PREP2: A biomarker-based algorithm for predicting upper limb function after stroke

Abstract

Objective: Recovery of motor function is important for regaining independence after stroke, but difficult to predict for individual patients. Our aim was to develop an efficient, accurate, and accessible algorithm for use in clinical settings. Clinical, neurophysiological, and neuroimaging biomarkers of corticospinal integrity obtained within days of stroke were combined to predict likely upper limb motor outcomes 3 months after stroke.

Methods: Data from 207 patients recruited within 3 days of stroke [103 females (50%), median age 72 (range 18-98) years] were included in a Classification and Regression Tree analysis to predict upper limb function 3 months poststroke.

Results: The analysis produced an algorithm that sequentially combined a measure of upper limb impairment; age; the presence or absence of upper limb motor evoked potentials elicited with transcranial magnetic stimulation; and stroke lesion load obtained from MRI or stroke severity assessed with the NIHSS score. The algorithm makes correct predictions for 75% of patients. A key biomarker obtained with transcranial magnetic stimulation is required for one third of patients. This biomarker combined with NIHSS score can be used in place of more costly magnetic resonance imaging, with no loss of prediction accuracy.

Interpretation: The new algorithm is more accurate, efficient, and accessible than its predecessors, which may support its use in clinical practice. While further work is needed to potentially incorporate sensory and cognitive factors, the algorithm can be used within days of stroke to provide accurate predictions of upper limb functional outcomes at 3 months after stroke. www.presto.auckland.ac.nz.

Figure 1

CART analysis for patients with a SAFE score ≥ 5 within 72 h poststroke. All of these patients are MEP+.

Figure 2

CART analyses of patients with a SAFE score < 5 at 72 h poststroke. (A) Both TMS and MRI biomarkers available. The analysis selects sensorimotor tract (SMT) lesion load to differentiate between MEP− patients who will have a Limited versus Poor upper limb outcome. (B) TMS but no MRI biomarkers available. The analysis selects NIHSS score to differentiate between MEP− patients who will have a Limited versus Poor upper limb outcome.

Figure 3

The PREP2 algorithm predicts upper limb functional outcome at 3 months poststroke. The four possible upper limb outcomes are color‐coded. The colored dots depict the proportion of patients expected to achieve each color‐coded outcome, depending on their pathway through the algorithm, based on the results of the CART analysis. Patients who achieve a SAFE score of five or more within 72 h of stroke symptom onset, and are less than 80 years old, are most likely to have an Excellent upper limb outcome. Patients who achieve a SAFE score of five or more within 72 h of stroke symptom onset and are 80 years old or more, are most likely to have an Excellent upper limb outcome provided their SAFE score is at least 8; otherwise they are likely to have a Good upper limb outcome. Patients whose SAFE score is less than 5 at 72 h after stroke symptom onset need TMS to determine MEP status in the paretic upper limb, a key biomarker of corticospinal tract integrity. If a MEP can be elicited (MEP+) approximately 5 days poststroke then the patient is likely to have at least a Good upper limb outcome. If a MEP cannot be elicited, the NIHSS score obtained 3 days poststroke can be used to predict either a Limited outcome if the score is less than 7, or a Poor outcome if the score is 7 or more.