Background and Purpose Falls are a major complication in inpatient stroke rehabilitation. An important issue in preventive strategies is the early identification of those at risk. This study aimed at assessing the fall-prediction accuracy of an easily administered fall risk index in stroke rehabilitation.
Methods A consecutive series of 135 patients admitted to a geriatric stroke rehabilitation unit was studied. A score on the Downton fall risk index was obtained from the admission assessment data and used as a predictive indicator of the risk of falls. The patients' falls were prospectively recorded during their rehabilitation stay. The correlation between falls and the predicted risk was assessed by means of survival analysis and a multiple regression analysis, adjusting for the time of observation.
Results The risk of falls as a function of the time observed was significantly greater among those predicted to be at high risk (index score ≥3) than among the others (P=.009, log-rank test; odds ratio, 2.9). Furthermore, the number of falls during rehabilitation stay was moderately correlated (R=.57) with the fall risk index sum when adjusted as for the time of observation. The sensitivity of the fall prediction as to outcome was 91%, whereas the specificity was limited to 27%.
Conclusions A moderately high correlation was found between the predicted and the observed risk of falls in stroke rehabilitation when the Downton fall risk index was used. However, a low specificity rate limits the accuracy of the prediction.
Falls are one of the most frequent complications among stroke patients in rehabilitation.1 In a geriatric stroke inpatient rehabilitation unit, 39% of the patients fell, and 17% suffered injury during their stay.2 It therefore seems urgent to find ways of preventing falls, or rehabilitation objectives will be seriously jeopardized among individuals prone to repeated falls.
Identifying high-risk individuals is a major task of prevention programs, and fall risk factors in different populations of stroke patients have already been studied to some extent.3 4 5 However, one assumption is that the cumulative effect of multiple risk factors, concurrently present and compounding each's contribution to the risk of falls, would contribute more to the tendency of falls than would the potential effect of each factor alone.6 On the basis of this assumption, a number of fall risk scoring systems have been presented.5 7 8 These systems screen for well-established risk factors. The present risk factors are then added together by means of different algorithms, and a score taken to correlate positively with the fall risk is obtained. The Tinetti fall risk index7 seems too complex to be convenient in clinical practice, although its accuracy concerning fall prediction has been well documented. Both the Downton fall risk index8 and the Fall Assessment Questionnaire5 offer satisfactory content validity as they are screening for well-documented risk factors.6 8 9 They also appear to be very easily administered in clinical practice. The Fall Assessment Questionnaire has shown a moderately high correlation with falls among male right-hemisphere stroke patients in rehabilitation,5 but the Downton index has not been tested on an independent patient population or among stroke patients.
The purpose of our study, therefore, was to investigate the fall prediction accuracy of the easily administered Downton fall risk index in a stroke patient population undergoing inpatient rehabilitation.
Subjects and Methods
The study was performed at the stroke rehabilitation unit of the geriatric clinic at Umeå University Hospital, Sweden, which is a 24-bed unit specializing in stroke rehabilitation. Patients admitted to this unit are the most severely affected third of the survivors of those admitted to the hospital for acute stroke in the catchment area of Umeå University Hospital. They all need further rehabilitation and hospital stay after the acute phase. The patients are admitted without selection from acute care clinics, usually 2 to 4 weeks after their strokes.
All patients (n=142) admitted for rehabilitation after cerebrovascular accidents or other clinically similar conditions from November 1, 1991, through October 31, 1992, were included in the study. Seven patients, who were completely immobile and bedridden throughout their entire stay and who made no locomotion efforts, were subsequently excluded because they were judged to be at no risk of falling, nor was rehabilitation possible. Thus, 135 patients remained in the study; their basic characteristics are summarized in Table 1⇓. Performance of activities of daily living was assessed according to the Katz index,10 and cognitive state was examined with the Mini-Mental State Examination (MMSE).11 The diagnoses were based on clinical examination and CT results in accordance with the routines of the stroke unit of the Umeå University Hospital, the criteria of which have been published previously.12 Clinical observations of signs of dyspractic behavior, visuospatial hemineglect, and aphasia were made and then rated in team consensus after admission assessments.
The patients were studied from admission to the rehabilitation unit until discharge or death. They were thus studied for a median time of 48 days, ranging from 3 to 289 days with an interquartile range of 22 to 76 days. Falls were defined as incidents in which the subject unintentionally came to rest on a level below knee height. All such incidents that took place during the hospital stay and that came to the knowledge of the nursing staff were reported on special fall report forms. The incidence rate, characteristics, and injury consequences of these falls have previously been presented in detail.2
In a previous study of falls among stroke patients in geriatric rehabilitation,2 we collected data on the falls and characteristics of the patients on admission to the rehabilitation unit. From these data, it was possible to obtain a slightly modified Downton index score representing the patients' state on admission. Index items and their prevalence among the patients who fell during their rehabilitation stay and those who did not are presented in Table 2⇓. The history of previous falls during the preceding year was obtained from medical records, from the subjects themselves during admission interviews, or from family members or caregivers. Medication data were obtained from medical records. Visual and hearing impairments were noted if the subject was not able to read a short text in 10-mm block letters at reading distance or perceive a conversation in a normal voice at a distance of 1 m. Limb impairment was assessed by a physiotherapist and defined as the presence of amputated limbs or signs of extremity pareses or sensory impairments by use of the Brunnström-Fugl-Meyer assessment scale.13 Subjects were categorized as cognitively impaired (ie, confused as labeled by the index) if they scored ≤23 points in the MMSE. Regarding gait, subjects were classified as unable if, in the rehabilitation team documentation, they were described as unable to walk, even with physical assistance. If the team expressed that physical assistance or supervision was required, subjects were classified as unsafe with/without aids and as safe/normal (ie, safe with or without aids) if no such precautions were judged necessary.
In three cases, it was not possible to obtain an index sum, since data were missing for one or two items. However, for two of these individuals, it was still possible to determine whether they would have scored ≥3 points, ie, the cutoff score proposed to indicate a high fall risk. The mean±SD sum of the fall risk index score was 3.8±1.5 (range, 0 to 9; n=132), and 107 of 134 patients (80%) scored ≥3 points.
Taking into account the wide range of the patients' length of stay, the Kaplan-Meier method of survival estimation was used to describe the risk of falling during the rehabilitation stay as a function of time.14 In accordance with this method, the time from admission to the first fall (if any) was calculated for each case, and cases were censored either at the time of discharge or death. The log-rank test was used to compare the Kaplan-Meier distributions of those scoring ≥3 on the fall risk index and those scoring less.14 Furthermore, the difference in index score sums, and in the prevalence of separate fall risk factors, between fallers and nonfallers was assessed using the Mann-Whitney U test and the χ2 test, respectively. Finally, a multiple linear regression analysis was performed to assess the correlation of the number of falls during rehabilitation (dependent variable) with the fall risk index sum, adjusted for the length of the observation time (independent variables). The SYSTAT software package was used for computerized statistical analyses.15
A total of 142 falls occurred during the patients' rehabilitation stay, which corresponds to a rate of 178 falls per 10 000 patient days, or 6.5 falls per patient and year. As for the individual patients, 53 of 135 (39%) fell at least once, and 37 (27%) fell twice or more. The number of falls among fallers ranged from 1 to 12. Of the 135 patients, 23 (17%) were injured in connection with a fall, and 4 of these sustained serious injury (fractures to the radius, pelvis, and proximal femur and a fatal intracranial hemorrhage).
The risk of falling as a function of the time of observation is shown in Fig 1⇓. At 8 weeks of hospitalization, almost half (46%) of the patients would have fallen (95% confidence interval, 34% to 57%).
The median fall risk index sum (interquartile range) was 4 (3 to 5) for fallers and 3 (2 to 4) for nonfallers (P=.016; Mann-Whitney U test). In Fig 2⇓, the different Kaplan-Meier distributions of patients scoring ≥3 and <3 on the index are shown. The difference was statistically significant (P=.009, log-rank test), and the odds ratio for falls during rehabilitation stay among patients scoring ≥3 on the index to those scoring less was 2.9. However, 5 of the 53 fallers were predicted not to be at high risk (index sum <3), which limits the sensitivity of the prediction to 91% (48/53). Likewise, only 22 of 82 nonfallers were correctly predicted, corresponding to a specificity rate of 27%. The overall accuracy of the prediction was 52% (70/134).
In Table 2⇑, the scores of individual index items among fallers and nonfallers are shown. Regarding many of the items, both groups scored more or less the same. More substantial differences were found in the use of antidepressants and in the prevalence of impairments of limbs, mental state, and gait. Other variables, not included in the Downton index yet of potential interest regarding fall prediction in stroke patients, were also analyzed. Significantly more fallers than nonfallers presented signs of visuospatial hemineglect (64% versus 36%; χ2, 10.3; P=.001) and of dyspraxia (49% versus 25%; χ2, 8.0; P=.005). Regarding the localization of brain lesions (right or left hemisphere) and the prevalence of dysphasia, no significant differences were found between fallers and nonfallers.
The fall risk index sum was moderately associated (R=.57) with the number of falls during rehabilitation stay when the regression model had been adjusted for the length of the observation period (Table 3⇓).
The Downton fall risk index showed a moderate association between the predicted risk and the observed outcome regarding falls in stroke patients undergoing inpatient rehabilitation. From a preventive point of view, the identification of patients prone to falls, as well as the identification of individual risk factors that would be the focus of targeted intervention measures, is of great importance. In a study of a community population of the elderly, Tinetti et al16 showed that a targeted strategy, screening for specific risk factors subsequently addressed by programs aimed at either modification or compensation, significantly reduced the fall risk.
The sensitivity of the Downton index prediction compares with that of the Fall Assessment Questionnaire, supplemented with a measure for behavioral impulsivity, as demonstrated in a population of male right-hemisphere stroke patients in an inpatient rehabilitation setting.5 The specificity rate, however, was much lower, which explains the low overall accuracy of prediction in our study. The limited specificity is illustrated by the fact that no less than 80% of the subjects were classified as liable to fall, whereas 39% were observed to fall during their rehabilitation stay. When a higher cutoff level (≥4 points) was used, the overall accuracy of the prediction slightly improved, but the sensitivity decreased to 70%, which is not acceptable from a preventive point of view. On the other hand, the great variation in observation time (length of stay ranging from 2 to 289 days, and a quarter of the patients staying ≤3 weeks) is a limitation to our study and must be seen as a potential confounder in the analyses of predictive accuracy with regard to sensitivity and especially specificity rates.
The above objections notwithstanding, it is plausible that adjustments to the Downton index would increase its accuracy in fall risk prediction among stroke patients in rehabilitation. This would also be expected because the index was apparently intended to be used on a more general elderly population and not specifically on stroke patients. Many of the index items, including those concerning previous falls and medication, showed no or only minor differences between the fallers and nonfallers in our study population. In some cases, the differences were even opposite to the expected result. Probably some of these index items are therefore less appropriate in a stroke population. It is likely that the item of previous falls, for instance, becomes less important as a risk factor in this population, since the stroke itself creates such a dramatic change in the individual's functional state. On the other hand, inability to walk, an item not considered a risk factor in the index, appeared to generate an important predictive value. It has previously been found that a significant number of falls among stroke patients and other patients in rehabilitation occur from a sitting position, mainly in wheelchairs.2 17 Apparently, nonwalkers are also at risk of falling. Other factors of potential interest in this respect, as shown in our additional analyses, would be visuospatial hemineglect and dyspraxia. One further factor, not accounted for in the Downton index, that has proved to increase the predictive accuracy of the Fall Assessment Questionnaire is impulsive behavior among right-hemisphere stroke patients.5
Furthermore, no explicit operational definitions were provided with the Downton index, so we specified the definitions we used for the scores in this study. Some of the considerations in our interpretation of the risk factors intended to be focused on by the items presented require special attention, since they might be taken as minor modifications to the index. Regarding mental state, we preferred the term “cognitively impaired” to “confused.” For practical and cultural reasons, we also used the well-established MMSE cutoff score of ≤23/30 points as a diagnostic criterion indicating cognitive impairment, instead of the suggested cutoff score of <7/10 points of the Abbreviated Mental Test score.18 Regarding gait, we interpreted the term “normal” in this connection to mean safe gait without walking aids rather than a normal gait pattern without any pathological signs. Thereby, the terms would also be mutually exclusive.
We believe our studied population is fairly representative of stroke patients undergoing subacute hospital rehabilitation. The moderate association found between fall prediction and outcome when the Downton fall risk index was used is promising with regard to the potential use of a simple assessment method to identify individuals liable to fall and risk factors that could be addressed by targeted intervention measures. However, adjustments to the index are obviously needed to increase the accuracy of prediction in a stroke population.
This study was supported by grants from the Borgerskapet i Umeå Research Foundation, the Federation of County Councils in Sweden, the 1987 Foundation for Stroke Research, the Joint Committee of the Northern Health Region of Sweden, the Gun and Bertil Stohne's Foundation, the Swedish Society of Medicine (reference no. 340.0), Umeå University's Fund for Medical Research, and the Västerbotten County Council. The authors would like to thank Mai Mattsson, MD, and Ulrica Radsjö, RPT, for their contribution to this study.
- Received May 3, 1996.
- Revision received July 1, 1996.
- Accepted July 11, 1996.
- Copyright © 1996 by American Heart Association
Dromerick A, Reding M. Medical and neurological complications during inpatient stroke rehabilitation. Stroke. 1994;25:358-361.
Nyberg L, Gustafson Y. Patient falls in stroke rehabilitation: a challenge to rehabilitation strategies. Stroke. 1995;26:838-842.
Downton JH. Falls in the Elderly. London, UK: Edward Arnold; 1993:64-80, 128-130.
Katz S, Ford AB, Moskowitz RW, Jackson BA, Jaffe MW. Studies of illness in the aged. The index of ADL: a standardized measure of biological and psychosocial function. J Am Med Assoc. 1963;185:914-919.
Strand T, Asplund K, Eriksson S, Hägg E, Lithner F, Wester PO. A non-intensive stroke unit reduces functional disability and the need for long-term hospitalization. Stroke. 1985;16:29-34.
Dawson-Saunders B, Trapp RG. Basic and Clinical Biostatistics. 2nd ed. East Norwalk, CT: Appleton & Lange; 1994:193-195, 200-201.
SYSTAT: Statistics, Version 5, 2nd Edition. Evanston, Ill: SYSTAT Inc; 1992.
Qureshi KN, Hodkinson HM. Evaluation of a ten-question mental test in the institutional elderly. Age Ageing. 1974;3:152-157.