Related Experiment Video
Updated: Jul 1, 2025
A Method for Quantifying Upper Limb Performance in Daily Life Using Accelerometers
Published on: April 21, 2017
Assessing Clinically Important Differences During Rehabilitation for Stroke: A Pilot Study Evaluating Anchor and
Erin Y Harmon1, Jerome Niyirora2, Amy E Teale1
1James A. Eddy Memorial Foundation Research Institute, Sunnyview Rehabilitation Hospital, Schenectady, NY.
Objective:
To determine clinically important differences (CIDs) on Section GG physical functioning scores on the Centers for Medicare and Medicaid Services (CMS) Inpatient Rehabilitation Facility Assessment Instrument (IRF-PAI) for patients with stroke, using anchor and distribution-based approaches.
Design:
Pilot prospective observational cohort study.
Setting:
Inpatient rehabilitation facility.
Participants:
Patients with stroke (N=208).
Interventions:
Physicians assessed improvements during rehabilitation using the modified Rankin scale (mRS). Improvements (≥1 point) on the mRS were used as the anchor for establishing CIDs.
Main Outcome Measures:
Classically summed and Rasch transformed Section GG change scores associated with clinically important improvements on the mRS.
Results:
A total of 166 patients (79.8%) improved ≥1 point on the mRS. Change scores of 27, 9, and 16 on Section GG total physical functioning (self-care + mobility), self-care, and mobility/walk scales, respectively, had high sensitivity (0.82-0.85) but low specificity (0.52-0.69) in identifying patients improving on the mRS. Positive predictive values ranged from 0.87 to 0.91, and negative predictive values ranged from 0.42 to 0.52. Total physical functioning and selfcare anchor-derived change scores were similar to the reliable change index (RCI [2.77 × SEM]), calculated as 28 and 10 points, respectively, whereas anchor-derived mobility/walk scale change scores were equivalent to 1.96 × SEM. Exploratory Rasch modeling identified 3 Section GG subscales (R-Self-Care, R-Mobility, and R-Walking). Improvements on the R-Walking subscale were most correlated with mRS improvements (ρ=-0.47); however, accuracy of CID estimates was not improved.
Conclusions:
Cut-off scores obtained using the mRS anchor aligned with more robust estimates of change, as estimated by distribution-based measures. While patients achieving anchor-derived cut-offs have a high probability of mRS improvement, change scores may fail to detect clinically meaningful improvements at these same thresholds. Alternative criteria for determining MCID/CIDs, should be explored. Rasch models require further validation.
