Is more better? Using metadata to explore dose-response relationships in stroke rehabilitation

Abstract

Background and purpose: Neurophysiological models of rehabilitation and recovery suggest that a large volume of specific practice is required to induce the neuroplastic changes that underlie behavioral recovery. The primary objective of this meta-analysis was to explore the relationship between time scheduled for therapy and improvement in motor therapy for adults after stroke by (1) comparing high doses to low doses and (2) using metaregression to quantify the dose-response relationship further.

Methods: Databases were searched to find randomized controlled trials that were not dosage matched for total time scheduled for therapy. Regression models were used to predict improvement during therapy as a function of total time scheduled for therapy and years after stroke.

Results: Overall, treatment groups receiving more therapy improved beyond control groups that received less (g=0.35; 95% confidence interval, 0.26-0.45). Furthermore, increased time scheduled for therapy was a significant predictor of increased improvement by itself and when controlling for linear and quadratic effects of time after stroke.

Conclusions: There is a positive relationship between the time scheduled for therapy and therapy outcomes. These data suggest that large doses of therapy lead to clinically meaningful improvements, controlling for time after stroke. Currently, trials report time scheduled for therapy as a measure of therapy dose. Preferable measures of dose would be active time in therapy or repetitions of an exercise.

Keywords: rehabilitation; stroke; therapy.

Figure 1

Funnel plot (A) showing effect-sizes (g) as a function of precision (standard error). Asymmetry was not significant. Forest plot (B) showing the effect-sizes and 95% confidence intervals for each study and the summary effect-size from the random-effects model. Positive values show a difference in favor of increased time scheduled for therapy. RE=random effects.

Figure 2

Observed effect-size (g) for each study as a function of additional time scheduled for therapy (A) and as a function of years post-stroke (B).

Figure 3

Predicted effect-size (ĝ) as a function years post-stroke (x-axis) and select values of additional time scheduled for therapy (separate lines). Model 3 (A) includes the linear effect of time scheduled for therapy. Model 4 (B) includes the linear and quadratic effects of time scheduled for therapy. The dashed black line (+0 hrs) represents the predicted effect-size when no additional time is scheduled for therapy between treatment and control groups.