Predicting models for arm impairment: External validation of the Scandinavian models and identification of new predictors in post-acute stroke settings

Abstract

BACKGROUND:

Post-stroke arm impairment at rehabilitation admission as predictor of discharge arm impairment was consistently reported as extremely useful. Several models for acute prediction exist (e.g. the Scandinavian), though lacking external validation and larger time-window admission assessments.

OBJECTIVES:

(1) use the 33 Fugl-Meyer Assessment-Upper Extremity (FMA-UE) individual items to predict total FMA-UE score at discharge of patients with ischemic stroke admitted to rehabilitation within 90 days post-injury, (2) use eight individual items (seven from the Scandinavian study plus the top predictor item from objective 1) to predict mild impairment (FMA-UE≥48) at discharge and (3) adjust the top three models from objective 2 with known confounders.

METHODS:

This was an observational study including 287 patients (from eight settings) admitted to rehabilitation (2009-2020). We applied regression models to candidate predictors, reporting adjusted R², odds ratios and ROC-AUC using 10-fold cross-validation.

RESULTS:

We achieved good predictive power for the eight item-level models (AUC: 0.70-0.82) and for the three adjusted models (AUC: 0.85-0.88). We identified finger mass flexion as new item-level top predictor (AUC:0.88) and time to admission (OR = 0.9(0.9;1.0)) as only common significant confounder.

CONCLUSION:

Scandinavian item-level predictors are valid in a different context, finger mass flexion outperformed known predictors, days-to-admission predict discharge mild arm impairment.

Keywords

Arm recovery arm functioning prognostic models prediction post-acute

1 Introduction

Hemiparesis of the contralateral upper limb is one of the most common deficits after stroke (Liu et al., 2021). Manifestations of upper extremity motor impairment include muscle weakness, joint laxity and impaired motor control (Lee et al., 2018) inducing disabilities in common daily activities such as reaching or picking up objects (Ham et al., 2022) affecting quality of life (Lieshout et al., 2020) and independence in daily living (Ghaziani et al., 2018). Therefore, improving upper limb function is a core element of rehabilitation after stroke to maximize recovery (Wattchow et al., 2018).

Prognostic models in stroke rehabilitation can assist clinicians in estimating the probability of an individual patient to achieve a favorable outcome over a specific period and guide the selection of the most appropriate intervention methods for the patient (Stinear et al., 2019).

In a recent study reporting on the development and validation of a predictive model for outcome after stroke rehabilitation, Scruitinio et al. included patients admitted to stroke rehabilitation units≤90 days from stroke occurrence (Scrutinio et al., 2015). The same time window has been consistently used in subsequent studies (Scrutinio et al., 2017).

Different combinations of tests assessing proximal (shoulder abduction, elbow flexion, placing the hand on the top of the head) and distal (finger extension, grip strength) function in the affected arm early (mostly within 7 days) post-stroke have been reported as multivariable prognostic models (Nijland et al., 2010; Persson et al., 2015; Snickars et al., 2017). Nevertheless, the need for external validation (i.e. in different contexts, by different authors) of already proposed predictors, as well as for development and validation of additional predictors available in clinical practice, still remains in both acute and post-acute settings (Harvey et al., 2015; Kwah et al., 2016 and Ghaziani et al., 2020).

Potential confounders such as, age, sex, stroke severity as measured using the National Institutes of Health Stroke Scale (NIHSS), aphasia, smoking habits, living-arrangement, time since injury to rehabilitation admission, were reported in previous research (Persson et al., 2015; Snickars et al., 2017 and Ghaziani et al., 2020).

The 33-items’ upper-extremity (UE) portion of the Fugl-Meyer assessment (FMA-UE), is the most frequently used outcome scale to measure post-stroke motor recovery of the UE, worldwide (Tauchi et al., 2021). It is considered as gold standard for evaluation of sensorimotor impairment after stroke and is the only impairment level measure recommended for stroke trials (Cecchi et al., 2021). The FMA-UE is well-established internationally, clinically feasible and shows excellent reliability, validity and responsiveness (Duncan Millar et al., 2019). The study by Shelton et al. was one of the first studies that used motor impairment (admission FMA-UE) to predict discharge motor impairment in order to focus on restorative strategies versus the use of compensatory techniques and assistive devices (Shelton et al., 2001). Estimating a patient’s future discharge scores early in a stroke rehabilitation program helps clinicians set realistic rehabilitation goals and anticipate needs for additional care or medical equipment at discharge (Harari et al., 2020).

Predictors of upper limb motor impairment including direct measurement of neurophysiological and neuroimaging biomarkers were recently reported (e.g. Boccuni, et al., 2019; Du et al., 2018, Stinear et al., 2017). Nevertheless, as recently reported (Ghaziani et al., 2020) such models are dependent on expensive equipment, often not available in regular clinical practice. Ghaziani et al. proposed in the Scandinavian study (EES/SALGOT, n = 223) easily conducted tests i.e., time efficient and easy to perform, (namely, seven individual items from the FMA-UE assessed during the first week post-stroke) to predict favorable arm recovery. This study showed that the presence of shoulder abduction, elbow extension, and finger extension predicts a moderate-to-mild motor impairment at 6 months post-stroke (Ghaziani et al., 2020).

We hypothesized that easily conducted tests such as the individual FMA-UE items assessed at rehabilitation admission, scarcely explored in the post-acute phase (≤90 days since injury to admission) could contribute as significant predictors of discharge FMA-UE total score and of motor functional outcome categories (e.g. mild, moderate-severe) in different geographical settings, involving different rehabilitation systems with different types of therapy.

Therefore, in this study we analyzed data from different settings, involving different therapies, in different contexts, aiming to (1) use each of the 33 individual FMA-UE items to predict total FMA-UE score at discharge of patients with ischemic stroke admitted to rehabilitation≤90 days since stroke onset, (2) select eight FMA-UE items (seven from the Scandinavian study plus the item with the highest predictive power from objective 1) and use each of them to predict mild impairment at discharge and (3) taking as starting point the 3 models with the highest predictive power identified in objective 2, adjust them with previously reported confounders to identify the most relevant predictors of mild impairment.

2 Methods

2.1 Study design

We conducted a retrospective observational cohort study enrolling patients from two independent data sources. The first data source involves post-acute patients with ischemic stroke admitted to the rehabilitation unit of the Acquired Brain Injury Department of Institut Guttmann (IGH), a neurological rehabilitation center in Barcelona, Spain. Recruitment period was from March 2018 to December 2020. All patients admitted at the rehabilitation unit are referred from different acute care setting hospitals. The rehabilitation program includes five daily hours of intensive treatment oriented towards cognitive, swallowing, behavioral and physical impairments as well as training in activities of daily life living.

The second data source involves participants recruited during inpatient rehabilitation, from 7 sites in the United States metropolitan areas of Los Angeles, Atlanta and Washington D.C from June 2009 to March 2014. This dataset is registered at the ClinicalTrials.gov under the title Arm Rehabilitation Study After Stroke (ICARE) (Winstein et al., 2016) with NCT identifier 00871715. The ICARE investigators tested 3 different arm therapy interventions: Accelerated Skill Acquisition Program (ASAP), Behavioral: Dose-Equivalent Usual & Customary Care (DEUCC) and Behavioral: Usual and Customary Care (UCC).

The ICARE dataset was provided by the National Institute of Neurological Disorders and Stroke (NINDS). The protocol for accessing NINDS data and the NINDS Data Request Form is listed in the Supplementary Materials.

This study conforms to the STROBE Guidelines (“Strengthening the Reporting of Observational Studies in Epidemiology”) (STROBE 2022).

2.2 Participants

Eligible participants were enrolled in the study using the following inclusion criteria: (a) first-ever ischemic stroke; (b) age≥18 years at rehabilitation admission; (c) impaired arm function measured at rehabilitation admission (±7 days) after stroke (<66 points on the FMA-UE); (d) admission to the stroke unit within 90 days after stroke onset.

Patients were excluded if one of the following criteria was present: (a) injury/condition prior to the stroke that limited the use of the affected arm; (b) severe multi-impairment or diminished physical condition prior to stroke; (c) short life expectancy; and (d) not able to communicate.

2.3 Outcomes

To address objective 1, the outcome to be predicted was the total FMA-UE score at rehabilitation discharge. The FMA-UE includes 33 items divided into 4 subscales: shoulder/elbow (A, 18 items), wrist (B, 5 items), hand (C, 7 items) and coordination/speed (D, 3 items) designed to measure impairment from proximal to distal and synergistic to isolated voluntary movement. The 33 items that constitute the FM-UE are scored on an ordinal scale of 0 (absent), 1 (partial impairment), and 2 (no impairment), resulting in a range of possible scores from zero to 66 (Fugl-Meyer et al., 1975).

To address the second and third objectives, appropriate FMA-UE total score cutoff points needed to be used. Different cut-off points have been previously reported to identify, e.g. moderate-to-mild or mild residual motor impairment determined by the FMA-UE total score. Ghaziani et al. (2020) used an FMA-UE score≥32 to predict moderate-to-mild motor impairment and FMA-UE score≥58 to predict mild impairment, at six months after stroke onset. Woytowicz et al. (2017) defined 31– 47 for moderate and 48– 66 for mild at 6 months. Similarly, Hoonhorst et al. (2015) used 32-47 for limited and 48-66 for notable or full performance.

At post-acute phase 25±7 days from stroke onset, Plantin et al. (2021) used FMA-UE score≥48 for mild impairment. Further studies and details are presented in Supplementary Table 1.

Preliminary screening of our data showed that only 12.5% of our patients scored≤32 points at discharge. Similarly, only 27.1% scored≥58 points at discharge, meanwhile 57.1% scored≥48 points, yielding to a more balanced cut-off. Therefore, in this work we used 48-66 for mild impairment and≤47 for severe and moderately-severe impairment.

2.4 Candidate predictors

To address objective 1 we used as candidate predictors all 33 items from FMA-UE protocol (FMA-UE PROTOCOL, 2021). Each item is described in Table 1 as will be used along this study. Items superscripted with an * were those used in the Scandinavian study (Ghaziani et al., 2020, and also in previous related research, e.g. Snickars et al., 2017).

Table 1
The 33-individual items that constitute the FMA-UE

A. UPPER EXTREMITY I. Reflex activity

Flexors: biceps and finger flexors (at least one) FM_BICEP

Extensors: triceps FM_TRICEP

Subtotal I (max 4)

II. Volitional movement within synergies, without gravitational help

Flexor synergy – Shoulder - retraction FM_FS_RET

Flexor synergy – Shoulder - elevation FM_FS_ELV

Flexor synergy – Shoulder - abduction (90°) FM_FS_ABD*

Flexor synergy – Shoulder – external rotation FM_FS_EXT

Flexor synergy – Elbow – flexion FM_FS_ELF

Flexor synergy – Forearm – supination FM_FS_SUP

Extensor synergy – Shoulder - adduction/internal rotation FM_ES_SHAD

Extensor synergy – Elbow - extension FM_ES_EXT*

Extensor synergy – Forearm - pronation FM_ES_FPR

Subtotal II (max 18)

III. Volitional movement mixing synergies, without compensation

Hand to lumbar spine FM_MS_HAND

Shoulder flexion 0°- 90° FM_MS_SHF

Pronation-supination FM_MS_PSUP*

Subtotal III (max 6)

IV. Volitional movement with little or no synergy

Shoulder abduction 0 - 90° FM_MOS_SAB

Shoulder flexion 90° - 180° FM_MOS_SFL

Pronation/supination FM_MOS_PRO

Subtotal IV (max 6)

V. Normal reflex activity

Biceps, triceps,finger flexors FM_NR

Subtotal V (max 2)

Total A (max 36)

B. WRIST Stability at 15° dorsiflexion (Elbow at 90°, Shoulder at 0°) FM_W_SE9*

Repeated dorsifexion / volar flexion (Elbow at 90°, Shoulder at 0°) FM_W_FE9

Stability at 15° dorsiflexion (Elbow at 0°, Shoulder at 30°) FM_W_SE3

Repeated dorsifexion / volar flexion FM_W_FE3

Circumduction FM_W_CIR

Total B (max 10)

C. HAND Finger Mass flexion FM_H_FMF

Finger Mass extension FM_H_FME*

Hook grasp FM_H_GRASP1

Thumb adduction FM_H_GRASP2

Pincer grasp, opposition FM_H_GRASP3*

Cylinder grasp FM_H_GRASP4*

Spherical grasp FM_H_GRASP5

Total C (max 14)

D. COORDINATION/SPEED Tremor FM_CS_TRE

Dysmetria FM_CS_DYS

Time FM_CS_SPE

Total D (max 6)


A. UPPER EXTREMITY	I. Reflex activity
	Flexors: biceps and finger flexors (at least one)	FM_BICEP
	Extensors: triceps	FM_TRICEP
	Subtotal I (max 4)
	II. Volitional movement within synergies, without gravitational help
	Flexor synergy – Shoulder - retraction	FM_FS_RET
	Flexor synergy – Shoulder - elevation	FM_FS_ELV
	Flexor synergy – Shoulder - abduction (90°)	FM_FS_ABD*
	Flexor synergy – Shoulder – external rotation	FM_FS_EXT
	Flexor synergy – Elbow – flexion	FM_FS_ELF
	Flexor synergy – Forearm – supination	FM_FS_SUP
	Extensor synergy – Shoulder - adduction/internal rotation	FM_ES_SHAD
	Extensor synergy – Elbow - extension	FM_ES_EXT*
	Extensor synergy – Forearm - pronation	FM_ES_FPR
	Subtotal II (max 18)
	III. Volitional movement mixing synergies, without compensation
	Hand to lumbar spine	FM_MS_HAND
	Shoulder flexion 0°- 90°	FM_MS_SHF
	Pronation-supination	FM_MS_PSUP*
	Subtotal III (max 6)
	IV. Volitional movement with little or no synergy
	Shoulder abduction 0 - 90°	FM_MOS_SAB
	Shoulder flexion 90° - 180°	FM_MOS_SFL
	Pronation/supination	FM_MOS_PRO
	Subtotal IV (max 6)
	V. Normal reflex activity
	Biceps, triceps,finger flexors	FM_NR
	Subtotal V (max 2)
	Total A (max 36)
B. WRIST	Stability at 15° dorsiflexion (Elbow at 90°, Shoulder at 0°)	FM_W_SE9*
	Repeated dorsifexion / volar flexion (Elbow at 90°, Shoulder at 0°)	FM_W_FE9
	Stability at 15° dorsiflexion (Elbow at 0°, Shoulder at 30°)	FM_W_SE3
	Repeated dorsifexion / volar flexion	FM_W_FE3
	Circumduction	FM_W_CIR
	Total B (max 10)
C. HAND	Finger Mass flexion	FM_H_FMF
	Finger Mass extension	FM_H_FME*
	Hook grasp	FM_H_GRASP1
	Thumb adduction	FM_H_GRASP2
	Pincer grasp, opposition	FM_H_GRASP3*
	Cylinder grasp	FM_H_GRASP4*
	Spherical grasp	FM_H_GRASP5
	Total C (max 14)
D. COORDINATION/SPEED	Tremor	FM_CS_TRE
	Dysmetria	FM_CS_DYS
	Time	FM_CS_SPE
	Total D (max 6)

FMA-UE: Fugl-Meyer Assessment - Upper Extremity. Items superscripted with an * were those used in the Scandinavian study (Ghaziani et al., 2020).

To address objective 3, taking as starting point the 3 models with the highest predictive power identified in objective 2, we used potential confounders reported in related research (Persson et al., 2015; Snickars et al., 2017 and Ghaziani et al., 2020) such as stroke severity (NIHSS), age, sex, smoking habits, living-arrangement.

Both our candidate predictors and outcomes co-vary with the time post-stroke, therefore our potential confounders include time since injury to rehabilitation admission and length of stay.

2.5 Statistical analyses

Statistical analyses were performed using in R-v3.5.1 (64 bits). The significance level was set at p≤0.05. χ² test was used to compare differences between groups of categorical variables and the Wilcoxon signed rank test for continuous and ordered variables.

To address objective 1, each FMA-UE item was included in a multiple regression analysis using the enter method to predict FMA-UE total score at discharge.

To address objective 2, we used the FMA-UE individual items with the highest predictive power obtained from objective 1) measured using the model’s explained variance (Adjusted R²) to predict mild impairment at discharge and entered them in a logistic regression with dependent variable dichotomized as mild impairment (FMA-UE≥48) or moderate-to-severe (FMA-UE<48) impairment.

To address objective 3, we selected from objective 2 the top three models with the highest AUC and adjusted each model using known confounders (e.g. age, time since injury to admission) with dependent variable dichotomized as mild impairment (FMA-UE≥48) or moderate to severe (FMA-UE<48).

Spearman correlation tests between independent variables were performed, as preliminary analysis to identify collinear variables. To test the models’ goodness of fit and accuracy, AUC-ROC curves were used. AUC was calculated by means of 10-fold cross validation in a test set, therefore we independently partitioned initial data in training set (65%) and test set (35%), as in similar related research using the caret R package (caret, 2022). McFadden (r2.m), Cox and Snell (r2.cs), Nagelkerke (r2.n), and Estrella (r2.e) were used to assess the importance of variables (dominance analysis, 2022). No correction for multiple tests was performed.

2.6 Ethical considerations

The study follows the Declaration of Helsinki and was approved by the Ethics Committee of Clinical Research of Institut Guttmann. The participants are anonymized and non-identifiable.

3 Results

From March 2018 to December 2020 a total of 134 patients with first-ever stroke were admitted to the rehabilitation unit of the Institut Guttmann hospital and assessed at admission and discharge using the FMA-UE, 71 of them with ischemic stroke. After excluding 18 with more than 90 days since stroke onset to rehabilitation admission, 3 with more than 7 days since admission to assessment, 2 younger than 18 years old at the moment of admission, 1 with injury/condition prior to the stroke that limited the use of the affected arm and 1 with severe multi-impairment or diminished physical condition prior to stroke; 46 patients were included in the study.

The ICARE dataset included 361 recruited between 5 and 106 days post-stroke (Winstein et al., 2016), 288 of them with ischemic stroke. After excluding 39 with more than 90 days since stroke onset to rehabilitation admission and 2 younger than 18 years old at the moment of admission, 241 patients were left to be included in the study. Therefore, the total number of included patients was 287.

Table 2 presents patients’ characteristics at admission, for each group. The median age was 60 years with no differences between both groups and the proportion of men was 57.1%. The MCID-9 (Minimal Clinically Important Difference at 9 points) (Arya et al., 2011) and MCID-12 (Minimal Clinically Important Difference at 12 points) (Hiragami et al., 2019) were 30.1% and 20.3% for patients in the severe-moderately-severe group, showing the margin of meaningful and clinically important improvement for patients in that group. MCID-9 and MCID-12 were significantly larger in the mild group as shown in Table 2, though for such patients FMA-UE total score was also significantly higher at admission (p < 0.001).

Table 2
Patient characteristics for all (n = 287) patients stratified in two groups (severe-moderately severe and mild) according to their total FMA-UA score at rehabilitation discharge

TOTAL (n = 287) <48 Severe- moderately-severe (n = 123) ≥48 Mild (n = 164) p

Male, n(%) 164 (57) 72 (58) 92 (56) 0.679

Age at admission, median (Q1-Q3) 60(51-69) 60(54-68) 61(50-70) 0.664

Younger than 65 years, n(%) 181 (63) 76 (61) 105 (64) 0.698

NIHSS at admission, median(Q1-Q3) 4(2-5) 5(3-7) 3(2-4) <0.001

NIHSS categorization, n(%) <0.001

Mild 221 (78) 76 (63) 145 (90)

Moderately severe 51 (18) 37 (30) 14 (8)

Severe 8 (2) 7 (5) 1 (0)

Affected side (Left), n(%) 146 (50) 68 (55) 78 (47) 0.195

Dominance (Right), n(%) 258 (89) 109 (88) 149 (90) 0.534

Aphasia, n(%) 35 (12) 18 (14) 17 (10) 0.274

Smoking habits (at admission), n(%) 42 (14) 21 (17) 21 (12) 0.311

COGNITVE-FIM at admission 26(22-30) 26(20-30) 26.5 (22-30) 0.307

MOTOR-FIM at admission 42(34-50) 40(32-49) 43(37-52) 0.014

TOTAL-FIM at admission 67(59-78) 64(55-75) 70(60-80) 0.013

Days since injury to admission, median (Q1-Q3) 41(28-57) 53(36.5-67) 33(25-44) <0.001

FMA-UE at admission, median (Q1-Q3) 41(32-49) 31(22-37) 47(42-52.5) <0.001

FMA-UE at discharge, median (Q1-Q3) 51(40.5-58) 38(26.5-44) 57(54-61) <0.001

LOS, days, median (Q1-Q3) 124(116-133) 125(114-132) 124(117-134) 0.325

Living arrangement, n(%) 0.519

Alone 67 (23) 31 (25) 36 (22)

With a spouse, other relative or friend 220 (76) 92 (74) 128 (78)

Data source n(%) <0.001

ICARE 241 (84) 92 (74) 149 (90)

IGH 46 (16) 31 (25) 15 (9)

MCID-12, n(%) 100 (34) 25 (20) 75 (45) <0.001

MCID-9, n(%) 135 (47) 37 (30) 98 (59) <0.001

	TOTAL (n = 287)	<48 Severe- moderately-severe (n = 123)	≥48 Mild (n = 164)	p
Male, n(%)	164 (57)	72 (58)	92 (56)	0.679
Age at admission, median (Q1-Q3)	60(51-69)	60(54-68)	61(50-70)	0.664
Younger than 65 years, n(%)	181 (63)	76 (61)	105 (64)	0.698
NIHSS at admission, median(Q1-Q3)	4(2-5)	5(3-7)	3(2-4)	<0.001
NIHSS categorization, n(%)				<0.001
Mild	221 (78)	76 (63)	145 (90)
Moderately severe	51 (18)	37 (30)	14 (8)
Severe	8 (2)	7 (5)	1 (0)
Affected side (Left), n(%)	146 (50)	68 (55)	78 (47)	0.195
Dominance (Right), n(%)	258 (89)	109 (88)	149 (90)	0.534
Aphasia, n(%)	35 (12)	18 (14)	17 (10)	0.274
Smoking habits (at admission), n(%)	42 (14)	21 (17)	21 (12)	0.311
COGNITVE-FIM at admission	26(22-30)	26(20-30)	26.5 (22-30)	0.307
MOTOR-FIM at admission	42(34-50)	40(32-49)	43(37-52)	0.014
TOTAL-FIM at admission	67(59-78)	64(55-75)	70(60-80)	0.013
Days since injury to admission, median (Q1-Q3)	41(28-57)	53(36.5-67)	33(25-44)	<0.001
FMA-UE at admission, median (Q1-Q3)	41(32-49)	31(22-37)	47(42-52.5)	<0.001
FMA-UE at discharge, median (Q1-Q3)	51(40.5-58)	38(26.5-44)	57(54-61)	<0.001
LOS, days, median (Q1-Q3)	124(116-133)	125(114-132)	124(117-134)	0.325
Living arrangement, n(%)				0.519
Alone	67 (23)	31 (25)	36 (22)
With a spouse, other relative or friend	220 (76)	92 (74)	128 (78)
Data source n(%)				<0.001
ICARE	241 (84)	92 (74)	149 (90)
IGH	46 (16)	31 (25)	15 (9)
MCID-12, n(%)	100 (34)	25 (20)	75 (45)	<0.001
MCID-9, n(%)	135 (47)	37 (30)	98 (59)	<0.001

FMA-UE: Fugl-Meyer Assessment – Upper Extremity; LOS: length of stay in rehabilitation; NIHSS: National Institutes of Health Stroke Scale; FIM: Functional Independence Measure. MCID (Minimal Clinically Important Difference) is defined as “the smallest difference in score in the domain of interest which patients perceive as beneficial and which would mandate, in the absence of troublesome side effect and excessive cost, a change in the patient’s management”. MCID-12 is an MCID = 12 points; MCID-9 is an MCID = 9 points.

Supplementary Table 2 shows a strong negative correlation (r=-0.54, p < 0.001) between NIHSS at admission and FMA-UE total score at admission. Length of stay (LOS) was very weakly correlated to FMA-UE (FM-ADM in Supplementary Table 2) at admission (r = 0.12, p < 0.05) and LOS was not correlated to FMA-UE at discharge, either not correlated to FM gain.

3.1 Item-level predictors of total FMA-UE at discharge (objective 1)

Table 3 presents for each item from subscale A, the obtained coefficients (95% CI), level of significance and adjusted R², with two items exceeding 50% of explained variance: elbow extension (FM_ES_EXT*) and forearm pronation (FM_ES_FPR).

Table 3
Predictive models for FMA-UE score at discharge for each candidate predictor item (FMA – subscale A). Linear regression models

Item Category n Coefficient (95% CI) p Adjusted R²

FM_BICEP 0: None 14 (4.9%) 1.7

2: Elicited 273 (95.1%) 1.0 (-7.1; 9.1) 0.801

FM_TRICEP 0: None 35 (12.2%) 1.8

2: Elicited 252 (87.8%) -3.2 (-8.5;2.1) 0.235

FM_FS_RET 0: None 34 (11.8%) 31.8

1: Partial 126 (43.9%) 20.8(16.1;25.6) <0.001

2: Full 127 (44.3%) 27.9(23.2;32.6) <0.001

FM_FS_ELV 0: None 22 (7.7%) 38.2

1: Partial 105 (36.6%) 24.3(18.8;29.7) <0.001

2: Full 160 (55.7%) 34.4(29.1;39.6) <0.001

FM_FS_ABD* 0: None 31 (10.8%) 35.8

1: Partial 105 (36.6%) 22.0(17.1;26.8) <0.001

2: Full 151 (52.6%) 29.9(25.3;34.6) <0.001

FM_FS_EXT 0: None 36 (12.5%) 44.3

1: Partial 146 (50.9%) 25.4(21.3;29.5) <0.001

2: Full 105 (36.6%) 32.9(28.6;37.2) <0.001

FM_FS_ELF 0: None 19 (6.6%) 41.9

1: Partial 67 (23.3%) 26.5(20.6;32.4) <0.001

2: Full 201 (70.0%) 37.5(32.0;42.9) <0.001

FM_FS_SUP 0: None 40 (13.9%) 45.1

1: Partial 169 (58.9%) 27.9(24.0;31.7) <0.001

2: Full 78 (27.2%) 31.4(27.1;35.7) <0.001

FM_ES_SHAD 0: None 28 (9.8%) 49.9

1: Partial 106 (36.9%) 27.4(23.0;31.9) <0.001

2: Full 153 (53.3%) 36.7(32.4;41.0) <0.001

FM_ES_EXT* 0: None 33 (11.5%) 52.2

1: Partial 122 (42.5%) 27.9(23.9;32.0) <0.001

2: Full 132 (46.0%) 35.9(31.9;39.9) <0.001

FM_ES_FPR 0: None 31 (10.8%) 52.3

1: Partial 121 (42.2%) 30.4(26.2;34.5) <0.001

2: Full 135 (47.0%) 36.8(32.7;40.9) <0.001

FM_MS_HAND 0: None 39 (13.6%) 43.8

1: Partial 167 (58.2%) 26.4(22.5;30.4) <0.001

2: Full 81 (28.2%) 32.1(27.7;36.4) <0.001

FM_MS_SHF 0: None 79 (27.5%) 37.3

1: Partial 101 (35.2%) 16.6(13.0;20.1) <0.001

2: Full 107 (37.3%) 22.8(19.4;26.3) <0.001

FM_MS_PSUP* 0: None 49 (17.1%) 39.6

1: Partial 161 (56.1%) 22.3(18.5;26.0) <0.001

2: Full 77 (26.8%) 28.5(24.3;32.7) <0.001

FM_MOS_SAB 0: None 79 (27.5%) 35.8

1: Partial 108 (37.6%) 16.1(12.6;19.7) <0.001

2: Full 100 (34.8%) 22.7(19.1;26.3) <0.001

FM_MOS_SFL 0: None 139 (48.4%) 27.4

1: Partial 122 (42.5%) 15.0(11.9;18.2) <0.001

2: Full 26 (9.1%) 19.0(13.6;24.4) <0.001

FM_MOS_PRO 0: None 96 (33.4%) 38.1

1: Partial 160 (55.7%) 18.6(15.6;21.6) <0.001

2: Full 31 (10.8%) 24.0(19.2;28.8) <0.001

FM_NR 0: None 270 (94.1%) 1.8

1: Partial 5 (1.7%) 5.8(-7.4;19.11) 0.386

2: Full 12 (4.2%) 11.4(2.7;20.1) 0.009

Item	Category	n	Coefficient (95% CI)	p	Adjusted R²
FM_BICEP	0: None	14 (4.9%)			1.7
	2: Elicited	273 (95.1%)	1.0 (-7.1; 9.1)	0.801
FM_TRICEP	0: None	35 (12.2%)			1.8
	2: Elicited	252 (87.8%)	-3.2 (-8.5;2.1)	0.235
FM_FS_RET	0: None	34 (11.8%)			31.8
	1: Partial	126 (43.9%)	20.8(16.1;25.6)	<0.001
	2: Full	127 (44.3%)	27.9(23.2;32.6)	<0.001
FM_FS_ELV	0: None	22 (7.7%)			38.2
	1: Partial	105 (36.6%)	24.3(18.8;29.7)	<0.001
	2: Full	160 (55.7%)	34.4(29.1;39.6)	<0.001
FM_FS_ABD*	0: None	31 (10.8%)			35.8
	1: Partial	105 (36.6%)	22.0(17.1;26.8)	<0.001
	2: Full	151 (52.6%)	29.9(25.3;34.6)	<0.001
FM_FS_EXT	0: None	36 (12.5%)			44.3
	1: Partial	146 (50.9%)	25.4(21.3;29.5)	<0.001
	2: Full	105 (36.6%)	32.9(28.6;37.2)	<0.001
FM_FS_ELF	0: None	19 (6.6%)			41.9
	1: Partial	67 (23.3%)	26.5(20.6;32.4)	<0.001
	2: Full	201 (70.0%)	37.5(32.0;42.9)	<0.001
FM_FS_SUP	0: None	40 (13.9%)			45.1
	1: Partial	169 (58.9%)	27.9(24.0;31.7)	<0.001
	2: Full	78 (27.2%)	31.4(27.1;35.7)	<0.001
FM_ES_SHAD	0: None	28 (9.8%)			49.9
	1: Partial	106 (36.9%)	27.4(23.0;31.9)	<0.001
	2: Full	153 (53.3%)	36.7(32.4;41.0)	<0.001
FM_ES_EXT*	0: None	33 (11.5%)			52.2
	1: Partial	122 (42.5%)	27.9(23.9;32.0)	<0.001
	2: Full	132 (46.0%)	35.9(31.9;39.9)	<0.001
FM_ES_FPR	0: None	31 (10.8%)			52.3
	1: Partial	121 (42.2%)	30.4(26.2;34.5)	<0.001
	2: Full	135 (47.0%)	36.8(32.7;40.9)	<0.001
FM_MS_HAND	0: None	39 (13.6%)			43.8
	1: Partial	167 (58.2%)	26.4(22.5;30.4)	<0.001
	2: Full	81 (28.2%)	32.1(27.7;36.4)	<0.001
FM_MS_SHF	0: None	79 (27.5%)			37.3
	1: Partial	101 (35.2%)	16.6(13.0;20.1)	<0.001
	2: Full	107 (37.3%)	22.8(19.4;26.3)	<0.001
FM_MS_PSUP*	0: None	49 (17.1%)			39.6
	1: Partial	161 (56.1%)	22.3(18.5;26.0)	<0.001
	2: Full	77 (26.8%)	28.5(24.3;32.7)	<0.001
FM_MOS_SAB	0: None	79 (27.5%)			35.8
	1: Partial	108 (37.6%)	16.1(12.6;19.7)	<0.001
	2: Full	100 (34.8%)	22.7(19.1;26.3)	<0.001
FM_MOS_SFL	0: None	139 (48.4%)			27.4
	1: Partial	122 (42.5%)	15.0(11.9;18.2)	<0.001
	2: Full	26 (9.1%)	19.0(13.6;24.4)	<0.001
FM_MOS_PRO	0: None	96 (33.4%)			38.1
	1: Partial	160 (55.7%)	18.6(15.6;21.6)	<0.001
	2: Full	31 (10.8%)	24.0(19.2;28.8)	<0.001
FM_NR	0: None	270 (94.1%)			1.8
	1: Partial	5 (1.7%)	5.8(-7.4;19.11)	0.386
	2: Full	12 (4.2%)	11.4(2.7;20.1)	0.009

Items superscripted with an * were those used in the Scandinavian study (Ghaziani et al., 2020).

Coefficients are interpreted as the difference in the predicted value for each one-unit difference in the predictor. A one unit difference represents switching from the reference category (0 = none function) to the other. Considering e.g. elbow extension (FM_ES_EXT*), if elbow extension assessed at admission changes from none function to partial function, then our model predicts that 27.9 units of extra FMA-UE total score at discharge will be achieved.

Similarly, Table 4 presents the obtained results from subscales B, C and D with one item finger mass flexion (FM_H_FMF) exceeding 50% of explained variance.

Table 4

Predictive models for FMA-UE score at discharge for each candidate predictor item (FMA – subscales B, C, D). Linear regression models

Item	Category	n	Coefficient (95% CI)	p	Adjusted R²
FM_W_SE9*	0: None	52 (18.1%)			44.2
	1: Partial	155 (54.0%)	23.5(19.9;27.0)	<0.001
	2: Full	80 (27.9%)	29.0(25.0;32.9)	<0.001
FM_W_FE9	0: None	43 (15.0%)			47.6
	1: Partial	165 (57.5%)	24.7(21.0;28.4)	<0.001
	2: Full	79 (27.5%)	33.1(29.0;37.1)	<0.001
FM_W_SE3	0: None	95 (33.1%)			33.4
	1: Partial	148 (51.6%)	16.7(13.6;19.9)	<0.001
	2: Full	44 (15.3%)	22.6(18.2;27.0)	<0.001
FM_W_FE3	0: None	90 (31.4%)			39.3
	1: Partial	149 (51.9%)	18.5(15.4;21.6)	<0.001
	2: Full	48 (16.7%)	24.3(20.2;28.4)	<0.001
FM_W_CIR	0: None	54 (18.8%)			43.2
	1: Partial	216 (75.3%)	24.3(20.9;27.7)	<0.001
	2: Full	17 (5.9%)	32.6(26.4;38.9)	<0.001
FM_H_FMF	0: None	24 (8.4%)			55.3
	1: Partial	141 (49.1%)	31.6(27.2;35.9)	<0.001
	2: Full	121 (42.2%)	41.9(37.5;46.4)	<0.001
FM_H_FME*	0: None	28 (9.8%)			60.1
	1: Partial	152 (53.0%)	33.0(29.2;36.9)	<0.001
	2: Full	107 (37.3%)	42.0(38.0;46.0)	<0.001
FM_H_GRASP1	0: None	105 (36.6%)			22.0
	1: Partial	107 (37.3%)	11.5(7.9;15.1)	<0.001
	2: Full	75 (26.1%)	17.5(13.5;21.5)	<0.001
FM_H_GRASP2	0: None	44 (15.3%)			47.3
	1: Partial	167 (58.2%)	25.3(21.6;28.9)	<0.001
	2: Full	76 (26.5%)	32.4(28.4;36.5)	<0.001
FM_H_GRASP3*	0: None	93 (32.4%)			27.9
	1: Partial	135 (47.0%)	15.7(12.3;19.1)	<0.001
	2: Full	59 (20.6%)	19.4(15.2;23.6)	<0.001
FM_H_GRASP4*	0: None	56 (19.5%)			48.9
	1: Partial	103 (35.9%)	21.9(18.3;25.4)	<0.001
	2: Full	128 (44.6%)	28.5(25.1;31.9)	<0.001
FM_H_GRASP5	0: None	48 (16.7%)			48.7
	1: Partial	108 (37.6%)	24.1(20.4;27.9)	<0.001
	2: Full	131 (45.6%)	29.9(26.3;33.5)	<0.001
FM_CS_TRE	0: None	38 (13.2%)			39.8
	1: Partial	58 (20.2%)	28.8(24.0;33.6)	<0.001
	2: Full	191 (66.6%)	27.9(23.8;32.0)	<0.001
FM_CS_DYS	0: None	44 (15.3%)			36.7
	1: Partial	67 (23.3%)	25.6(21.1;30.2)	<0.001
	2: Full	176 (61.3%)	25.3(21.4;29.3)	<0.001
FM_CS_SPE	0: None	121 (42.2%)			26.8
	1: Partial	100 (34.8%)	15.1(11.7;18.6)	<0.001
	2: Full	66 (23.0%)	16.9(13.0;20.8)	<0.001

Figure 1 ranks all 33 FMA-UE items, visually showing for each of them the obtained adjusted R², highest values are shown to the right of the Figure. Therefore, the overall top predictor was finger mass extension (FM_H_FME*) followed by finger mass flexion (FM_H_FMF), the former was included in the Scandinavian study (Ghaziani et al., 2020) whereas the latter was not.

Fig. 1

Obtained adjusted R² for each of the FMA-UE items.

3.2 Predictors of mild motor impairment (FMA-UE≥48) at discharge: Unadjusted models (objective 2)

Table 5 presents the obtained ORs and AUC (95% CI) for the 7 items reported in the Scandinavian study (Ghaziani et al., 2020) and for finger mass flexion (FM_H_FMF), the one with the highest adjusted R² from Section 3.1. In order to interpret these results, considering e.g. elbow extension (FM_ES_EXT*), OR = 13.6 indicates that the odds to achieve mild motor impairment at discharge is 13.6 times higher with partial function at admission than with none function.

Table 5
Unadjusted logistic regression models of FMA-UE at discharge for 7 selected candidate predictor items used in the Scandinavian study (superscripted with an ) plus one item (FM_H_FMF) with the highest predictive power as determined in Tables 3 and 4

Item Category <48 n (%) ≥48 n (%) OR (95% CI) p AUC (95% CI) Sensitivity. Specificity

FM_FS_ABD 0 29 (23.6) 2 (1.2) 0.72 (0.64-0.81) 0.69 0.67

1 56 (45.5) 49 (29.9) 12.6(3.5;81.0) <0.001

2 38 (30.9) 113 (68.9) 43.1(12.2;274.7) <0.001

FM_ES_EXT * 0 31 (25.2) 2 (1.2) 0.80(0.72-0.87) 0.81 0.70

1 65 (52.8) 57 (34.8) 13.6(3.8;86.2) <0.001

2 27 (22.0) 105 (64.0) 60.3(16.8;387.1) <0.001

FM_MS_PSUP* 0 40 (32.5) 9 (5.5) 0.75(0.67-0.83) 0.42 0.93

1 69 (56.1) 92 (56.1) 5.9(2.8;13.7) <0.001

2 14 (11.4) 63 (38.4) 20.0(8.2;53.2) <0.001

FM_W_SE9* 0 45 (36.6) 7 (4.3) 0.77(0.70-0.85) 0.32 0.96

1 64 (52.0) 91 (55.5) 9.1(4.1;23.3) <0.001

2 14 (11.4) 66 (40.2) 30.3(11.9;87.1) <0.001

FM_H_FME* 0 27 (22.0) 1 (0.6) 0.80(0.72-0.87) 0.89 0.60

1 79 (64.2) 73 (44.5) 24.9(5.1;450.6) <0.001

2 17 (13.8) 90 (54.9) 11.9(5.2;51.3) <0.001

FM_H_GRASP3* 0 66 (53.7) 27 (16.5) 0.70(0.61-0.78) 0.53 0.80

1 45 (36.6) 90 (54.9) 4.8(2.7;8.8) <0.001

2 12 (9.8) 47 (28.7) 9.6(4.5;21.5) <0.001

FM_H_GRASP4* 0 49 (39.8) 7 (4.3) 0.79(0.72-0.87) 0.40 0.95

1 48 (39.0) 55 (33.5) 8.0(3.5;20.9) <0.001

2 26 (21.1) 102 (62.2) 27.4(11.7;72.9) <0.001

FM_H_FMF 0 23 (18.7) 1 (0.6) 0.82(0.75-0.89) 0.89 0.71

1 81 (65.9) 61 (37.2) 17.3(3.5;314.1) <0.001

2 19 (15.4) 102 (62.2) 123.4(23.8;2275) <0.001

Item	Category	<48 n (%)	≥48 n (%)	OR (95% CI)	p	AUC (95% CI)	Sensitivity.	Specificity
FM_FS_ABD*	0	29 (23.6)	2 (1.2)			0.72 (0.64-0.81)	0.69	0.67
	1	56 (45.5)	49 (29.9)	12.6(3.5;81.0)	<0.001
	2	38 (30.9)	113 (68.9)	43.1(12.2;274.7)	<0.001
FM_ES_EXT *	0	31 (25.2)	2 (1.2)			0.80(0.72-0.87)	0.81	0.70
	1	65 (52.8)	57 (34.8)	13.6(3.8;86.2)	<0.001
	2	27 (22.0)	105 (64.0)	60.3(16.8;387.1)	<0.001
FM_MS_PSUP*	0	40 (32.5)	9 (5.5)			0.75(0.67-0.83)	0.42	0.93
	1	69 (56.1)	92 (56.1)	5.9(2.8;13.7)	<0.001
	2	14 (11.4)	63 (38.4)	20.0(8.2;53.2)	<0.001
FM_W_SE9*	0	45 (36.6)	7 (4.3)			0.77(0.70-0.85)	0.32	0.96
	1	64 (52.0)	91 (55.5)	9.1(4.1;23.3)	<0.001
	2	14 (11.4)	66 (40.2)	30.3(11.9;87.1)	<0.001
FM_H_FME*	0	27 (22.0)	1 (0.6)			0.80(0.72-0.87)	0.89	0.60
	1	79 (64.2)	73 (44.5)	24.9(5.1;450.6)	<0.001
	2	17 (13.8)	90 (54.9)	11.9(5.2;51.3)	<0.001
FM_H_GRASP3*	0	66 (53.7)	27 (16.5)			0.70(0.61-0.78)	0.53	0.80
	1	45 (36.6)	90 (54.9)	4.8(2.7;8.8)	<0.001
	2	12 (9.8)	47 (28.7)	9.6(4.5;21.5)	<0.001
FM_H_GRASP4*	0	49 (39.8)	7 (4.3)			0.79(0.72-0.87)	0.40	0.95
	1	48 (39.0)	55 (33.5)	8.0(3.5;20.9)	<0.001
	2	26 (21.1)	102 (62.2)	27.4(11.7;72.9)	<0.001
FM_H_FMF	0	23 (18.7)	1 (0.6)			0.82(0.75-0.89)	0.89	0.71
	1	81 (65.9)	61 (37.2)	17.3(3.5;314.1)	<0.001
	2	19 (15.4)	102 (62.2)	123.4(23.8;2275)	<0.001

Items superscripted with an * were those used in the Scandinavian study (Ghaziani et al., 2020).

As shown in Table 5, the top 3 predictive items were elbow extension (FM_ES_EXT*), finger mass extension (FM_H_FME*) and finger mass flexion (FM_H_FMF).

3.3 Predictors of mild motor impairment (FMA-UE≥48) at discharge: Adjusted models (objective 3)

Table 6 presents the obtained ORs and AUC (95% CI) for the top 3 predictors from section 3.2. Finger mass flexion (FM_H_FMF) item yielded the highest AUC = 0.88 (0.82-0.94) with sensitivity and specificity = 0.83. The only other significant independent variable was the time since stroke onset to rehabilitation admission with an OR = 0.9 indicating that each additional increase of one day in time to admission is associated with a 10% decrease in the odds of achieving mild motor impairment at discharge.

Table 6
Adjusted models of the dichotomized FMA-UE at discharge for each of the top 3 candidate predictor items

OR (95% CI) p AUC (95% CI) Sensitivity Specificity

FM_ES_EXT_01 28.7(6.6;207.1) <0.001 0.85(0.78-0.93) 0.71 0.89

FM_ES_EXT_02 107.9(24.2;801.4) <0.001

Sex.Female 1.21(0.6;2.2) 0.522

Age 0.9(0.9;1.1) 0.339

TSI 0.9 (0.9;1.0) <0.001

LOS 0.9(0.9;1.0) 0.319

Aphasia 1.0(0.4;2.8) 0.900

Smoke. 2.2(0.8;6.2) 0.123

Living with a spouse, 2.0(1.0;4.1) 0.042

other relative or friend

FM_H_FME_01 37.5(6.1;741.3) <0.001 0.88(0.81-0.94) 0.71 0.90

FM_H_FME_02 231.3(36.3;4662.5) <0.001

Sex.Female 1.2(0.6;2.2) 0.483

Age 0.9(0.9;1.1) 0.038

TSI 0.9(0.9;1.0) <0.001

LOS 0.9(0.9;1.0) 0.485

Aphasia 0.8(0.2;2.2) 0.692

Smoke. 1.2(0.4;3.7) 0.647

Living with a spouse, 1.8(0.9;3.6) 0.092

other relative or friend

FM_H_FMF_01 21.5(3.5;424.8) 0.005 088(0.82-0.94) 0.83 0.83

FM_H_FMF_02 143.2(22.6;2881) <0.001

Sex.Female 1.3(0.7;2.5) 0.315

Age 0.9(0.9;1.0) 0.460

TSI 0.9(0.9;1.0) <0.001

LOS 0.9(0.9;1.0) 0.480

Aphasia 0.8(0.3;2.3) 0.795

Smoke. 1.1(0.4;3.1) 0.786

Living with a spouse, 1.8(0.9;3.6) 0.099

other relative or friend

	OR (95% CI)	p	AUC (95% CI)	Sensitivity	Specificity
FM_ES_EXT_01	28.7(6.6;207.1)	<0.001	0.85(0.78-0.93)	0.71	0.89
FM_ES_EXT_02	107.9(24.2;801.4)	<0.001
Sex.Female	1.21(0.6;2.2)	0.522
Age	0.9(0.9;1.1)	0.339
TSI	0.9 (0.9;1.0)	<0.001
LOS	0.9(0.9;1.0)	0.319
Aphasia	1.0(0.4;2.8)	0.900
Smoke.	2.2(0.8;6.2)	0.123
Living with a spouse,	2.0(1.0;4.1)	0.042
other relative or friend
FM_H_FME_01	37.5(6.1;741.3)	<0.001	0.88(0.81-0.94)	0.71	0.90
FM_H_FME_02	231.3(36.3;4662.5)	<0.001
Sex.Female	1.2(0.6;2.2)	0.483
Age	0.9(0.9;1.1)	0.038
TSI	0.9(0.9;1.0)	<0.001
LOS	0.9(0.9;1.0)	0.485
Aphasia	0.8(0.2;2.2)	0.692
Smoke.	1.2(0.4;3.7)	0.647
Living with a spouse,	1.8(0.9;3.6)	0.092
other relative or friend
FM_H_FMF_01	21.5(3.5;424.8)	0.005	088(0.82-0.94)	0.83	0.83
FM_H_FMF_02	143.2(22.6;2881)	<0.001
Sex.Female	1.3(0.7;2.5)	0.315
Age	0.9(0.9;1.0)	0.460
TSI	0.9(0.9;1.0)	<0.001
LOS	0.9(0.9;1.0)	0.480
Aphasia	0.8(0.3;2.3)	0.795
Smoke.	1.1(0.4;3.1)	0.786
Living with a spouse,	1.8(0.9;3.6)	0.099
other relative or friend

TSI: time since stroke onset to rehabilitation admission; LOS: length of stay.

For this model we performed a dominance analysis using four different metrics (r2.m, r2.cs, r2.n, r2.e). Figure 2 presents the results for the Nagelkerke (r2.n) metric (very similar to those obtained using the other three metrics), clearly showing the highest dominance of the finger mass flexion (FM_H_FMF) item and the time since injury to rehabilitation admission.

Fig. 2

Results of dominance analysis.

We identified the optimal cut-off for the number of days to discriminate between mild (FMA-UE≥48) and moderate-to-severe (FMA-UE<48) motor impairment, obtaining forty-two days using two different methods (see Supplementary Figures 1 and 2).

4 Discussion

In this study we used each of the 33 individual FMA-UE items to predict total FMA-UE score at discharge of patients with ischemic stroke admitted to post-acute rehabilitation within 90 days from stroke occurrence to rehabilitation admission. We identified the top individual predictors, some of them previously reported in related research such as FM_H_FME*, meanwhile others not such as FM_H_FMF, FM_ES_FPR. FM_ES_SHAD.

We performed a twofold external validation (involving two merged datasets, one European and the other from U.S) of the 7 items reported in the Scandinavian study (Ghaziani et al., 2020). We extended their results to a larger time span from injury to admission considering first each of the individual items and then adjusting the top 3 models with known confounders. We showed good predictive power in both cases with AUC: 0.70-0.82 for the unadjusted models and AUC: 0.85-0.88 for the adjusted models. The large sample size (n = 287) strengthens the internal validity of our results. In Supplementary Table 1, the related studies are included, of which the only large sample was (n = 460) (Hoonhorst et al., 2015).

Besides, we identified an additional FMA-UE item, finger mass flexion (FM_H_FMF) which yielded the highest unadjusted AUC = 0.82 as well as the highest adjusted AUC = 0.88. Finger mass extension (FM_H_FME*) previously reported in the Scandinavian study (Ghaziani et al., 2020) also yielded an adjusted AUC = 0.88, but to our best knowledge finger mass flexion was never proposed before as individual predictor of mild motor impairment.

In relation to the time since injury to rehabilitation admission, Scruitinio et al. recently reported it to be significantly associated with functional independence gain at discharge (as measured using the FIM within 90 days since injury to admission) (Scrutinio et al., 2015). The authors concluded that this finding “can be a relevant issue for clinicians involved in decision making about admission to rehabilitation hospital and policy makers”. In their case, the reported model’s adjusted R² was 0.2.

Most previous models for prediction of arm recovery included assessments performed within the first week post-stroke (Nijland et al., 2010; Persson et al., 2015; Snickars et al., 2017; Ghaziani et al., 2020; Stinear et al., 2012; Kwakkel et al., 2003; Winters et al., 2016).

Specially in developing countries, the public rehabilitation system is under high demand and has limited infrastructure. As recently reported, patients wait for longer periods and the recommended early stroke rehabilitation does not occur (de Athayde Costa et al., 2020) Besides, neurological assessments performed at acute care hospitalization may not be available at admission to specialized rehabilitation centers.

Some previous studies report larger time windows. Plantin et al. (2021) included patients within two to six weeks from onset (mean time since injury to admission was 25±7 days). Feys et al. (2000) included patients within 13 and 38 days after the onset of stroke (22.7±6.6 days). Shelton et al. (2001) reported 17±12 days of an initial, unilateral, hemispheric, ischemic stroke. Time since injury to admission was not reported as significant predictor of arm impairment in any of them.

In our case mean time to admission was 43±19 days, we first found it moderately (negative) correlated to FMA-UE at discharge (Supplementary Table 2) and then we found it as the only confounder common to all 3 adjusted models of mild impairment, all 3 of them indicating that each additional increase of one day in time to admission is associated with a 10% decrease in the odds of achieving mild motor impairment at discharge.

One of the main limitations of this study is the natural correlation between the FMA-UE items at admission and at discharge; also reported in the Scandinavian study. The dichotomization into severe- moderately-severe and mild is based on the execution of activities, not from the ICF- body function domain to which the predictors belong (Ghaziani et al., 2020).

In this study we analyzed data from different settings, involving different therapies, in different contexts. The ICARE investigators reported on 3 different arm therapy interventions (in the United States) meanwhile IGH implements standard treatment (in Europe). The effects of the different applied interventions were not included as confounding factors in the reported results, in order to focus on the item-level predictors and on previously reported confounders. Nevertheless, as an initial step we analyzed the effect of the different settings. In Supplementary Table 3 the data sources (IGH. ICARE) are included as confounder in the model with the highest AUC (using the finger mass flexion item). As shown in the table, the data source variable was found quasi-significant (p = 0.062). Dominance analysis presented in Supplementary Figure 3 shows the data source variable (DATA) to be less dominant than the top 3 presented in Fig. 2 (time since injury and Fugl-Meyer items). Similar results were obtained for the other models presented in Table 6 when including the data source variable. The effect of the different interventions involves several particularities (e.g. intensities, sessions, technical support), leaving room for an specific analysis in future research.

Finally, AUC was calculated by means of 10-fold cross validation (Kuhn et al., 2008), in a test set, we independently partitioned initial data in training set (65%) and test set (35%), nevertheless our results may in turn require an external validation.

5 Conclusions

We showed good predictive power results in the external validation of seven previously reported FMA-UE items, assessed at admission, as predictors of mild motor impairment at discharge considering a wider time window (frequently encountered in several clinical settings) since ischemic stroke onset to rehabilitation admission. We identified an additional item (finger mass flexion) as top item-level predictor of mild impairment at discharge not previously reported in related research. Finally, we significantly associated each delayed day since injury to rehabilitation admission with a specific decrease in the odds of achieving mild motor impairment at discharge.

Footnotes

Acknowledgments

This research was partially funded by EU H2020 PRECISE4Q - Personalized Medicine by Predictive Modeling in Stroke for better Quality of Life (Grant Agreement 777107 – Research and Innovation Action).

Declaration of interest

The authors report no conflict of interest.

Data availability statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

The supplementary files are available from .

References

Arya, K. N., Verma, R. & Garg, R.K. (2011). Estimating the minimal clinically important difference of an upper extremity recovery measure in subacute stroke patients. Top Stroke Rehabil. 18(Suppl 1), 599–610. doi: 10.1310/tsr18s01-599. PMID: 22120029.

Boccuni, L., Meyer, S., D’cruz, N. et al. (2019). Premotor dorsal white matter integrity for the prediction of upper limb motor impairment after stroke. Sci Rep 9, 19712. https://doi.org/10.1038/s41598-019-56334-w.

Brown, A. W., Therneau, T. M., Schultz, B. A., Niewczyk, P. M. & Granger, C. V. (2015). Measure of functional independence dominates discharge outcome prediction after inpatient rehabilitation for stroke. Stroke 46(4), 1038–44. 10.1161/STROKEAHA.114.007392. Epub 2015 Feb 24. PMID: 25712941.

caret: Classification and Regression Training. Misc functions for training and plotting classification and regression models. https://cran.r-project.org/web/packages/caret/index.html (last accessed March, 2023).

Cecchi, F., Carrabba, C., Bertolucci, F., Castagnoli, C., Falsini, C., Gnetti, B., Hochleitner, I., Lucidi, G., Martini, M., Mosca, I. E., Pancani, S., Paperini, A., Verdesca, S., Macchi, C. & Alt Murphy, M. (2021). Transcultural translation and validation of Fugl-Meyer assessment to Italian. Disabil Rehabil 43(25), 3717–3722. doi: 10.1080/09638288.2020.1746844. Epub 2020 May 1. PMID: 32356509.

de Athayde Costa E Silva, A., Viana da Cruz Júnior, A. T., da Silva Souza, G. & Callegari, B. (2020). Positive Balance Recovery in Ischemic Post-Stroke Patients with Delayed Access to Physical Therapy. Biomed Res Int, 2020, 9153174–10.1155/2020/9153174. PMID: 32047819; PMCID: PMC7003254.

Dominance Analysis for multiple regression models: Ordinary Least Squares, Generalized Linear Models, Dynamic Linear Models and Hierarchical Linear Models. https://github.com/cran/dominanceanalysis (last accessed March, 2023).

Du, J., Yang, F., Zhang, Z., Hu, J., Xu, Q., Hu, J., Zeng, F., Lu, G. & Liu, X. (2018). Early functional MRI activation predicts motor outcome after ischemic stroke: a longitudinal, multimodal study. Brain Imaging Behav 12(6), 1804–1813. doi: 10.1007/s11682-018-9851-y. PMID: 29766355.

Duncan Millar, J., van Wijck, F., Pollock, A. & Ali, M. (2019). Outcome measures in post-stroke arm rehabilitation trials: do existing measures capture outcomes that are important to stroke survivors, carers, and clinicians? Clin Rehabil, 33(4), 749. doi: 10.1177/0269215518823248. Epub 2019 Jan 15. PMID: 30646750; PMCID: PMC6429625.

10.

Feys, H., De Weerdt, W., Nuyens, G., van de Winckel, A., Selz, B. & Kiekens, C. (2000). Predicting motor recovery of the upper limb after stroke rehabilitation: value of a clinical examination. Physiother Res Int 5(1), 1–18. doi: 10.1002/pri.180. PMID: 10785907.

11.

FMA-UE PROTOCOL Rehabilitation Medicine, University of Gothenburg [Last accessed 30-04-2022] https://www.gu.se/sites/default/files//FM-UE%eng%%PROTOCOL.pdf.

12.

Fugl-Meyer, A. R., Jääskö, L., Leyman, I., Olsson, S. & Steglind, S. (1975). The post-stroke hemiplegic patient. 1. a method for evaluation of physical performance. Scand J Rehabil Med 7(1), 13–31. PMID: 1135616.

13.

Ghaziani, E., Couppé, C., Siersma, V., Søndergaard, M., Christensen, H. & Magnusson, S.P. (2018). Electrical Somatosensory Stimulation in Early Rehabilitation of Arm Paresis After Stroke: A Randomized Controlled Trial. Neurorehabil Neural Repair 32(10), 899–912. doi: 10.1177/1545968318799496. Epub 2018 Sep 25. PMID: 30251591.

14.

Ghaziani, E., Couppé, C., Siersma, V., Christensen, H., Magnusson, S.P., Sunnerhagen, K.S., Persson, H.C. & Alt Murphy, M. (2020). Easily Conducted Tests During the First Week Post-stroke Can Aid the Prediction of Arm Functioning at 6 Months. Front Neurol 10, 1371. doi: 10.3389/fneur.2019.01371. PMID: 31993016; PMCID: PMC6962352.

15.

Ham, Y. & Shin, J. H. (2022). Efficiency and usability of a modified pegboard incorporating computerized technology for upper limb rehabilitation in patients with stroke. Top Stroke Rehabil 29, 1–9. doi: 10.1080/10749357.2022.2058293. Epub ahead of print. PMID: 35348031.

16.

Harari, Y., O’Brien, M. K., Lieber, R. L. & Jayaraman, A. (2020). Inpatient stroke rehabilitation: prediction of clinical outcomes using a machine-learning approach. J Neuroeng Rehabil, 17(1), 71. doi: 10.1186/s12984-020-00704-3. PMID: 32522242; PMCID: PMC7288489.

17.

Harvey, R. L. (2015). Predictors of functional outcome following stroke. Phys Med Rehabil Clin N Am, 26, 583–98. doi: 10.1016/j.pmr.2015.07.002.

18.

Hiragami, S., Inoue, Y. & Harada, K. (2019). Minimal clinically important difference for the Fugl-Meyer assessment of the upper extremity in convalescent stroke patients with moderate to severe hemiparesis. J Phys Ther Sci 31(11), 917–921. doi: 10.1589/jpts.31.917. Epub 2019 Nov 26. PMID: 31871377; PMCID: PMC6879402.

19.

Hoonhorst, M. H., Nijland, R. H., van den Berg, J. S., Emmelot, C. H., Kollen, B. J. & Kwakkel, G. (2015). How Do Fugl-Meyer Arm Motor Scores Relate to Dexterity According to the Action Research Arm Test at 6 Months Poststroke? Arch Phys Med Rehabil 96(10), 1845–9. doi: 10.1016/j.apmr.2015.06.009. Epub 2015 Jul 2. PMID: 26143054.

20.

Kwah, L. K. & Herbert, R. D. (2016). Prediction of walking and arm recovery after stroke: a critical review. Brain Sci 6, 53. doi: 10.3390/brainsci6040053.

21.

Kwakkel, G., Kollen, B. J., van der Grond, J. & Prevo, A. J. (2003). Probability of regaining dexterity in the flaccid upper limb: impact of severity of paresis and time since onset in acute stroke. Stroke 34, 2181–6. doi: 10.1161/01.STR.0000087172.16305.CD.

22.

Lee, S. Y., Im, S. H., Kim, B. R. & Han, E. Y. (2018). The Effects of a Motorized Aquatic Treadmill Exercise Program on Muscle Strength, Cardiorespiratory Fitness, and Clinical Function in Subacute Stroke Patients: A Randomized Controlled Pilot Trial. Am J Phys Med Rehabil 97(8), 533–540. doi: 10.1097/PHM.0000000000000920. PMID: 29533252.

23.

Lieshout, E. C. C. V., van de Port, I. G., Dijkhuizen, R. M. & Visser-Meily, J. M. A. (2020). Does upper limb strength play a prominent role in health-related quality of life in stroke patients discharged from inpatient rehabilitation? Top Stroke Rehabil 27(7), 525–533. doi: 10.1080/10749357.2020.1738662. Epub 2020 Mar 9. PMID: 32148183.

24.

Liu, Z., Xin, H. & Chopp, M. (2021). Axonal remodeling of the corticospinal tract during neurological recovery after stroke. Neural Regeneration Research 16(5), 939–943. https://doi.org/10.4103/1673-5374.297060.

25.

Nijland, R. H., van Wegen, E. E., Harmeling-van der Wel, B. C., Kwakkel, G. & Investigators, E. (2010). Presence of finger extension and shoulder abduction within 72 hours after stroke predicts functional recovery: early prediction of functional outcome after stroke: the EPOS cohort study. Stroke 41, 745–50. doi: 10.1161/STROKEAHA.109.572065.

26.

Persson, H. C., Murphy, M., Danielsson, A., Lundgren-Nilsson, A. & Sunnerhagen, K. S. (2015). A cohort study investigating a simple, early assessment to predict upper extremity function after stroke – a part of the SALGOT study. BMC Neurol 15, 92. doi: 10.1186/s12883-015-0349-6.

27.

Plantin, J., Verneau, M., Godbolt, A. K., Pennati, G. V., Laurencikas, E., Johansson, B., Krumlinde-Sundholm, L., Baron, J. C., Borg, J. & Lindberg, P. G. (2021). Recovery and Prediction of Bimanual Hand Use After Stroke. Neurology 97(7), e706–e719. doi: 10.1212/WNL.0000000000012366.

28.

Scrutinio, D., Monitillo, V., Guida, P., Nardulli, R., Multari, V., Monitillo, F., Calabrese, G. & Fiore, P. (2015). Functional Gain After Inpatient Stroke Rehabilitation: Correlates and Impact on Long-Term Survival. Stroke 46(10), 2976–80. doi: 10.1161/STROKEAHA.115.010440. Epub 2015 Sep 3. PMID: 26337968.

29.

Scrutinio, D., Lanzillo, B., Guida, P., Formica, R., Russo, G., Guarnaschelli, C., Ferretti, C. & Calabrese, G. (2017). Development and Validation of a Predictive Model for Functional Outcome After Stroke Rehabilitation: The Maugeri Model. Stroke 48(12), 3308–3315. doi: 10.1161/STROKEAHA.117.018058. Epub 2017 Oct 19. PMID: 29051222.

30.

Shelton, F. D., Volpe, B. T. & Reding, M. (2001). Motor impairment as a predictor of functional recovery and guide to rehabilitation treatment after stroke. Neurorehabil Neural Repair 15(3), 229–37. doi: 10.1177/154596830101500311. PMID: 11944745.

31.

Snickars, J., Persson, H. C. & Sunnerhagen, K. S. (2017). Early clinical predictors of motor function in the upper extremity one month post-stroke. J RehabilMed 49, 216–22. doi: 10.2340/16501977-2205.

32.

Stinear, C. M., Barber, P. A., Petoe, M., Anwar, S. & Byblow, W. D. (2012). The PREP algorithm predicts potential for upper limb recovery after stroke. Brain 135(Pt 8), 2527–35. doi: 10.1093/brain/aws146. Epub 2012 Jun 10. PMID: 22689909.

33.

Stinear, C. M., Byblow, W. D., Ackerley, S. J., Smith, M. C., Borges, V. M. & Barber, P.A. (2017). PREP: A biomarker-based algorithm for predicting upper limb function after stroke. Ann Clin Transl Neurol 4(11), 811–820. doi: 10.1002/acn3.488. PMID: 29159193; PMCID: PMC5682112.

34.

Stinear, C. M., Smith, M. C. & Byblow, W. D. (2019). Prediction Tools for Stroke Rehabilitation. Stroke 50(11), 3314–3322. doi: 10.1161/STROKEAHA.119.025696. Epub 2019 Oct 15. PMID: 31610763.

35.

STROBE [Last accessed 30-04-2022] https://strobe-statement.org/index.php?id=available-checklists.

36.

Tauchi, Y., Kyougoku, M., Takahashi, K., Okita, Y. & Takebayashi, T. (2021). Dimensionality and item-difficulty hierarchy of the Fugl-Meyer assessment of the upper extremity among Japanese patients who have experienced stroke. Top Stroke Rehabil 20, 1–9. doi: 10.1080/10749357.2021.1965797. Epub ahead of print. PMID: 34414858.

37.

Wattchow, K. A., McDonnell, M. N. & Hillier, S.L. (2018). Rehabilitation Interventions for Upper Limb Function in the First Four Weeks Following Stroke: A Systematic Review and Meta-Analysis of the Evidence. Arch Phys Med Rehabil, 99(2), 367–382. doi: 10.1016/j.apmr.2017.06.014. Epub 2017 Jul 20. PMID: 28734936.

38.

Winstein, C. J., Wolf, S. L., Dromerick, A. W., Lane, C. J., Nelsen, M. A., Lewthwaite, R., Cen, S. Y. & Azen, S. P. (2016). Interdisciplinary Comprehensive Arm Rehabilitation Evaluation (ICARE) Investigative Team. Effect of a Task-Oriented Rehabilitation Program on Upper Extremity Recovery Following Motor Stroke: The ICARE Randomized Clinical Trial. JAMA 315(6), 571–81. doi: 10.1001/jama.2016.0276. PMID: 26864411; PMCID: PMC4795962.

39.

Winters, C., Kwakkel, G., Nijland, R. & vanWegen, E. (2016). When does return of voluntary finger extension occur post-stroke? a prospective cohort study. PLoS ONE 11, e0160528. doi: 10.1371/journal.pone.0160528.

40.

Woytowicz, E. J., Rietschel, J. C., Goodman, R. N., Conroy, S.S., Sorkin, J. D., Whitall, J. & McCombe Waller, S. (2017). Determining Levels of Upper Extremity Movement Impairment by Applying a Cluster Analysis to the Fugl-Meyer Assessment of the Upper Extremity in Chronic Stroke. Arch Phys Med Rehabil 98(3), 456–462. doi: 10.1016/j.apmr.2016.06.023. Epub 2016 Aug 9. PMID: 27519928; PMCID: PMC5299057.