Early prediction of long-term upper limb spasticity after stroke

Objective: To identify predictors and the optimal time point for the early prediction of the presence and severity of spasticity in the upper limb 12 months poststroke. Methods: In total, 117 patients in the Gothenburg area who had experienced a stroke for the first time and with documented arm paresis day 3 poststroke were consecutively included. Assessments were made at admission and at 3 and 10 days, 4 weeks, and 12 months poststroke. Upper limb spasticity in elbow flexion/extension and wrist flexion/extension was assessed with the modified Ashworth Scale (MAS). Any spasticity was regarded as MAS $1, and severe spasticity was regarded as MAS $2 in any of the muscles. Sensorimotor function, sensation, pain, and joint range ofmotion in the upper limbwere assessedwith the Fugl-Meyer assessment scale, and, together with demographic and diagnostic information, were included in both univariate and multivariate logistic regression analysis models. Seventy-six patients were included in the logistic regression analysis. Results: Sensorimotor function was the most important predictor both for any and severe spasticity 12 months poststroke. In addition, spasticity 4 weeks poststroke was a significant predictor for severe spasticity. The best prediction model for any spasticity was observed 10 days poststroke (85% sensitivity, 90% specificity). The best prediction model for severe spasticity was observed 4 weeks poststroke (91% sensitivity, 92% specificity). Conclusions: Reduced sensorimotor function was the most important predictor both for any and severe spasticity, and spasticity could be predicted with high sensitivity and specificity 10 days poststroke. Neurology® 2015;85:873–880 GLOSSARY ADL5 activities of daily living; ARAT5 Action Research Arm Test; CI5 confidence interval; FMA-UE5 Fugl-Meyer Assessment Upper Extremity Scale; MAS 5 modified Ashworth Scale; NIHSS 5 NIH Stroke Scale; NLR 5 negative likelihood ratio; PLR 5 positive likelihood ratio; ROM 5 range of motion; SALGOT 5 Stroke Arm Longitudinal Study at the University of Gothenburg. Upper limb spasticity has been found to be associated with reduced arm function and low levels of independence, and with a 4-fold increase in direct care costs during the first year poststroke. The prevalence of upper limb spasticity in all patients 12 months poststroke varies from 17% to 38% and was found to be 46% in patients with initial impaired arm function. It has been found that 4%– 13% of patients need treatment for spasticity 6–12 months poststroke. Previous studies during the first 10 days poststroke have identified several predictors for spasticity 3–12 months poststroke, e.g., reduced sensorimotor function and activities of daily living (ADL), muscle weakness, left-sided paresis, and smoking. These studies were relatively small (n5 47), assessed patients several days after stroke onset, or assessed spasticity in both upper and lower limbs. Whether early assessments of upper limb function and impairments can predict the occurrence and degree of upper limb spasticity 12 months poststroke with good accuracy is uncertain. The optimal time for early prediction of upper limb spasticity 12 months poststroke is also unknown. This information is of clinical relevance, as patients with an increased risk of developing spasticity-related impairments, complications, and From the Institute of Neuroscience and Physiology (A.O., A.D., M.A.M., H.C.P., K.S.S.), Rehabilitation Medicine, Sahlgrenska Academy, University of Gothenburg, Sweden; and Sunnaas Rehabilitation Hospital (A.O.), Nesoddtangen, Norway. Go to Neurology.org for full disclosures. Funding information and disclosures deemed relevant by the authors, if any, are provided at the end of the article. The Article Processing Charge was paid by the authors. This is an open access article distributed under the terms of the Creative Commons Attribution License 4.0 (CC BY), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. © 2015 American Academy of Neurology 873 a 2015 American Academy of Neurology. Unauthorized reproduction of this article is prohibited. increased disability may be identified. The study aims were to identify predictor variables and the optimal time for early prediction of any spasticity and severe spasticity in the upper limb 1 year poststroke. METHODS All patients with first-ever stroke in an 18-month period in 2009–2010 who were admitted to the largest of 3 acute stroke units at the Sahlgrenska University Hospital, Gothenburg, Sweden, within 3 days after stroke onset were eligible for consecutive screening for inclusion in the present study, which was a part of the Stroke Arm Longitudinal Study at the University of Gothenburg (SALGOT). In the SALGOT study, the recovery of upper extremity function was investigated in a nonselected sample during the first year poststroke. All included patients had ischemic or hemorrhagic stroke for the first time, were over 18 years old, and had impaired upper extremity function, which was assessed at day 3 with the Action Research Arm Test (ARAT) (0–57) and defined as ,57 points. The study sample size estimation (n 5 88) for SALGOT was to determine a medium change of 6 points (10%) on ARAT, with a power of 0.8 and a significance level of 0.05. With an expected dropout rate of 30%, the aim was to include 120 patients. Standard protocol approvals, registrations, and patient consents. Study approval was provided by the Regional Ethics Committee of the Western region of Sweden (Registration number 225/08), and written informed consent was obtained. The study is registered at www.clinicaltrials.gov (NCT 01115348). Assessment procedure. In SALGOT, the patients were assessed 9 times during the first year: at admission; at 3 and 10 days; at 3, 4, and 6 weeks; and at 3, 6 and 12 months poststroke. In the current study, data from admission, 3 and 10 days, 4 weeks, and 12 months were used. Predominantly, the assessments were carried out by 3 physiotherapists and were performed according to a standardized protocol. A majority of the assessments were performed at the university hospital. If traveling was not possible for the patient, the assessments were conducted in the patient’s home, nursing home, or rehabilitation unit. Variables. Predictor variables collected at admission (day 0). Clinical characteristics and assessments routinely registered at admission during the acute stage of stroke were selected as potential predictor variables (age, sex, ischemic or hemorrhagic stroke, side of stroke, and smoking in the last 3 months). Stroke localization was classified using the Oxfordshire Classification and ischemic stroke was classified after cause of lesion using the Trial of Org 10172 in Acute Stroke Treatment criteria. The initial severity of stroke and arm paresis was assessed with the 0to 42point ordinal NIH Stroke Scale (NIHSS) and the NIHSS arm subscale (0–4), respectively. NIHSS arm was treated as a categorical variable with 0 as the reference category. Predictor variables collected at 3 and 10 days and 4 weeks poststroke. Common clinical assessment scales of sensorimotor impairments assessed at 3 and 10 days and 4 weeks poststroke were selected as potential predictors. Sensorimotor function in the upper limb was assessed with the motor function part (sections A–D) of the Fugl-Meyer Assessment Upper Extremity Scale (FMA-UE). The FMA-UE (sections A–D) includes 33 active motor function tests, where a higher score indicates a better performance (0–66). The nonmotor domains of the same scale (sections H–J) were used to assess sensation (0–12), joint pain (0– 24), and range of motion (ROM) during passive joint motions (0–24); lower scores indicate reduced sensation, more pain, and reduced ROM, respectively. Spasticity in elbow flexors, elbow extensors, wrist flexors, and wrist extensors were assessed with the 6-level modified Ashworth Scale (MAS). The original MAS categories were reordered into integers between 0 and 5, to incorporate the score 11. MAS were dichotomized, and spasticity was considered to be present if the MAS score was $1 in any of these muscle groups. Dependent variables. At 12 months poststroke, spasticity was reassessed in a similar way as previously. Any spasticity was considered if the MAS score was $1, and severe spasticity was considered if the MAS score was $2 in any of the muscle groups. Statistical methods. Continuous and normally distributed variables are presented with means and SDs. Ordinal and nonnormally distributed variables are presented with medians and 1st and 3rd quartile (Q1–Q3). Univariate logistic regression analyses were used to assess the relationship between the potential predictor variables and the outcome variable. The predictor variables were tested for correlations, and when 2 variables had a high correlation (r . 0.8), one was omitted before multivariate logistic regression analysis. The multivariate logistic regression analyses were used to predict the presence of (A) any spasticity and (B) severe spasticity at 12months poststroke. In both (A) and (B), 4 models (A1–A4 and B1–B4) were found and compared (figure 1). The criteria for including a potential predictor variable in the multivariate logistic regression analyses were as follows: (1) significant predictor identified in previous studies, with the condition that if these variables were not found predictive in models A/B1, they were not included in models A/ B2–4, and (2) univariate logistic regression p value ,0.20. Age and sex were included in all models. The variables in the multivariate logistic regression analysis are shown in figure 1. In the multivariate logistic regression analysis, the enter method was used stepwise and nonsignificant variables were removed manually one at the time, to ensure that only significant variables (p, 0.05) were included in the final model. To control for possible nonlinearity between summed ordinal predictors and the dependent variable, a multivariate model with the squared predictor variables (FMA-UE and Sensation) was tested. Nonsignificant associations were interpreted as nonlinearity not shown. The results are presented with unstandardized coefficients, p values, and odds ratios with 95% confidence intervals (95% CIs). The sensitivity (%), specificity (%), positive likelihood ratio (PLR), negative likelihood ratio (NLR), and the corresponding 95% CI for each of the models was calculated using MedCalc for Windows, version 12.7.7.0 (MedCalc Software, Ostend, Belgium). All other statistical calculations were performed using IBM SPSS statistics 21.0 (IBM, Armonk, NY). The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines were followed. RESULTS Participants. In total, 763 patients were admitted to the stroke unit during the inclusion period, 117 of whom were included in the study and assessed at day 3 (SD 1). There were no significant differences between the participating and nonparticipating patients in terms of sex or NIHSS score. The nonparticipating patients were significantly older (76.0 [SD 13.1] vs 69.2 [SD 13.2] years, p , 0.001), and fewer of them had hemorrhagic infarction (6% vs 16% [x 5 4.92, p 5 0.027]). 874 Neurology 85 September 8, 2015 a 2015 American Academy of Neurology. Unauthorized reproduction of this article is prohibited. In total, 76 patients were assessed 12 months poststroke (35% dropout) and included in the logistic regression analysis. The 2 main causes for dropout were death (n 5 14) and study withdrawal (n 5 7). The clinical characteristics of the 76 patients included in the logistic regression analysis are shown in table 1. At 3 and 10 days and 4 weeks, 24%, 43%, and 46% of the 76 patients were assessed with any spasticity, respectively. At 12 months, 46% and 29%were assessed as having any spasticity and severe spasticity, respectively. Prediction of any spasticity 12 months poststroke. The univariate logistic regression analysis for all potential predictor variables for any spasticity and severe spasticity is presented in tables e-1 and e-2 on the Neurology® Web site at Neurology.org. The multivariate logistic regression analysis results are presented in table 2. In models A2 and A3, the FMA-UE was a significant predictor. In model A4, age at stroke onset was a significant predictor in addition to FMA-UE, and higher FMA-UE scores and higher age were associated with reduced probability for spasticity. The sensitivity, specificity, PLR, and NLR of model A3 indicated that this model had the highest predictive value (table 3). The fit of prediction model A3 was assessed in a scatterplot of the predicted probabilities in relation to FMA-UE at day 10 (figure 2). The squared FMA-UE variable (FMA-UE) was not significant in models A2–A4, and nonlinearity could not be shown. Prediction of severe spasticity 12 months poststroke. The results from multivariate logistic regression models B1–B4 are shown in table 2. Sensorimotor function (FMA-UE) was a significant predictor in all models, except B1. Lower FMA-UE scores (models B2–B4) were associated with an increased predicted probability for severe spasticity. In model B2 reduced sensation and in model B4 spasticity at 4 weeks poststroke were additional significant predictors. The sensitivity, specificity, PLR, and NLR of model B4 indicated that this model had the highest predictive value (table 3). The squared variables (FMA-UE and Sensation) were not significant predictors in any of the models B2–B4, thus nonlinearity could not be shown. DISCUSSION The present study demonstrated that any spasticity was best predicted with variables collected at day 10 poststroke and severe spasticity was best predicted with variables collected 4 weeks poststroke. Lower sensorimotor function score, as identified with the FMA-UE, consistently and significantly predicted both any spasticity and severe spasticity at 12 months poststroke. Age at stroke onset was a significant predictor 4 weeks poststroke, with higher age associated with reduced probability for spasticity. The presence of upper limb spasticity 4 weeks poststroke was a significant predictor for severe spasticity. These findings support previous studies reporting paresis and reduced ADL function to be significant predictors for spasticity poststroke. However, none of the previous studies used the FMA-UE to assess sensorimotor function, and thus, a direct comparison cannot be made with those studies. Additionally, none of the former studies compared different prediction models for spasticity at 12 months poststroke based on the assessments at different time points relatively early after stroke onset, as was performed in the present study. Predicting upper limb spasticity with a relatively high accuracy based on early assessments may have high clinical relevance, as patients with an increased risk of developing spasticity-related impairments may be identified early and monitored more closely in order to implement appropriate interventions. In the present study, patients with higher age were predicted to have reduced probability for spasticity. This finding may be in accordance with a previous Figure 1 The multivariate logistic regression analyses models At day 3, Oxfordshire category and spasticity was included only in model B2. FMA-UE 5 Fugl-Meyer Assessment Upper Extremity scale; NIHSS5NIH Stroke Scale; ROM5 range of motion. Neurology 85 September 8, 2015 875 a 2015 American Academy of Neurology. Unauthorized reproduction of this article is prohibited. study where more severe spasticity was found in younger patients 3 months poststroke, but not 18 months poststroke. Muscle force generation from tendon reflexes has been found to be slower and weaker with increasing age and if this also is the case for tonic reflexes associated with spasticity, spastic responses may be weaker in older patients. In the present study, there was no association between the side of weakness and spasticity 1 year after stroke, as described previously. There may be several reasons for this discrepancy; for instance, there were methodologic differences between the studies, as the Tone Assessment Scale was used to assess spasticity, and spasticity and weakness were assessed in both arms and legs as opposed to only upper limb spasticity and sensorimotor function in the present study. An association between smoking and severe spasticity has been found previously, which could not be confirmed here. The authors discussed that their model, which included smoking, had an event per variable of 7, compared to the recommended minimum of 10. Model B1 showed that patients with lacunar stroke had a lower OR for severe spasticity than those with other stroke locations. Therefore, the Oxford Table 1 Demographic and clinical data of the patients included in the logistic regression analysis (n 5 76) Characteristics Values