|Home > RNJ > 2007 > November/December > Discriminating High Fall Risk on an Inpatient Rehabilitation Unit|
Discriminating High Fall Risk on an Inpatient Rehabilitation Unit
The objective of this study was to identify on admission the most discriminating fall predictors for patients to an inpatient rehabilitation unit. Medical information from 34 patients who fell over a consecutive 7-month period and 102 controls (1:3 ratio) matched for diagnosis, age, and gender was analyzed to identify a set of best predictors. Admission mobility and problem solving FIM™ scores accounted for 17% of variance in whether a fall occurred during the admission. After statistically deriving optimal cutoff thresholds for decision making, high fall risk was retroactively assigned to patients. Logistic regression revealed increased odds of having fallen by 5.1 times for poorer mobility and 2.4 times for poorer problem solving. The practical benefits of the evidence-based risk assessment were discussed.
Patients on an inpatient rehabilitation unit (IRU) typically have multiple fall-risk factors compared to general hospital patients or persons at risk in the community. The presence of multiple risks makes it difficult to discriminate those likely to fall. Many of the measures used to assess fall risk and predict falls have been validated for the community, skilled nursing and extended care facilities, or general hospital settings. Generalization of instruments and scores across settings is often limited (Oliver, Daly, Martin, & McMurdo, 2004; Perell et al., 2001). The aim of this study was to identify a set of variables that could discriminate those who fell on an IRU from those who did not, despite the presence of multiple risk factors.
Unintentional falls remain the leading cause of nonfatal injuries in the United States (Centers for Disease Control and Prevention [CDC], 2006). Several recent studies provided large N overall hospital fall rates of 2.45 to 3.73 falls per 1000 patient days (Dunton, Gajewski, Tauton, & Moore, 2004; Halfon, Eggli, Van Melle, & Vagnair, 2001; Hitcho et al., 2004). Fall rates for inpatient rehabilitation have not been as systematically estimated. Rates vary from 2.92 to 15.9 falls per 1000 patient days and reflect selected diagnoses, older studies with long rehabilitation lengths of stay, data from care in multiple countries, and other factors (Aisen, Iverson, Schwalbe, Weaver, & Aisen, 1994; Morse, 1996; Nyberg & Gustafson, 1996; Rogers, 1994; Suzuki et al., 2005; Sze, Wong, Leung, & Woo, 2001). Many studies of inpatient rehabilitation unfortunately report rates using different measures, which make comparisons difficult. It is clear that rates of falls in inpatient rehabilitation are higher than those of the general hospital.
Studies with IRU samples have identified multiple medical, functional, and cognitive factors associated with higher fall risk (e.g., Juneja, Czyrny, & Linn, 1998; Rapport, Hanks, Millis, & Deshpande, 1998; Teasell, McRae, Foley, & Bhardwaj, 2002). Attempts to devise a risk profile or IRU-specific measure have resulted with complex multiple-factor indices (Nyberg & Gustafson, 1997; Vassallo, Sharma, Briggs, & Allen, 2003). Olsson, Lofgren, Gustafson, and Nyberg (2005) recently described the difficulty in trying to replicate the validity of one such complex index in IRU patients. An alternative to complex measurement would be to discriminate falls in a more controlled study.
Although studies of falls in IRU patients have been both retrospective and prospective, randomized assignment to a “fall condition” is not possible. No study, though, has applied a matched case-control design to control for common factors known to affect fall risk. Several patient characteristics from the literature that consistently affect fall risk in the community, inpatient settings, and IRUs have been a person’s diagnosis, age, and, less consistently, gender. Right-brain cerebrovascular accident (CVA), older age, and being female have been associated with greater fall risk than comparison groups (e.g., Bueno-Cavanillas, Padilla-Ruiz, Jimenez-Moleon, Peinado-Alonso, & Galvez-Vargas, 2000; Grant & Hamilton, 1987; Halfon et al., 2001; Perell et al., 2001). In this study, IRU patients who fell were matched with controls of the same diagnosis, age, and gender.
Poor discrimination with the assessment component of the hospital-wide plan was the clinical impetus for the study. The hospital plan used an evidence-based Morse Scale cutoff score of 45 to determine high risk (Morse, 1996, 2006). Our experience had been that 75%–90% of the IRU patients were then designated as “high risk.” Implementing the intensive hospital fall prevention plan (designed for a much smaller proportion of patients per unit) was not feasible. Our aim was to develop an assessment tool for discriminating those at greatest fall risk, without discounting the relative fall risk in all rehabilitation patients.
The first hypothesis was that by controlling for diagnosis, age, and gender, one or two functional characteristics would be sufficient to discriminate those who fell from matched controls. A second hypothesis was that cutoff scores for the significant predictors identified in the test of the first hypothesis would create a feasible clinical tool for determining fall risk.
Thirty-four patients who fell on an acute inpatient hospital rehabilitation unit over a continuous 7-month period were the target sample in this study (fall group). A fall was operationally defined as “unintentionally coming to rest on the ground, floor, or other lower level” (Buchner et al., 1993). If the patient lost balance and was lowered to the floor by a helper or was found on the floor, both the attended or unattended situation was considered a fall. A comparison group of 102 rehabilitation patients during the same period were matched with the fall group on diagnosis (including subtype such as right- vs. left-brain stroke), age (as close as possible), and gender, in that order (matched comparison group). The comparison group consisted of three patients matched to each fall patient to enhance statistical power. Data from retrospective chart reviews and scores from the 1999 version of the FIM™ (Uniform Data System for Medical Rehabilitation, 1999) instrument were used. Length of stay, time since onset of diagnosis, and the 18 admission-FIM™ item-scores scaled from 1 (total assistance) to 7 (complete independence) were used as potential predictors. The study received medical center institutional review board approval prior to any data collection from medical records. Analyses were performed on a de-identified research data set.
Procedure. All patients received usual care, and no interventions were performed beyond the hospital-wide fall prevention plan. The data for this study were collected retrospectively with the existing fall prevention plan being implemented.
Analyses. Using SPSS for Windows (Release 12.0), potential predictors were entered as independent variables in a stepwise discriminant analysis to predict membership in the two groups (fall vs. comparison). This method is similar to a multiple regression, but the aim of the analysis is accurate classification of a categorical dependent measure. A canonical correlation expresses the degree of association between the sets of independent and dependent variables in a discriminant analysis. The canonical correlation is parallel to a multiple R in this study, but discriminant analysis uses the alternative term because it can correlate a set of independent variables with a set of dependent variables. Anticipating one to three variables that would discriminate the groups, signal detection, or decision-making analyses (Ward, Marx, & Barry, 2000; Zarin, 2000) would identify optimal cutoff scores using sensitivity (ability to detect the fall group) and specificity (ability to detect the comparison group). This approach involved exploring the range of correct hit rates for sensitivity and correct rejection rates for specificity, and deciding on the optimal set of values for the two. Although high scores on both sensitivity and specificity would be ideal, a sensitive measure was more important with the practical objective of identifying persons at risk for falls. Logistic regression identified the odds ratios associated with the retrospective prediction of fall risk using the variables with cutoff scores. Although discriminant and logistic regression methods overlap in their aim, each approach had features relevant to this study (Press & Wilson, 1978).
Table 1 lists the participant characteristics on the matching variables. As the three statistical tests indicated, there were no group differences, even though a few individuals could not be matched exactly on all three variables. The table also summarizes group differences on several global outcome variables not used as predictors. Those individuals who fell had lower total admission and discharge FIM™ total scores, suggesting these individuals had greater functional impairment from the onset of rehabilitation. There were nonsignificant trends for a shorter length of stay and lower length of stay (LOS) efficiency for those who fell. The LOS efficiency measure is calculated by the total FIM™ score change divided by the length of stay. A patient who is more efficient in rehabilitation progress would make greater average FIM™-score gains per day and would regain the functional independence necessary for discharge rapidly. Thus, it is one of the major indicators of a positive rehabilitation outcome (e.g., Ottenbacher et al., 2004).
The resulting canonical correlation for the stepwise discriminant analysis of the two best predictive variables, mobility and problem-solving FIM™ scores, was .326 (χ2 = 14.927, p = .001). Correct classification was 66.2%, which was significantly greater than 50% for chance. The mean discriminant-function scores and confidence intervals were –0.59 (95% CI = –0.85, –0.33) and 0.20 (95% CI = –0.01, 0.41) for the fall and comparison groups, respectively. The fall group had significantly lower scores than the comparison group. The confidence intervals of the means did not overlap, with zero indicative of significant discrimination between groups. These analyses employed predictors as continuous variables on 7-point scales, but better practical utility could be achieved with the cutoff scores for higher versus lower fall risk.
Signal detection analyses of the admission mobility and problem-solving FIM™ scores provided sensitivity and specificity scores relative to chance discrimination at .50. As an additional reference point, the discriminant function of the 7-point FIM™ scores yielded a sensitivity of .68 and specificity of .66. The use of a cutoff of admission mobility FIM™ score as maximal or total assistance (a score of 1 or 2 versus higher scores) improved sensitivity to .88, although specificity dropped to chance at .42. The optimal cutoff for the admission problem solving FIM™ score of moderate or more direction (a score of 3 or less versus higher scores) was .62 for both sensitivity and specificity. The use of both these cutoff scores for admission FIM™ scores should enhance prediction of fall risk over not using this information.
Figure 1 breaks down the sample groups by predicted fall risk from low with FIM™ scores above cutoff levels (more independent) on both scores to the highest combined risk with FIM™ scores below the cutoffs (more dependent or impaired) on both scores. A chance relationship would be 25% in each cell. The very low number of the patients who fell in the low-risk group (5.9%) and the high number in the highest predicted risk group (55.9%) accounted for the strength in the relationship. Both of these cell frequencies were two or more standard residual units from expectation and statistically significant.
Logistic regression with the two cutoff-score versions of mobility and problem solving coded as one for at or below cutoff (associated with greater risk) or zero for above cutoff (reference) tested the strength of relationships to predict fall group membership. Both predictors were statistically significant (p < .05). Analysis with the variables as FIM™ scores or as dichotomous cutoff scores did not differ in strength of relationship, suggesting the use of cutoff scores, which would be more useful in clinical decision making, did not sacrifice any loss of predictive power. The Negelkerke estimate of R² of .167 for predictors as cutoff scores was better than the squared canonical correlation coefficient of .108 for the two predictors as full FIM™ scale scores. The odds ratios and 95% confidence intervals for the two independent variables as cutoff scores indicated that persons who were maximal to total assistance on admission had 5.140 times the odds for having a fall than those in the comparison group who did not fall (CI = 1.667, 15.848). Persons requiring moderate or more direction for problem solving on admission had 2.400 times the odds for all fall (CI = 1.049, 5.492). The odds ratios are multiplicative, indicating 12.326 times the odds if both factors were present. If the interaction is included in the logistic regression, with being above on both cutoff scores (low risk) as reference and low mobility only, low problem solving only or low on both (highest risk) as independent variables, only the interaction was significant.
Measuring risk in its full complexity may be counterproductive when trying to discriminate or predict those who fell or might fall from those who do not on an IRU. Nyberg and Gustafson (1996) tested an index based on medical and functional characteristics summed to the presence of up to 11 risk factors to predict fall risk in stroke rehabilitation patients with a high sensitivity of .91, but low specificity of .27. Olsson et al. (2005) failed to replicate these findings, and ended up with a modified scale of only three items: impaired balance, visual hemi-inattention, and male gender. The cumulative effect of revised risk scale was replicated in two samples and yielded hazard ratios (and 95% CI) of 1.8 (1.4 – 2.4) in the model fit sample and 1.9 (1.4 – 2.7) in the validation sample. The current study yielded similar findings to Olsson et al. (2005) with (1) only two necessary predictors (independent of gender, which was controlled for), mobility and problem solving, (2) the use of widely used FIM™ scores as predictors, (3) greater discrimination of risk (reported odds ratios, with combined odds for fall of 12.3 with both factors present), (4) more modest sensitivity (.88 and .62 for the mobility and cognitive respectively), and (5) better, but still poor specificity (.42 and .62, respectively). Thus, fewer factors can increase correct identification of those at high fall risk over more complex models.
Most IRU patients have many risk factors, but the between-person variation in combinations makes use of the complexity infeasible. As an informal comparison of the multiple risks for the persons in this study, participants were considered “impaired” or not for each of the 15 ranked factors from the Perell at al. (2001) review. Impairment was defined as demonstrating impairment on specific tests such as manual muscle testing or a score of 4 (minimal assist) or lower when measured by FIM™ item. Using a sum of the presence or absence of impairment on any factor from the Perell et al. list, the two groups did not differ (Fall: M = 12.00, SD = 1.18; Comparison: M = 11.68, SD = 2.31; t < 1.0, NS). Over both groups, 94% had 10 or more of the 15 risk factors.
Although the results from this study did not provide a definitive prediction of patients admitted to an IRU who are likely to fall during their stay, patients with impaired mobility and problem solving beyond the defined cutoff levels was associated retrospectively with greater risk. The study aim was the utility of common, predominant factors. We confirmed the discriminating power of mobility (whether walking or via wheelchair) and cognition over other predictors. These results were consistent with findings in major literature reviews and specific studies with IRU samples. The two factors are present and usually at or near the top of fall-risk factor lists (Bueno-Cavanillas et al., 2000; Halfon et al., 2001; Oliver et al., 2004; Perell et al., 2001). Impaired mobility and cognition are also components of more complex fall-risk indices (Morse, 1996, 2006; Olsson et al., 2005).
Alexander (1996) discussed multiple ways in which cognition and gait might interact, which might explain the greater importance of the mobility and problem solving variables over others. In the presence of cognitive impairment, gait speed is slower, gait speed does not increase as much by time of rehabilitation discharge, and there is increased step-to-step variability. From the other point of view, for a person with poorer mobility, there is slower cognitive reaction to environmental change and poorer correct reaction to circumstances. White matter ischemic changes (leukoaraiosis) have been associated with both poorer gait and cognition even when other risk factors have been controlled (Podgorska, Hier, Pytlewski, & Czlonkowska, 2002). Some neurological basis of the gait-cognition interaction is likely.
Other studies have focused on specific components of cognition that have been associated with fall risk, most particularly attention, environmental perception, and reaction time, and executive functions (Mayo, Korner-Bitensky, & Kaizer, 1990; Rapport et al., 1993, 1998). The nature of the relationship among specific cognitive impairments, mobility, and falls needs further study. The mobility– cognition interaction is also consistent with descriptive or epidemiological studies of the circumstances of hospital falls (Hitcho et al., 2004, Kerzman, Chetrit, Brin, & Toren, 2004; Mion et al., 1989; Rogers, 1994). Fall situations typically involve a patient getting up from bed, climbing over bedrails, or need to use the toilet. Even when attended, external obstacles or sudden situations requiring quick reaction can lead to a fall.
The current study had its limitations. The sample was small, and data were from a short time interval. We did find the sample to be representative of the IRU population during that period, but replication with a larger or multisite sample would be beneficial. The study was also retrospective. Mobility and cognition FIM™ scores technically could only discriminate those who fell from those who did not and do not predict in the prospective sense. The set of independent variables was also small and perhaps unfair to other potentially predictive factors discarded after the preliminary analyses. The choice in the current study was to (1) build on preliminary results from a pilot study that evaluated the relative utility of more than 200 measures, (2) force a result that would avoid complexity and yield significant factors, (3) examine potential factors that have demonstrated past fall prediction, and (4) use only common measures of widespread use in IRUs. That the predictors, even though few, that were significant were also frequently high on other risk factor lists does support the validity of the results. Finally, even though the results are easily translated into practice, good evidence requires replication in other samples.
Subsequent to the research study, we implemented a change in the assessment component of the fall prevention plan for the IRU at our facility (Table 2). If both mobility and cognition were impaired on the cutoff scores as well as a score greater than 45 on the Morse Scale, an individual was deemed to be high risk. Patients impaired on both cutoff scores were always above the hospital-wide high score on the Morse Scale. It was more feasible to adapt the existing hospital fall prevention plan than to create a separate one for the IRU. The approved revision of the fall prevention policy and procedure for the IRU documented the adapted assessment of high fall risk and rationale. The intensive fall prevention intervention included nondescript orange indicators outside a patient’s door, an orange identification wristband, and orange signs with fall prevention reminders for the patient in the person’s own language and for staff. Other patient-specific recommendations other than chart or Kardex alerts could also be used with the orange theme. Family members were also educated to the fall plan. The orange served as a cue to all staff for prompt response to any patient needs, preventive toileting, and not leaving such persons out of visible monitoring. Once on high-risk precautions, only a team decision documented as part of the weekly team conference would release the person from high-risk status. With a month of staff awareness training, the unit fall rate had decreased. Compared to a 3-year fall rate of 6.6 per 1,000 patient days prior to using the new assessment tool, the fall rate for the subsequent 12 months was 5.7. This was not statistically significant because of the large variance in falls across the months prior, but the decreased rate with less variability after use of the new decision tool was a desirable outcome.
Due to a general high risk for patients in an IRU relative to other populations, we retained “universal fall precautions” for all rehabilitation patients. These general precautions might include verbal or written reminders to ask for assistance, educating family not to assist patients in mobility unless trained, regular checks on the patient for personal needs such as toileting, and access to a call light or other notification system. The orange theme was reserved only for those with high risk. To attempt an intensive fall prevention plan on virtually everyone had been prohibitive. Yet the challenge in a rehabilitation environment is to find a realistic alternative. On the one hand, therapy is essential for functional independence by pushing patients’ limits to strengthen muscles, increase functional ability, and improve awareness and cognition. Carryover of techniques outside of therapy is essential for good outcomes. On the other hand, patients sometimes get into or put themselves into situations that challenge them beyond their means without staff support. Falls are one risk associated with this essential limit pushing. The balance point of personal safety and maximal functional outcome is not easy to achieve for patients with new disabilities. The use of the new discriminative tool for determining the highest risk did reduce the high risk solely on the Morse Scale from 87% to 33% using the additional criteria. With the safety of all patients in mind, the intensive fall prevention plan became more feasible for those assessed as highest risk.
About the Authors
Michael J. Gilewski, PhD, is assistant professor of physical medicine at Loma Linda University. Address correspondence to him at 11406 Loma Linda Drive, Loma Linda, CA 92354–3711 or firstname.lastname@example.org.
Pamela Roberts, MSHA OTR/L, is a manager of rehabilitation, psychology and neurology at the Department of Rehabilitation, Cedars-Sinai Medical Center, Los Angeles, CA.
Jodi Hirata, MPT, is a physical therapist at the Department of Rehabilitation, Cedars-Sinai Medical Center, Los Angeles, CA.
Richard Riggs, MD, is a medical director of the Department of Rehabilitation and chairman of the Department of Physical Medicine at Cedars-Sinai Medical Center, Los Angeles, CA.
AUTHOR NOTES: A preliminary version of this paper was on the program of the annual meeting of the American Academy of Physical Medicine and Rehabilitation, New Orleans, September 2001. The research study was approved by the Cedars-Sinai Medical Center Institutional Review Board, study number 3617.
Aisen, M. L., Iverson, D., Schwalbe, C., Weaver, B., & Aisen, P. S. (1994). Falls on a neurorehabilitation unit: Reassessment of a prevention program. Journal of the American Paraplegia Society, 17, 179–182.
Alexander, N. B. (1996). Gait disorders in older adults. Journal of the American Geriatrics Society, 44, 434–451.
Buchner, D. M., Hornbrook, M. C., Kutner, N. G., Tinetti, M. E., Ory, M. G., Mulrow, C. D., et al. (1993). Development of the common data base for the FICSIT trials. Journal of the American Geriatrics Society, 41, 297–308.
Bueno-Cavanillas, A., Padilla-Ruiz, F., Jimenez-Moleon, J. J., Peinado-Alonso, C. A., & Galvez-Vargas, R. (2000). Risk factors in falls among the elderly according to extrinsic and intrinsic precipitating causes. European Journal of Epidemiology, 16, 849–859.
Centers for Disease Control and Prevention. (2006). 10 leading causes of nonfatal injury, United States: 2005, all races, both sexes, all cases. Table from WISQARS interactive database. Retrieved October 22, 2006, from http://www.cdc.gov/ncipc/wisqars
Dunton, N., Gajewski, B., Tauton, R. L., & Moore, J. (2004). Nurse staffing and patient falls on acute care hospital units. Nursing Outlook, 52, 53–59.
Grant, J. S., & Hamilton, S. (1987). Falls in a rehabilitation center: A retrospective and comparative analysis. Rehabilitation Nursing, 12, 74–76.
Halfon, P., Eggli, Y., Van Melle, G., & Vagnair, A. (2001). Risk of falls for hospitalized patients: a predictive model based on routinely available data. Journal of Clinical Epidemiology, 54, 1258–1266.
Hitcho, E. B., Krauss, M. J., Birge, S., Dunagan, W. C., Fischer, I., Johnson, S., et al. (2004). Characteristics and circumstances of falls in a hospital setting: a prospective analysis. Journal of General Internal Medicine, 19, 732–739.
Juneja, G., Czyrny, J. J., & Linn, R. T. (1998). Admission balance and outcomes of patients admitted for acute inpatient rehabilitation. American Journal of Physical Medicine and Rehabilitation, 77, 388–393.
Kerzman, H., Chetrit, A., Brin, L., & Toren, O. (2004). Characteristics of falls in hospitalized patients. Journal of Advanced Nursing, 4, 223–229.
Mayo, N. E., Korner-Bitensky, N., & Kaizer, F. (1990). Relationship between response time and falls among stroke patients undergoing physical rehabilitation. International Journal of Rehabilitation Research, 13, 47–55.
Mion, L. C., Gregor, S., Buettner, M., Chwirchak, D., Lee, O., & Paras, W. (1989). Falls in the rehabilitation setting: Incidence and characteristics. Rehabilitation Nursing, 14, 17–22.
Morse, J. M. (1996). Preventing patient falls. Newbury Park, CA: Sage.
Morse, J. M. (2006). The safety of safety research: The case of patient fall research. Canadian Journal of Nursing Research, 38, 72–86.
Nyberg, L., & Gustafson, Y. (1996). Using the Downton Index to predict those prone to falls in stroke rehabilitation. Stroke, 27, 1821–1824.
Nyberg, L., & Gustafson, Y. (1997). Fall prediction index for patients in stroke rehabilitation. Stroke, 28, 716-721.
Oliver, D., Daly, F., Martin, F. C., & McMurdo, M. E. T. (2004). Risk factors and risk assessment tools for falls in hospital in-patients: A systematic review. Age and Ageing, 33, 122–130.
Olsson, E., Lofgren, B., Gustafson, Y., & Nyberg, L. (2005). Validation of a fall risk index in stroke rehabilitation. Journal of Stroke and Cerebrovascular Diseases, 14, 23–28.
Ottenbacher, K. J., Smith, P. M., Illig, S, B., Linn, R. T., Ostir, G. V., & Granger, C. V. (2004). Trends in length of stay, living setting, functional outcome, and mortality following medical rehabilitation. Journal of the American Medical Association, 294, 1687–1695.
Perell, K. L., Nelson, A., Goldman, R. L., Luther, S. L., Prieto-Lewis, N., & Rubenstein, L. Z. (2001). Fall risk assessment measures: an analytic review. Journals of Gerontology Series A: Biological Sciences and Medical Science, 56A, M761–766.
Podgorska, A., Hier, D. B., Pytlewski, A., & Czlonkowska, A. (2002). Leukoaraiosis and stroke outcome. Journal of Stroke and Cerebrovascular Diseases, 11, 336–340.
Press, S. J., & Wilson, S. (1978). Choosing between logistic regression and discriminant analysis. Journal of the American Statistical Association, 73, 699–705.
Rapport, L. J., Hanks, R. A., Millis, S. R., & Deshpande, S. A. (1998). Executive functioning and predictors of falls in the rehabilitation setting. Archives of Physical Medicine and Rehabilitation, 79, 629–633.
Rapport, L. J., Webster, J. S., Flemming, K. L., Lindberg, J. W., Godlewski, M. C., Brees, J. E., et al. (1993). Predictors of falls among right-hemisphere stroke patients in the rehabilitation setting. Archives of Physical Medicine and Rehabilitation, 74, 621–626.
Rogers, S. (1994). Reducing falls in a rehabilitation setting: A safer environment through team effort. Rehabilitation Nursing, 19, 274–276.
Suzuki, T., Sonoda, S., Misawa, K, Saitoh, E., Shimizu, Y., & Kotake, T. (2005). Incidence and consequence of falls in inpatient rehabilitation of stroke patients. Experimental Aging Research, 31, 457–469.
Sze, K-H, Wong, E., Leung, H. Y., & Woo, J. (2001). Falls among Chinese stroke patients during rehabilitation. Archives of Physical Medicine and Rehabilitation, 82, 1219–1225.
Teasell, R., McRae, M., Foley, N., & Bhardwaj, A. (2002). The incidence and consequences of falls in stroke patients during impatient rehabilitation: Factors associated with high risk. Archives of Physical Medicine and Rehabilitation, 83, 329–333.
Uniform Data System for Medical Rehabilitation (UDSMR; 1999). The FIMware® user guide and self-guided training manual (version 5.20). Buffalo, NY: UDSMR. (FIM and FIMware are trademarks of Uniform Data System for Medical rehabilitation, a division of UB Foundation Activities, Inc.)
Vassallo, M., Sharma, J. C., Briggs, R. S. J., & Allen, S. C. (2003). Characteristics of early fallers on elderly patient rehabilitation wards. Age and Ageing, 32, 338–342.
Ward, M. M., Marx, A. S., & Barry, N. N. (2000). Identification of clinically important changes in health status using receiver operating characteristics curves. Journal of Clinical Epidemiology, 53, 279–284.
Zarin, D. A. (2000). Considerations in choosing, using, and interpreting a measure for a particular clinical context. In A. Rush, H. A. Pincus, M. B. First, D. Blacker, J. Endicott, S. J. Keith, et al. (Eds.), Handbook of psychiatric measures (pp. 15–21). Washington, DC: American Psychiatric Association.