The overall CCR of 60% found in Study I was statistically significantly lower than the overall accuracy of 75% found by Stinear et al23 (15% difference, 95% CI 3-27%, P< 0.001, Chi2 test). Three possible explanations for the lower CCR in Study I should be mentioned. First, differences in study populations may contribute to the
lower CCR.105 In the original study, patients were included within days of stroke.
Thus, a part of these patients would have been either too severely or too mildly affected to be referred to in-patient rehabilitation 2 weeks after stroke. Conse-quently, patients in the present Study I may represent a more selected sample with respect to both UL impairment as well as other factors, e.g. motivation and co-morbidity, which may affect UL prognosis. Thus, the included patients still experienced UL impairment two weeks after stroke, but on the other hand, were expected to benefit from rehabilitation.
Second, that the CCR of 75% in the original study was computed from the training data used for developing it. Despite Stinear et al using pruning and cross valida-tion to prevent data from being over-fitted, it might still be difficult to achieve an equally high accuracy in a subsequent data set.105,106 A third possible explanation for the overall lower accuracy, and particularly for the large number of patients for whom the algorithm was too optimistic, is the decreased room for improve-ment two weeks after stroke. Most spontaneous biological recovery occurs early after stroke, and while the room for increases in UL function scores is high during the first days after stroke, it declines during the course of recovery.107,108 When the prediction is obtained at a later point in time, patients will be closer to their maxi-mally achievable UL function.
Results from the EPOS cohort study corroborate that the chance of recovery of UL function declines the longer the severe impairment lasts.12 In the EPOS study 188 patients with stroke were included. The return of some dexterity on ARAT (i.e.,
≥10 points) at 6 months was predicted by shoulder abduction and finger exten-sion measured within 72 hours after stroke.12 Retesting the model on days 5 and 9 showed that the probability of regaining dexterity remained 98% for those pa-tients who were able to extend their fingers and abduct their shoulders, whereas the probability declined from 25% to 14% for those who did not satisfy either of these criteria.12
In the present population, MEP status was established using a belly-to-tendon placement of electrodes. This electrode placement will generally capture MEPs from not only one, but several muscles from the hand and forearm79,109 and en-sures, that not only activity limited to specific muscles, but any activity in the corticospinal tract is more likely to be captured. Other studies have used different TMS procedures110-112 and e.g. used the resting motor threshold and recruitment curves steepness over the primary motor hand area (slope ratio between the ipsilesional and contralesional hemisphere) as a measure of corticomotor excit-ability.110,111 This alternative method may provide a more detailed measure of corticomotor excitability. Still, in the present study, the TMS protocol by Stienar et al.77 was followed as to ease comparison of results.89 Additionally, other studies suggest that MEP amplitudes and latencies to single TMS pulses have adequate reliability for both healthy volunteers113 and certain patient populations.114 The Stroke Recovery and Rehabilitation Roundtable recommends MEP as a bio-marker for use in clinical trials for stratification purposes.2 Still, its clinical ap-plication can be debated. In a recent review, Kim and Winstein suggest that TMS should be used in combination with clinical tests to predict UL function.37 In line with this a review by Hayward et al.36 concluded that the presence of MEPs as indexed by TMS was the only biomarker associated with better motor outcome, though with large inter-individual variability.36 The value of MEP status as a sup-portive tool for prognosis has also been suggested by other research, though not always unambiguous.22,23,35-41,115
The unsatisfactory overall accuracy of PREP2 as applied in the present Study I can be mainly attributed to the poor accuracy for patients predicted Good based on MEP+. Studies have shown that the presence of MEP+ day 3-7 after stroke indicates a good prognosis and improves the prediction accuracy for UL func-tion.2,23,29,38,39,43 In contrast, the presence of MEP+ two weeks after stroke as in Study I is not as promising. This discrepancy in findings may be attributed to the above-mentioned spontaneous biological recovery. On the other hand, MEP-
in-dicates a poor prognosis for UL function both when obtained at two weeks after stroke as in the present study, or when obtained on day 3-7. Still, even patients with MEP- may improve. When MEP- is obtained within the first days of stroke, 15-20% of the patients recover at least some UL function.32,39,117 In Study I as well, three of the 12 MEP- patients (25%) ended up in the outcome category Limited instead of the predicted category Poor. These findings indicate that in some of the patients with very poor UL function early after stroke, at least some regeneration of the corticospinal pathways or some compensation for the loss of functionality may occur. However, the category Limited still indicates a very compromised UL function.
The use of TMS may be an obstacle for the clinical use of the PREP2 algorithm.
Patients have to be screened for contraindications for TMS, and a substantial part may not be available. In the present Study I, 8 patients could not have the predic-tion obtained either due to contraindicapredic-tions or due to technical issues. Further-more, TMS can be relatively expensive and it requires trained staff. On the other hand, TMS is relatively easy and fast to perform. In Study I, MEP status could be established within ½ hour, including the setup of equipment.