Predicting activities after stroke: what is clinically relevant?

Authors

  • G. Kwakkel,

    Corresponding author
    1. Centre of Excellence for Rehabilitation Medicine, Rehabilitation Centre ‘De Hoogstraat’, Utrecht, The Netherlands
    2. Department of Rehabilitation and Sports Medicine, Rudolf Magnus Institute of Neuroscience, UMC, Utrecht, The Netherlands
    • Department of Rehabilitation Medicine, Research Institute MOVE, VU University Medical Center (VUmc), Amsterdam, The Netherlands
    Search for more papers by this author
  • B. J. Kollen

    1. Department of General Practice, University Medical Centre Groningen, University of Groningen, The Netherlands
    Search for more papers by this author

  • Conflict of interest: None declared.

Correspondence: Gert Kwakkel, Department of Rehabilitation Medicine, VU University Medical Centre, de Boelelaan 1117, 1081 HV Amsterdam. The Netherlands.

E-mail: g.kwakkel@vumc.nl

Abstract

Knowledge about factors that determine the final outcome after stroke is important for early stroke management, rehabilitation goals, and discharge planning. This narrative review provides an overview of current knowledge about the prediction of activities after stroke. We reviewed the pattern of stroke recovery for functions and activities, the impact of spontaneous recovery on activities, and the measurement of improvement in general. We explored the activities profiles during the chronic phase and predictors for activities of daily living independence after stroke, and finally, we discussed where to from here? Mathematical regularities explain the nonlinear patterns of recovery, making the outcome of activities of daily living highly predictable. Initial severity of disability and extent of improvement observed within the first weeks poststroke are important indicators of the outcome at six-months. The sequence of progress in activities is almost fixed in time. Studies showed that most motor recovery is almost completed within 10 weeks poststroke. On average, stroke recovery plateaus three- to six-months after onset. Strong evidence was found that age and scores on scales assessing severity of neurological deficits in the early poststroke phase are strongly associated with the final basic activities of daily living outcome after three-months poststroke. The validated prediction models using simple algorithms, such as National Institutes of Health Stroke Scale or Barthel Index, need to be implemented in rehabilitation services and used for stratifying stroke patients in trials. Future studies should investigate the accuracy of dynamic models that includes time poststroke to optimize the application of prediction rules in individuals with stroke.

Introduction

Stroke recovery is heterogeneous in terms of outcome, and it is estimated that 25% to 74% of the 50 million stroke survivors worldwide require some assistance or are fully dependent on caregivers for activities of daily living (ADL) after their stroke [1]. In addition to medical management after acute stroke to prevent further cerebral damage, early stroke rehabilitation is initiated with the ultimate goal of achieving better recovery in terms of body functions and activities in the first months after stroke, and to reduce disability and handicap during the years that follow [2]. Knowledge about factors that determine the final outcome in terms of activities after stroke is important for early stroke management, in order to set suitable rehabilitation goals, enable early discharge planning, and correctly inform patients and relatives. The current trend to shorten the length of stay in hospital stroke units, as well as the increasing demand for efficiency in the continuity of stroke care, imply that knowledge about the prognosis for the outcome in terms of basic activities such as dressing, mobility, and bathing is crucial to optimize stroke management in the first months poststroke. A number of observational studies suggest that the degree of recovery in terms of impairments and activities after stroke is already largely defined within the first days after stroke onset [3-10]. This finding also suggests that the effectiveness of therapy is not only determined by selecting the most effective therapy but also depends on selecting the most appropriate patients for that specific therapy. Moreover, many evidence-based therapies such as constraint-induced movement therapy (CIMT) or modified versions of it, upper limb robotics, functional electrostimulation of the arm, and early supported discharge policies by a stroke team are heavily dependent on an appropriate selection of stroke patients that may benefit most from a particular intervention [11]. Hence, the establishment of an adequate prognosis by a stroke rehabilitation team will increase the efficiency of stroke services and reduce costs. From a patient's perspective, effective prognostics enable health care professionals to respond to changes that occur over time, to estimate the feasibility of the short-term and long-term treatment goals, and to provide correct information to patients and their families [11].

This narrative review provides an overview of current knowledge about recovery and prediction of activities after stroke.

Body functions and activities

The development over time of body functions (i.e. impairments) and activities (i.e. disabilities) after stroke is characterized by clear diversity. Some patients show hardly any improvement even in the long term, whereas other patients fully recover within hours or days post stroke. Even though the outcome of stroke patients is heterogeneous and individual recovery patterns differ, clear mathematical regularities (i.e. logistics and sigmoidal) have been found in these nonlinear patterns of recovery, making the outcome in terms of body functions and activities highly predictable [8, 12-17].

In addition, a number of cohort studies have shown that the initial severity of disability as well as the extent of improvement observed within the first days or weeks poststroke are important indicators of the outcome at six-months after stroke [4, 12-14, 18]. The time course after stroke is characterized by larger improvements during the first weeks poststroke than in the postacute phases beyond three-months after stroke, reflecting common underlying intrinsic mechanisms known as ‘spontaneous neurological recovery’ [11, 18-21].

Another striking feature supporting the existence of a predefined biological pattern in time is the observation that the sequence of progress in activities, as assessed e.g. with the Barthel Index (BI), is almost fixed in time. Hierarchical scaling procedures of the BI show that in about 80% of all patients with a first-ever middle cerebral artery stroke, progress of activities follows the same sequence of BI items [22, 23].

Skills that allow the use of compensation strategies, such as grooming, recover earlier than more complex skills such as dressing and climbing stairs. The fact that the recovery of activities after stroke follows a fixed hierarchy is not limited to ADL outcomes measured with instruments like BI or Functional Independence Measure (FIM) [24], but has also been found with the Stroke Impact Scale [25] and the National Institutes of Health Stroke Scale (NIHSS) in acute stroke [26], as well as for the recovery of the upper limb function measured with the ABILHAND questionnaire [27] or the Action Research Arm Test (ARAT) [28]. These findings support the notion that defining milestones may serve as an important part of multidisciplinary stroke management [29-31] in order to define realistically attainable treatment goals.

Spontaneous neurological recovery

Longitudinal regression modeling of change scores has shown that most motor recovery is almost completed within four- to 10 weeks poststroke [16]. This finding is in agreement with the patterns observed in a number of prospective cohort studies [3-5, 12, 14]. What all these studies and others have in common, regardless of the measures they used, is the observation that the greatest gain occurs within the first three-months after stroke. The fact that the greatest motor recovery occurs within a limited time window after injury is entirely congruent with observations in animal models, in which converging data at the molecular, cellular, physiological, and behavioral levels suggest a limited time window of heightened plasticity and increased receptivity to training regimens [32]. However, stroke models in animals such as rodents are limited by the fact that most studies are conducted over the short term and do not measure long-term recovery, rodents tend to recover completely within six- to eight-weeks of even devastating stroke, and a day in the life of a rodent is not likely to match a day in the life of a human. These issues limit what we can currently take away from animal stroke models.

At the body function level, intrinsic, spontaneous neurological recovery can be defined as the degree of neurological improvement of body functions, such as synergism, attention, strength that is generated by the progress of time alone [16]. These spontaneous nonlinear changes in impairments within the first four- to eight-weeks poststroke result in activities showing concomitant instantaneous improvements as well. This definition does not preclude the possibility that heightened homeostatic plasticity in the spontaneous recovery time window could also allow for faster learning of compensatory strategies while performing tasks that is less dependent on recovery from impairment alone [32]. In the absence of observational studies without intervention, one could argue that a certain degree of recovery is due to therapeutic intervention. Although this cannot be ruled out totally, the consistency of recovery patterns over all kind of therapeutic interventions suggests that the type of intervention does not affect the limited time window of 10 weeks within which recovery occurs. Furthermore, multilevel regression modeling of longitudinal data in stroke survivors has learned that time itself is an independent factor for progress of recovery [33]. Exploration of differences in variances within and between groups over time could provide additional support for this assumption.

Measurement of spontaneous neurological recovery

Unfortunately, the underlying mechanisms responsible for these spontaneous, time-dependent changes early poststroke are still poorly understood. With that, there is no uniform definition of this intrinsic, spontaneous neurological recovery and no method to measure its contribution to the overall pattern of recovery directly. In addition, ordinal-scaled measures are implemented to provide estimates of perceived improvements in clinical functions and activities. However, differences in item-responsiveness and ceiling effects of used outcomes may be responsible for plateauing recovery pattern.

In the present review, we will define spontaneous neurological recovery as the amount of improvement in terms of body functions and activities that is determined by the progress of time alone. Using this concept, there are two ways of demonstrating the contribution of progress of time to the nonlinear recovery pattern after stroke. The first involves applying an individual curve fitting analysis, while the second involves uses random coefficient analysis of change scores [33].

Another way of investigating the impact of spontaneous neurological recovery on the observed time-dependent change in functions and activities is by applying longitudinal regression analysis to observed change scores [34-36]. For this purpose, recently, a new longitudinal first-order regression model for stroke rehabilitation was introduced [34-36]. This model, based on within-subject change scores, was used to investigate the contribution of the progress of time to the recovery of body functions and activities poststroke.

For example, progress of time, corrected for age, gender, type, and hemisphere of stroke and type of intervention accounted for eight-points (or 40%) of change on the total BI score. In other words, adding eight-points to the initial BI score measured at the fifth to seventh days poststroke will produce a valid conservative estimate of the final BI at six-months poststroke [9, 16]. This finding further confirms the close relationship between the initial deficit in terms of activities and the final outcome at six-months poststroke. The same 40% gain was found by Ng et al. [37] between admission FIM and FIM at discharge in 2213 subjects. This 40% gain was irrespective of vascular territory (anterior, middle or posterior, cerebellar, or brainstem), age, risk factors for stroke or hemisphere [37]. Patients with multiple strokes in more than one vascular territory did not follow this rule and showed significant less improvement in the motor and cognitive FIM scores [37].

Activities during the chronic phase

Although long-term prospective cohort studies after stroke are scarce [38], it is generally accepted that, on average, patients seem to ‘plateau’ in terms of recovery three- to six-months after onset. The Copenhagen Stroke Study also showed that recovery of activities in terms of the BI was completed in 12·5 weeks from stroke onset in 95% of the patients (n = 1197) [4]. Although the time course of body functions followed a pattern similar to that of activities, body functions preceded the pattern of recovery of activities by an average of two-weeks and reached a plateau phase sooner. This cohort study suggests that a reliable prognosis can be made within the first three-months poststroke and even in patients with severe strokes. In contrast, significant recovery of body functions and activities should not be expected after the first six-months after stroke onset at group level, however, prospective studies beyond six-months suggest that about 5% to 10% of all patients will show further improvement in upper or lower limb function and ADL, whereas 15% to 25% shows a significant decline at activity level [39].

Can we predict individual activities after stroke?

At best, prognostic models only provide valid estimates of the risk for patients with characteristics that are similar to those in the study population. A slightly different sample is generated each time the population is resampled because data sets are based on samplings from a population, and not on the population itself. A predictive model should be able to determine the future outcome regarding a particular measure for a single patient within an acceptable margin of error. However, accurate prognostic models providing 100% certainty about the outcome or future of an individual stroke patient are not yet available in rehabilitation medicine. A clinical prognosis for an individual stroke patient is based on the examination of so-called ‘prognostic factors’ (‘markers’ or ‘predictors’) that have been found to be associated with the final outcome in a representative sample of stroke patients. In most cases, these prognostic markers (i.e. predictors or determinants) are based on multivariate (or multivariable) regression models. However, the determinants derived from these regression or association models will never be 100% predictive. As a consequence, these determinants should be used as markers in estimating a patient's future at a certain time poststroke. In other words, using these determinants or predictors for individual stroke patients should always be done with a degree of skepticism, keeping in mind that exceptions to the prediction rules exist.

Prediction of ADL independence

A systematic review of 48 studies that aimed to predict ADL outcome showed that the BI and modified Rankin Scale (mRS) were the two activity level outcome measures most frequently used in prognostic stroke studies. Despite the fact that only a small proportion of the included studies was of high quality, i.e. six out of 48 (12·5%) [40], strong evidence was found that age and scores on scales assessing severity of neurological deficits in the early poststroke phase, such as the NIHSS and Canadian Neurological Scale (CNS), are strongly associated with the final basic ADL outcome beyond three-months poststroke [41]. The same finding was shown in a recent prospective study conducted in six-countries in Europe (n = 2033). Accurate prediction with high precision of independent survival at three-months was obtained by the initial BI plus age and the NIHSS for all centers [42].

The systematic review of 48 prognostic studies of Veerbeek et al. ( Table 1, [40]) also showed that gender and the presence of risk factors for stroke, such as atrial fibrillation, did not significantly predict the outcome in terms of basic ADL. Conspicuously, imaging data for the prediction of ADL outcome proved to be of limited value when compared with the contribution of clinical variables alone [40].

Table 1. Quality assessment of reports of prognostic studies [40]
Outcome strategiesScaleCriteria
  1. Y, positive, 1 point; N, negative, 0 points; ?, partial/unknown; SD, standard deviation; CI, confidence interval; IQR, interquartile range; SE, standard error of the mean; OR, odds ratio; RR, Relative Risk; HR, Hazard Ratio; ROC, receiver operating characteristic; HL, Hosmer-Lemeshow Statistic; PPV, Positive Predictive Value; NPV, Negative Predictive Value.
Evaluation of Study design
D1Source population and recruitmentY/N/?Positive when sampling frame (e.g. hospital based, community based, primary care) and recruitment procedure (place and time period, method used to identify sample) are reported.
D2Inclusion and exclusion criteriaY/?Positive if both the inclusion and exclusion criteria are explicitly described.
D3Important baseline key characteristics of study sampleY/?Positive if the following key characteristics of the sample are described: gender, age, type, localization, number of strokes, and stroke severity. Number of strokes is adequate when at least ‘a history of stroke’ or ‘recurrent stroke’ is reported.
D4Prospective designY/N/?Positive when a prospective design was used, or in case of a historical cohort in which prognostic factors were measured, before the outcome was determined.
D5Inception cohortY/N/?Positive if observation started at a uniform time point within two-weeks after stroke onset.
D6Information about treatmentY/N/?Positive if information on treatment during observation period is reported (e.g. medical or paramedical, usual care, randomized, etc.)
Study attrition   
A1Loss to follow-upY/N/?Positive if loss to follow-up during period of observation did not exceed 20%.
A2Reasons for loss to follow-upY/N/?Positive if reasons for loss to follow-up are specified, or there was no loss to follow-up.
A3Methods to deal with missing dataY/N/?Positive if adequate method of dealing with missing values was used in case of missing values (e.g. multiple imputation), or there were no missing values.
A4Comparison of completers and noncompletersY/N/?Positive if article mentions that there are no significant differences between participants who completed the study and those who did not, concerning key characteristics of gender, age, type, and severity and candidate predictors and outcome, or if there was no loss to follow-up.
Predictor measurement   
P1Definition of predictorsY/?Positive if the article clearly defines or describes all candidate predictors (concerning both clinical and demographic features).
P2Measurement of predictors reliable and validY/N/?Positive if ≥1 candidate predictor was measured in a valid and reliable way, or referral is made to other studies which have established reliability and validity.
P3Coding scheme and cutoff pointsY/N/?Positive if coding scheme for candidate predictors was defined, including cutoff points and rationale for cutoff points, or if there was no dichotomization or classification.
P4Data presentationY/N/?Positive if frequencies or percentages or mean (SD/CI), or median (IQR) are reported for all candidate predictors.
Outcome measurement   
O1Outcome(s) definedY/N/?Positive when a clear definition of the outcome(s) of interest is presented.
O2Measurement of outcome(s) reliable and validY/N/?Positive when outcome was measured in a valid and reliable way, or reference is made to other studies which have established reliability and validity.
O3Coding scheme and cutoff points describedY/N/?Positive if the coding scheme of the outcome is given, including cutoff points and rationale for cutoff points, or if there was no dichotomization.
O4Appropriate end-points of observationY/N/?Positive if observation was obtained at a fixed moment after stroke onset; negative if observation was made at discharge.
O5Data presentationY/N/?Positive if frequencies or percentages or mean (SD/CI) or median (IQR) are reported for the outcome measure.
Statistical analysis   
S1Strategy for model building describedY/N/?Positive if the method of the selection process for multivariable analysis is presented (e.g. forward, backward selection, including P-value).
S2Sufficient sample sizeY/N/?Positive if the number of patients with a positive or negative outcome (event) per variable in the logistic regression analysis was adequate, i.e. equal to or exceeding 10 events for each variable in the multivariable model (events per variable), or in case of linear regression analysis n ≥ 100.
S3Presentation of univariate analysisY/N/?Positive if univariate crude estimates and confidence intervals (β/SE, OR/CI, RR, HR) are reported. Negative when only P-values or correlation coefficients are given, or if no tests were performed at all.
S4Presentation of multivariable analysisY/N/?Positive if point estimates with confidence intervals (β/SE, OR/CI, RR, HR) are reported for the multivariable models.
S5Continuous predictorsY/N/?Positive if continuous predictors were not dichotomized in the multivariable model.
Clinical performance/validity   
C1Clinical performanceY/?Positive if article provides information concerning at least one of the following performance measures: discrimination (e.g. ROC), calibration (e.g. HL statistic), explained variance, clinical value (e.g. sensitivity, specificity, PPV, NPV)
C2Internal validationY/?Positive if appropriate techniques were used to assess internal validity (e.g. cross-validation, bootstrapping); negative if split-sample method was used.
C3External validationY/?Positive if the prediction model was validated in a second independent group of stroke patients.

The less than optimal prediction of BI at six-months for patients assessed within 72 h in our study may have been caused by the instability of neurological deficits, which is manifested by the neurological worsening observed in approximately 25% of all patients during the first 24–48 h after stroke [43]. (Fig. 1) However, a parallel study focusing on the timing of an assessment of neurological deficits by the NIHSS in the same population resulted in no significant differences between days 2, 5, and 9 [9], which makes neurological worsening within this period unlikely. (Fig. 2) A more plausible explanation could be that observers find it difficult to determine the patient's actual performance in terms of basic ADLs when the patient is still bedridden. As a consequence, an assessment within 72 h poststroke will underestimate their actual performance. In line with the recommendation by Kasner [44], our findings suggest that even in individuals with a minor stroke who are bedridden during the first few days after stroke, the BI will underestimate outcome scores, making the BI an unsuitable instrument to measure disability within the first three-days poststroke.

Figure 1.

Graphic presentation of receiver operating characteristic (ROC) analyses of the timing of the assessment of Barthel Index (BI) on days 2, 5, and 9 for the outcome in terms of dichotomized BI scores (≥19) after six-months (n = 206).

Figure 2.

Graphic presentation of receiver operating characteristic analyses of the moment of timing of the assessment of National Institutes of Health Stroke Scale scores for the outcome of BI (≥19) at six-months after stroke.

Other determinants reported in valid prospective cohort studies suggest that not only baseline ADL factors such as sitting balance but also urinary incontinence, severity of hemiplegia, comorbidity, consciousness at admission, cognitive status, and depression are independent factors that contribute to the outcome in terms of ADL beyond six-months [42, 45-47]. Interestingly, several studies showed that these simple models did not improve by the use of CT-derived radiological variables [48] or magnetic resonance imaging [49, 50], whereas a recent study suggested that larger number of predictors including scales and comorbidities are better than simpler models [51].

Regaining walking ability

Regaining independent gait is considered a primary goal in stroke rehabilitation. A number of prospective cohort studies have shown that approximately 60% [4, 33, 52] to 80% [52] of stroke patients are able to walk independently at six-months poststroke. Various prognostic studies suggest that age [53, 54], severity of sensory and motor dysfunction of the paretic leg [54, 55], homonymous hemianopia [54, 55], urinary incontinence [5, 53], sitting balance [5, 7, 56-58], initial disability in ADL and ambulation [4, 5, 7], level of consciousness on admission [53], and the number of days between stroke onset and first assessment [10] are independently associated with gait outcomes six-months after stroke [59]. The increasing accuracy of prediction over time may reflect underlying intrinsic neurological mechanisms of recovery such as elevation of diaschisis after stroke [16, 60]. Comparing these findings with those of other studies is difficult due to the lack of prognostic studies investigating the accuracy of prediction within 72 h. However, a number of prospective studies have shown that muscle strength of the hemiplegic leg [54, 55] and sitting balance [5, 56], when measured in the second to fourth weeks after stroke, are significantly associated with improvement of walking ability [7] and achieving independent gait [7, 57] at six-months. Obviously, the early control of sitting balance as a prerequisite for regaining standing balance and gait is an important factor for the final outcome at six-months [57]. The importance of balance control for gait is also supported by the study of Kollen et al. [33], who showed that improvement in standing balance was the most important variable associated with improvement of gait performance as measured with the Functional Ambulation Categories (FAC) [33].

Because the proportion of false positives (≈7%) was clearly smaller than the proportion of false negatives (≈27%) within two-days poststroke, our study suggests that our model is generally somewhat pessimistic, and illustrates that some patients with an initially poor sitting balance and a severe paresis of the hemiplegic limb will nevertheless regain independent gait [7]. This finding is supported by a number of recent longitudinal studies showing that gait recovery is closely related to learning to use compensatory movement strategies [61-63]. Obviously, the aforementioned adaptation strategies already start as soon as patients learn to accomplish tasks within the first weeks poststroke.

Regaining dexterity

Although prospective epidemiological studies are lacking, findings of a number of prospective cohort studies suggest that 33% to 66% of stroke patients with a paretic upper limb do not show any recovery in upper limb function six-months after stroke [64, 65]. Depending on the outcome measures used, 5% to 20% achieve full recovery of the upper paretic limb in terms of activities at six-months [6, 64, 65].

In order to better understand the functional prognosis of the upper limb, we tested the probability of impairment being reduced and dexterity being regained at six-months using logistic regression analysis in patients who had an almost flaccid upper limb in the first week poststroke and no dexterity as assessed the Fugl-Meyer (FM) arm score, as shown in Fig. 3 [6]. We found that only those patients with some early reduction of impairment in the upper paretic limb had greater gains later: patients showing some (synergistic) movement in the upper limb within four-weeks poststroke had a 94% chance of improving their ARAT score, whereas this probability remained below 10% in those who failed to show any return of motor control.

Figure 3.

Probability of achieving some dexterity at 6 months post stroke. Within the first 3 to 4 weeks, a critical time window was present in which the outcome in terms of dexterity (dichotomized into ARAT < 10 points or ARAT ≥ 10) was determined. Optimal prediction was based on the Fugl-Meyer scores of the paretic arm and the motricity index score of the leg (MI-leg) [6].

This study, with repeated measurements over time, suggests that there is a critical time window in which the final outcome in terms of dexterity is still not entirely determined. In fact, it is the same limited time window that has been found in animal studies for an upregulation of growth-promoting factors, resulting in synapse strengthening and activity-dependent rewiring of neuronal networks to compensate for tissue lost to injury [32].

Obviously, the reservation of some voluntary finger extension reflects the importance of the presence of some intact fibers of the corticospinal tract system in the affected hemisphere that can activate distal arm and hand muscles [66, 67], assuming that the forearm and hand lack bilateral innervation from both hemispheres [68]. In the same vein, Coupar et al. [69] showed recently, after reviewing 58 prognostic studies of upper limb recovery, that the initial severity of motor impairment or function was found to be the most important predictor for upper limb recovery after stroke.

Knowledge about the early prediction of final functional outcome for the upper limb function is paramount for the implementation of effective stroke management. In particular, subsequent multidisciplinary rehabilitation services may be optimized based on the probability of regaining function, in view of the fact that many evidence-based therapies for the upper paretic limb, including CIMT, require some return of voluntary wrist and finger extension [70, 71]. This finding also suggests that evidence-based practice is not only a matter of applying the most effective therapy for a particular patient but is also about selecting the appropriate patients to be offered this specific therapy.

Summary

Findings of prognostic research in the field of stroke rehabilitation might improve early stroke management decisions like discharge and multidisciplinary intervention planning at acute and subacute stroke units. This in turn may allow multidisciplinary rehabilitation services to be optimized based on the probability of reducing impairment or regaining ADLs, walking ability, or upper limb function. The validated prediction models using simple algorithms often based on existing outcome scales, such as NIHSS or BI, need to be implemented in rehabilitation services and used for stratifying stroke patients in trials. The sequence of progress in activities is almost fixed in time, whereas longitudinal studies with repeated measurements over time showed that most motor recovery is almost completed within 10 weeks poststroke. On average, stroke recovery plateaus three- to six-months after onset. Strong evidence was found that age and scores on scales assessing severity of neurological deficits in the early poststroke phase are strongly associated with the final basic ADL outcome after three-months poststroke. Sitting and standing balances as well as muscle strength of the lower paretic limb are found to be strong predictors for walking ability when assessed during the first days poststroke, whereas voluntary finger extension is the most powerful indicator for outcome of dexterity at six-months poststroke.

It should be noted, however, that the present review has some limitations. First, the present review is not a systematic, but a narrative review of published work in this field. With that, the choice of selected clinical factors is somewhat biased as these factors are mainly selected on their predictive validity as well as their simplicity for clinical use in hospital stroke units. However, these prognostic tools are currently recommended for the management of acute stroke victims in the Dutch guidelines for physiotherapy. Furthermore, most prognostic models published so far did not involve non-stroke-related factors that may affect outcomes, such as financial and social status and premorbid functional levels. Finally, probably prognostic factors that are harder to measure, such as motivation, are not investigated yet and reported in the literature [72].

Future studies should meet the key methodological criteria for valid prognostic research and should externally validate their prediction models before implementation [73] In addition, time-dynamic models are needed to optimize the precision of prediction rules in individuals irrespective of the timing of assessment poststroke. Neuroimaging and neurophysiological techniques such as repetitive transcranial magnetic stimulation (rTMS), diffusion tensor imaging (DTI) may be used to optimize clinical models by taking the reversibility of neurological damage and functional reorganization into account [74]. Finally, to develop prediction models in patient subgroups and to improve the precision of validated models, large cohorts are needed and the development of a world wide network.

Ancillary