| Study |
Bias |
| Randomisation process |
Deviations from intended interventions |
Missing outcome data |
Measurement of the outcome |
Selection of the reported results |
Overall |
| Authors' judgement |
Support for judgement |
Authors' judgement |
Support for judgement |
Authors' judgement |
Support for judgement |
Authors' judgement |
Support for judgement |
Authors' judgement |
Support for judgement |
Authors' judgement |
Support for judgement |
| Subgroup 1.3.1 Rehabilitation‐related activities |
| Abizanda 2011 |
Low risk of bias |
Allocation sequence was random (computer generated random numbers). Allocation occurred after enrolment by an investigator not involved in the participants' clinical management. Higher number of participants admitted with a stroke in the intervention group (39 vs. 16), in addition, the 'Others' subgroup the Barthel index is higher in the control group than the intervention group (29.1 vs. 23.1). This is thought to be compatible with chance. |
Low risk of bias |
Its assumed participants and the Occupational Therapist (OT) delivering the intervention were aware of the assigned intervention allocation. All participants received their allocated intervention. Therefore, even though intention to treat analysis not specified, analysis deemed appropriate. |
Low risk of bias |
Given the nature of this population (older adults during and after acute hospitalisation) we considered a threshold of 90% of participants with data as sufficient. We felt this would be consistent with measurements outside of a trial context. When accounting for mortality, 93% and 91% assessed at discharge in intervention and control arms respectively. |
Low risk of bias |
The use of the Confusion Assessment Method (CAM) to assess for delirium is considered appropriate, and there were no differences in measurement or ascertainment of the outcomes between groups. The OT assessor was blinded. |
Low risk of bias |
The data analyst was blinded, and followed a pre‐specified statistical plan. It is not thought that results were from multiple outcome measurements or multiple analyses. |
Low risk of bias |
The study is judged to be at low risk of bias in all domains for this outcome. |
| Asplund 2000 |
Low risk of bias |
The method of allocation sequence is not described, but the use of sealed envelopes suggests a random component was used. Participant characteristics were balanced. |
Some concerns |
Participants and those delivering the interventions were aware of intervention assignments. Twenty‐five patients were excluded due to them not meeting the set eligibility criteria, however, these: protocol violations are not expected to influence the effect estimate of the outcome as per protocol analyses used. The per protocol analyses was not thought to have a substantial impact on the result given the main reason for exclusion was inappropriate recruitment. Excluding this 25, the other 6 exclusions represent approximately 1% of the total sample size. |
Low risk of bias |
Data at discharge from hospital was available for 98% of participants. |
High risk of bias |
The confusion assessment method (CAM) instrument was considered appropriate, and there were no differences in the measurement or ascertainment between groups. The assessors were not blinded, and it was therefore considered likely that knowledge of the intervention could influence the outcome, given the likely strong belief in the benefits of the intervention ward. |
Some concerns |
A pre‐specified statistical plan was not found, but it is not thought that the results were from multiple outcome measures or multiple analyses. |
High risk of bias |
The study is judged to be at high risk of bias due to measurement of the outcome. |
| Subgroup 1.3.2 Structured exercise |
| Brown 2016 |
Low risk of bias |
No description of randomisation sequence generation, but the paper describes a block randomisation strategy and sequence concealed (sealed envelopes). The patient characteristics appear well‐balanced between groups. |
Low risk of bias |
We assume both participants and those delivering the interventions were aware of intervention assignments. 6 participants did not receive allocated intervention, this is considered to be consistent with what could occur outside the trial context, and intention to treat analysis was used. |
Low risk of bias |
No missing data. |
High risk of bias |
The Confusion Assessment Method (CAM) score is considered appropriate, and there were no differences in the measurement or ascertainment between groups. The assessors were blinded at admission, discharge and follow up, but not the research assistants as they were delivering the intervention. “The research assistants also used the CAM at each patient visit throughout the hospital stay to ensure that patients in either group did not develop incident delirium.” Therefore, assessors not considered to be blinded, and it is likely given that there is judgement involved in scoring of the CAM, that knowledge of the intervention could influence the scoring of the CAM given the likely strong beliefs in the benefits of the intervention ward. |
Some concerns |
The study protocol details statistical analysis plan which did not refer to assessment of delirium after baseline assessments, but it is not thought that the results were from multiple outcome measures or multiple analyses. |
High risk of bias |
The study is judged to be at high risk of bias in measurement of the outcome. |
| Subgroup 1.3.3 Progressive resistance exercise |
| Jeffs 2013 |
Low risk of bias |
Sequence concealment was achieved via a member of the research team not involved in recruitment and using sealed envelopes. There was no information regarding method of generating random sequence, other than to say that there was randomisation, given details of allocation concealment, assumption made that appropriate method used. Patient characteristics between groups appear well‐balanced. |
Low risk of bias |
Participants and those delivering the interventions were likely to be aware of intervention allocations. One participant was not allocated to a group due to an administrative error and 35 participants did not receive the intervention as planned (17 in the intervention group). This is thought to be consistent with what could occur outside of the trial context. Intention to treat analysis was used. |
Low risk of bias |
No missing data. |
Low risk of bias |
The Confusion Assessment Method (CAM) is considered appropriate, and there were no differences in the measurement or ascertainment between groups. The assessors were blinded. |
Some concerns |
A pre‐specified statistical plan was not found, but it is not thought that the results were from multiple outcome measures or multiple analyses. |
Some concerns |
The study is judged to raise some concerns in the selection of the reported results. |
| Martinez‐Velilla 2019 |
Low risk of bias |
Allocation sequence was random (www.randomizer.org) and personal communication with author confirms allocation concealment. Baseline differences between groups are thought to be compatible with chance. |
Low risk of bias |
Both participants and clinicians delivering care were aware of assigned interventions. There is no evidence of deviations from intended interventions and intention to treat analysis used. |
Some concerns |
Assuming that those that discontinued the study did not provide Barthel Index (BI) outcome data, and accounting for mortality, 84% of intervention group and 86% of control group had outcome data. Although there were participants who discontinued the study due to medical deterioration, which could have and is considered likely to have biased the results (i.e. medical deterioration being associated with poor functional outcomes), the numbers were relatively low (only 6% of the sample) and were well‐balanced between groups. We therefore feel that this was unlikely to bias the results. |
Low risk of bias |
The use of the Confusion Assessment Method (CAM) to measure delirium is considered appropriate, and there were no differences in the measurement or ascertainment between groups. The assessors were blinded. |
Low risk of bias |
The trial protocol reflects the analyses as that were conducted, and it is not thought that the results were from multiple outcome measures or multiple analyses. |
Some concerns |
The study is judged to raise some concerns due to missing outcome data. |
| McCullagh 2020 |
Low risk of bias |
Allocation sequence was random (computer‐generated) and allocation sequence was concealed. Baseline differences are thought to be compatible with chance. |
Low risk of bias |
Both participants and clinicians delivering care were aware of assigned interventions. There is no evidence of deviations from intended interventions and intention to treat analysis used. |
Low risk of bias |
90% of intervention group and 94% of the control group had outcome data. Given the nature of this population (older adults during and after acute hospitalisation) we considered a threshold of 90% of participants with data as sufficient. |
Low risk of bias |
The use of the Six‐item Cognitive Impairment Test (6CIT) to assess for delirium is considered appropriate, and there were no differences in the measurement or ascertainment between groups. The assessors were blinded. |
Low risk of bias |
The trial protocol reflects the analyses as that were conducted, and it is not thought that the results were from multiple outcome measures or multiple analyses. |
Low risk of bias |
The study is judged to be at low risk of bias in all domains for this outcome. |
| Mudge 2008 |
High risk of bias |
Pseudo‐randomisation, allocation was based on admitting unit and bed availability, and admitting unit was determined by a rotating roster. No evidence of baseline differences in patient characteristics other than those thought to be compatible with chance. |
Low risk of bias |
Both participants and clinicians delivering care were aware of assigned interventions. There is no evidence of deviations from intended interventions and intention to treat analysis used. |
Low risk of bias |
No missing data. |
Low risk of bias |
Identifying delirium according to chart review using validated methodology is considered appropriate, and there were no differences in the measurement or ascertainment between groups. The assessors were blinded. |
Some concerns |
A pre‐specified statistical plan was not found, but it is not thought that the results were from multiple outcome measures or multiple analyses. |
High risk of bias |
The study is judged to be at high risk of bias due to the randomisation process. |
