Centralized Open-Access Research (COAR): A Database for Stroke Rehabilitation. Keith Lohse,
1 PhD ,
Lara Boyd, PT
2 PhD ,
and Catherine Lang, PT
3 PhD
1Motor
Learning Lab, School of Kinesiology, Auburn University 2Brain and Behaviour Lab, Department of Physical Therapy, University of British Columbia 3Neurorehabilitation Research Lab, Washington University School of Medicine in St. Louis
0. General description of the project and database.
1. Therapy is effective, but there is significant Experimental Groups heterogeneity in outcomes.
2. Group, latency, and time in therapy explain variation in terminal therapy outcomes.
Control Groups Control Groups n = 51 (6 na) Random Effects Model: d = 0.46, 95% CI [0.35, 0.56] τ2 = 0.091 Test of Heterogeneity: Q(50) = 172.49, p < 0.001 I2 = 68.92% (variance between studies/total variance) H2 = 3.22 (total variance/sample variance)
Prospero Registration No. CRD42014009010 There is a problem in the information architecture of physical medicine and rehabilitation for stroke. Randomized controlled trials (RCTs) are published across a spectrum of journals, by different authors, with varied publication formats and standards. This makes data integration very difficult. We are creating an open-access, living database with demographic, methodological, and outcome data from RCTs of physical and occupational therapies for adults with stroke.
Weighted meta-regression using maximum likelihood estimatation5 revealed:
Experimental Groups n = 61 (8 na) Random Effects Model: d = 0.78, 95% CI [0.66, 0.90] τ2 = 0.15 Test of Heterogeneity: Q(60) = 228.79, p < 0.001 I2 = 74.78% (variance between studies/total variance) H2 = 3.97 (total variance/sample variance)
2,892 titles were, and 398 manuscripts are currently being, screened for inclusion. Here we present “pilot” data extraction of 59 randomized and controlled studies, constituting 126 independent groups of participants. Data are based on summary statistics from n = 3,746 adults with stroke. Having a centralized repository for the summary statistics from RCTs allows researchers and clinicians to a) quickly and easily visualize data, b) generate hypotheses, c) facilitate data sharing, d) provide an efficient/searchable index of related studies, e) lead to more accurate/comprehensive power analyses for clinical trial design.
Cohen’s d
4. Inclusion/exclusion criteria lack standardization.
? ? ?
Criterion A
0.619
[0.542, 0.696]***
Group.c
0.180
[0.025, 0.334]*
Time50.c
0.005
[0.002, 0.008]**
lnDPS.c
-0.068
[-0.107, -0.030]***
Group X Time50
-0.003
[-0.001, 0.003]
Group X lnDPS
0.059
[-0.018, 0.136]
Time50 X lnDPS
-0.002
[-0.003, -0.001]*
Group X Time50 X lnDPS
0.002
[-0.001, 0.005]
Test for Residual Heterogeneity: QE(86) = 215.16, p < .0001
(Effect-size variance1 was estimated using r = 0.5, [based on available data]. Analyses used the “metafor” package in R4,5).
3. A lack of common-data elements (CDEs) is a barrier to analysis.
Intercept
tau = 0.2339 (square root of estimated tau^2 value) I2 = 53.21% (residual heterogeneity / unaccounted variability) H2 = 2.14 (unaccounted variability / sampling variability) r2 = 48.15% (amount of heterogeneity accounted for)
There is significantly greater variation between effects than expected by chance alone. Sources of variability: 1. Population (age, DPS) 2. Intervention (type, dose) 3. Outcome (participation, activity, function)
Cohen’s d
Estimate 95% CI
*p < 0.05, **p < 0.01, ***p < 0.001
There is a statistically significant improvement from the beginning to the end of an intervention for control and experimental groups.
We present initial results showing the utility of the database, identify current barriers to constructing a flexible/powerful information architecture, and potential solutions for overcoming these barriers.
Coefficient
Control Group
n = 100
Experimental Group
n = 30
Test of Moderators (all coefficients): QM(7) = 38.83, p < .0001
6. Text mining may provide insights into “conventional therapy” control conditions in RCTs. *stop words, “control-” and “group-” excluded.
*stop words, “control-”, “patient-”, “group-”, “therap-”, “treatment” excluded.
Criterion B Criterion C Criterion D
5. There is a lack of detail in descriptions of control therapies. a) There is considerable variability in demographic and outcome DEs across studies, complicating analysis. (How can we deal with multiplicity?) b) Dose of therapy is currently underreported as a methodological DE. (Time scheduled for therapy is a rough estimate at best.) c) Correlations between time-points are key DE that are currently completely missing. (Required for variance calculations1.)
a) Words regarding ideological approaches (e.g., Bobath, NDT) or therapeutic practices (e.g., functional training, walking, activities of daily living) were marginally common. b) Words describing the professional delineations of therapy teams are more common (e.g., occupational therapy, physical/physiotherapy). (All text mining analyses were conducted using the “tm” and “wordcloud” packages in R2,3,4.) Using the text from intervention descriptions only in the Method sections of RCTs, there is a lack of detail in reporting control therapies. Control therapies are likely clinically sound and theoretically motivated, but without these details, practices and theoretical underpinnings of control therapies are unknowable.
References 1. Borenstein, M., Hedges, L.V., Higgins, J.P.T., & Rothstein, H.R. (2009). Introduction to Meta-Analysis. West Sussex, UK: John Wiley & Sons, Ltd. 2. Fellows, I. (2014). wordcloud: Word Clouds. R package version 2.5. http://CRAN.R-project.org/package=wordcloud 3. Meyer, D., Hornik, K., & Feinerer, I. (2008). Text Mining Infrastructure in R. Journal of Statistical Software, 25(5): 1-54. URL: http://www.jstatsoft.org/v25/i05/. 4. R Core Team (2014). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/. 5. Viechtbauer, W. (2010). Conducting meta-analyses in R with the metafor package. Journal of Statistical Software, 36(3), 1-48. URL http://www.jstatsoft.org/v36/i03/.
