Forster 2015.

Study characteristics
Methods	Aim of study: an effectiveness‐implementation hybrid trial assessing clinical effectiveness and implementation. Authors wanted to determine the effectiveness of implementing a new postdischarge system of care (LoTS) targeting longer‐term problems experienced by stroke survivors and their carers. Study design: cluster randomised trial Unit of randomisation: stroke care co‐ordinator service Mean cluster size: intervention 26.7 patients (range 2–45), control 28.5 (range 15–46) Unit of analysis: patient Sample size calculation: study was designed with 90% power at 5% significance to detect a clinically relevant difference of 2.5 points in the primary outcome (GHQ‐12) based on a previous study. 40 clusters involving 800 patients overall (20 patients per cluster) was determined to be needed, accounting for an estimated 25% loss to follow‐up and clustering; an inflation factor of 1.95 was derived from the maximum cluster size of 20 and an ICC > 0.05. The study was able to identify 32 eligible clusters willing to participate, so planned for each cluster to recruit 25 patients providing 88% power, assuming equal cluster size and ≤ 25% loss to follow‐up. To minimise unequal recruitment clusters, maximum number of patients per service was capped at 45.
Participants	Healthcare professionals: 29 clusters of stroke care co‐ordinators or stroke care co‐ordinator services. Number and demographics of staff within clusters not provided. 15 intervention clusters, 14 control clusters Patients: stroke survivors (n = 800) Experimental group: n = 401, 53.6% men, mean age 70.9 (SD 13.18) years Control group: n = 399, 54.6% men, mean age 72.5 (SD 12.84) years Ethnicity: Intervention group: white 96.8% Control group: white 97.5% Carers: n = 208 Experimental group: n = 108, 32.5% men, mean age 61.0 (SD 15.02) years Control group: n = 100, 32% men, mean age 61.4 (SD 14.07) years Setting: community‐based setting: outpatients and home‐based care Country: UK
Interventions	Targeted behaviour change: for staff to use a new service model (the LoTS system) with stroke survivors living in the community Theory used: the MRC Framework was used for the evaluation of a complex intervention. No implementation theory mentioned. Category of implementation intervention: delivery arrangements and implementation strategies Multifacetedintervention: yes Tailoring to barriers: not described Description of intervention: co‐ordination of care and management of care processes, interprofessional education, educational materials (see Appendix 2 for full details of intervention) Delivery: centrally based, face‐to‐face training by experienced clinicians Frequency: 2 × 1‐day workshops approximately 1 month apart Duration of intervention: 4‐month period of 'hands‐on experience' following training (or 12 months LoTS care delivery to patients) Control: no intervention; care delivered as usual by stroke care co‐ordinators as determined by local policy and practices
Outcomes	Main outcome: Patient outcomes: improved stroke survivor psychological well‐being measured by the GHQ‐12 Secondary outcome: Patient outcomes: FAI Barthel Index EQ‐5D Patient LUNS questionnaire Death Hospital readmission Institutionalisation Total costs Cost‐effectiveness/cost utility Carer outcomes: GHQ‐12 CBS Satisfaction Death Institutionalisation Follow‐up: 6 and 12 months after recruitment Loss of clusters and individuals: no clusters lost to follow‐up once participation began. Patients lost to follow‐up at 6 and 12 months: Intervention group: 6 months: n = 91 (22.7%), 12 months: 120 patients (32.8%) Control group: 6 months: n = 99 (24.8%), 12 months: 131 patients (29.9%) Adjusted for clustering for each outcome: yes Method of cluster adjustment for each outcome: 2‐level linear model used for primary and secondary analysis, models were adjusted for patient‐level covariates and stroke unit‐level covariates. ICC reported for each outcome: yes, for primary and secondary outcomes
Notes	Outcomes used in this review: EQ‐5D, GHQ‐12, Barthel Index, FAI, death, hospital readmission We pooled death rates from this trial as per guidance in the Cochrane Handbook for Systematic Reviews of Interventions Section 23.1.4.1 and applied a design effect to the number of participants and events to account for the clustering (Higgins 2019). Unit of analysis error: no Ethical approval and informed consent obtained: yes, ethics approval from Leeds West and Scotland Research Ethics Committees. Written informed consent obtained from patients and carers (if appropriate) Funding source: NIHR (Programme Grants for Applied Research Programmed; RP‐PG‐0606‐1128) and The Stroke Association (TSA 2006/15). Declarations of interest: none declared or detected. Authors stated in 'Sources of funding' this is independent research. Contact with author? yes Additional outcome data provided from author? no Trial registration: ISRCTN67932305
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Stroke units were randomly assigned (1:1) to groups (process not described), stratified by quality of stroke unit, annual number of referrals, stroke care co‐ordinators working alone or within community‐based team. Clusters randomised in 2 phases due to delays in research approval. Method of obtaining balanced randomisation based on a peer‐reviewed published algorithm for cluster randomisation.
Allocation concealment (selection bias)	Low risk	Cluster randomisation was performed centrally by a CTRU. Recruitment of trial participants was by independent research staff who were blinded to which group they were recruiting to. Screening and recruitment data were collected and analysed.
Blinding of participants and personnel (performance bias) All outcomes	Low risk	States stroke care co‐ordinators were unaware which of their patients had consented to participate. Stroke participants and their carers likely blinded though not specifically reported. Intervention (care plan) was designed so that it replaced previous patient documentation and became embedded in standard practice; participants were new stroke patients so likely not to notice change in standard practice.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Outcomes were self‐report measures from patients and carers who were likely blinded.
Incomplete outcome data (attrition bias) All outcomes	Low risk	No clusters lost to follow‐up. Response rates for patient‐reported outcomes were comparable: 75.2% in control and 77.3% in intervention at 6 months and 67.2% in control and 70.1% in intervention at 12 months.
Selective reporting (reporting bias)	Low risk	Prospectively registered trial with published protocol, prespecified primary and secondary outcomes reported.
Other bias	Unclear risk	Recruitment bias: low risk Incorrect analysis: low risk Similar baseline characteristics: low risk, baseline characteristics of patients balanced across control and intervention groups. Differences between groups in language or cognitive impairment and length of inpatient stay, which authors noted. Language and cognitive impairment were accounted for in statistical modelling. Similar baseline outcome measures: low risk, some imbalance in baseline outcome measures between intervention and control groups; however, the authors used a multilevel statistical model to adjust for patient‐level covariates, e.g. baseline Barthel Index and GHQ‐12 score. Reliability of primary outcome measures: unclear risk, validity of GHQ‐12 version not fully established in a stroke population, authors only stated measure was short and easy to complete and was consistent with the high prevalence of psychological symptoms after stroke. Adequate protection against contamination: allocation by stroke care co‐ordinator service. Stratification addressed co‐ordinators working individually or within a team.