Scope
This resource covers systematic reviews and meta-analyses of home-based rehabilitation for adults after stroke. Thirty-two reviews contributed to the umbrella review. The present release contains 66 direct-comparison evidence units derived from RCT-level data.
Each reanalysis preserves a specific intervention, comparator, outcome construct, data form, and assessment time. Different active comparators are not silently combined.
Evidence selection and overlap
Candidate reviews were compared by PICO fit, search recency, methodological quality, data completeness, and statistical integrity. RCT identities were harmonised through trial-family mapping to avoid counting the same participants more than once.
Shared controls, multiple treatment arms, duplicate reviews, and repeated time points were resolved before analysis. Values that could not be verified were excluded rather than reconstructed from a pooled diamond.
Uniform RCT-level reanalysis
Arm-level sample size, mean, and standard deviation were used to calculate Hedges g. Positive values are oriented toward home-based rehabilitation benefit.
Arm-level events and totals were used to calculate risk ratios. Direction is interpreted according to whether the recorded event is beneficial or adverse.
Random-effects models were used consistently. Each result reports a pooled estimate and 95% confidence interval; heterogeneity and a 95% prediction interval are shown when estimable. Forest plots use one visual system across all 66 analyses.
Evidence certainty
The legacy GRADE table refers to earlier evidence units and is awaiting reassessment against the final standardised reanalyses. Therefore, the current public release displays Not Reported for all 66 reanalysis units. It does not recode missing certainty as Very low.
Very confident in the estimated effect.
Moderately confident; the true effect may differ.
Limited confidence in the estimated effect.
These remain separate labels in all detail views.
Interpretation rules
Colour communicates the effect signal; dots communicate certainty. Statistical significance alone does not establish clinical importance. Standardised effects support consistent display across scales but do not replace instrument-specific minimal clinically important differences.
- Clinically relevant benefit: the display classification indicates a larger favourable standardised signal.
- Statistical benefit: a favourable statistical signal of smaller magnitude.
- No clinically relevant effect: the estimate remains near the predefined display null zone.
- Uncertain: the confidence interval does not support a clear direction.
- Harm: a statistically or clinically relevant signal favouring the comparator.
Creator-managed updates
The creator console uses authenticated owner access. Every creation, edit, publication, or archive action writes a complete immutable snapshot to the version log. Concurrent edits are rejected through optimistic version checks. When a numeric result changes without a replacement image, the older forest plot is removed from the live record to prevent result–figure drift.
Limitations
The platform inherits limitations of the included reviews and original trials, including sparse long-term evidence, inconsistent intervention descriptions, variable comparators, and incomplete adverse-event reporting. These are direct comparisons, so effect magnitudes must not be interpreted as a treatment ranking.
Reference framework
The information architecture was informed by the EBI-ADHD evidence platform and its accompanying BMJ umbrella review. This HomeStroke implementation uses original code, visual design, and stroke-specific evidence data.
EBI-ADHD evidence platform ↗BMJ umbrella review ↗