly introduce bias because it is rarely missing randomly, the central team allocated far more time and resources to follow-up than had been planned.

Reasons for the low response rate

We did not formally assess the reasons for the low rate of response to the postal questionnaires, but the following issues contributed:

• Centres providing incomplete or inaccurate postal addresses for participants. Increasingly, stroke services admit patients acutely to one hospital and then transfer the patients to another hospital, or community rehabilitation centre, for ongoing care. We collected the discharge address on our discharge form, and this was often completed when patients moved from the centre in which they were recruited. Downstream facilities often did not inform us that the patient had then been discharged to an address other than their original home address. Our team spent a lot of time tracking down these patients to obtain their current address.

  Anecdotally, patients had often not opened our postal questionnaires because they believed them to be circulars. Any indication of the content of the letter on the outside of the envelope (e.g. contains FOCUS trial questionnaires) could have an impact on patient confidentiality and was, therefore, not used.

• Our 6- and 12-month follow-up questionnaires included the SIS. This is long, containing 59 items in several domains. In one domain, the Likert scaling for three of the nine items is reversed (i.e. good outcomes have low instead of high scores in the 1–5 range). Patients found this confusing, and they often entered internally inconsistent information, which we then queried by telephone.

• The burden of carrying out telephone follow-ups to clarify or complete information received by post, or to complete the whole follow-up by telephone, was considerable, and was increased as a result of several factors, including:

  • Patients changing telephones from landlines to mobiles, which meant that our co-ordinating team spent a lot of time communicating with centres, downstream health-care facilities, patients’ GPs and proxies to obtain up-to-date contact information.

  • Patients and proxies increasingly not answering calls from unknown numbers, which are often assumed to be marketing calls or scams. In many cases, patients or their proxies were telephoned at different times of the day and week, often on multiple occasions, before contact was actually made. To overcome this barrier, we often texted the recipient in advance of a phone call to increase the likelihood that they would answer. In future trials, it would be useful at the time of recruitment to enter the trial co-ordinating centre’s number into the patients’ and/or proxies’ mobile phones as a contact so they could make a more informed decision about whether or not to answer a telephone call from the trial centre.

• General practitioners and their staff varied greatly in the assistance they would provide in completing the GP questionnaires and in helping us contact patients. Most were very helpful but:

  • Some refused to provide information because they were unaware that the patient had provided written consent for them to do so. We had routinely sent GPs a copy of the patient information and the completed consent form with a covering letter at the time of recruitment. It appears that this was often filed and, therefore, it was not obvious to the staff when they received a request 6 or 12 months later. Some patients had moved house and changed GPs, which meant that the original trial documentation was not necessarily available to the new GP.

  • Some GPs reported that they were simply too busy to help.

  • Some felt that the fee of £54 negotiated with the primary care network was insufficient, and that they required a much larger sum to complete a follow-up questionnaire. We occasionally paid a little more, but usually we obtained the data via an alternative route.

Unblinding

The emergency unblinding procedure was carried out for only three patients. One was on the request of a coroner after the patient died, a second was on the request of the sponsor, for a suspected unexpected serious adverse reaction, and the third was because the responsible clinician felt that knowledge of the treatment would significantly alter their management of the patient. All of the patients had been allocated fluoxetine.

Adherence

The primary measure of adherence was the estimated duration of study medication (interval in days from first to last dose of study medication) based on all available data. Capsule counts were available in 398 (25.6%) of those allocated fluoxetine and 410 (26.4%) of those allocated placebo. The patients returned a median of 32 [interquartile range (IQR) 10–135] capsules in the fluoxetine group and 33 (IQR 11–139) capsules in the placebo group. Our primary measure of adherence was available in 1417 (91%) patients in each group. The median duration of treatment was 185 (IQR 149–186) days in the fluoxetine group and 183 (IQR 136–186) days in the placebo group. The median delay between randomisation and first dose was 0 (IQR 0–1) days in both treatment groups. A total of 1519 (97%) participants in the fluoxetine group and 1494 (96%) participants in the placebo group received their first dose by day 2 after randomisation (Table 8). Table 9 shows the number and proportion of patients meeting our eligibility criteria and different levels of adherence to the study medication. Patients stopping the trial medication because of perceived adverse effects within the first 90 days was only marginally more common in the fluoxetine group (n = 143, 9.1%) than in the placebo group (n = 122, 7.8%). About two-thirds of patients took the study medication for at least 150 days.

Problems with adherence

We identified some reasons for non-adherence to the trial medication. These included:

• Recruiters not checking that they had someone named on the delegation log who could sign the initial prescription. Usually this just led to a short delay in starting the medication, but was more complicated if the patient was discharged or moved hospital prior to receiving the first dose.

• Delays in obtaining the trial medication from pharmacies, especially when the patient was randomised later in the day.

• In discharge letters to GPs, it was sometimes unclear that patients were in a placebo-controlled trial and that patient had been given their trial medication. This sometimes led GPs to prescribe fluoxetine or nursing home staff to ask for fluoxetine to be prescribed on admission or when a patient’s supply was running low.

• Patients, or their families or nursing home staff, often wanted the trial medication in a blister pack, in a dosette box provided by pharmacists or in liquid form and so they would then request that the GPs prescribe fluoxetine – GPs would comply with this. Patients and relatives could put their trial medication into a dosette box, but we were unable to identify any means of getting pharmacies to put the trial medication into dosette boxes or blister packs. Thus, some patients allocated to the placebo group were inadvertently given fluoxetine by their GP.

We introduced a safety alert letter (see www.journalslibrary.nihr.ac.uk/programmes/hta/130430/#) after patient 1000 was recruited. This was automatically generated when we received the discharge form (see www.journalslibrary.nihr.ac.uk/programmes/hta/130430/#). The introduction of the safety fax had no effect on the frequency of inappropriate prescribing of antidepressants, and fluoxetine in particular, in the randomised patients.

The 24/7 helpline was very helpful in reducing these problems; in many cases, patients, their families or health-care staff rang the helpline to clarify the situation and we were able to reduce the non-adherence by talking with all of these groups.

Confirmation of safety outcome events and data cleaning

This was carried out by the data manager and co-chief investigator (MD) prior to unblinding of the treatment code. We did not have an event or outcome adjudication committee because there is evidence that these do not significantly improve data quality. 58 We aimed to obtain information from patients, their relatives, GPs and hospitals (usually via the research co-ordinators) to confirm the event, its nature and its date. We established some rules that were applied so that assumptions were made consistently.

Monitoring

We monitored the quality and integrity of the accumulating clinical data in accordance with a protocol agreed with the study sponsors [the Academic and Clinical Central Office for Research and Development (ACCORD) representing the University of Edinburgh and NHS Lothian], which involved central statistical monitoring, supplemented by on-site monitoring and detailed source data verification in the co-ordinating centre and triggered visits when patterns in the data at a centre seemed anomalous. All baseline data and in-hospital and 6- and 12-month outcome data were subject to verification checks built into the randomisation and data management system. In practice, almost all monitoring visits to centres were triggered by the occurrence of protocol deviations and violations, rather than any concerns about data quality.

Closeout

All trial monitoring activities, including those for trial closure, were carried out in compliance with the agreed FOCUS monitoring plan and in accordance with the trial sponsor’s standard operating procedures to ensure that all study-related activities were reconciled, recorded and reported at the end of the trial in accordance with the trial protocol and all applicable regulatory requirements and that they complied with good clinical practice.

Remote closeouts were conducted by the FOCUS central team, using a combination of sponsor-approved checklists (see the project web page: www.journalslibrary.nihr.ac.uk/programmes/hta/130430/#; accessed 7 May 2020) and trial-specific instructions and documents for reconciliation of the investigator site file for completeness. A completed closeout checklist was available for all centres. Bespoke per-site reports were provided to each site to reconcile all essential documentation required for the investigator site file in preparation for archiving. The site files included site delegation logs, documentation of staff qualifications and training (good clinical practice, curricula vitae), IMP management and reconciliation, consent confirmation, randomisation records, documentation of violations, deviations, serious adverse events and SUSARs. A final closeout visit to the co-ordinating centre and trial master file review was conducted by the trial sponsor in preparation for final closure and archiving of all essential trial master file documents.

Introduction

The statistically significant excess of bone fractures (see Table 13) that we observed is potentially important for several reasons. First, many observational studies have demonstrated an association between SSRI use and fractures. 27,28 However, the question of whether or not this association is causal has remained. RCTs provide stronger evidence of causality than any other research method. Second, if fractures affect patients’ functional outcomes, then this excess might have offset some beneficial effects of fluoxetine on functional outcomes. Third, if SSRIs, and fluoxetine specifically, cause fractures, the following question arises: what are the mechanisms by which they have this effect? Is it owing to an increased risk of falling, which could be through several possible mechanisms, or the proposed effects of SSRIs on bone density, or both, or neither?

We have, therefore, carried out some post hoc analyses in an attempt to answer the following questions:

• What sort of fractures occurred after stroke and were they likely to have had an impact on patients’ function?

• Might the increased risk of fractures, along with their associated loss of function, have offset the beneficial effects of fluoxetine on neurological recovery?

• What baseline factors were associated with fracture risk?

• Does the temporal pattern of fractures in the FOCUS trial provide any useful information about the relative importance of the potential mechanisms of SSRI-induced fractures?

Methods

We have published a brief account of further analyses and results that attempts to address these questions. 59 We extracted further information from our trial database while remaining blind to treatment allocation. Here we present the post hoc analyses in more detail. Having identified the excess risk of bone fractures in our prespecified analyses,52 we also prespecified further analyses to address our additional research questions about fractures. We coded baseline medications to distinguish non-SSRI antidepressants, blood pressure lowering, bone density reducing (e.g. glucocorticoids) or bone density increasing (e.g. calcium, vitamin D and bisphosphonates) medications and those that might increase risk of falls (e.g. tranquillisers). In addition, we extracted available data on the fracture site and any associated falls or seizures. We had not systematically collected data that indicated the side of the body where fracture(s) occurred. We compared the number of fractures occurring in those with and without specific characteristics but formally tested each variable by plotting Kaplan–Meier survival curves in those with and without each characteristic and compared these with the log-rank statistic. We did not formally test differences in fracture risk where numbers were small (i.e. fewer than six). We included all variables with a p-value of < 0.1 into a Cox proportional hazards model to identify independent predictors of fracture risk. We repeated these analyses focusing on fractures only at sites typically associated with low bone density (i.e. neck of femur, wrist and vertebrae).

Results

Temporal pattern of fractures

The Kaplan–Meier curve comparing fracture risk in the two treatment groups is shown in Figure 5. The risks appear to diverge early after randomisation, which might suggest that an increased risk of falling associated with taking fluoxetine contributed to the excess of fractures. A delayed divergence might have suggested that the excess was mainly owing to effects on bone density, which presumably would not have occurred immediately.

FIGURE 5.

Kaplan–Meier curves to 6 months, with number of subjects at risk, comparing the risk of fracture in those allocated fluoxetine and placebo where patients dying or being lost to follow-up were censored. Reproduced from Dennis et al. 59 © American Heart Association, Inc.

Reproduced from Dennis et al. 59 © American Heart Association, Inc.

Introduction

We undertook a within-trial economic analysis to estimate the cost-effectiveness of fluoxetine on an intention-to-treat basis. The primary treatment effect in the economic analysis was estimated using an individual-level regression model for average (mean) incremental costs and incremental survival times over 12 months after randomisation.

Methods

Results

Health outcomes

Mean survival times up to 12 months were 334 days (11 months) for both treatment groups. The pattern and distribution of EQ-5D-5L dimensions and levels were similar at 6 months, with little change at 12 months (Table 27). The frequency of ‘no problems’ (i.e. level 1) and ‘problems’ (i.e. levels 2 to 5) across the five dimensions at 6 months revealed a slight advantage in mobility and usual activity for the placebo group and less anxiety and depression in the fluoxetine group. These differences diminished by 12 months. The 10 most frequent EQ-5D-5L profiles accounted for around one-fifth of profiles and there was a high degree of concordance between treatment groups at 6 and 12 months (Table 28). EQ-5D-5L index values averaged around 0.47 at 6 months with no significant difference between the treatment groups; at 12 months, there was a slight increase in the mean index value to 0.50 but no sign of difference between patients allocated to fluoxetine or placebo (Table 29). Given the similar survival trajectories and EQ-5D-5L index values up to 12 months, there is no evidence of a difference in quality-adjusted survival times.

Cost-effectiveness

Table 30 and Figure 6 present the cost-effectiveness results at 12 months. Although there is a slight difference of around £500 in the mean total costs and a very small increase (< 1 day) in survival times for the fluoxetine group, neither of these differences are significant. The ICER of £2609 per day is not significantly different from zero, as illustrated in the cost-effectiveness plane (see Figure 6). The joint distribution of incremental costs and incremental survival days is essentially centred on zero.

FIGURE 6.

Cost-effectiveness plane with means centred.

Finally, given the statistically significant increase in fracture risk at 6 months in the fluoxetine group, exploratory analysis suggests that this is likely to account for around 37% of the difference in mean total costs reported above as patients who sustained a fracture reported a £13,330 (95% CI £8453 to £18,207) increase in mean total costs at 12 months.

Introduction

A 2012 Cochrane systematic review of SSRIs for stroke recovery25 suggested that SSRIs, including fluoxetine, reduced disability in stroke patients even if they did not have depression, but poor methodological quality of the trials probably introduced bias. In general, systematic reviews and meta-analyses should be updated as soon as there are new studies that might change the conclusions of the review. Thus, we, in collaboration with many other individuals (see Appendix 3), conducted an updated meta-analysis focusing specifically on the role of fluoxetine for stroke recovery, rather than on all SSRIs. We have included the results of the FOCUS trial and any other RCTs completed since the earlier review. 25 We sought to determine whether or not fluoxetine, at any dose, given within the first year after stroke to patients who did not have to have mood disorders at randomisation, compared with usual care or placebo, reduced disability and dependency at the end of treatment, reduced neurological deficits and fatigue, and improved motor function, mood and cognition at the end of treatment and follow-up, with the same number of or fewer adverse effects. The methods and results of this updated systematic review have been published63 and are included here to put the FOCUS trial results into context.

Methods

Results

From the database searches, we identified 3412 references, removed 426 duplicates, screened 2988 references and assessed 499 full texts for eligibility (Figure 7). Three published papers had the same grant number,68–70 very similar inclusion criteria and recruited patients from the same hospital during overlapping time periods; one appeared to be the 3-year follow-up data70 from one of the earlier publications. 69 Thus, we included the publication with the largest number of patients reporting our prespecified outcomes69 and categorised the other two68,70 as ‘awaiting assessment’ pending further information. We identified three further new eligible trials from the database searches66,71,72 and one by contact with experts. 73 We also included the FOCUS trial. 55

FIGURE 7.

Flow diagram showing selection of studies. Reproduced from Mead et al. 63 International Journal of Stroke. Copyright © 2019 by SAGE Publications. Reprinted by permission of SAGE Publications, Ltd.

Reproduced from Mead et al. 63 International Journal of Stroke. Copyright © 2019 by SAGE Publications. Reprinted by permission of SAGE Publications, Ltd.

These six new trials (n = 3710)55,66,69,71–73 were added to seven eligible trials12,16,17,74–78 (n = 435) identified in the 2012 Cochrane review25 (total of 13 completed trials, n = 4145, Table 31). One further registered trial was withdrawn because it recruited no patients. 79

Several ongoing RCTs together aim to recruit about 3775 patients.

Risk of bias

There were four high-quality trials (n = 3283) with a low risk of bias across important quality criteria17,55,66,72 (Figure 8). One terminated early having recruited six patients, and reported no deaths. 72

FIGURE 8.

Risk of bias. +, low risk of bias; ?, unclear risk of bias; –, high risk of bias. Reproduced from Mead et al. 63 International Journal of Stroke. Copyright © 2019 by SAGE Publications. Reprinted by permission of SAGE Publications, Ltd.

Reproduced from Mead et al. 63 International Journal of Stroke. Copyright © 2019 by SAGE Publications. Reprinted by permission of SAGE Publications, Ltd.

Results of studies and synthesis of results

Co-primary outcomes: independence (Figure 9) and disability at end of treatment (Figure 10)

FIGURE 9.

Forest plot: mRS (0–2) at the end of treatment. df, degrees of freedom; M–H, Mantel–Haenszel. Reproduced from Mead et al. 63 International Journal of Stroke. Copyright © 2019 by SAGE Publications. Reprinted by permission of SAGE Publications, Ltd.

Reproduced from Mead et al. 63 International Journal of Stroke. Copyright © 2019 by SAGE Publications. Reprinted by permission of SAGE Publications, Ltd.

FIGURE 10.

Forest plot: disability at the end of treatment. df, degrees of freedom. IV, inverse variance. Reproduced from Mead et al. 63 International Journal of Stroke. Copyright © 2019 by SAGE Publications. Reprinted by permission of SAGE Publications, Ltd.

Reproduced from Mead et al. 63 International Journal of Stroke. Copyright © 2019 by SAGE Publications. Reprinted by permission of SAGE Publications, Ltd.

Three trials (n = 3249) reported independence. 17,55,66 Fixed-effects meta-analysis found no difference in the proportion who were independent (36.6% fluoxetine vs. 36.7% control, RR 1.00, 95% CI 0.91 to 1.09; p = 0.99; I² = 78%) and no difference in disability (seven trials, n = 3404, SMD 0.05, 95% CI –0.02 to 0.12; p = 0.15; I² = 81%).

Two other trials73,78 reported improvements in mRS in the fluoxetine group but the data were in a format that could not be used in the meta-analysis and the authors did not respond to our requests for clarification.

Random-effects models demonstrated a small but statistically significant benefit of fluoxetine for disability (SMD 0.34, 95% CI 0.04 to 0.64; p = 0.03; I² = 81%), and a higher RR (RR 1.87, 95% CI 0.74 to 4.56; p = 0.19; I² = 78%) of being independent than the fixed-effects models because of the greater weight given to smaller positive trials.

Primary question: does the routine, early administration of fluoxetine (20 mg o.d.) for 6 months after an acute stroke improve patients’ functional outcome?

Our results do not support the hypothesis that a course of fluoxetine improves patients’ outcomes. We have considerable confidence that this is the correct conclusion because of the trial’s methodological strengths.

The strengths of the study, supporting the internal validity of the results, are that bias was minimised by central randomisation without any prospect of foreknowledge; blinding of patients, carers and outcome assessment (with only three episodes of unblinding); very few losses to follow-up (< 1%); and published prespecified intention-to-treat analyses. The small difference in the number of patients stopping the trial medication for perceived adverse effects suggests that unblinding owing to adverse effects was unlikely to have had a significant effect on our results. In any case, expectation bias would normally be expected to bias the result in favour of the active treatment. Random error was also minimised by randomising a large number of patients and high rates of follow-up, providing much greater statistical power than previous similar trials.

However, it is important to consider whether or not methodological factors could have caused us to miss an improvement in outcome. Possible explanations might include:

• Inadequate power to detect a small but clinically significant improvement in functional outcome. Our sample size estimate indicated that we had 90% probability of detecting an effect equivalent to a COR of 1.23 at the 5% significance level. The planned total sample size for the FOCUS, AFFINITY and EFFECTS trials,62 6000 patients, provides 90% power to detect a smaller effect size equivalent to a COR of 1.16. Our estimate of effect in the FOCUS trial was a COR of 0.951 (95% CI 0.839 to 1.079; p = 0.439; these 95% CIs, which include neither 1.16 nor 1.23, make it very unlikely that we have overlooked a treatment effect of the size that we had considered likely or of clinical importance).

• Inclusion of the wrong type of stroke patients. For instance, those included in the FLAME trial,17 on average, had more severe strokes and all had motor deficits. Our prespecified subgroup analyses (see Table 11) did not identify any interaction between our overall effect and any subgroup. We specifically explored whether or not there was a benefit for patients who had a motor deficit at baseline; there was not.

The external validity of the results, at least for the UK stroke population, is supported by the large number of participating hospitals throughout the UK. Compared with unselected stroke patients admitted to UK hospitals (Table 34), there were a few differences to the subjects enrolled in the trial. 80,81 The patients enrolled in the FOCUS trial had slightly more severe strokes (NIHSS 6 vs. 4), which probably reflected the inclusion criteria that required patients to have a neurological deficit persisting at the time of enrolment. Also, 60% were male, compared with a UK average of 50%: an unexplained but common observation in stroke trials. 82 Patients enrolled were slightly younger (mean 71.4 years) than unselected patients who were admitted to hospital with stroke in the UK, which might partly explain the male preponderance, with older women being under-represented. Many of the studies included in the previously published systematic review of RCTs of fluoxetine were from China; non-white patients made up < 5% of those recruited in the FOCUS trial. The ongoing AFFINITY trial is recruiting in Vietnam and will include a much larger proportion of Asian patients. 62

Secondary questions

Discussion of the post hoc analyses relating to fractures occurring during the 6-month treatment period

The most common site of fractures among patients with stroke enrolled in the FOCUS trial and assigned fluoxetine or placebo for 6 months was the neck of femur, and most fractures were in sites associated with osteoporosis; almost all resulted from a fall. Greater age, female sex and allocation to fluoxetine were independent predictors of subsequent fractures. An increased risk of falling is likely to explain much of the excess risk of fractures because most fractures were associated with a fall. Falls with injury were more common in the fluoxetine group than in the placebo group [n = 120 (7.67%) vs. n = 94 (6.01%); p = 0.0663], although the difference did not quite reach statistical significance and the risks in the two treatment groups diverged early after randomisation. No other baseline factors analysed had statistically significant associations with fracture risk.

These findings in an adequately powered prospective randomised trial confirm that the association between SSRI and fracture risk noted in observational studies is real and is not due to methodological biases, and that SSRI medication is likely to be causative. Our finding that greater age and female sex are associated with a greater fracture risk confirms the findings of previous observation studies in stroke. 28–30 The risk of fractures appears to increase soon after randomisation (see Figure 5) and most fractures were related to falls (see Table 21), suggesting that the predominant mechanism explaining the excess of fractures is likely to be the non-significant excess of falls seen in the FOCUS trial compared with the control group (see Table 13). This might be due to effects on cognition, co-ordination, balance or activity levels. However, we cannot exclude a contribution from fluoxetine’s possible effect on bone density.

These predefined secondary analyses have several limitations. The number of fractures in the first 6 months was modest, which means that we were only able to identify powerful predictors of fracture risk. Our only baseline indicators of bone density were previous fractures and the use of medications to reduce bone density loss at baseline. We had no direct measures of bone density. In addition, the effect of fluoxetine, and other medications, may have been diluted by non-adherence or changes to medication after randomisation. We did not collect data on current medication at the time of the fracture. Also, we did not systematically collect fractures beyond 6 months so cannot determine whether or not the effect of fluoxetine on fracture risk persists, as it might if it causes osteoporosis, or whether or not the risk subsides after stopping if it caused falls by affecting balance, etc., or increased activity by reducing depression. We did not systematically collect data on the side of the fracture so we could not reliably confirm previous findings that fractures most often affect the side of any weakness. 30

The cost-effectiveness of fluoxetine

There was no evidence that fluoxetine improves health outcomes, and it did not have a significant effect on the costs of health care, even though it reduced the frequency of new depression and increased the risk of fractures. Treatment with fluoxetine is cheap, but treatment is not justified by the clinical or economic outcomes.

The FOCUS results in the context of all similar randomised controlled trials

The updated systematic review of fluoxetine for stroke recovery identified 13 trials that recruited > 4000 patients, of which four trials (n = 3283) were of high methodological quality.

There were no differences between groups for the co-primary outcomes of dependency and disability. Fluoxetine was associated with better neurological scores at the end of treatment, lower depression scores (fewer symptoms of depression) and fewer diagnoses of depression, although the effect sizes were all small and there was substantial heterogeneity. There was a higher risk of seizures with fluoxetine. However, when only high-quality trials were considered, the only statistically significant difference between groups was lower depression scores (fewer symptoms) at the end of treatment.

We used fixed-effects models as these give appropriate weight to larger trials. The sensitivity analysis using random-effects models found a large benefit of fluoxetine on independence (RR 1.87) because of the disproportionate weight given to smaller trials. Fixed- and random-effects models produced only slightly different effect sizes for depression scores.

Previous meta-analyses suggested that fluoxetine might reduce dependency and disability if given early after stroke. 24,25,85 This meta-analysis, which includes many more patients than previous reviews, has not confirmed these promising effects. Although one of the reviews85 strongly recommended fluoxetine to promote neurological recovery, this recommendation was based on the results of just four reports,17,68,74,75 only one of which was high quality. 17

Thus, these data do not support the routine prescription of fluoxetine early after stroke in order to reduce dependency and disability. Clinicians and patients may wish to consider the routine use of fluoxetine early after stroke for its small effects on reducing the incidence of new depression, but this would need to be weighed up against the excess of seizures and bone fractures. The most important depression-related outcome after stroke is suicide, which is about double what it is in the non-stroke population; however, it is rare, and for that reason the RCTs shed no light on whether or not antidepressant treatment reduces the risk.

There are some limitations at the study and outcome level: only four trials were of high methodological quality, and not all had been registered prospectively or they reported the same outcomes. Furthermore, different scales were used for the same outcome; although we used SMD to combine data, the interpretation of SMD is not intuitive, and clinicians prefer to know the effect size on a familiar scale (e.g. Functional Independence Measure). Two large ongoing trials (AFFINITY and EFFECTS62) are using the same measures as those used in the FOCUS trial. A future meta-analysis will report mean differences for continuous data.

We did not register the review in PROSPERO, but we used almost the same methods as the 2012 Cochrane review, which had a prospectively published protocol. 25 We used sensitive searches developed by Cochrane Stroke,86 and there was complete retrieval of identified research, no language restrictions and inclusion of unpublished data.

About three-quarters of the patients were from the FOCUS trial. There was quite marked heterogeneity, even for the high-quality trials (see Table 33); this might be explained by the different types of patients and health-care settings. Five of the low-quality trials were from China; the three reporting disability all found favourable effects of fluoxetine. As the evidence base increases, it may be possible to undertake meta-regression analyses to determine the factors (e.g. country, health-care setting and trial quality) associated with good outcome.

Conclusions

The results of the FOCUS trial show that, in the context of the UK NHS, routine use of fluoxetine does not improve functional outcomes in patients with a recent stroke. There was evidence that its use reduces the risk of developing new episodes of depression after stroke, but that this is at the expense of an increased risk of adverse effects, including bone fractures.

Ongoing trials (AFFINITY and EFFECTS62), which used almost identical methods to those of the FOCUS trial, and the investigators with whom we have closely collaborated, will report in 2020. An individual patient data meta-analysis including data from FOCUS, AFFINITY and EFFECTS is planned. This will indicate whether or not the findings of the FOCUS trial are generalisable to other countries, ethnic groups and health-care systems. They will also help to provide more precise, and generalisable, estimates of the risks associated with several months of fluoxetine treatment in a predominantly elderly population. Such data are likely to be relevant to the use of fluoxetine, and perhaps other SSRIs, in similar populations (i.e. elderly patients with comorbidities). This knowledge could be of use when counselling patients about the pros and cons of starting treatment with fluoxetine.

Most previous RCTs of fluoxetine, and other SSRIs, have been carried out in patients with depression, who are often much younger than those recruited into stroke trials, and have only tested short courses with the purpose of demonstrating an improvement in mood. 87 There is a good case for studying the effects of prolonged courses of fluoxetine, not just those on mood, in older patients who are likely to be more susceptible to adverse effects. Such information would be useful in balancing the risks against the perceived benefits of using SSRIs for a variety of indications.

In advance of the FOCUS trial results becoming available, we carried out systematic reviews to determine whether or not SSRIs might be of benefit in conditions other than stroke through the same mechanisms that we thought might underpin the hypothesised benefit in stroke. All the reviews suggested promising effects of SSRIs. 88–90 However, before embarking on large trials in these other conditions, the relevance of the neutral results of the FOCUS trial needs to be considered carefully.

Contributions of authors

Professor Martin Dennis (https://orcid.org/0000-0003-1148-8972) (Stroke Physician, University of Edinburgh, UK) was co-chief investigator; participated in the TSC; was involved in the design of the trial; collected, verified and analysed data; and drafted this report.

Professor John Forbes (https://orcid.org/0000-0002-8255-3762) (Health Economist, University of Limerick, Ireland) participated in the TSC, was involved in the design of the trial and analysed health economic data.

Ms Catriona Graham (https://orcid.org/0000-0003-1889-712X) (Unblinded Statistician, University of Edinburgh, UK) participated in the TSC, was involved in the design of the trial, wrote the first draft of the statistical analysis plan and verified and analysed data.

Professor Maree Hackett (https://orcid.org/0000-0003-1211-9087) (Psychologist, University of Sydney, Australia) was involved in the trial design and helped carry out relevant systematic reviews.

Professor Graeme J Hankey (https://orcid.org/0000-0002-6044-7328) (Neurologist, University of Western Australia, Australia) was involved in the trial design and helped carry out relevant systematic reviews.

Professor Allan House (https://orcid.org/0000-0001-8721-8026) (Psychiatrist, University of Leeds, UK) was involved in the trial design and advised on the management of depression within the trial.

Professor Stephanie Lewis (https://orcid.org/0000-0003-1210-2314) (Blinded Statistician, University of Edinburgh, UK) was involved in the trial design and advised on the statistical analysis plan.

Professor Erik Lundström (https://orcid.org/0000-0002-5313-9052) (Neurologist, Karolinska Institutet, and Uppsala University, Sweden) was involved in the design of the trial.

Professor Peter Sandercock (https://orcid.org/0000-0001-8484-0135) (Neurologist, University of Edinburgh, UK) chaired the TSC of the initial phase.

Professor Gillian Mead (https://orcid.org/0000-0001-7494-2023) (Stroke Physician/Geriatrician, University of Edinburgh, UK) was co-chief investigator, participated in the TSC, was involved in the design of the trial and data collection, and co-ordinated the systematic review of the RCTs.

All members of the writing committee listed here have commented on the analyses and drafts of this report and have seen and approved the final version of the report.

Publications

Mead G, Hackett ML, Lundstrom E, Murray V, Hankey GJ, Dennis M. The FOCUS, AFFINITY and EFFECTS trials studying the effect(s) of fluoxetine in patients with a recent stroke: a study protocol for three multicentre randomised controlled trials. Trials 2015;16:369.

Graham C, Lewis S, Forbes J, Mead G, Hackett ML, Hankey GJ, et al. The FOCUS, AFFINITY and EFFECTS trials studying the effect(s) of fluoxetine in patients with a recent stroke: statistical and health economic analysis plan for the trials and for the individual patient data meta-analysis. Trials 2017;18:627.

Dennis M, Forbes J, Graham C, Hackett ML, Hankey GJ, House A, et al. Fluoxetine and fractures after stroke: exploratory analyses from the FOCUS trial. Stroke 2019;50:11.

FOCUS Trial Collaboration. Effects of fluoxetine on functional outcomes after acute stroke (FOCUS): a pragmatic, double-blind, randomised, controlled trial. Lancet 2019;393:265–74.

Mead GE, Legg L, Tilney R, Hsieh CF, Wu S, Lundström E, Hankey GJ. Fluoxetine for stroke recovery: meta-analysis of randomized controlled trials [published online ahead of print October 17 2019]. Int J Stroke 2019.

Data-sharing statement

All data requests should be submitted to the corresponding author for consideration. Access to available anonymised data may be granted following review and appropriate agreements being in place.

Patient data

This work uses data provided by patients and collected by the NHS as part of their care and support. Using patient data is vital to improve health and care for everyone. There is huge potential to make better use of information from people’s patient records, to understand more about disease, develop new treatments, monitor safety, and plan NHS services. Patient data should be kept safe and secure, to protect everyone’s privacy, and it’s important that there are safeguards to make sure that it is stored and used responsibly. Everyone should be able to find out about how patient data are used. #datasaveslives You can find out more about the background to this citation here: https://understandingpatientdata.org.uk/data-citation.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health and Social Care. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health and Social Care.

Co-chief investigators

Martin Dennis and Gillian Mead.

Writing group

Martin Dennis (chairperson), John Forbes, Catriona Graham, Maree Hackett, Graeme J Hankey, Allan House, Stephanie Lewis, Erik Lundström, Peter Sandercock and Gillian Mead.

FOCUS Trial Co-ordinating Centre

Rosemary Anderson, David Buchanan, Ann Deary, Jonathan Drever, Ruth Fraser, Catriona Graham, Karen Innes, Connor McGill, Aileen McGrath, David Perry, Pauli Walker and Carol Williams.

Telephone 6- and 12-month follow-up (numbers conducted)

Martin Dennis (1394), Carol Williams (227), Gillian Mead (220), Rosemary Anderson (173), Yvonne Chun (60), Lynn Dinsmore (50), Emma Maschauer (46), Wellcome Trust Clinical Research Facility (Greig Fraser, Katherine Lawrence and Alison Shaw) (27), Amanda Barugh (15), Shadia Mikhail (8), Gordon Blair (6), Ingrid Hoeritzauer (6) and Maggie Scott (4).

Trial Steering Committee

Stroke Association funded phase: Peter Sandercock (chairperson), Steff Lewis (statistician), Judith Williamson (patient involvement), Martin Dennis (co-chief investigator), Gillian Mead (co-chief investigator), John Forbes (health economist), Graeme J Hankey (AFFINITY), Maree Hackett (AFFINITY), Veronica Murray (EFFECTS) (deceased), Karen Innes (trial manager) and Ray French (sponsor representative).

NIHR funded phase: David Stott (independent chairperson), David Burgess (lay member), Jonathan Emberson (independent statistician), Graham Ellis, Pippa Tyrrell, Judith Williamson (patient involvement), Martin Dennis (co-chief investigator), Gillian Mead (co-chief investigator), Karen Innes (trial manager) and Ray French (sponsor representative).

Co-applicants on funding applications

Independent Data Monitoring Committee

Peter Langhorne, Fiona Reid and Helen Rodgers.

Other activities

Marian Brady (Glasgow Caledonian University) helped produce aphasia-friendly versions of the patient information leaflect and consent form (see Appendix 2). Dan Morales reviewed information sent to GPs.

Investigational medicinal product

Manufactured by Unichem (Mumbai, India).

Sourced through Niche Generics Ltd (Hitchin, UK) and Discovery Pharmaceuticals Ltd (Castle Donington, UK).

Packaged and distributed by Bilcare Ltd (Pune, India), then Sharp Clinical Services (Tredegar, UK).

Participating centres

We have listed each hospital with the total number of patients recruited [n], followed by the names of the local PI(s) and other significant contributors in that centre. The hospitals are ordered depending on the numbers recruited.

Aim

Development of accessible versions of participant information sheets and consent forms for the FOCUS trial.

Methods

Initial drafts of accessible versions of the standard FOCUS information sheets and consent forms were prepared for review by people with aphasia based on established methodologies91,92 and sound ethical principles93 by an experienced speech and language therapist and stroke rehabilitation researcher (Marian Brady).

Accessible versions of the standard materials were drafted:

• participant information sheets

• participant consent forms.

Two meetings were planned with Speakeasy (Bury, UK) members in Dundee on 28 October and again on 4 November 2011. Version 1 of the accessible materials were sent in advance of the first meeting to members of the Speakeasy group (peer support group for people with aphasia following stroke) based in Dundee. These were sent in advance to provide sufficient time for the Speakeasy members with language impairment to review and prepare in advance of the first meeting. The process of drafting and review with comments was scheduled to continue to contribute to the iterative development of accessible materials until the Speakeasy group felt that they were suitable to be used in the trial with people with aphasia.

Results

The members reviewed version 1 of the proposed accessible materials in advance of the meeting. Eight members attended. The materials were reviewed on a statement-by-statement basis with reference (where necessary) to the standard materials. In this way, we ensured that the essence of the statement contained in the standard format was retained in the accessible version, and made comments.

Version 2 of the accessible materials was prepared in response to the group’s comments. Version 2 was reviewed by the FOCUS trial team at the University of Edinburgh, including the PIs, trial manager and other trial staff. They raised some feasibility issues regarding the version 2, which was in colour and on A4 sheets. They queried whether printing the accessible versions in black and white (considerably cheaper than the colour) would be acceptable to the group and to people with aphasia. They also highlighted that printing in black and white would allow sites to easily photocopy the materials, which would support implementation of the trial at sites. They also asked whether or not the A4 size of the materials could be reduced to an A5 booklet, again to support implementation.

Version 3 of the materials was shared with the Speakeasy group (black and white and A5 version), with a specific response requested in relation to the above issues via the post to members who had participated in the first meeting. The Speakeasy members responded that although colour printing would be preferable, for cost reasons the black-and-white versions were acceptable. They also felt that although larger A4 was preferable, the A5 version would also be acceptable. All other changes were accepted, and no further changes were requested. A second scheduled face-to-face meeting was agreed to be unnecessary and was cancelled.

The ethics committee responded positively to the accessible versions of information sheets and consent forms and agreed that accessible versions could be made available to all potential trial participants as deemed required by trial staff. The ethics committee’s response to the materials was fed back to the Speakeasy members.

Discussion and conclusions

In addition, and beyond the FOCUS trial, the accessible materials used in this trial contributed to the NIHR template materials supporting the involvement of people with aphasia in stroke research. 94

Reflections/critical perspective

Involvement of people with aphasia supported the development of accessible materials to support the ethical inclusion of people with aphasia in the FOCUS trial. The perspective of people with a language impairment is vital to the development of such materials as it is not possible for people with intact language skills to perceive all the challenges experienced with a language impairment.

Preparation of accessible versions of materials in stroke research is increasingly standard practice (Big CACTUS95 and ICONS II96). There is some suggestion that reliance on proxy consent may contribute to selection bias in the participant recruitment. 97

Contributors to updated systematic review

Gillian E Mead, MD, Fellow of Royal College of Physicians of Edinburgh, Professor of Stroke and Elderly Care Medicine, University of Edinburgh, Edinburgh, UK.

Lynn Legg, Doctor of Philosophy (PhD), Senior Researcher, Department of Medicine for the Elderly, Royal Alexandra Hospital, Paisley, UK.

Russel Tilney, Doctor of medicine (MD), Department of Medicine, Mater Dei Hospital, Malta.

Cheng Fang Hsieh, MD, Division of Geriatrics and Gerontology, Department of Internal Medicine and Department of Neurology, Kaohsiung Medical University Hospital, Kaohsiung, Taiwan.

Simiao Wu, MD, PhD, Department of Neurology, West China Hospital, Sichuan University, Chengdu, China.

Erik Lundström, Associate Professor of Neurology, Karolinska Institutet, Department of Clinical Neuroscience; and Adjunct Senior Lecturer at Department of Neuroscience, Neurology, Uppsala University, Uppsala, Sweden.

Ann Sofie Rudberg, MD, Department of Clinical Neuroscience, Karolinska Institutet, Stockholm, Sweden.

Mansur Kutlubaev, MD, PhD, Department of Neurology, G.G. Kuvatov Republican Clinical Hospital; Department of Neurology, Bashkir State Medical University, Ufa, Russia.

Martin S Dennis, Professor of Stroke Medicine, University of Edinburgh, Edinburgh, UK.

Babak Soleimani, MB, Bachelor of Medicine and Bachelor of Surgery, Clinical Fellow, NHS Lothian, UK.

Amanda Barugh, PhD, Consultant Stroke Physician, NHS Lothian, UK.

Maree L Hackett, PhD, Programme Head Mental Health, Associate Professor, Faculty of Medicine, University of New South Wales, Sydney; Professor of Epidemiology, The University of Central Lancashire, UK; The George Institute for Global Health, Australia.

Graeme J Hankey, MD, Fellow of the Royal Australasian College of Physicians, Professor of Neurology, Medical School, The University of Western Australia, WA, Australia.

Acknowledgements

Joshua Cheyne of Cochrane Stroke ran the literature searches.

Maureen Harding obtained articles for full-text review.

Dr Juan Marquez Romero provided unpublished data from the fluoxetine for motor recovery after acute intracerebral hemorrhage (FMRICH) trial.

Professor Chollet responded to requests for information about other new trials.

Dr Jan Bembenek and Professor Anna Czlonkowska provided further information about FOCUS-Poland.

Professor Peter Sandercock commented on the manuscript.

Contributions

Gillian Mead conceived the study, screened references, extracted data, assessed risk of bias, carried out the analyses and wrote the first draft.

Lynn Legg searched for selected studies for inclusion, collected data, assessed risk of bias, managed studies through the review process and contributed to the final version.

Russel Tilney, Cheng Fang Hsieh, Simiao Wu, Erik Lundström, Ann-Sofie Rudberg, Mansur Kutlubaev, Babak Soleimani and Amanda Barugh screened citations, retrieved potentially relevant papers and screened their eligibility for the systematic review, assisted with data extraction and drafted the manuscript for submission.

Maree Hackett extracted data and edited the final manuscript.

Graeme J Hankey and Martin Dennis conceived the study, provided expertise in relation to analysis methods and edited the draft paper.

Funding

This review was not specifically funded. Maree L Hackett held a National Health and Medical Research Council (Australia) Career Development Fellowship, level 2 (reference APP1141328) (2018–21).

Search strategy for updated systematic review

The following searches were carried out (Cochrane Stroke Group Trials Register; date searched: 17 July 2018):

• Cochrane Central Register of Controlled Trials (CENTRAL) (via The Cochrane Library) (date range searched: 2018, issue 6).

• MEDLINE (via Ovid) (date range searched: 1948 to 17 July 2018).

• EMBASE (via Ovid) (date range searched: 1980 to 17 July 2018).

• Cumulative Index to Nursing and Allied Health Literature (CINAHL) (via EBSCOhost) (date range searched: 1982 to 17 July 2018).

• Allied and Complementary Medicine (AMED) (via Ovid) (date range searched: 1985 to 17 July 2018)

• PsycINFO (via Ovid) (date range searched: 1967 to 17 July 2018).

• US National Institutes of Health Ongoing Trials Register [via ClinicalTrials.gov (www.clinicaltrials.gov)].

• World Health Organization International Clinical Trials Registry Platform (apps.who.int/trialsearch) (date searched: 26 June 2018).