Descriptive analysis was used for all variables, and results are presented as the means, standard deviations and percentages. Significant differences between fallers and non-fallers were assessed using independent Student’s t-test for continuous variables and χ2 test analysis for categorical variables. The Mann-Whitney U test was
used to detect mean differences between groups when variable distributions were not
normal. Linear correlations between continuous variables were calculated using Pearson’s correlation test. Multivariate logistic regression (MLR) analysis was
performed using a forward stepwise method with an entry criteria of P =0.1 to identify the factors that were independently associated with falls. Two models were developed based on variables with statistical significance from bivariate analysis and clinical interests. Adjusted Odds Ratios (aOR) were acquired from the estimated coefficients and presented with the corresponding 95% confidence interval (CI) of the ratio.
The predictive accuracy of the model in discriminating fallers and non-fallers was assessed using sensitivity and specificity. The optimal cutoff point with the highest sensitivity and specificity for each model was defined as the Youden index [30,31].
A receiver operating characteristics (ROC) curve was plotted to assess the
discrimination of the generated multivariate logistic models. The area under the curve (AUC) of the ROC was also calculated for each model to determine the fitness of individual MLR analysis. An AUC value below 0.5 was considered no discrimination, 0.7≦AUC≦0.8 was considered acceptable discrimination, 0.8≦AUC≦0.9 was considered excellent discrimination, and 0.9≦AUC≦1.0 was considered outstanding discrimination [32]. Commercial statistical software, SPSS version 13.0, was used,
and a two-tailed P <.05 was considered significant.
Results
A total of 140 subjects were enrolled, and 112 subjects (60 men and 52 women) completed the study. The mean age, height, and body weight of the subjects were 69.6
±10.3 years old (range, 45-89 years old), 158.1 ± 6.7 cm (range, 143–175 cm), and 61.2 ± 9.9 kg (range, 41–85 kg), respectively. Approximately half (50.8%) of all the subjects were right hemiplegic patients, and 88.4% of the subjects suffered stroke due to infarction.
Subjects were further divided into non-faller and faller groups depending on whether the subject experienced falls during the follow-up period. A total of 37 patients who experienced falls were classified into the faller group, and 75 subjects were classified into the non-faller group.
No significant differences were found in baseline measurements of age, gender, height, body weight, stroke affected side, stroke type, mental status, ambulation aids, or medications between faller and non-faller groups. However, physical and
psychological assessments revealed that the faller group exhibited higher MAS and GDS and lower FIM and mFES scores compared to the non-faller group (Table 2-1).
These physical and psychological assessments indicated that the faller group exhibited
higher muscle tone, more severe depression, poor overall activity performance of daily life and lower confidence.
Table 2-1 Baseline measurements of the study subjects.
Benzodiazepines (%) 39(0.35) 23(30.7) 16(43.2) 0.189
Hypoglycemic (%) 12(0.11) 8(10.7) 4(10.8) 0.981
Antihypertensives (%) 45(0.40) 26(34.7) 19(51.4) 0.090 Values are % or mean ± SD.
MAS, Modified Ashworth Scale; MMSE, Mini-Mental State Exam; FIM, Functional Independence Measure; mFES, modified Fall Efficacy Scale; GDS, Geriatric
Depression Scale
We used an unbiased quantification using a computerized system to measure the balance and gait abilities in patients post-stroke to provide objective analyses. These computerized measurements were considered to be more objective tools than the traditional assessments [33].
The abilities of balance and gait were different between faller and non-faller groups. Computerized gait assessment revealed that the faller group exhibited slower walking velocity and fewer cadences compared to the non-faller group (P < .001) (Table 2-2). The temporal asymmetry ratios for ASY_ss, ASY_ds, and ASY_step were significantly greater (approximately twofold) in the faller group (P < .05). These results indicated that the faller group exhibited more severe asymmetry gait than the non-faller group.
The faller group exhibited larger COP_area and greater COP_ml in computerized balance assessments (P < .01). These results demonstrated that the faller group exhibited worse postural sway in the mediolateral direction and area compared to the non-faller group. Therefore, the computerized gait and balance assessments may be used to accurately predict fall in the faller group.
Table 2-2 Comparison of balance and gait parameters in study subjects. ASY_ss, asymmetry ratio of single support time; ASY_ds, asymmetry ratio of double support time; ASY_swing, asymmetry ratio of single swing time; ASY_stance, asymmetry ratio of stance time; ASY_step, asymmetry ratio of step time; COP, center of pressure; ml: medial-lateral; ap: anterior-posterior.
Correlation analysis was also performed based on the results in Table 2-2 to
determine the risk factors for predicting fall occurrence. Correlations between gait and balance variables were evaluated (Table 2-3). All parameters of the temporal
asymmetry ratios negatively correlated with walking velocity and cadence. The COP_ml and COP_area exhibited a low-to-medium positive correlation with all parameters of the temporal asymmetry ratios. Therefore, the computer automatically selected ASY_ss and COP_ml to represent the gait and balance assessments,
respectively, for further analysis.
Table 2-3 Correlation coefficients of balance and gait parameters (n = 112).
Variables Cadence Velocity Trajectory of COP
ml ap area
Cadence
(steps/min) 1.00 0.31† −0.34‡ −0.16 −0.32‡
Velocity (m/s) 0.31† 1.00 −0.10 −0.02 −0.08
Asymmetry Ratio
ASY_ss −0.62‡ −0.26† 0.40‡ 0.09 0.28†
ASY_ds −0.50‡ −0.20* 0.23* 0.12 0.23*
ASY_swing −0.54‡ −0.30† 0.49‡ 0.20* 0.48‡
ASY_stance −0.50‡ −0.23* 0.34‡ 0.12 0.34‡
ASY_step −0.61‡ −0.23* 0.32‡ 0.14 0.29†
Trajectory of COP
COP_ml (mm) −0.34‡ −0.10 1.00 0.34‡ 0.82‡
COP_ap (mm) −0.16 −0.02 0.34‡ 1.00 0.70‡
COP_area
(mm2) −0.32‡ −0.08 0.82‡ 0.70‡ 1.00
ASY_ss, asymmetry ratio of single support time; ASY_ds, asymmetry ratio of double support time; ASY_swing, asymmetry ratio of single swing time; ASY_stance, asymmetry ratio of stance time; ASY_step, asymmetry ratio of step time; COP, center of pressure; ml: Medial-Lateral; ap: Anterior-Posterior
*P < .05; † P < .01; ‡ P < .001
Correlations between computerized gait and balance assessments and other physical or psychological assessments were further analyzed. The MAS of the gastrocnemius exhibited a low-to-medium positive correlation with COP_ml, ASY_ss, and GDS (Table 2-4). FIM also exhibited a medium negative correlation with MAS. This correlation analysis demonstrated that FIM negatively correlated with most of the physical and psychological assessments. The strength of the correlation was low-to-moderate between variables (Table 2-3and Table 2-4), but most correlations revealed significant
differences. These results were used as variables for the subsequent MLR analysis.
Table 2-4. Correlation between predictors of risk of falls in stroke subjects (n = 112).
Variables GDS FIM ASY_ss COP_ml MAS_gas
GDS 1.00 −0.48‡ 0.17 0.42‡ 0.39‡
FIM −0.48‡ 1.00 −0.46‡ −0.33‡ −0.34‡
ASY_ss 0.17 −0.46‡ 1.00 0.39‡ 0.20*
COP_ml 0.42‡ −0.33‡ 0.39‡ 1.00 0.26‡
MAS_gas 0.39‡ −0.34‡ 0.20* 0.26‡ 1.00
GDS, Geriatric Depression Score; FIM, Functional Independence Measure; ASY_ss, asymmetry ratio of single support; COP_ml, center of pressure in mediolateral direction;
MAS_gas, Modified Ashworth Score of the gastrocnemius
* P < .05; ‡ P < .001.
The variables in Table 2-3 and Table 2-4 were used for MLR analyses to determine the risk factors for predicting fall in stroke patients. Two models were subsequently generated. Table 5 shows that the significant predictors of fall occurrence (with P<0.05) in stroke patients were as follows in model I of the MLR analysis: (1) GDS (adjusted OR, 1.4; 95% CI, 1.2–1.8; P = .001); (2) gait asymmetry (ASY_ss) [aOR, 2.2; 95% CI, 1.2–3.8; P =.006]; and (3) spasticity of the gastrocnemius (aOR, 3.2; 95% CI, 1.4–7.3; P
=.006). The sensitivity and specificity of this model were 82.6% and 86.5%,
respectively, with a Youden index of 0.69. The model I analysis suggested that GDS, Gait Asymmetry (Single Support), and Spasticity (Gastrocnemius) were strong
predictors for fall in stroke patients.
Notably, the commonly used measurement for regular functional assessment during stays in the rehabilitation unit, FIM,[34] was not automatically selected as one of the predictors after the MLR analysis in model I. This result may be attributed to the results that GDS exhibited the strongest negative correlation with FIM (-0.48) in the correlation analysis between risk factors in stroke subjects (Table 2-4). Therefore, GDS was
excluded in another round of MLR analysis, and prediction model II was generated.
Table 2-5 shows that the predictors of determining fall occurrence in model II included (1) FIM (aOR, 0.9; 95% CI, 0.9-1.0; P = .002), (2) gait asymmetry (ASY_ss) (aOR, 3.6;
95% CI, 1.4-9.2; P =.009), and (3) postural sway (mediolateral, COP_ml) (aOR, 1.7;
95% CI, 1.0-2.7; P = .033). Model II also exhibited relatively high sensitivity (76.9%) and specificity (75.7%) with a Youden index of 0.53, but the sensitivity and specificity were lower than model I.
The ROC curves of the two models (Figure 2-3) for predicting falls in stroke patients were plotted to discriminate the two multivariate logistic models presented in Table 2-5.
The ROC analysis revealed that model I (AUC value: 0.856) was better fitted than model II (AUC value 0.815). However, both models exhibited excellent fitness to
predict fall occurrence in stroke patients with high sensitivity and specificity, with AUC values greater than 0.8[32].
Table 2-5. Multivariate logistic regression for predictors of accidental falls.
Model Factor Coefficient
(ß)
Adjusted odds
ratio (95% CI) P value I
Geriatric Depression Scale 0.361 1.4 (1.2-1.8)a 0.001 Gait Asymmetry (Single Support) 0.783 2.2 (1.2-3.8)b 0.006 Spasticity (Gastrocnemius) 1.164 3.2 (1.4-7.3)a 0.006
Youden Index = 0.69; Sensitivity = 82.6%, Specificity = 86.5%
II
Functional Independence Measure −0.090 0.9 (0.9–1.0)a 0.002 Gait Asymmetry (Single Support) 1.267 3.6 (1.4–9.2)b 0.009 Postural Sway (Mediolateral) 0.518 1.7 (1.0–2.7)a 0.033
Youden Index = 0.53; Sensitivity = 76.9%, Specificity = 75.7%
a: predicted change in odds for a unit increase in corresponding variables
b: predicted change in odds for a standard deviation (SD = 0.3) in corresponding variable
Figure 2-3 The ROC curves for predicting the occurrence of falls in stroke patients using models I and II. AUCs were 0.856 and 0.815, respectively. Arrowheads indicate the identified optimal cutoffs (Youden Index) for these prediction models (0.69 in model I and 0.53 in model II).
Discussion
To our knowledge, this study is the first to include physical and psychological
variables for determining the predictive risk factors of fall in stroke patients. The results underscore the significance of quantitative gait and balance assessments before
discharge from rehabilitation units for predicting fall in stroke subjects by comparing the functional and baseline variables between the faller and non-faller groups of stroke subjects.
The faller group exhibited slower walking speed, asymmetrical gait, unstable balance, and lower functional performance than the non-faller group at baseline. Thirty-seven of the 112 enrolled subjects had at least one falling accident within 6 months after a stroke in this study (33% fall incidence).
Impaired gait symmetry, depression, and higher abnormal muscle tone were found in stroke patients who experienced falls. Prediction models for falls in stroke patients were developed using these physical and psychological parameters. The current findings provide sufficient information for predicting future falls, and early intervention strategies may be implemented to prevent falls in stroke patients.
2.4.1 Assessment of Falls
Previous studies reported that the “gold standard” for collecting information on falls (e.g., prospective collection with calendars or postcards, regular reminders, and follow-up telephone calls) was prone to errors (e.g., memory, forgetting to write diaries and ambiguous definitions of fall)[14]. To minimize these types of errors in this study, falls were recorded regularly by nurses during home visits 4, 12, and 24 weeks post-discharge and by subjects’ self-report. Recordings of fall history, environmental risk exam, and medical consultations were performed during the interviews with each subject. One advantage of the interview was to provide better interaction between subjects and research team workers. Therefore, subjects could fully understand the risk of falls and the ultimate goal of this study to prevent fall occurrence.