Effects of Intensity of Rehabilitation After Stroke

A Research Synthesis

1. Gert Kwakkel, MSc; 
2. Robert C. Wagenaar, PhD; 
3. Tim W. Koelman, PT;
4. Gustaaf J. Lankhorst, MD, PhD; 
5. Johan C. Koetsier, MD, PhD

Abstract

1. 1. Background and Purpose A research synthesis was performed to (1) critically review controlled studies evaluating effects of different intensities of stroke rehabilitation in terms of disabilities and impairments and (2) quantify patterns by calculating summary effect sizes. The influences of organizational setting of rehabilitation management, blind recording, and amount of rehabilitation on the summary effect sizes were calculated.

      Methods A Medline literature search was performed for a critical review of the literature. The internal and external validity of the studies was evaluated. In addition, a meta-analysis was performed by applying the fixed (Hedges’s g) effects model.

      Results The effects of different intensities of rehabilitation were studied in nine controlled studies involving 1051 patients. Analysis of the methodological quality revealed scores varying from 14% to 47% of the maximum feasible score. Meta-analysis demonstrated a statistically significant summary effect size for activities of daily living (0.28±0.12). Lower summary effect sizes (0.19±0.17) were found for studies in which experimental and control groups were treated in the same setting compared with studies in which the two groups of patients were treated in different settings (0.40±0.19). Variables defined on a neuromuscular level (0.37±0.24) showed larger summary effect sizes than variables defined on a functional level (0.10±0.21). Weighting individual effect sizes for the difference in amount of rehabilitation between experimental and control groups resulted in larger summary effect sizes for activities of daily living and functional outcome parameters for studies that were not confounded by organizational setting.

      Conclusions A small but statistically significant intensity-effect relationship in the rehabilitation of stroke patients was found. Insufficient contrast in the amount of rehabilitation between experimental and control conditions, organizational setting of rehabilitation management, lack of blinding procedures, and heterogeneity of patient characteristics were major confounding factors.
      Key Words:
      • activities of daily living
      • cerebrovascular disorders
      • meta-analysis
      • rehabilitation
      In a 1989 review of efficacy of rehabilitation after stroke, Dobkin¹ asserted that “convincing evidence concerning the therapeutic usefulness of stroke rehabilitation does not yet exist.” Other reviewers claim that any of the available approaches in physical and occupational therapy will improve the patient’s functional status,^2 3although spontaneous neurological recovery appears to account for most of the improvement in functional ability.^4 5 Wagenaar and Meyer^6 7 concluded that the effects of rehabilitation methods are highly specific, indicating that improvements on the impairment level do not generalize to disabilities (ADL). In addition, they found that expert care has a positive effect on functional recovery compared with traditional care. However, in their opinion it remains unclear which part of expert care (eg, team care, active family participation, special staff education, early start of treatment, and/or intensity of treatment) has caused this effect. Recently, Gladman and colleagues⁸ stated that “till so far the knowledge about what it is in the ‘black box’ of a stroke unit that is effective is lacking.” The results of a few controlled studies suggest that an early start of intensive treatment is an important aspect of expert care.^4 6 7
      Langhorne et al^9 10 11 showed in a meta-analysis combining the findings of 10 randomized controlled trials that stroke rehabilitation wards statistically significantly reduce mortality (approximately 28%) in comparison to general medical wards. In a similar approach, Ottenbacher and Jarnell¹² demonstrated that programs of focused stroke rehabilitation may improve functional performance for some patients who have sustained a stroke. They have combined the findings of 36 studies on the effects of stroke rehabilitation wards as well as different methods of rehabilitation in one summary effect size.
      In a more recent meta-analysis, Langhorne et al¹³ combined the results of seven randomized trials on the effects of different intensities of physical therapy and reported a small but significant reduction in mortality as well as significant improvements in ADL and impairments as a result of higher intensities of treatment. Sensitivity analysis revealed that the organizational setting in which the physiotherapy was delivered was an important confounding factor; studies in which the experimental and control conditions were applied in different settings (confounded studies) resulted in a smaller overall treatment effect than studies carried out in one setting (unconfounded studies).
      The purpose of this article was to add a critical review of studies evaluating the efficacy of different intensities of stroke rehabilitation to trace variables that may influence rehabilitation outcome. The impact of these factors was then analyzed by including them as an independent variable in the calculation of the summary effect size. Since the present study includes two more studies than the analysis of Langhorne et al,¹³ first the effect size of different intensities of rehabilitation in terms of disabilities and impairments will be evaluated.
      Previous SectionNext Section

Materials and Methods

Study Identification

1. 1. A Medline literature search was performed for the period 1966 to 1995 (CD-ROM, Medical Subject Headings, volume 36, 1996). We used the key words (MeSH) stroke, cerebrovascular disorders, dose-response relationship, effectiveness, cost-effectiveness, rehabilitation, therapy, physical therapy, physiotherapy, occupational therapy, and exercise therapy. References presented in relevant publications were examined, and abstracts published in proceedings of conferences were searched. Investigators of studies were contacted if more information about the trial was needed. Studies had to meet the following criteria: (1) patients suffering from a stroke are studied; (2) the effects of different intensities of PT and/or OT are evaluated; (3) the study concerns true or quasi-experimentation¹⁴ ; (4) the rehabilitation outcome is measured in terms of ADL; and (5) the study is published in a journal or book.

Critical Review

1. 1. A methodological quality score was developed with items recommended by the Potsdam standards¹⁵ and other investigators in this field^{16 17 18 19 20} to identify independent variables (eg, elements of study design and contrast in amount of therapy) that might modify the overall effect size.

      The following items were evaluated: (1) randomization or matching procedures; (2) blinding procedures; (3) description of dropouts and intention-to-treat analysis; (4) reliability and validity of assessment instruments; (5) control for cointervention(s); (6) comparability of baseline patient characteristics; and (7) control for amount of therapy (see “Appendix”). To each item a binary weight (0/1) was attached. In total 16 items were scored. Two reviewers (G.K., E. van Wegen) assessed the methodological quality of each study independently. Names of author(s), institution(s), and journal were masked to establish independent extraction of relevant data. Interrater reliability of individual items was assessed with Cohen’s κ with adjustment for tied ranks. In a meeting the reviewers tried to accomplish agreement on differences in scoring. When disagreement persisted, a third reviewer (R.C.W.) made the final decision.

Summary Effect Sizes

1. 1. The effect size g_i (Hedges’s g) of each individual study was calculated by the difference between means of experimental and control group divided by the average population standard deviation (SD_i).²¹ To estimate SD_i for g_i’s, baseline estimate standard deviations of control and experimental groups were pooled.^21 22Alternatively, Hedges’s g_i estimates were obtained from probability and t values.²³Since the g_i’s tend to overestimate the population effect size in studies with a small number of patients, a correction was made to obtain an unbiased estimator g^u.²¹²³ The impact of sample size was addressed by estimating a weighting factor w_i for each study, assigning larger weights to effect sizes from studies with larger study samples and thus smaller variances. Subsequently, g^u’s of individual studies were averaged, resulting in a weighted summary effect size (¯T^u).^{21 22 23 24} Finally, the w_i’s were combined to estimate the variance of the summary effect size T̄^u.²⁴

      On the disability level, effect size T̄^u was computed for global outcome of ADL (eg, Barthel Index score) as well as separate domains of functional outcome (eg, dexterity, walking performance, and walking velocity). On the impairment level, neuromuscular outcome variables were used for assessing recovery of hemiplegia (eg, muscle strength and synergism).

      In one study the effects of three treatment conditions were compared.²⁵ Only the comparisons of the two experimental conditions with the control condition were included. Including two of three possible comparisons in one overall summary effect size would have introduced “dependency of findings,” which would violate further statistics. Therefore, summary effect sizes were also calculated including each comparison from the same study separately.

      The homogeneity (or heterogeneity) test statistic (Q statistic) of each set of effect sizes was examined to determine whether studies shared a common effect size of which the variance could be explained by sampling error alone.^21 26 The fixed effects model was used to decide whether a summary effect size was statistically significant.^27 28 If a significant heterogeneity in summary effect sizes was found, a random effects model was applied.

      Post hoc analyses were performed for the organizational setting of rehabilitation management,¹³ the amount of rehabilitation given in the experimental and control group, and the effects of blinding. The post hoc analysis for organizational setting was performed by comparing the summary effect sizes of studies in which all patients were treated in the same setting of rehabilitation with those studies in which experimental and control groups were treated in different settings. In a similar vein, a post hoc analysis for blinding was performed.

      A sensitivity analysis was performed on the amount of rehabilitation by weighting the additional time spent (t_i) on PT and OT in the experimental group compared with the control group by dividing the difference by the total amount of rehabilitation in the experimental group. In this way, studies with a proportionally greater difference in amount of rehabilitation between groups were given more weight than studies in which this contrast was small. For all outcome variables, the critical value for rejecting H₀ was set at .05.
      Previous SectionNext Section

Results

1. 1. A total of 11 studies (3 preexperiments,^28 29 30 1 quasi- experiment,³¹ and 8 true experiments^{25 32 33 34 35 36 37 38} ) were identified as being relevant (Table 1⇓). The preexperiments were excluded because of a lack of experimental control.

      View this table:

      • In this window
       
      • In a new window

      Table 1.

      Study Characteristics of Trials on Intensity of Stroke Rehabilitation

Critical Review

1. 1. The results of the methodological quality score of the 9 trials involving 1051 patients are presented in Table 2⇓. Initially there was disagreement between two independent reviewers on 10 (7%) of the 144 criteria scored. Cohen’s κ was 0.86. The methodological quality score varied from 13%^32 33 34 to 47%^35 37 38 of the maximum feasible score.

      View this table:

      • In this window
       
      • In a new window

      Table 2.

      Application of Methodological Criteria to Studies Investigating the Effects of Intensity of Rehabilitation

      Eight studies were classified as randomized trials,^{25 32 33 34 35 36 37 38} of which only 2 studies reported the method of randomization.^33 34 The observers were blinded in 4 studies,^{35 36 37 38} and dropouts were described for experimental and control groups separately in 7 studies.^{25 33 34 35 36 37 38} However, none of these controlled trials reported intercurrent dropouts or applied an intention-to-treat analysis. Three studies were confounded by the organizational setting. In 3 studies the patients in the experimental and control groups were comparable for age, initial ADL index, and type of stroke.^35 36 37

      On average, the intensive rehabilitation group received daily almost twice as much PT (ie, 48.4 minutes) and OT (ie, 44 minutes) as the control group (ie, 23.4 and 18.5 minutes, respectively) (Table 1⇑). However, this difference in amount of rehabilitation was larger in the unconfounded studies (n=6)^{25 32 35 36 37 38} than in confounded studies (n=3).^31 33 34 The contrast t_i between rehabilitation intensities of unconfounded and confounded studies was 0.68 and 0.34, respectively.

Meta-analysis

ADL

1. 1. The unbiased overall summary effect size T̄^u for the 9 studies was 0.28 (CI, ±0.12) SDU (Fig 1⇓ and Table 3⇓). The test statistic for heterogeneity was not statistically significant. Controlling for dependency of estimated effect sizes as a result of including two multiple comparisons from the study of Smith et al²⁵ showed almost similar summary effect sizes (0.28 to 0.26 SDU) as well as standard errors (±0.13). Weighting each study by factor t_i resulted in a slight increase of overall summary effect size T̄^u from 0.28 (CI, ±0.12) to 0.34 (CI, ±0.15) SDU (Table 3⇓).

      

      View larger version:

      • In this page
       
      • In a new window

      • Download as PowerPoint Slide

      Figure 1.

      Ninety-five percent CIs of effect sizes for outcome of ADL. Positive effect sizes indicate that intensive rehabilitation was more effective for ADL than control conditions; negative effect sizes indicate that the control regimen was more effective.

      View this table:

      • In this window
       
      • In a new window

      Table 3.

      Categorical Fixed Effects Model for Confounded and Unconfounded Studies Investigating the Effects of Intensity of Rehabilitation in Terms of ADL

      The summary effect sizes T̄^u for unconfounded and confounded studies were 0.19 (CI, ±0.17) and 0.40 (CI, ±0.19) SDU, respectively (Table 3⇑). No significant heterogeneity was found for confounded and unconfounded studies separately or between these two groups of studies. Controlling for dependency of estimated effect sizes as a result of including two multiple comparisons from the study of Smith et al²⁵ decreased the summary effect sizes of unconfounded studies to 0.12 (CI, ±0.18) and 0.19 SDU (CI, ±0.17), respectively.

      A marked difference was observed in overall effect size for ADL in studies with and without blind assessment recording, that is, 0.05 (CI, ±0.23) versus 0.38 (CI, ±0.23). Three of 5 studies in which assessment was not blinded were confounded by management of experimental and control groups at separate locations.

      Weighting each study by a factor t_i resulted for the unconfounded studies in a statistically significant summary effect size T̄^u for ADL; that is, T̄^u increased from 0.19 (CI, ±0.17) to 0.25 (CI, ±0.19) SDU (Table 3⇑). The unbiased summary effect size of confounded studies changed from 0.40 (CI, ±0.19) to 0.45 (CI, ±0.22) SDU. The homogeneity test statistic for ADL scores decreased slightly in both confounded and unconfounded studies by weighting the summary effect sizes by a factor t_i, but remained not statistically significant. In addition, heterogeneity was not statistically significant between these two groups of studies. Weighting the four blinded studies by factor t_i hardly changed the summary effect size (ie, 0.1 [CI, ±0.26]).

Functional and Neuromuscular Outcome Parameters

1. 1. Only one study investigating the effects of intensity of rehabilitation on neuromuscular outcome was confounded because patients were treated at separate locations³⁴ (Fig 2⇓ and Table 4⇓). The studies in which functional outcome was assessed were not confounded by organizational setting. In addition, all studies were blindly recorded. The unbiased summary effect size T̅^u’s for functional and neuromuscular scores were estimated at 0.10 (CI, ±0.21) and 0.37 (CI, ±0.24) SDU, respectively. No significant heterogeneity was found. To control for organizational setting and blinding procedures, a post hoc analysis for neuromuscular scores revealed comparable summary effect sizes at 0.35 (CI, ±0.30) and 0.36 (CI, ±0.31), respectively).

      

      View larger version:

      • In this page
       
      • In a new window

      • Download as PowerPoint Slide

      Figure 2.

      Ninety-five percent CIs of effect sizes for outcome of neuromuscular and functional parameters. Positive effect sizes indicate that intensive rehabilitation was more effective than control conditions; negative effect sizes indicate that the control regimen was more effective.

      

      View larger version:

      • In this page
       
      • In a new window

      • Download as PowerPoint Slide

      Figure 3.

      Appendix

      View this table:

      • In this window
       
      • In a new window

      Table 4.

      Categorical Fixed Effects Model for Studies Evaluating the Effects of Intensity of Rehabilitation in Terms of Neuromuscular and Functional Outcome

      Weighting each study by a time factor t_i resulted in a significant summary effect size T̄^u for variables defined on a functional level; that is, an increment from 0.10 (CI, ±0.25) to 0.36 (CI, ±0.22) SDU was found. On the neuromuscular level the summary effect size decreased from 0.37 (CI, ±0.23) to 0.32 (CI, ±0.25) SDU after weighting for t_i. When we weighted the blindly recorded studies only (n=3), a slightly higher summary effect size was found (0.36; CI, ±0.33).
      Previous SectionNext Section

Discussion

1. 1. In a number of critical reviews it has been suggested that an early start of intensive rehabilitation may be an important part of expert care in stroke patients.^{4 5 6 7 8 39}In the present research synthesis small⁴⁰ but statistically significant improvements in ADL as well as in neuromuscular and functional outcome variables were found as a result of higher intensities of rehabilitation. These results support the findings of Langhorne et al, who applied Fisher’s inverse χ² test and odds ratios. After probability values of the 9 studies included in the present study were combined with Fisher’s inverse χ² test,^21 41 significant improvements in ADL as well as neuromuscular and functional outcome parameters as a result of higher intensities of rehabilitation were found.

      In the present study Hedges’s g method was applied, because ADL, muscle strength,^32 34 35 synergism,^37 38 dexterity,^35 38 and walking velocity^36 37 are assessed on an ordinal or interval scale. This method allows for sensitivity analysis by weighting effect sizes for covariates such as amount of rehabilitation. The clinical relevance of obtained or reported summary effect sizes remains an important problem.

      The overall unbiased summary effect size of 0.28 SDU for ADL is smaller than the summary effect size of 0.57 SDU found by Ottenbacher and Jarnell.¹² Even the application of a random effects model in the present study resulted in almost comparable effect sizes (ie, 0.29; CI, ±0.13 SDU) (Table 5⇓).

      View this table:

      • In this window
       
      • In a new window

      Table 5.

      Categorical Random Effects Model for Confounded and Unconfounded Studies Investigating the Effects of Intensity of Rehabilitation in Terms of ADL

      As in the study of Langhorne et al,¹³ a difference was found in summary effect size between studies in which experimental and control groups were managed in the same setting (unconfounded studies) compared with the management of experimental and control groups in different settings (confounded studies). However, in the present study the effect size was smaller in the unconfounded studies (0.19; CI, ±0.17) than in the confounded studies (0.40; CI, ±0.19). The higher summary effect size for confounded studies compared with that for unconfounded studies may be the result of comparing the effects of stroke rehabilitation wards with those of general medical wards in 2 of 3 studies.^33 34Combining the findings of these 2 studies resulted in a summary effect size of 0.50 (CI, ±0.33) SDU. Smith et al⁴² and Kalra et al^43 44 demonstrated improved ADL and reduced hospital stay in stroke rehabilitation wards compared with patients admitted to general medical wards, while the amount of therapy was almost similar in both experimental conditions. Factors such as early onset of therapy,^{42 43 44 45}better education of staff members,⁴⁵ better organization of stroke care,^43 44 and family participation^45 46 may explain differences in better outcome of stroke rehabilitation wards.^6 7 8

      The low to moderate summary effect size found for the studies investigating the effects of intensity of rehabilitation may be caused by heterogeneity of the patient population, limited responsiveness of used assessment instruments, and insufficient modulation of intensity of therapy.⁴⁷ With respect to the heterogeneity of the patient population, it should be noted that the number of previous strokes, lesion size, and localization of stroke have an important impact on functional recovery.^48 49 50 Only one study³⁸ treated stroke as a single diagnostic category.

      Ottenbacher and Jarnell¹² were unable to find a statistically significant correlation coefficient (r=.11) between length or extent of therapy and effect size. In the present study, weighting the additional amount of rehabilitation in the experimental group compared with the control group in the individual studies increased the summary effect sizes for ADL in the unconfounded studies from 0.19 (±0.17) to 0.25 (±0.18) SDU. This finding provides further evidence for the presence of an intensity-effect relationship and suggests that this relationship is more likely to occur when the difference in intensity of rehabilitation between experimental conditions is sufficiently large.

      Remarkable in our research synthesis was the observation that in studies conducted in North America^{31 32 37 38} the amount of rehabilitation in both experimental and control groups was twice that of European studies.^{25 33 34 35 36} It should be noted, however, that the actual amount of rehabilitation has been adequately controlled in only one study.³⁷ Treatment days,³⁴ frequency of treatment,³⁴ or amount of treatment without correction for duration of admission²⁵ are only rough indicators for intensity of therapy (see Reference 29²⁹ for discussion).

      Another important result is that the summary effect size for outcome variables defined on the neuromuscular level is almost three times as high (ie, 0.37 SDU) as the summary effect size for functional outcome parameters (ie, 0.10 SDU). This finding may reflect the higher responsiveness of assessment instruments for neuromuscular functioning and supports the assumption that improvements on an impairment level are not unequivocally related to improvements in disability.^6 7 37However, weighting the additional amount of rehabilitation in the experimental group compared with the control group in individual studies resulted in a significant summary effect size for functional outcome variables, providing more evidence for the presence of an intensity-effect relationship. However, lower (weighted) summary effect sizes were obtained when the four blinded studies (with the highest methodological quality^{35 36 37 38} ) were analyzed separately (see also Reference 12¹² ). The latter finding suggests that observation bias may also explain part of the reported differences in summary effect sizes.

      In summary, the present research synthesis demonstrates small but statistically significant improvements in terms of ADL and functional outcome parameters. However, generalization of the results of the present research synthesis is difficult because of the low methodological quality of the included studies. Further research on the effects of intensity of physical and occupational therapy is necessary. These studies should experimentally control for (1) sufficient contrast in amount of therapy spent, (2) organizational setting of rehabilitation management, and (3) specification of patient characteristics such as type and localization of stroke, number of previous strokes, and initial ADL score. In addition, these studies should adhere to the methodological principles, especially blind recording, as described in the present research synthesis.
      Previous SectionNext Section

Acknowledgments

1. 1. This study is part of a research project supported by a grant from The Netherlands Heart Foundation (project reference No. 93.134). The authors thank Walter Cambach and Erwin van Wegen for their contribution to the data analysis.
      Previous SectionNext Section

Selected Abbreviations and Acronyms

1. 1. ADL	=	activities of daily living
      CI	=	confidence interval
      OT	=	occupational therapy
      PT	=	physical therapy
      SDU	=	standard deviation units
      Previous SectionNext Section

Footnotes

1. 1. • Reprint requests to Gert Kwakkel, MSc, Department of Physical Therapy, University Hospital Vrije Universiteit, PO Box 7057, 1007 MB Amsterdam, Netherlands.
      • Received March 24, 1997.
      • Revision received May 13, 1997.
      • Accepted May 13, 1997.
      • Copyright © 1997 by American Heart Association
      Previous Section

       

References

1. 1. 1. ↵ 
         Dobkin BH. Focused stroke rehabilitation programs do not improve outcome. Arch Neurol. 1989;41:701-703.
      2. ↵ 
         Reding MJ, McDowell FH. Focused stroke rehabilitation programs improve outcome. Arch Neurol. 1989;46:700-701.

          

         CrossRefMedline
      3. ↵ 
         Ernst E. A review of stroke rehabilitation and physiotherapy. Stroke.1990;21:1081-1085.

          

         Abstract/FREE Full Text
      4. ↵ 
         Wade DT, Langton-Hewer R. Rehabilitation after stroke. In: Toole JF, ed.Handbook of Clinical Neurology, Vol 11: Vascular Diseases, Part III. New York, NY: Elsevier Science Publishers BV; 1989:233-254.
      5. ↵ 
         Lind KA. Synthesis of studies on stroke rehabilitation. J Chron Dis.1982;35:133-149.

          

         CrossRefMedline
      6. ↵ 
         Wagenaar RC, Meyer OG. Effects of stroke rehabilitation, I: a critical review of the literature. J Rehabil Sci. 1991;4:61-73.
      7. ↵ 
         Wagenaar RC, Meyer OG. Effects of stroke rehabilitation, II: a critical review of the literature. J Rehabil Sci. 1991;4:97-109.
      8. ↵ 
         Gladman J, Barer D, Langhorne P. Specialist rehabilitation after stroke. Br Med J. 1996;312:1623-1624.

          

         FREE Full Text
      9. ↵ 
         Langhorne P, Williams BO, Gilchrist W, Howie K. Do stroke units save lives?Lancet. 1993;342:395-398.

          

         CrossRefMedline
      10. ↵ 
          Dennis M, Langhorne P. So stroke units save lives? Where do we go from here? Br Med J. 1994;309:1273-1277.

           

          Abstract/FREE Full Text
      11. ↵ 
          Langhorne P, Dennis MS, Williams BO. Stroke units: their role in acute stroke management. Vasc Med Rev. 1995;6:33-44.
      12. ↵ 
          Ottenbacher KJ, Jarnell S. The results of clinical trials in stroke rehabilitation research. Arch Neurol. 1993;50:37-44.

           

          CrossRefMedline
      13. ↵ 
          Langhorne P, Wagenaar RC, Partridge C. Physiotherapy after stroke: more is better? Physiotherapy Res Int. 1996;1:75-88.
      14. ↵ 
          Cook TD, Campbell DT. Quasi-experimentation. Boston, Mass: Houghton Mifflin Company; 1979.
      15. ↵ 
          Cook DJ. Methodologic guidelines for systematic reviews of randomized control trials in health care from the Potsdam consultation on meta-analysis. J Clin Epidemiol. 1995;48:167-171.

           

          CrossRefMedline
      16. ↵ 
          Chalmers TC, Smith H, Blackburn B, Silverman B, Schroeder B, Reitman D, Ambroz A. A method for assessing the quality of a randomized control trial.Control Clin Trials. 1981;2:31-49.

           

          CrossRefMedline
      17. ↵ 
          Cho MK, Bero LA. Instruments for assessing the quality of drug studies published in the medical literature. JAMA. 1994;272:101-104.

           

          CrossRefMedline
      18. ↵ 
          Detsky AS, Naylor D, Rourke KO, McGeer AJ, Abbe KAL. Incorporating variations in the quality of individual randomized trials into meta-analysis. J Clin Epidemiol. 1992;45:255-265.

           

          CrossRefMedline
      19. ↵ 
          Jenicek M. Meta-analysis in medicine. J Clin Epidemiol. 1989;42:35-44.

          CrossRefMedline
      20. ↵ 
          Schultz KF, Calmers I, Hayes RJ, Altman DG. Empirical evidence of bias.JAMA. 1995;273:408-412.

           

          CrossRefMedline
      21. ↵ 
          Hedges LV, Olkin I. Statistical Methods for Meta-analysis. Orlando, Fla: Academic Press, Inc; 1985.
      22. ↵ 
          Wolf FM. Meta-analysis: Quantative Methods for Research Synthesis.London, UK: Sage Publications; 1986.
      23. ↵ 
          Rosenthal R. Parametric measures of effect size. In: Cooper H, Hedges LV, eds. The Handbook of Research Synthesis. New York, NY: Russell Sage Foundation; 1994;16:231-244.
      24. ↵ 
          Shadish WR, Haddock CK. Combining estimates of effect size. In: Cooper H, Hedges LV, eds. The Handbook of Research Synthesis. New York, NY: Russell Sage Foundation; 1994;18:261-282.
      25. ↵ 
          Smith DS, Goldberg E, Ashburn A, Kinsella G, Sheikh K, Brennan PJ, Meade TW, Zutshi DW, Perry JD, Reeback JS. Remedial therapy after stroke: a randomized controlled trial. Br Med J. 1981;282:5-9.
      26. ↵ 
          Hedges LV. Fixed effects models. In: Cooper H, Hedges LV, eds. The Handbook of Research Synthesis. New York, NY: Russell Sage Foundation; 1994;19:285-300.
      27. ↵ 
          Raudenbush SW. Random effects model. In: Cooper H, Hedges LV, eds. The Handbook of Research Synthesis. New York, NY: Russell Sage Foundation; 1994;20:301-323.
      28. ↵ 
          Brocklehurst JC, Andrews K, Richards B, Laycock PJ. How much physical therapy for patients with stroke? Br Med J. 1978;1:1307-1310.
      29. ↵ 
          Wade DT, Skilbeck CE, Langton-Hewer R, Wood VA. Therapy after stroke: amounts, determinants and effects. Int Rehabil Med. 1984;6:105-110.

          Medline
      30. ↵ 
          Nugent JA, Schurr KA, Adams RD. A dose-response relationship between amount of weight-bearing exercise and walking outcome following cerebrovascular accident. Arch Phys Med Rehabil. 1994;75:399-402.

          CrossRefMedline
      31. ↵ 
          Keith A, Wilson DB, Gutierrez P. Acute and subacute rehabilitation for stroke: a comparison. Arch Phys Med Rehabil. 1995;76:495-500.

           

          CrossRefMedline
      32. ↵ 
          Stern PH, McDowell F, Miller JM, Robinson M. Effects of facilitation exercise techniques in stroke rehabilitation. Arch Phys Med Rehabil. 1970;51:526-531.

          Medline
      33. ↵ 
          Peacock PD, Riley CHP, Lampton, TD, Raffel SS, Walker JS. The Birmingham Stroke Epidemiology and Rehabilitation Study. In: Stewart GT, ed. Trends in Epidemiology. Springfield, Ill: Charles C Thomas, Publisher; 1972:231-345.
      34. ↵ 
          Sivenius J, Pyorala K, Heinonen OP, Salonen JT, Riekkinen P. The significance of intensity of rehabilitation of stroke: a controlled trial. Stroke. 1985;16:928-931.

           

          Abstract/FREE Full Text
      35. ↵ 
          Sunderland A, Tinson DJ, Fletcher D, Hewer RL, Wade DT. Enhanced physical therapy improves recovery of arm function after stroke: a randomised controlled trial. J Neurol Neurosurg Psychiatry.. 1992;55:530-535.

          Abstract/FREE Full Text
      36. ↵ 
          Wade DT, Collen FM, Robb GF, Warlow CP. Physiotherapy intervention late after stroke and mobility. Br Med J. 1992;304:609-613.
      37. ↵ 
          Richards CL, Malouin F, Wood-Dauphinee S, Williams JI, Bouchard JP, Brunet D. Task-specific physical therapy for optimization of gait recovery in acute stroke patients. Arch Phys Med Rehabil. 1993;74:612-620.

           

          CrossRefMedline
      38. ↵ 
          Werner RA, Kessler S. Effectiveness of an intensive outpatient rehabilitation program for postacute stroke patients. Am J Phys Med Rehabil. 1996;75:114-120.

           

          CrossRefMedline
      39. ↵ 
          Dombovy ML, Bach-Y-Rita P. Clinical observations on recovery from stroke. In: Waxman SG, ed. Functional Recovery in Neurological Disease. New York, NY: Raven Press; 1988;47:265-276.
      40. ↵ 
          Cohen J. Statistical Power Analysis for the Behavioural Sciences. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc; 1988.
      41. ↵ 
          Becker BJ. Combining significance levels. In: Cooper H, Hedges LV, eds. The Handbook of Research Synthesis. New York, NY: Russell Sage Foundation; 1994;15:215-230.
      42. ↵ 
          Smith ME, Garraway WM, Smith DL, Akthar AJ. Therapy impact on functional outcome in a controlled trial of stroke rehabilitation. Arch Phys Med Rehabil.1982;63:21-24.

           

          Medline
      43. ↵ 
          Kalra L, Dale P, Crome P. Improving stroke rehabilitation: a controlled study.Stroke. 1993;24:1462-1467.

           

          Abstract/FREE Full Text
      44. ↵ 
          Kalra L. The influence of stroke unit rehabilitation on the functional recovery from stroke. Stroke. 1994;25:821-825.

           

          Abstract/FREE Full Text
      45. ↵ 
          Hamrin E II. Early activation in stroke: does it make a difference? Scand J Rehabil Med. 1982;14:101-109.

           

          Medline
      46. ↵ 
          Glass TA, Matchar DB, Beleya M, Feussner JR. Impact of social support on outcome of first stroke. Stroke. 1993;24:76-80.

           

          Abstract/FREE Full Text
      47. ↵ 
          Matyas TA, Ottenbacher KJ. Confounds of insensitivity and blind luck: statistical conclusion validity in stroke rehabilitation clinical trials. Arch Phys Med Rehabil. 1993;74:559-565.

           

          CrossRefMedline
      48. ↵ 
          Chae J, Zorowitz R, Johnston MV. Functional outcome of hemorrhagic and nonhemorrhagic stroke patients after in-patient rehabilitation. Am J Phys Med Rehabil. 1996;75:177-182.

           

          CrossRefMedline
      49. ↵ 
          Task Force Recommendations. Symposium recommendations for methodology in stroke outcome research. Stroke. 1990;21(suppl II):II-68-II-73.
      50. ↵ 
          Kwakkel G, Wagenaar RC, Kollen BJ, Lankhorst GJ. Predicting disability in stroke: a critical review of the literature. Age Ageing. 1996;25:479-489.

          Abstract/FREE Full Text