Cross-Validation of Machine Learning Models for the Functional Outcome Prediction after Post-Stroke Robot-Assisted Rehabilitation.

The state of the art is still lacking an extensive analysis of which clinical characteristics are leading to better outcomes after robot-assisted rehabilitation on post-stroke patients. Prognostic machine learning-based models could promote the identification of predictive factors and be exploited as Clinical Decision Support Systems (CDSS). For this reason, the aim of this work was to set the first steps toward the development of a CDSS, by the development of machine learning models for the functional outcome prediction of post-stroke patients after upper-limb robotic rehabilitation. Four different regression algorithms were trained and cross-validated using a nested 5×10-fold cross-validation. The performances of each model on the test set were provided through the Median Average Error (MAE) and interquartile range. Additionally, interpretability analyses were performed, to evaluate the contribution of the features to the prediction. The results on the two best performing models showed a MAE of 13.6 [13.4] and 13.3 [14.8] on the Modified Barthel Index score (MBI). The interpretability analyses highlighted the Fugl-Meyer Assessment, MBI, and age as the most relevant features for the prediction of the outcome. This work showed promising results in terms of outcome prognosis after robot-assisted treatment. Further research should be planned for the development, validation and translation into clinical practice of CDSS in rehabilitation. Clinical relevance- This work establishes the premises for the development of data-driven tools able to support the clinical decision for the selection and optimisation of the robotic rehabilitation treatment.

Gender differences in post-stroke functional outcome at discharge from an intensive rehabilitation hospital.

BACKGROUND: Gender differences in stroke functional recovery after rehabilitation are poorly investigated. Our aim was to compare functional outcomes at discharge from an intensive rehabilitation hospital after stroke in men and women, and to analyze their prognostic factors. METHODS: Retrospective observational study of consecutive stroke patients discharged from an intensive neurological rehabilitation hospital, from January 2018 to June 2019. Modified Rankin Scale (mRS) at discharge was the main outcome measure. RESULTS: Among the 208 included patients (mean age 73.4 ± 13.6 years), 105 (50.5%) were women. Women were significantly older (75.3 ± 13.8 vs. 71.4 ± 13.2 years, respectively, p = 0.041), and less often had a history of smoking habit (27% vs. 50%, p < 0.001). No gender differences emerged for vascular risk factors and comorbidities, pre-stroke functional status, length of hospital stay, stroke type, and number of clinical deficits. At admission to the rehabilitation hospital, mRS score distributions were not different (p = 0.795). At discharge, mRS score distributions and destinations did not differ between men and women (p = 0.391, p = 0.785, respectively). A significant interaction between gender and the change in mRS score from admission to discharge was found (F = 6.6, p = 0.011) taking into account age, stroke type, and number of initial clinical deficits. Dividing the cohort according to age, elderly women showed a better functional recovery compared to men. CONCLUSIONS: At admission to an intensive rehabilitation hospital, men and women presented a similar functional and clinical status and a substantial overlap of functional recovery after stroke. At higher ages, the potential for recovery appeared better in women compared to men.

Comparison between Ischemic and Hemorrhagic Strokes in Functional Outcome at Discharge from an Intensive Rehabilitation Hospital.

Comparison studies on recovery outcomes in ischemic (IS) and hemorrhagic strokes (HS) have yielded mixed results. In this retrospective observational study of consecutive IS and HS patients, we aimed at evaluating functional outcomes at discharge from an intensive rehabilitation hospital, comparing IS vs. HS, analyzing possible predictors. Modified Rankin Scale (mRS) at discharge was the main outcome. Out of the 229 patients included (mean age 72.9 ± 13.9 years, 48% males), 81 had HS (35%). Compared with IS (n = 148), HS patients were significantly younger (75 ± 12.5 vs. 68.8 ± 15.4 years, p = 0.002), required longer hospitalizations both in acute (23.9 ± 36.7 vs. 35.2 ± 29.9 days, p = 0.019) and rehabilitation hospitals (41.5 ± 31.8 vs. 77.2 ± 51.6 days, p = 0.001), and had more severe initial clinical deficit (mean number of neurological impairments: 2.0 ± 1.1 vs. 2.6 ± 1.4, p = 0.001) and mRS scores at admission (p = 0.046). At discharge, functional status change, expressed as mRS, was not significantly different between IS and HS (F = 0.01, p = 0.902), nor was the discharge destination (p = 0.428). Age and clinical severity were predictors of functional outcome in both stroke types. On admission in an intensive rehabilitation hospital, HS patients presented a worse functional and clinical status compared to IS. Despite this initial gap, the two stroke types showed an overlapped trajectory of functional recovery, with age and initial stroke severity as the main prognostic factors.

Reliability, validity and discriminant ability of a robotic device for finger training in patients with subacute stroke.

BACKGROUND: The majority of stroke survivors experiences significant hand impairments, as weakness and spasticity, with a severe impact on the activity of daily living. To objectively evaluate hand deficits, quantitative measures are needed. The aim of this study is to assess the reliability, the validity and the discriminant ability of the instrumental measures provided by a robotic device for hand rehabilitation, in a sample of patients with subacute stroke. MATERIAL AND METHODS: In this study, 120 patients with stroke and 40 controls were enrolled. Clinical evaluation included finger flexion and extension strength (using the Medical Research Council, MRC), finger spasticity (using the Modified Ashworth Scale, MAS) and motor control and dexterity during ADL performance (by means of the Frenchay Arm Test, FAT). Robotic evaluations included finger flexion and extension strength, muscle tone at rest, and instrumented MAS and Modified Tardieu Scale. Subjects were evaluated twice, one day apart, to assess the test-retest reliability of the robotic measures, using the Intraclass Correlation Coefficient (ICC). To estimate the response stability, the standard errors of measurement and the minimum detectable change (MDC) were also calculated. Validity was assessed by analyzing the correlations between the robotic metrics and the clinical scales, using the Spearman's Correlation Coefficient (r). Finally, we investigated the ability of the robotic measures to distinguish between patients with stroke and healthy subjects, by means of Mann-Whitney U tests. RESULTS: All the investigated measures were able to discriminate patients with stroke from healthy subjects (p < 0.001). Test-retest reliability was found to be excellent for finger strength (in both flexion and extension) and muscle tone, with ICCs higher than 0.9. MDCs were equal to 10.6 N for finger flexion, 3.4 N for finger extension, and 14.3 N for muscle tone. Conversely, test-retest reliability of the spasticity measures was poor. Finally, finger strength (in both flexion and extension) was correlated with the clinical scales (r of about 0.7 with MRC, and about 0.5 with FAT). DISCUSSION: Finger strength (in both flexion and extension) and muscle tone, as provided by a robotic device for hand rehabilitation, are reliable and sensitive measures. Moreover, finger strength is strongly correlated with clinical scales. Changes higher than the obtained MDC in these robotic measures could be considered as clinically relevant and used to assess the effect of a rehabilitation treatment in patients with subacute stroke.

Upper Limb Robotic Rehabilitation After Stroke: A Multicenter, Randomized Clinical Trial.

BACKGROUND AND PURPOSE: After stroke, only 12% of survivors obtain complete upper limb (UL) functional recovery, while in 30% to 60% UL deficits persist. Despite the complexity of the UL, prior robot-mediated therapy research has used only one robot in comparisons to conventional therapy. We evaluated the efficacy of robotic UL treatment using a set of 4 devices, compared with conventional therapy. METHODS: In a multicenter, randomized controlled trial, 247 subjects with subacute stroke were assigned either to robotic (using a set of 4 devices) or to conventional treatment, each consisting of 30 sessions. Subjects were evaluated before and after treatment, with follow-up assessment after 3 months. The primary outcome measure was change from baseline in the Fugl-Meyer Assessment (FMA) score. Secondary outcome measures were selected to assess motor function, activities, and participation. RESULTS: One hundred ninety subjects completed the posttreatment assessment, with a subset (n = 122) returning for follow-up evaluation. Mean FMA score improvement in the robotic group was 8.50 (confidence interval: 6.82 to 10.17), versus 8.57 (confidence interval: 6.97 to 10.18) in the conventional group, with no significant between-groups difference (adjusted mean difference -0.08, P = 0.948). Both groups also had similar change in secondary measures, except for the Motricity Index, with better results for the robotic group (adjusted mean difference 4.42, P = 0.037). At follow-up, subjects continued to improve with no between-groups differences. DISCUSSION AND CONCLUSIONS: Robotic treatment using a set of 4 devices significantly improved UL motor function, activities, and participation in subjects with subacute stroke to the same extent as a similar amount of conventional therapy. Video Abstract is available for more insights from the authors (see the Video, Supplemental Digital Content 1, available at: http://links.lww.com/JNPT/A291).

Prognostic value of motor evoked potentials in motor function recovery of upper limb after stroke.

OBJECTIVE: To determine the prognostic value of clinical assessment and motor evoked potentials for upper limb strength and functional recovery after acute stroke, and to establish the possible use of motor evoked potentials in rehabilitation. DESIGN: A prospective study. SUBJECTS: Fifty-two patients with hemiparesis were enrolled one month post-stroke; 38 patients concluded the study at 12 months. METHODS: Motor evoked potentials were recorded at baseline and after one month. Upper limb muscular strength (Medical Research Council Scale, MRC) and functional tests (Frenchay Arm Test, Barthel Index) were used as dependent outcome variables 12 months later. Motor evoked potentials were classified as present or absent. Predictive values of motor evoked potentials and MRC were evaluated. RESULTS: At 12 months, patients with baseline recordable motor evoked potentials showed a good functional recovery (positive predictive value 94%). The absence of motor evoked potentials did not exclude muscular strength recovery (negative predictive value 95%). Motor evoked potentials had a higher positive predictive value than MRC only in patients with MRC < 2. CONCLUSION: Motor evoked potentials could be a supportive tool to increase the prognostic accuracy of upper limb motor and functional outcome in hemiparetic patients, especially those with severe initial paresis (MRC < 2) and/or with motor evoked potentials absent in the post-stroke acute phase.

Gait in hemiplegia: evaluation of clinical features with the Wisconsin Gait Scale.

OBJECTIVE: To assess the ability of the Wisconsin Gait Scale to evaluate qualitative features of changes in hemiplegic gait in post-stroke patients. DESIGN: A prospective observational study. SUBJECTS: Ten healthy subjects and 56 hemiplegic outpatients, more than 12 months post-stroke, consecutively admitted in a rehabilitation centre. METHODS: Patients were videotaped while walking at a comfortable speed. Quantitative and clinical gait parameters were derived from videotaped walking tasks at admission and at the end of a period of rehabilitation training. Qualitative features were assessed using the Wisconsin Gait Scale. Functional status was rated through the modified Barthel Index. RESULTS: After training, the median Wisconsin Gait Scale score improved significantly (28 vs 26.5; p = 0.003). In particular, "weight shift to paretic side" and patterns during the swing phase of the affected leg were improved. Gait velocity (0.3 vs 0.4 m/sec; p = 0.001) and stride length (77 vs 85 cm; p = 0.0002) increased significantly, whereas number of steps (25 vs 23; p = 0.004), stride period (2.5 vs 2.3 sec; p = 0.04), and stance period (2.1 vs 2 sec; p = 0.03) of the unaffected side were reduced. The Barthel Index score increased (71 vs 78; p = 0.005). CONCLUSION: The Wisconsin Gait Scale is a useful tool to rate qualitative gait alterations of post-stroke hemiplegic subjects and to assess changes over time during rehabilitation training. It may be used when a targeted and standardized characterization of hemiplegic gait is needed for tailoring rehabilitation and monitoring results.

Application of a volar static splint in poststroke spasticity of the upper limb.

OBJECTIVE: To evaluate clinical and neurophysiologic effects of 3-month reflex inhibitory splinting (RIS) for poststroke upper-limb spasticity. DESIGN: Pretest-posttest trial. SETTING: Outpatient rehabilitation center. PARTICIPANTS: Forty consecutive patients with hemiplegia and upper-limb spasticity after stroke that had occurred at least 4 months before. INTERVENTION: Patients wore an immobilizing hand splint custom-fitted in the functional position for at least 90 minutes daily for 3 months. MAIN OUTCOMES MEASURES: Patients underwent measurement of (1) spasticity at the elbow and wrist according to Modified Ashworth Scale; (2) passive range of motion (PROM) at the wrist and elbow; (3) pain at the shoulder, elbow, and wrist using a visual analog scale; (4) spasms; and (5) comfort and time of splint application. The instrumental measure of spasticity was the ratio between the maximum amplitude of the H-reflex and the maximum amplitude of the M response (Hmax/Mmax ratio). RESULTS: A significant improvement of wrist PROM (F=8.92, P=.001) with greater changes in extension than in flexion, and a reduction of elbow spasticity (F=5.39, P=.002), wrist pain (F=2.89, P=.04), and spasms (F=4.33, P=.008) were observed. The flexor carpi radialis Hmax/Mmax ratio decreased significantly (F=4.2, P=.007). RIS was well tolerated. CONCLUSIONS: RIS may be used as an integrative treatment of poststroke upper-limb spasticity. It can be used comfortably at home, in selected patients without functional hand movements, and in cases of poor response or tolerance to antispastic drugs.

Evaluation of upper-limb spasticity after stroke: A clinical and neurophysiologic study.

OBJECTIVES: To assess upper-limb spasticity after stroke by means of clinical and instrumental tools and to identify possible variables influencing the clinical pattern. DESIGN: Descriptive measurement study of a consecutive sample of patients with upper-limb spasticity after stroke. SETTING: Neurorehabilitation hospital. PARTICIPANTS: Sixty-five poststroke hemiplegic patients. INTERVENTIONS: Not applicable. Main outcome measures Upper-limb spasticity, as assessed clinically (Modified Ashworth Scale [MAS], articular goniometry) and neurophysiologically (maximum H-reflex [Hmax], maximum M response [Mmax], Hmax/Mmax ratio). RESULTS: Poorer MAS scores were associated with lower passive range of motion (PROM) values at the wrist ( P =.01) and elbow ( P =.002). The flexor carpi radialis Hmax/Mmax ratio correlated directly with MAS scores at the wrist ( P =.005) and correlated inversely with PROM. The presence of pain in the fingers, wrist, and elbow was significantly associated only with lower PROM values at the wrist. CONCLUSIONS: Upper-limb spasticity is involved in the development of articular PROM limitation after a stroke. Pain appears to be related to PROM reduction as well, but the exact causal relationship between these 2 factors is still unclear. The MAS and the Hmax/Mmax ratio correlated when evaluating poststroke spasticity; they characterize 2 different aspects of spasticity, clinical and neurophysiologic, respectively, and they could be used as an integrated approach to study and follow poststroke patients.