A total of 91 patients were eligible for Study I (see Figure 5 flow diagram for details). Characteristics at baseline and baseline assessments for the patients included in Study I are displayed in Table 4. The 12 patients not included were not statistically significant different for any of the demographic characteristics or stroke details. However, they were statistically significant different for FMA, ARAT and SAFE score at baseline. This is expected and reflects that the majority of pa-tients not included in Study I were those with more severe UL impairments and a SAFE score below 5, excluded due to not having TMS performed, which prevented the obtainment of a baseline prediction (see Figure 5 flow diagram for details).
Table 4. Demographic Characteristics and Baseline Assessments for Study I Included
Patients (n= 91) Excluded
Patients (n=12) P-value
Age, years, mean (SD) 64.0 (10.6) 66.3 (7.8) 0.46
Sex, female/ male, n% 39 (43%) / 52 (57%) 5 (42%)/ 7 (58%) 0.94
Days since stroke, mean (SD) 13.4 (1.6) 13.8 (1.6) 0.35
Stroke type, ischemic/ haemorrhagic, n (%) 73 (80%) / 18 (20%) 9 (75%) / 3 (25%) 0.67 Side of paresis, left/ right n (%) 53 (58%) / 38 (42%) 6 (50%)/ 6 (50%) 0.59
Dominant UL affected, n (%) 42 (46%) 6 (50%) 0.80
Stroke confirmed on imaging, n (%) 90 (99%) 11 (92%) 0.09
Stroke location
FIM cognitive score, median (IQR) 24 (19- 30) 24 (20- 28) 0.58
FMA4, median (IQR,) 42 (16- 53) 11 (4- 29) 0.017*
Bilateral stimulation affected, n (%) 27 (31%) 3 (25%) 0.69
Two point stimulation affected, n (%) 49 (54%) 5 (42%) 0.43
UL pain present, n (%) 28 (31%) 3 (25%) 0.68
UL pain intensity, median (IQR) 0 (0- 4) 0 (0- 2) 0.63
Neglect present, n (%) 21 (24%) 3 (27%) 0.79
FAC8, median (IQR) 1 (0- 4) 0 (0- 3) 0.38
*The included and excluded patients were significantly different. SD: Standard deviation. IQR:
Inter quartile range. 1Stroke thrombolysis/ thrombectomy rates were calculated for patients with ischaemic stroke only. 2NIHSS: National Institute of Health Stroke Scale Score. The NIHSS score for 12 patients who were MEP- was a median of 13 (IQR 7-15, min 9, max 21). 3FIM: Functional Inde-pendence Measure. 4FMA: Fugl-Meyer Motor Assessment Upper Extremity score. 5ARAT: Action Research Arm Test. 6SAFE: Shoulder Abduction Finger Extension. 7MEP: Motor-evoked potentials.
MEP was assessed in 40 patients with a SAFE score below 5 and assumed present in all patients with a SAFE score ≥5. MEP was assessed a mean of 13.4 days after stroke (SD 1.7, min 11, max 18). 8FAC: Functional Ambulation Classification. Source: Modified from Study I89.
Prediction of future UL function
Information on SAFE score, MEP status, age and NIHSS score was used to predict future UL function.89
At baseline the mean SAFE score was 5 (SD 2.8, min 0, max 9) and was obtained 13.4 days (SD 1.6, min 10, max 18) after stroke. Twelve patients were MEP- and 26 were MEP+. For 38 of 91 patients (42%) the SAFE score was < 5 and MEP status was established.89 Fifty patients (55%) had the prediction Excellent UL function, 29 (32%) had the prediction Good UL function, three patients (3%) had the prediction Limited UL function, and nine patients (10%) had the prediction Poor UL function (Table 5).89
Follow-up assessments on ARAT were obtained a mean of 91 days after stroke (SD 3.8, min 84, max 99). At follow-up the median ARAT score was 50 (IQR 33-55, min 0, max 57). The median within group improvement in ARAT scores was 17 (IQR 3 - 27, P < 0.001).89
At follow-up patients were grouped according to the actually achieved ARAT score. Forty-four patients were included in the category Excellent UL function, 22 in Good, 13 in Limited, and 12 patients in Poor UL function (Table 5). The ARAT score was a median of 55 (IQR 54-56, min 51, max 57) in the category Excellent, a median of 42 (IQR 39-49, min 24, max 50) in the category Good, a median of 31 (IQR 24-32, min 17, max 33) in the category Limited, and a median 0 (IQR 0-2, min 0, max 6) in the category Poor UL function. 89
UL prediction accuracy
The overall CCR was 60% (95% CI 51-71).89 Twenty-six of 91 patients (29%) did not achieve as high UL function as predicted and hence the prediction was too opti-mistic. On the contrary, for 10 of 91 patients (11 %) the prediction was too pes-simistic and actual UL function at 3 months exceeded the predicted. Twenty-eight
of the 36 of patients (78%) for whom the prediction was inaccurate achieved an actual outcome category adjacent to the predicted category, e.g. they were pre-dicted Limited, but ended up in the outcome category Good.
Table 5. Predicted and Actual ARAT Categories and Agreement Between Them
Predicted ARAT category at baseline
Actual ARAT outcome category at 3 months
Excellent Good Limited Poor Total, n (%)
Excellent 37 10 3 0 50 (55%)
Good 7 10 8 4 29 (32%)
Limited 0 1 1 1 3 (3%)
Poor 0 1 1 7 9 (10%)
Total 44 (48%) 22 (24%) 13 (14%) 12 (13%) 91 (100%) Green: Patients for whom the outcome category was equivalent to the predicted category (n=55)
Yellow: Patients for whom the outcome category was adjacent to the predicted category (n=28)
Orange: Patients for whom the outcome category was two categories away from the pre-dicted category (n=8)
Red: Patients for whom the outcome category was three categories away from the pre-dicted category (n=0)
Source: Replicated from Study I89
Within the four categories, CCR was a maximum of 78% (95% CI 43-95%) for pa-tients predicted Poor UL function followed by 74% (95% CI 60-84%) for those pre-dicted Excellent UL function. CCR was 35% (95% CI 20-53%) for patients prepre-dicted Good and 33% (95% CI 4-85%) for those predicted Limited UL function (Table 6).
For the 53 patients with a SAFE ≥ 5, CCR was 74% (95% CI 62-86%). For the 38
patients with a SAFE < 5, CCR was 42% (95% CI 26-58%)(Table 6).89 This low CCR was primarily caused by 26 patients who were predicted to have Good UL function based on MEP+ (Table 5, previous page). Many of these patients ended in another actual outcome category at 3 months (Table 5). In contrast to this, patients who were MEP- and predicted a Poor UL function generally also ended up in the pre-dicted outcome category (Table 5).89
Table 6. Accuracy of the Prediction Algorithm for UL Function 89 CCR (95% CI) Sensitivity
(95% CI) Specificity
(95% CI) CCR for SAFE
≥ 5 (95% CI) CCR for SAFE
< 5 (95% CI) Overall (n= 91) 60% (50; 71)
Excellent (n= 44) 74% (60; 84) 84% (70; 93) 72% (57; 84) 74% (62; 86) Good (n= 22) 35% (20; 53) 46% (24; 68) 73% (60; 83)
Limited (n= 13) 33% (4; 85) 8% (0; 36) 97% (91; 100) 42% (26; 58) Poor (n= 12) 78% (43; 95) 58% (28; 85) 98% (91: 100)
CCR: Correct Classification Rate; n= number of patients in outcome category Source: Replicated from Study I89
CART analysis
A CART analysis was conducted to examine if modifications of the included com-ponents could improve prediction accuracy when the PREP2 algorithm was ob-tained two weeks after stroke. The decision tree developed by CART analysis had an overall CCR of 66% (95% CI 56-76) (Figure 6).89
A total of 89 of 91 patients were included in the CART, as the NIHSS score was not available for two patients. Based on CART, the SAFE score was found to be the most important predictor. For patients with a SAFE ≥ 5 a prediction of either Good or Excellent UL function was made, based on age. For patients with a SAFE score
< 5 who were MEP+, the NIHSS score was needed to differentiate between Good and Limited UL function. Patients with a SAFE < 5 who were MEP- were predicted a future Poor UL function.89
CCR for each of the four categories was 67% (95% CI 37-87%) for Poor, 60% (95%
CI 20-90%) for Limited, 45% (95% CI 28-63%) for Good, and 80% (95% CI 66-89%) for the category Excellent. For patients with a SAFE ≥ 5, CCR was 78% (95% CI 67-89%); and for patients with a SAFE < 5, CCR was 50% (95% CI 34-66%).
Figure 6. CART Model for Prediction of UL Function
Available for the CART analysis were the PREP2 components: SAFE score, MEP status, age, and the NIHSS score.
Source: Replicated from Study I89
Sex, female/ male, n (%) 35 (40) / 52 (60) 9 (56)/ 7 (44) 0.23
Days since stroke, mean (SD) 13.3 (1.6) 13.9 (1.5) 0.16
Stroke type, ischemic/ hemorrhagic, n (%) 70 (80) / 17 (20) 12 (75)/ 4 (25) 0.62 Side of paresis, left/right, n (%) 47 (54) / 40 (46) 12 (75)/ 4 (25) 0.12
Dominant UL affected, n (%) 43 (49) 5 (31) 0.18
Stroke confirmed on imagining, n (%) 86 (99) 15 (94) 0.17
Stroke location
Cortical, n (%) 41 (47) 5 (31) 0.24
Subcortical, n (%) 41 (47) 10 (63) 0.26
Brainstem, n (%) 5 (6) 0 (0) 0.33
Thrombolisis1, n (%) 29 (41) 5 (42) 0.99
Thrombectomy1, n (%) 17 (24) 1 (8) 0.22
Premorbid able to walk (+/-walking aid), n (%) 86 (100) 15 (94) 0.02*
Premorbid living in own home, n (%) 87 (100) 16 (100)
First stroke, n (%) 79 (91) 14 (88) 0.68
Co-morbidity present, n (%) 61 (70) 14 (88) 0.15
Hypertension, n (%) 40 (46) 11 (69) 0.09
Coronary artery disease, n (%) 16 (18) 5 (31) 0.24
Diabetes, n (%) 7 (8) 3 (19) 0.18
Other neurological disease, n (%) 3 (3) 1 (6) 0.59
Current smoker, n (%) 26 (35) 5 (33) 0.89
BMI, median (IQR), 26 (23- 29) 29 (28- 33) 0.005*
FIM2, median (IQR) 72 (49- 85) 75 (57- 96) 0.66
FIM motor score, median (IQR) 48 (30- 57) 50 (43- 70) 0.49
FIM cognitive score, median (IQR) 24 (19- 29) 24 (15- 30) 0.68
Assessments at baseline
FMA3, median (IQR) 40 (14- 53) 25 (11- 49) 0.64
ARAT4, median (IQR) 17 (3- 39) 13 (4- 39) 0.86
SAFE5, median (IQR) 5 (2- 8) 5 (2- 8) 0.81
MEP6 not present, n (%) 9 (11) 3 (19) 0.28
Shoulder subluxation present, n (%) 18 (21) 1 (7) 0.20
Light touch affected, n (%) 41 (47) 7 (44) 0.80
Proprioception affected, n (%) 24 (28) 3 (20) 0.52
Bilateral stimulation affected, n(%) 26 (31) 4 (27) 0.76
Two point stimulation affected, n (%) 43 (51) 8 (50) 0.83
UL pain present, n (%) 26 (30) 5 (31) 0.91
UL pain intensity, median (IQR) 0 (0- 4) 0.0 (0- 3) 0.97
Neglect present, n (%) 23 (27) 1 (7) 0.09
FAC7, median (IQR) 1 (0- 4) 1.5 (0- 4) 0.57
*The included and excluded patients were significantly different. SD: Standard deviation. IQR:
Inter quartile range. 1Stroke thrombolysis/ thrombectomy rates were calculated for patients with ischaemic stroke only. 2FIM: Functional Independence Measure. 3FMA: Fugl-Meyer Motor Assess-ment Upper Extremity score. 4ARAT: Action Research Arm Test. 5SAFE: Shoulder Abduction Finger Extension. 6MEP: Motor-evoked potentials. MEP was assessed in 40 patients with a SAFE score