Etiology of Falls among Cognitively Intact Hospice Patients

Abstract

Background:

Falls can pose a serious threat to hospice patients receiving palliative care. Interventions to reduce falls have yielded minimal results among older patients. Falls among hospice patients provide a unique population from which a new approach to fall prevention may need to be established.

Objective:

The aim is to devise a forecasting model with which to predict the probability of a patient fall and evaluate whether the model predicts patient falls better than existing measures.

Methods:

Two hundred patients were randomly selected from one of the largest hospices in the United States. After patient admission, patient falls were followed-up via weekly calls until a fall, patient death, or hospice discharge occurred. Independent factors included demographic, functional status, environmental measures, symptoms, medications, attitudinal dispositions, and the use of an ambulatory aid.

Results:

Cognitively intact hospice patients who have a higher risk of falls are those who had a past history of a fall (p = 0.022), patients that are physically more functional as demonstrated by higher score on the Palliative Performance Scale (p = .039), patients with a greater “fear-of-losing-independence (p = 0.023),” those who try to “avoid asking for help (p = 0.005),” and those who “feel uneasy about asking for help (p = 0.05).” Patients who depend on ambulatory aids were less likely to fall (p = 0.06). The forecasting model predicted patient falls correctly in 78% of the patients observed.

Conclusions:

The current model predicted fall occurrence far better than the Morse Falls Scale and other functional status measures and may lead to a shift in fall prevention approaches among hospice patients.

Introduction

Life expectancy of individuals in the United States is at an all-time high and this phenomenon is associated with an increased number of people living with and dying from chronic debilitating diseases,¹ in which older adults are the fastest growing segment of the U.S. population. By year the 2040, the number of those 65 years and older is expected to be more than 77 million.² It has also been shown that patients have a higher prevalence of falls among those with an increased severity of illness or more advanced illness.^3,4 Hospice cares for patients with significant disease severity and advanced illness. Approximately 38.5% (or 963,000 of 2,500,000) deaths in the United States in 2008 occurred under hospice care.⁵ Therefore, hospice is one of the most realistic settings in which to learn about the etiology of patient falls among older adults with advanced illness and launch fall prevention and fall reduction strategies.

The prevalence of falls among the elderly is relatively high, as are the health care costs associated with them. According to Rubenstein et al.,⁶ nursing home patients often experience multiple falls; approximately 2.6 falls per person per year on an average. A study by Tinetti et al.⁷ indicates that one third of older adults more than 65 years of age fall each year in the United States. Of the latter, one quarter sustain moderate to severe injuries that reduce mobility and independence, and increase the risk of premature death. Older adults are hospitalized for fall-related injuries 5 times more often than they are for injuries from other causes. Women are nearly 3 times more likely than men to be hospitalized for a fall-related injury.⁸ Bone fractures are the most serious health consequences of falls, and 87% of all fractures among adults 65 years and older are due to falls. The total cost of all fall injuries for people 65 years and older in 1994 was estimated to be $20.2 billion.⁹ By year 2020, the cost of fall injuries is expected to reach $32.4 billion before adjusting for inflation.⁹ Without effective fall prevention strategies that work in real world settings, the number of fall injuries is expected to increase at alarming rates as the U.S. population ages, not to mention the health care costs associated with these falls.^10,11

Much of the literature on patient falls are descriptive studies on the incidence and prevalence of falls within specific groups, focusing largely on the seriousness, costs of health care, potential environmental hazards, and the outcome of falls rather than the determination of causal factors. Among community-dwelling elders, a meta-analysis of prognostic values of risk factors utilizing multivariate analysis identified only a history of fall and gait or balance problems as the sole precursors of falls. In this study, visual impairment, medications, limitations in activities of daily living and cognitive impairment did not consistently predict falls.¹² There has been limited research on fall-related studies among hospice patients and no study on the etiology of falls in this particular group of patients. Rawsky¹¹ conceptualized falls as an interaction between intrinsic (personal) and extrinsic (environmental) factors. Others have added that falls are more commonly related to age, gender, functional disability, disease status, physical strength, coordination ability, and level of alertness.^13–17 More recently, the Centers for Disease Control and Prevention (CDC)¹⁸ noted “a fall is a result of an interaction of intrinsic, extrinsic and environmental factors” without clear definitions pertaining to the boundaries of these domains. Most of the prior studies describing potential factors that contribute to falls have not been translated into preventive practice. As a result, the past research has had minimal influence on reducing the incidence and prevalence of falls among various types of patient groups in general and none among hospice patients. There are likely other contributing factors, such as attitudinal dispositions, symptom-related factors, and others among those with advanced illness that likely impact the risk of falls, which existing literature has not addressed adequately.

One of the objectives of this study is to develop a conceptual model on the etiology of falls among cognitively intact hospice patients. The model addresses several domains that are potentially meaningful for predicting hospice patient falls and delineates a logistic regression model for forecasting patient falls at the individual level using the variables embedded in these domains. The study will test which health-related symptoms are related to patient falls, how functional status influences patient falls, and whether patients' attitudes on the “fear-of-losing-independence” (FOLI) have a significant impact on the risk of falls. Finally, the predictive utility of Morse Scale¹⁹ has not been tested among hospice patients. The end product of this study is to develop an algorithm for predicting future patient falls using patient data collected at the time of initial patient assessment during hospice admission, where the overall predictive value generated is greater than those predicted by the Morse Scale score or other functional status scales that are commonly used.

Methods

Conceptualization on the etiology of falls among hospice patients

Realizing that factors that contribute to falls among cognitively intact hospice patients are somewhat different from those in the general population, a conceptual model on the etiology of falls among hospice patients is elaborated. It begins with an a priori assumption that patient falls are accidents and, as such, falls are more likely to happen to hospice patients who make unexpected or nonroutine moves, those who are more active/functional or more mobile, those who do not ask for help from or depend on others, those who do not use an ambulatory aid, and those who have the need to move beyond their normal course of activities. Certain precipitous symptoms may force patients to make nonroutine movements. Symptoms such as cough and dry mouth may lead patients to take nonroutine moves. It is also likely that past behavior is indicative of future behavior and, accordingly, it follows that the risk of falls is higher among patients with a history of falls prior to their admission to hospice than those without such incidents. Other symptoms may restrict the need to make sudden movements, including nausea, which might be less severe or less aggravated by limiting movement.

The conceptual model does not specify fall risk changes among those with different ethnic or racial backgrounds. The use of certain medications has been associated with a greater risk of falls among other samples studied.^20–24 This generally was thought to result from altered mental status increasing the risk of a fall. However, in the general hospice population, these patients may already be more globally impaired. In addition, no previous hospice patient study evaluated the risk of falls from medications from the perspective of whether utilization of the medication increased or decreased mobility and the subsequent risk of a fall. For example, a patient may feel better after taking antidepressants over time, and thus increase their mobility after successful use of these medications. For explorative purposes, eight types of medications were examined in evaluating the risk of falls among hospice patients: analgesics, antihypertensives, antipsychotics, diuretics, antidepressants, benzodiazepines, laxatives, and anticonvulsants.

The model on the etiology of falls assumes that the attitudinal makeup of hospice patients plays an important role, where accidental falls are more likely to occur among patients who have a fear of losing independence or fear of becoming dependent on others. Accordingly, hospice patients with a fear of losing independence are viewed as more inclined to avoid asking for help, even though help may be available at home. From the model, it may deduced that younger patients may be more likely to fall than older patients since the former are more likely to be or desire to be mobile and active. The model, however, does not guide whether patients who live alone or live with a spouse/partner or child are more or less likely to fall since the degree of help provided by the caregivers at home partly depends on the willingness of the patients to seek such help as well as the degree to which such help is available. Finally, the model assumes that hazard conditions found in the patients' living quarters may increase the risk of falls, while the protective conditions found in the home environments are expected to decrease the risk of falls.

In summary, demographic factors such as age and gender, functional status/mobility, past history of falls, patient symptoms, personal attitude, and environmental conditions are expected to have a significant impact on the prevalence of falls, while other factors such as most medications and patient ethnicity likely have little impact on patient falls. The significance of the model introduced is augmented by its ability to discriminate factors that may have increasing and decreasing risks on patient falls as well as those that have little or negligible impact.

Hypotheses

Deducing from the conceptual model on falls among cognitively intact hospice patients, it is hypothesized that the following variables will increase the risk of falls among those hospice patients: female; younger age; patients with greater mobility; patients who ambulate without help; patients who do not utilize ambulatory aids regularly; patients with a past history of falls and higher frequency of falls prior to hospice admission; patients with frequent symptoms such as cough and dry mouth that may induce some patients to take nonroutine movements to get water to quench their thirst; patients who take antidespressants, patients with a fear of losing independence; patients who purposely avoid asking for help; and patients who feel uneasy about asking for help from others. The independent variables that may decrease the risks of falls are patients with severe levels of distress from nausea; patients who routinely utilize ambulatory aids for walking; and patients who are less mobile, bed-bound. All together, 36 hypotheses are tested at the bivariate level.

Operationalization

The project dependent variable (DV) is measured by a binomial variable, where 1 and 0 are assigned to the patients with a fall and no fall, respectively. Table 1 enumerates all original and rescaled independent variables (IVs), variable names and labels, value labels, the mean, standard deviation, mini-max range, and sample size for each IV. All IVs are grouped into seven domains: demographics, functional status, history of a previous fall, patient symptoms, medications, attitudinal dispositions, and environmental conditions. For the assessment of functional impairment, this study includes three scales widely used in health care: Morse Scale on Falls,²⁵ Palliative Performance Scale,²⁶ and Activities of Daily Living Index.²⁷ For patients' symptom assessment, the Memorial Symptom Assessment Scale²⁸ is used. It lists 21 symptoms at three levels: frequency, severity, and symptom distress. Medication is operationalized by eight IVs. The attitudinal scale of fear of losing independence (FOLI) is measured by a new instrument using eight items scored with a Likert scale (Table 2). The α coefficient of reliability of FOLI Scale is 0.72. Hotelling's T test for equality of means among the items has registered a T value of 19.1 (p = 0.0005), indicating that the means of scale items are significantly different from each other.

Table 1.

List of Original and Rescaled Variables Identified in the Model

Variable name	Operationalization: Value labels (Prevalence rate)	Mini-max range	Mean	SD	n
Project dependent variable (1 DV)
Fall status	Fall observed at follow up Interview: 0 = no fall (73%); 1 = yes fall (27%)	0–1	0.27	0.45	200
Demographic variables (9 IVs)
Age	Age at hospice admission	45–94	74	11.32	200
Gender	1 = male (56%); 2 = female (44%)	1–2	1.44	0.50	200
Schooling in years	Schooling/education in years	4–20	11.95	2.75	200
Live alone	1 = yes, live alone (16.2%); 0 = no, live with others (83.8%)	0–1	0.16	0.370	197^a
Live with spouse/partner	1 = yes, live with spouse (50.3%); 0 = no (49.7%)	0–1	0.503	0.501	197^a
Live with child(ren)	1 = yes, live with child(ren) (27.4%); 0 = no (72.6%)	0–1	0.274	0.447	197^a
White	0 = not white 1 = white	0–1	0.805	0.397	200
Black	0 = not black 1 = black	0–1	0.110	0.314	200
Hispanic	0 = not Hispanic 1 = Hispanic	0–1	0.080	0.272	200
Functional status (4 IVs)
Morse Scale score	Total Morse Scale score measured at hospice admission (0 = normal; 125 = maximum impairment)	0–125	43.35	21.35	200
Palliative Performance Scale	Palliative Performance Scale score measured at hospice admission (0 = death; 100 = normal activity)	0–100	55.20	0.69	200
Activities of Daily Living (ADL) Index	ADL Index scale score measured at admission (1 = independent; 7 = dependent)	1–7	2.62	2.08	186^a
Use of ambulatory aid	0 = none or uses wheel chair (46.5%); 1 = use crutches, cane(s), walker or furniture (53.5%)	0–1	0.535	0.050	200
History of falls (2 IVs)
Past history of fall	0 = no fall (47%); 1 = yes (53%);	0–1	0.53	0.500	200
Number of falls year prior to hospice admission	0 = no fall past year (48%); 1 thru 30 (52%)	0–30	1.45	3.02	200
Symptoms (9 IVs)
Frequency of dry mouth	0 = none; 1 = rarely; 2 = occasionally; 3 = frequently; 4 = almost constantly	0–4	1.63	1.19	200
Severity of dry mouth	0 = none; 1 = slight; 2 = moderate; 3 = severe; 4 = very severe	0–4	1.05	0.86	200
Interaction of freq. and severity of dry mouth	How often dry mouth X How severe dry mouth	0–16	2.53	2.85	200
Frequency of dyspnea	0 = none;1 = rarely; 2 = occasionally; 3 = frequently; 4 = almost constantly	0–4	1.92	1.33	200
Severity of dyspnea	0 = none;1 = slight; 2 = moderate; 3 = severe; 4 = very severe	0–4	1.39	1.04	200
Interaction of frequency and severity of dyspnea	How often dyspnea X How severe dyspnea	0–16	3.90	4.12	200
Cross interaction of frequency of dry mouth and severity of dyspnea	How often dry mouth X How severe dyspnea	0–16	2.53	2.96	200
Frequency of coughing	0 = none;1 = rarely; 2 = occasionally; 3 = frequently; 4 = almost constantly	0–4	1.24	1.25	200
Distress of nausea	0 = not at all;1 = a little bit; 2 = somewhat; 3 = quite a bit; 4 = very much	0–4	0.55	0.819	200
Medication (8 IVs)
Analgesics	0 = not medicated; 1 = medicated	0–1	0.72	0.451	199^a
Antihypertensives	0 = not medicated; 1 = medicated	0–1	0.51	0.501	200
Antipsychotics	0 = not medicated; 1 = medicated	0–1	0.08	0.264	200
Diuretics	0 = not medicated; 1 = medicated	0–1	0.34	0.475	200
Antidepressants	0 = not medicated; 1 = medicated	0–1	0.25	0.434	200
Benzodiazepines	0 = not medicated; 1 = medicated	0–1	0.26	0.440	200
Laxatives	0 = not medicated; 1 = medicated	0–1	0.50	0.501	200
Anticonvulsants	0 = not medicated; 1 = medicated	0–1	0.04	0.196	200
Attitudinal/behavioral disposition (5 IVs)
HPC Scale: Fear-of-Losing-Independence	8-item scale with alpha coefficient reliability = 0.72; 8 = no fear at all; 40 = high fear of losing independence	8–40	22.1	3.94	200
I try to do things by myself to avoid asking for help	1 = strongly disagree; 2 = disagree; 3 = not sure; 4 = agree; 5 = strongly agree	1–5	3.58	0.98	200
I feel uneasy about asking for help from other people	1 = strongly disagree; 2 = disagree; 3 = not sure; 4 = agree; 5 = strongly agree	1–5	3.01	1.14	200
Interaction of “avoid asking” and “feeling uneasy about asking for help”	Multiplication of two scale values 1–5, where 1 = strongly disagree; 2 = disagree; 3 = not sure; 4 = agree; 5 = strongly agree	1–25	11.1	5.94	200
I attempt to complete tasks by myself	1 = strongly disagree; 2 = disagree; 3 = not sure; 4 = agree; 5 = strongly agree	1–5	3.88	0.824	200
Environmental/intrinsic risk/protective factors (2 IVs)
Risk conditions in the living quarter and intrinsic risks from a list of 19 items	Count of total number of risk conditions listed: 0 ( = 10%); 1 ( = 33%); 2 ( = 31.5%); 3 ( = 18.5%); 4 ( = 4%); 5 (= 1.5%); 6 ( = 1%); 7 + ( = 0.5%)	0–16	1.85	1.22	200
Protective conditions in the living quarter from a list of 10 items	Count of total number of protective conditions listed: 0 ( = 15%); 1 ( = 26%); 2 ( = 27%); 3 ( = 16%); 4 (= 10%); 5 ( = 4.5%); 6 + ( = 1.5%)	0–11	2.01	1.49	200

n less than 200 is due to data missing from patient charts.

SD, standard deviation; DV, dependent variable; IV, independent variable.

Table 2.

HPC Scale on Fear of Losing Indepedence Results from Item-Total Correlation

Variable r	Variable r
I pride myself about being independent 0.510
I attempt to complete tasks by myself 0.495	I have a fear of losing my own independence 0.353
I feel that I am mentally alert 0.279	I try to do things by myself to avoid asking for help 0.525
I feel that I can still do many things without help 0.327	I feel uneasy about asking for help from other people 0.256
I have a strong desire to be independent 0.452

Cronbach alpha = 0.72; n = 200; r = Item-Total Pearson's correlation coefficient; Hotelling's T test: T = 19.1, F = 50.7; df1 = 7; df2 = 193; p = 0.0005.

HPC, hospice and palliative care.

Environmental condition is measured by counting the presence of the following 15 risk conditions found in the patient's home at the time of the initial interview: wet floor, loose rug, cluttered area, slippery floor, inappropriate footwear, no footwear, unlighted electrical switches on the wall, poor lighting, unreachable light switches, no handy flashlight, needed items out of reach, presence of tubing, faulty/broken household items, faulty/broken medical support equipment and “other,” including four intrinsic risk conditions such as hearing and visual impairments, difficulty making needs known and expressing needs. Although these impairments are not truly environment factors, they are placed here secondary to factors that have an affect on the environmental outcomes and risks. The protective conditions are measured by counting the presence of following 10 conditions: visible hand rails, grab bars in toilet area, grab bars in shower area, mat with suction cups, bench/stool in shower area; steps tools with hand rails, bedside commode, bell/baby monitors, stool with handrails, and “other.” The listed environmental/home safety/hazard conditions are cited often as having potential impact on patient falls among elderly persons living in the community.^11,12,19

Data collection methods used

This study randomly selected patients from all newly admitted patients to LifePath Hospice, which is one of the largest not-for-profit hospices in the United States. Hospice patients who have consented to participate in the study were selected based on “first come-first served” basis. The sample included 200 patients: 81% white, 11% black, and 8% Hispanic. A total of 52.5% had cancer and 47.5% had noncancer illnesses; 17.5% had chronic obstructive pulmonary disease, 8.5% had heart failure, 6.5 had adult failure to thrive, and 15% other illnesses. All original data are captured in two stages. Data are extracted from existing hospice patient charts. This is followed by patient interviews conducted within 7 days of patient admission to hospice, and weekly follow-up phone calls until a fall, patient death, or hospice discharge occurred. Weekly calls determine whether falls had occurred for patients enrolled in the study with an end point of a patient fall, their death, or discharge from hospice.

Statistical power analysis

According to an institutional quality assurance report at the study site, the percent of patients who fell was 23.5% in 2005.²⁹ Thus, 47 patients in our sample are expected to fall during the current study period, which lasted for a year for data collection. For this hypothesis test, the Wald statistic is applied where the test statistic uses the χ² distribution. With n = 200, type I error set at 0.05, assuming small to medium effect size (ω = 0.30), with large degrees of freedom of 30 (representing large number of IVs), the statistical power³⁰ of χ² test statistic computed is 0.82. Thus, the probability of rejecting the hypothesis of no association (or no added explanation) is expected to be 0.82.

Patient inclusion/exclusion criteria

The subject inclusion criteria are: English speaking, noninstitutional community dwelling, and age 45 years or older. Patients are excluded if they have cognitive impairment with greater than four errors on the Short Portable Mental Status Questionnaire.³¹ Finally, the patient interviews were realized following a written consent signed by patients or their surrogates, if patients were physically unable to sign.

Data reduction strategy

The viability of each hypothesis is tested via bivariate logistic regression analysis (LRA),³² in which all directional hypotheses are tested using one-tailed test. Multivariate LRA is applied to test the significance of domain-specific models and a final model while applying two-tailed test, incorporating all IVs that were initially included in the conceptual model on the etiology of falls. It is widely accepted that there is no unique data reduction strategy, which results in the selection of the “best” model.³² Accordingly, the final model is selected among several alternative models based on the principles of interpretability, parsimony, ease of data acquisition, and a model that has the smallest “regret” with a minimum overall-percentage-correct (OPC) exceeding 75%.

In this study, the false-negative (i.e., predicted as no fall, but actually observed as a fall) is viewed as the greatest “harm” to the investigator/patient while the false-positive (i.e., predicted as a fall but actually observed as no fall) is considered as an acceptable or “tolerable” regret.^33–37 These measures, including the positive predictive value (PPV), the negative predictive value (NPV) as well as the OPC, all depend on which “cutoff” point one uses in the model selection. Table 1 lists one DV and 39 IVs. Of the latter, 35 IVs are measured in the original format while 4 are reconstituted interaction variables identified during the conceptualization stage. For each of the domains, domain-specific multivariate LRA applied to determine whether all of the IVs included in each domain have any significant impact on patient falls.

Results

Demographic variables

Table 3 lists the test results of 35 independent variables tested at the bivariate level. Within the demographic domain, 5 IVs have logistic regression coefficients (B) with negative signs, indicating that patients with these profiles have reduced odds of having a fall, while those with positive B have increased odds of having a fall. For example, age has negative B, which means older patients (with larger age value) are less likely to fall than younger patients (B = −0.016; odds = 0.984; p = 0.13). Female patients (with larger values) are more likely to fall than male patients (B = 0.462; odds = 1.588; p = 0.075). But, neither of these has shown significance at the 0.05 level. Family member/caregiver status variables have shown no significance at the 0.05 level. Similarly, none of other ethnic variables such as being white (B = 0.634; odds = 1.885; p = 0.161), black (B = −0.564; odds = 0.569; p = 0.329), or Hispanic (B = −0.508; odds = 0.602; p = 0.443) had any impact on patient falls. The PPV of the forecasting model by nine demographic variables is 50%, NPV is 72.8%, specificity is 97.2%, sensitivity is 1.9%, regret index (RI) score is 27.2%, and the OPC is 72.6%. The domain-specific multivariate logistic regression model has failed to show statistical significance at 0.05 level (χ² = 8.607; df = 9; n = 197; p = 0.474). Hence, it is generalized that demographic characteristics of hospice patients have random effect on hospice patient falls.

Table 3.

Bivariate Relationships between Patient Falls and Thirty-Five Independent Variables

Independent variables: Variable name/label	B (−∞ to +∞)	Odds change Exp(B)	SE of B	Wald ^g	Sig. (p)	Sample
Demographic variables (9 IVS): Regret Index^a = 27.2%; Specificity^b = 97.2%; Sensitivity^c = 1.85%; PPV^d = 50%; NPV^e = 72.8%; PC^f = 72.6% (χ² = 8.607; df = 9; n = 197; p = 0.474)
Age	−0.016	.984	0.014	1.27	0.130	200
Gender	0.462	1.59	0.320	2.08	0.075	200
Educational level	−0.009	0.991	0.058	0.022	0.441	200
Live alone	0.043	1.04	0.430	0.010	0.461	197
Live with spouse/partner	−0.218	0.804	0.320	0.465	0.248	197
Live with child(ren)	0.396	1.49	0.347	1.30	0.127	197
White	0.634	1.89	0.452	1.97	0.161	200
Black	−0.564	0.569	0.577	0.954	0.329	200
Hispanic	−0.508	0.602	0.661	0.590	0.443	200
Functional status (4 IVs): Regret Index = 19.6%; Specificity = 94.8%; Sensitivity = 29.4%; PPV = 68.2%; NPV = 80.4%; OPC = 76.9% (χ² = 15.396; df = 4; n = 186; p = 0.004)
Morse Fall Scale (MFS)	0.005	1.01	0.008	0.446	0.253	200
Palliative Performance Scale (PPS)	0.029	1.03	0.016	3.12	0.039	199
Activities of Daily Living Scale (ADL)	−0.140	0.869	0.087	2.60	0.054	200
Use of ambulatory aid	−.499	0.607	0.321	2.42	0.060	200
History of falls (2 IVs): Regret Index = 25.6%; Specificity = 99.3%; Sensitivity = 7.4%; PPV = 80%; NPV = 74.4%; OPC = 74.5% (χ² = 8.242; df = 2; n = 200; p = 0.016)
History of fall	0.666	1.946	0.330	4.08	0.022	200
Number of falls past year prior to hospice admission	0.155	1.168	0.066	5.44	0.010	200
Symptoms (5 IVs): Regret Index = 24.2%; Specificity = 96.6%; Sensitivity = 16.7%; PPV = 64.3%; NPV = 75.8%; OPC = 75% (χ² = 11.403; df = 5; n = 200; p = 0.044)
Interaction of freq. and severity of dry mouth	0.128	1.14	0.053	5.74	0.009	200
Interaction of freq. and severity of dyspnea	0.077	1.08	0.037	4.31	0.019	194
Cross interaction of freq. of dry mouth and severity of dyspnea	0.118	1.13	0.052	5.18	0.012	200
Frequency of coughing	0.269	1.31	0.127	4.50	0.017	200
Distress of nausea	−0.160	0.852	0.207	0.595	0.220	200
Medications (8 IVs): Regret Index = 27.4%; Specificity = 95.2%; Sensitivity = 1.9%; PPV = 22.2%; NPV = 72.6%; OPC = 70.4%χ² = 10.643; df = 8; n = 199; p = 0.301)
Analgesics	−0.009	0.991	0.348	0.001	0.979	199
Antihypertensives	0.251	1.29	0.320	0.613	0.493	200
Tranquilizers	0.328	1.39	0.573	0.328	0.567	200
Diuretics	0.072	1.08	0.335	0.046	0.830	200
Antidepressants	0.700	2.01	0.350	4.01	0.045	200
Benzodiazepines	0.252	1.29	0.355	0.505	0.477	200
Laxatives	−0.027	0.973	0.319	0.007	0.931	200
Anticonvulsants	−20.3^g	0.000^g	118^g	0.000^g	0.023	200
Attitudinal/behavioral disposition (5 IVs): Regret Index = 25.5%; Specificity = 97.9%; Sensitivity = 13%; PPV = 62.5%; NPV = 74.5%; OPC = 74.0% (χ² = 14.81; df = 5; n = 200; p = 0.011)
Fear of losing independence	0.084	1.09	0.042	3.98	0.023	200
Try to do things by myself to avoid asking for help	0.366	1.44	0.481	6.74	0.005	200
Feel uneasy about asking for help from other people	0.070	1.07	0.042	2.70	0.050	200
Interaction of avoiding and feeling uneasy about asking for help	0.013	1.01	0.004	10.5	0.001	200
Attempt to complete tasks by myself	−0.056	0.945	0.191	0.087	0.768	200
Enviromental risk/protective factors (2 IVs): Regret Index = 27%; Specificity = 100%; Sensitivity = 0%; PPV = 0%; NPV = 73.0%; OPC = 73% (χ² = .654; df = 2; n = 200; p = 0.721)
Risk condition in the living quarter	0.091	1.095	0.129	0.495	0.241	200
Protective conditions in the living quarter	−0.041	0.960	0.109	0.144	0.353	200

Regret Index refers to percent of patients who actually fell among all patients predicted as no fall.

Specificity is the percent of no fall correctly predicted among all patients who actually did not fall.

Sensitivity is the percent of falls correctly predicted among all patients who actually fell.

Positive predictive value is the percentage of patients predicted as a fall by the model and observed as a fall among all patients predicted as a fall by the model.

Negative predictive value is the % of patients observed as no fall among all patients predicted as no fall by the model.

Overall percentage correct is the percentage of correct classifications made (falls and no falls) by the model over all predictions made (or cell entries).

None of the eight patients who took anticonvulsants fell in the sample, which resulted in having a very large B and zeroes for odds change Exp(B) and Wald statistic.

B, logistic regression coefficient called logit, which is log of the odds. The odds of an event occurring is the ratio of the probability that it will occur to the probability that it will not.

S.E, Standard error of B.

Exp(B), Expected change in B expressed in odds, where Exp(B) less than 1 and more than 1 indicate decreased and increases odds ratio of having a fall, respectively.

Functional status measures

Four functional status measures were examined at the bivariate level. First, an attempt was made to determine the degree to which the Morse Fall Scale (MFS)²⁵ is capable of predicting a fall among hospice patients. From the conceptual model, it was expected that patients with a larger MFS value would have lower risk of falls. On the contrary, the MFS developer assumed a positive relationship between the Morse Scale score and patient falls in the general population. Table 3 shows that the MFS score had random effect on patient fall (B = 0.005; odds = 1.005; p = 0.253). However, the Palliative Performance Scale (PPS) has shown positive relationship, where the increasing level of patient's performance level (higher PPS score) is associated increased risk of fall (B = 0.029; odds = 1.029; p = 0.039). Similarly, the Activities of Daily Living (ADL) scale had nearly significant impact on patient falls (B = −0.140; odds = 0.869; p = 0.054), where greater degree of physical dependence (greater ADL score) is associated with lower risk of a fall. Finally, patient use of an ambulatory aid such as crutches, cane, walker, or furniture (i.e., manifested markers of physical dependence) was found to have a lower risk of a fall (B = −0.499; odds = 0.607; p = 0.06). The forced entry of all four IVs of functional status has resulted in the PPV at 68.2%, NPV at 22%, RI at 19.6%, specificity at 94.8%, sensitivity at 29.4%, and OPC at 76.9%. It is, therefore, generalized that cognitively intact hospice patients who are more functional are more likely to fall, whereas patients who routinely depend on an ambulatory aid are less likely to fall (χ² = 15.396; df = 4; n = 186; p = 0.004).

History of falls

Patients who had a history of falls prior to hospice admission (B = 0.666; odds = 1.946; p = 0.022) and those with more frequent falls during the past year were observed to be more likely to fall (B = 0.155; odds = 1.168; p = 0.01). In a multivariate LRA, these two IVs together have shown PPV at 80%, NPV at 74.4%, RI at 25.6%, specificity is 99.3%, sensitivity at 7.4%, and OPC at 74.5%. Accordingly, a history of past falls prior to hospice admission is considered to be one of the important determinants of future patient falls in hospice care (χ² = 8.242; df = 2; n = 200; p = 0.016).

Symptoms

As described by the conceptual model, domain-specific patient symptoms such as dry mouth and cough are expected to have an increased risk of falls. As shown in Table 3, the interaction of frequency and severity of dry mouth had a higher odds of having a fall (B = 0.128; odds = 1.136; p = 0.009). Similarly, the interaction of frequency and severity of dyspnea also had an increased effect on patient falls (B = 0.077; odds = 1.08; p = 0.019). A cross-symptom interaction of frequency of dry mouth and the severity of dyspnea had an increased odds of having falls (B = 0.118; odds = 1.125; p = 0.012), as well as more frequent coughing was associated with increased patient falls (B = 0.269; odds = 1.308; p = 0.017). The directions of association were all consistent with the directions specified by the conceptual model. Contrary to the model specification, the distress of nausea had no impact on patient falls (B = −0.160; odds = 0.852; p = 0.22) at the bivariate level, although the direction of the hypothesis is consistent with the model specification. In a multivariate LRA, these five MSAS symptom variables were able to register PPV at 64.3%, NPV at 75.8%, RI at 24.2%, specificity at 96.6%, sensitivity at 16.7%, and OPC at 75%. Based on these observations, it is generalized that patient symptoms that make patients move or engage in non-routine movements have a significant impact on patient falls (χ² = 11.403; df = 5; n = 200; p = 0.044).

Medications

None of the medications listed had a significant impact on patient fall except the following: Antidepressants had significant positive impact on patient falls (B = 0.700; odds = 2.014; p = 0.045), meaning that cognitively intact hospice patients taking antidepressants are more likely to fall than those who did not take these medications. While 25% of the sample patients took antidepressants, only 4% of these were tricyclic antidepressants. In addition, hospice patients taking anticonvulsants are observed to be less likely to fall than those who do not use the drug (B = −20.3; p = 0.023). In the sample, there were eight patients who took anticonvulsants and none fell during the study period. In a multivariate LRA, these eight medication variables were able to register PPV at 22.2%, NPV at 72.6%, RI at 27.4%, specificity is 95.2%, sensitivity at 1.9%, and OPC at 70.4% (χ² = 10.643; df = 9; n = 199; p = 0.301). Based on these observations, it is generalized that patient medications have an insignificant impact on patient falls among cognitively intact hospice patients.

Attitudinal/behavioral dispositions

Consistent with the conceptual model, patients who fall are more likely to have a greater fear-of-losing- independence, FOLI (B = 0.084; odds = 1.087; p = 0.023), avoid asking for help (B = 0.366; odds = 1.442; p = 0.005), feel uneasy about asking for help (B = 0.07; odds = 1.072; p = 0.05), and an interaction of the latter two IVs (B = 0.013; odds = 1.013; p = 0.001). Unexpectedly, however, patients who “attempt to complete tasks by themselves” had an insignificant impact on patient falls (B = −0.056; odds = 0.945; p = 0.768). It seems that patients who simply “attempt to do things by themselves”—by and of itself— is not a reliable estimator for predicting future falls since “being independent” or “trying to be independent” is not equivalent to having the FOLI. As such, “attempting to complete tasks themselves” is conceived not as important in explaining a patient fall as those who actually have the FOLI and purposely avoid asking for help from others. While employing all domain-specific IVs, the PPV is registered at 62.5%, NPV at 74.5%, RI at 25.5%, specificity is 97.9%, sensitivity at 13%, and OPC at 74%. Based on these results, it is shown that those cognitively intact hospice patients who have a FOLI and ask-avoidance dispositions have a greater likelihood of a fall than those without these attitudes and behavior (χ² = 14.81; df = 5; n = 200; p = 0.011).

Environmental and intrinsic risk/protective factors

Contrary to the model specifications, neither the number of risk conditions (B = 0.091; odds = 1.095; p = 0.241) nor the number of protective conditions found in the homes of hospice patients (B = −0.041; odds = 0.96; p = 0.353) had a significant impact on the patient falls, although the directions shown are consistent with the model specifications. While using these two IVs, the PPV reached is 0%, NPV is 73%, RI is 27%, specificity is 100%, sensitivity is 0%, and the OPC is 73%. There was not a great deal of variability in the risk and protective conditions found in the patients' home environments and, therefore, risk and protective conditions seen at homes are not important determinants of future fall status among these patients. Based on these observations, it is concluded that risk and protective conditions found in the living quarters of cognitively intact hospice patients have little impact on patient falls (χ² = 0.654; df = 2; n = 200; p = 0.721). This is contrary to the usual thinking, which assumes that there is a positive correlation between the incidence of falls and the numbers of hazard conditions found in the patients' living environments.

The final model

As noted, forecasting model selection must be based on an eclectic use of theory- and data-driven methods within a predetermined set of empirical criteria. Table 4 shows the final forecasting model chosen. It is an eight-variable model chosen by calibrating the optimum cutoff point for predicting falls, which was set at 0.44. Patients with cutoff point ≥0.44 are predicted as “fall” and those with <0.44 were predicted as “no fall.” Influential IVs were selected via forward stepwise method with probability-in (PIN) and probability-out (POUT) values set at 0.13 and 0.14, respectively. The final model chosen is a z prediction equation of the form: \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland,xspace}\usepackage{amsmath,amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document}\begin{align*} z = - 0.2771 + 0.574 \ \hbox{\rm (avoid asking for help)} + 0.281 \hbox{\rm (feel uneasy about asking for help)} + 0.986 \hbox{\rm (prior history of a fall)} + 0.149 \hbox{\rm (frequency of dry mouth} \times \hbox{\rm severity of dry mouth)} - 0.445 \hbox{\rm (nausea distress)} - 0.569 \hbox{\rm (use of ambulatory} \hbox{aid)} - 0.422 \hbox{\rm (attempt to complete tasks by myself).}\end{align*}\end{document}

Table 4.

Logistic Regression Model for Predicting Falls among Cognitively Intact Hospice Patients

Variables	B	S.E.	Wald	df	Sig.	Exp(B)
Constant	−2.77	0.984	7.935	1	0.005	0.063
Avoid asking for help	0.574	0.227	6.371	1	0.012	1.775
Uneasy asking for help	0.281	0.159	3.130	1	0.077	1.324
History of falls	0.986	0.372	7.032	1	0.008	2.680
Interaction of frequency and severity of dry mouth	0.149	0.060	6.246	1	0.012	1.160
Distress of nausea	−0.445	0.231	3.721	1	0.054	0.641
Patient uses ambulatory aid	−0.569	0.356	2.547	1	0.111	0.566
Complete tasks by myself	−0.422	0.239	3.131	1	0.077	1.775

OPC, 78%; (χ² = 29.62; df = 7; n = 200; p = 0.0005).

B = logistic regression coefficient or log odds, indicating the amount of change in log odds when the values of the other IVs remain the same; S.E., standard error of B; df, degrees of freedom; Sig., significance of Wald statistic using two-tailed test; Exp(B), odd change by a factor Exp(B) when B changes by one unit.

The probability of each patient having a fall, P(FALL), is computed in terms of z, where P(FALL) = 1 / (1 + 2.718^-z). Applying the z prediction equation to a patient John Doe, who strongly agreed (value = 5) to question 1 on “avoid asking”; agree (value = 4) to Q2 on “uneasy asking”; had a prior history of falls (value = 1) to Q3; responded as “frequent” (value = 3) to the question on dry mouth during the past week (Q4); “slight” (value = 1) to question on severity of dry mouth (Q5); had no distress (value = 0) on nausea (Q6); did not use ambulatory aid (value = 0) on Q7; and who disagreed (value = 2) to Q8 on the “attempt to complete tasks by myself,” the z score computed is 1.81 and the P(FALL) is 0.86. Hence, John Doe is considered highly likely to have a fall during his stay in hospice care. With further validation of this model, users will be in a position to predict the probability of falls for cognitively intact patients under hospice care and make correct fall predictions approximately 78% ((135 + 21)/200) of the time.

As shown by classification table (Table 5), the prevalence rate of a fall is 27% ( = 54/200). The diagonal cell entries are correct predictions made by the model where there are 135 patients who are true-negatives (TN) and 21 patients who are true-positives (TP). The remaining, off-diagonal, cells have 33 patients who are false-negatives (FN) and 11 patients who are false-positives (FP). The specificity of the final model is 92.5% ( = 135/146, i.e., the degree to which no fall is correctly predicted) and the sensitivity is 38.9% ( = 21/54). Positive predictive value (PPV), the probability that a person has actually fell given that the patient was predicted as positive (fall) by the forecasting model, is 65.6% ( = 21/32). Negative predictive value (NPV), the probability that a person did not fall given that the patient was predicted as negative (no fall) by the forecasting model, is 80.4% ( = 135/168). For this study, an optimal cutoff-point for the forecasting model is set at 0.44, which yielded the smallest RI score 19.6% ( = 33/168), the largest sensitivity score (38.9% = 21/54), and the largest NPV (80.4% = 135/168) when compared to all other domain-specific measurements. The final model chosen has seven IVs, one of which is an interaction variable. These seven variables together were able to predict patient fall status correctly and prospectively 78% of the time (χ² = 29.62; df = 7; n = 200; p = 0.0005) with the least amount of regret when compared to all other domain-specific models.

Table 5.

Crosstab of Hospice Patients Observed on Fall Status by Fall Classification Predicted by a Seven-Variable Forecasting Model

		Predicted fall status by forecasting model
Hospice patients observed		No fall	Yes fall	n % correct
Actual fall status at follow up	No Fall	135 (TN)	11(FP)Specificity = 92.5% = 135/146	146
	Yes Fall	33 (FN)	21 (TP)	54
	n % correct	168 NPV = 80.4% = 135/168	32 PPV = 65.6% = 21/32	Sensitivity = 38.9% = 21/54200 OPC = 78% = (135 + 21)/200

Negative predictive value (NPV) is 80.4% ( = 135/168). Positive predictive value (PPV) is 65.6% ( = 21/32). Regret index value is 19.6% ( = 33/168); Specificity of the testing model is 92.5% ( = 135/146); Sensitivity of the testing model is 38.89% ( = 21/54). OPC is 78% ( = (135 + 21)/200).

Discussion

In a meta-analysis on the etiology of falls, Tinetti and Kumar³⁸ have summarized 17 risk factors of falls originating from 33 independent studies conducted on falls among community-living older adults. From these, the risk factors cited as having the strongest impact on falls were: previous falls, balance impairment, decreased muscle strength, visual impairment, medications (>4 or psychoactive medication use), gait impairment, and depression. Review of interventions tested yielded mixed results. None of the studies described addressed the attitudinal attribute as a potential risk factor. While clinical practice guidelines have been published regarding the prevention of falls among community dwelling elders, there are no subset recommendations for the hospice population.³⁹ The current study, however, tested how demographics, patient's functional status, falls history, patient symptoms, medications utilized, patient attitudes on the FOLI, and environmental conditions are related to falls among hospice patients in a multivariate setting. The model on hospice patient falls formulated by this study is superior to any of the existing models that explain patient falls both on conceptual and empirical means, although the current model is applicable to cognitively intact hospice patients only.

Although the FOLI had a significant impact on patient falls at the bivariate level (B = 0.084; odds = 1.087; n = 200; p = 0.023), the original relationship found between FOLI and patient fall was rendered spurious, controlling for other independent variables included in the z-prediction equation (B = 0.014; odds = 1.014; n = 200; p = 0.873). This however does not mean that the FOLI has no impact on hospice patients' falls. It simply shows that the FOLI has significant interactions with other IVs chosen in the final model. Accordingly, any research on fall incidents among hospice patients that ignores the attitudinal construct is likely to have poorer predictive capability in forecasting hospice patients who fall. This suggests that fall prevention/intervention projects involving hospice patients will be more effective if hospice professionals help patients feel comfortable with asking for help when help is available and refraining from completing tasks independently just to avoid asking for help from others.

The eight-variable model is composed of three attitudinal questions, one on the history of falls, three on symptoms of illnesses, and one on a functionality measure involving the use of an ambulatory aid. However, none of the medication variables was included in the final model. It is very important to note that hospice patients who are mobile are more likely to fall than immobile hospice patients. This finding contradicts some of the traditional assumptions about falls, where falls are expected to occur more frequently among patients that are more functionally impaired, weak, and immobile. As all knowledge is always contextually determined,⁴⁰ it must be noted that the sampled population is already quite functionally impaired, weak and somewhat immobile. If any fall is to occur, it does happen to people who are more mobile than immobile because there will be no falls without any movements. Thus, it follows that falls in general will be higher among more active and mobile than inactive and immobile groups of people within a specific group within which falls are measured.

It is also important to note that individualized risk/protective factors observed in the patients' environments have no impact on falls among the sampled patients. Further research involving protective environmental factors in the homes of hospice patients is expected to have little impact on reducing the risk of falls in this patient group. Six of the 36 IVs hypothesized as having directional impact on patient falls in the bivariate model were not statistically significant at the 0.05 level. They are: age (p = 0.13), gender (p = 0.075), distress of nausea (p = 0.22), attempt to complete tasks by oneself (p = 0.77), risk (p = 0.24), and protective (p = 0.35) conditions found in the living quarters of hospice patients. Of these, the directions of logistic regression Bs are consistent with the conceptual model on the etiology of falls with the only exception of patients who attempt to “complete tasks by myself” (B = −0.056; p = 0.768). The study design measured observations of patient falls, which occurred at a later point in time following the initial patient assessment. Therefore, this study provided generalizations that are more valid than studies where the patient characteristics are extracted concurrently or among those who already had a fall.

In order to secure more valid responses, this study by necessity only included patients that are cognitively intact. While excluding cognitively impaired patients may have decreased generalizability of the study results, this also may have increased the validity and interpretation of the data responses received. One may argue that hospice patients with more cognitive impairment may be more likely to fall because they are less aware of their surroundings and functional limitations. The etiology of a fall in the current study specifically addresses against such a suggestion by noting that cognitively intact hospice patient falls are largely accidents that are more likely to happen to patients with better performance status. These patients are more likely to engage in non routine activities, have greater fear of losing independence, feel uneasy about and purposely avoid asking for help, suggesting cognitive factors influence patient falls in these patients. It is also possible that hospice patients with cognitive impairment might be so globally impaired that they lack spontaneous, nonroutine movements that typically might put them at risk of falls. It would appear that a separate study that included patients with cognitive impairment would likely test different hypotheses regarding the etiology of patient falls based on a different theoretical framework.

It is also obvious that there may be many other variables that may have influenced patient falls, such as drug interactions, other measures on patient classification and patient functional status. Also, it was thought that the study could have been improved had we had more patients who fell, yielding enough cases to establish more sensitive and contextually appropriate generalizations. Finally, the study was limited to a single hospice and, as such, would require validation of the model in other hospice settings.

In general, the final forecasting model chosen is simple but meaningful and, with additional validation, it may be useful to clinicians and non-clinicians. The conceptual model on the etiology of falls among cognitively intact patients is to be used to guide caregivers in evaluating the probability of such patients having a fall under hospice care. Addressing the risk factors in specific high-risk patients should improve the management of falls among hospice patients by providing higher quality care at home and limiting utilization of emergency room or hospital visits associated with these falls. This will decrease associated morbidity and mortality and ultimately result in significant cost savings. More than anything else, this study set forth a somewhat different approach and identified unique risk factors for falls not previously described in the literature, highlighting the need to evaluate specific subpopulations for different sets of risk factors that are contextually determined.

Footnotes

Acknowledgments

This work was supported by the funding from HPC Healthcare Inc. and the Center for Hospice, Palliative Care, & End-of-Life Studies at the University of South Florida. For those interested in the secondary analysis, direct requests to: kims@hpchealthcare.org.

Author Disclosure Statement

No competing financial interests exist.

References

Hoyert

, Kochanek

, Murphy

. Deaths: Final data for 1997. Natl Vital Stat Rep, 1999; 47:19.

U.S. Census Bureau, (NP-D1-A) Projections of the Resident Population by Age, Sex, Race, and Hispanic Origin: 1999 to 2100. Population Projections Program, January 2000.

Tinetti

, Inouye

, Gill

, Doucette

. Shared risk factors for falls, incontinence, and functional dependence. Unifying the approach to geriatric syndromes. JAMA, 1995; 273:1348–1353.

Rabin

. Falls and gait disorders. Abrams

, Beers

, Berkow

. The Merck Manual of Geriatrics, 2nd. Rahway, NJ: Merck Sharp & Dohme Research Laboratories, 1995; 65–78.

National Hospice and Palliative Care Organization. NHPCO Facts and Figures: Hospice Care in America, 2009. Alexandria, VA: National Hospice and Palliative Care Organization, 2009.

Rubenstein

, Josephson

, Robbins

. Falls in the nursing home. Ann Intern Med, 1994; 121,6:442–451.

Tinetti

, Speechley

, Ginter

. Risk factors for falls among elderly persons living in the community. N Engl J Med, 1988; 319,26:1701–1707.

Alexander

, Rivera

, Wolf

. The cost and frequency of hospitalization for fall-related injuries in older adults. Am J Public Health, 1992; 82,7:1020–1023.

Englander

, Hodson

, Terregrossa

. Economic dimensions of slip, fall injuries. J Forensic Sc, 1995; 41,5:733–746.

10.

Rizzo

, Friedkin

, Williams

, Nabors

, Acampora

, Tinetti

. Health care utilization and costs in a Medicare population by fall status. Med Care, 1998; 36,8:1174–1188.

11.

Rawsky

. Review of the literature on falls among the elderly. J Nurs Scholarsh, 1998; 30,1:47–52.

12.

Ganz

, Bao

, Shekelle

, Rubenstein

. Will my patient fall? JAMA, 2007; 297:77–86.

13.

Gehlsen

, Whaley

. Falls in the elderly: Part II, balance, strength, and flexibility. Arch Phys Med Rehab, 1990; 71:739–741.

14.

Groves

, Lavori

, Rosebaum

. In-hospital injuries of medical and surgical patients: The predictive effect of a prior injury. Psychosom Med, 1992; 54,3:264–274.

15.

Graafman

, Ooms

, Hofstee

, Bezemer

, Bouter

, Lips

. Falls in the elderly: A prospective study of risk factors and risk profiles. Am J Epidemiol, 1996; 143,11:1129–1136.

16.

Gluck

, Wientjes

, Rai

. An evaluation of risk factors for in-patient falls in acute and rehabilitation elderly care wards. Gerontology, 1996; 42:104–107.

17.

Mathers

, Weiss

. Incidence, characteristics of fall-related emergency department visits. Acad Emerg Med, 1998; 5,11:1064–1070.

18.

Centers for Disease Control and Prevention. CDC Fact Book 2000/2001. www.cdc.gov. 2010 October 5.

19.

Morse

. Preventing Patient Falls. Thousand Oaks, CA: Sage Publications, 1997.

20.

Joint Commission: FAQs for the 2007 National Patient Safety Goals [online]

2007. www.jointcommission.org/NR/rdonlyres/D4844675-25D7-4B5B-A47DC549D939F9E5/0/07_NPSG_FAQs_9.pdf. 2010 October 5.

21.

Centers for Disease Control and Prevention. The state of aging and health in America [report online] 2007. www.cdc.gov/aging/pdf/saha_2007.pdf. 2010 October 5.

22.

Pennsylvania Patient Safety Advisory. Medication assessment: One determinant of falls risk. Pa Patient Saf Advis, 2008; 5:16–18.

23.

Mycek

, Harvey

, Champe

. Pharmacology, 2nd. Philadelphia, PA: Lippincott Williams & Wilkins, 2000.

24.

Perell

, Nelson

, Goldman

, Luther

, Prieto-Lewis

, Rubenstein

. Fall risk assessment measures: an analytic review. J Gerontol Biol Med Sci, 2001; 56:761–766.

25.

Morse

, Morse

, Tylko

. Development of a scale to identify the fall-prone patient. Can J Aging, 1980; 8:366–377.

26.

Anderson

, Downing

, Hill

, Casorso

, Lerch

. Palliative performance scale: A new tool. J Palliat Care, 1996; 12:5–11.

27.

Katz

, Ford

, Maskowitz

, Jackson

, Jaffe

. Studies of illness in the aged: The index of ADL—a standardized measure of biological and psychosocial function. JAMA, 1963; 185:914–919.

28.

Portenoy

, Thaler

, Kornblith

. The Memorial Symptom Scale: An instrument for the evaluation of symptom prevalence, characteristics and distress. Eur J Cancer, 1994; 30A:1326–1336.

29.

HPC Healthcare, Inc. Falls Occurrence Report Fourth Quarter 2005. Tampa, FL, 2005.

30.

Cohen

. Statistical Power Analysis for the Behavioral Sciences. New York: Academics Press, 1977.

31.

Pfeiffer

. A short potable mental status questionnaire for assessment of organic brain deficit in elderly patients. J Am Geriatr Soc, 1975; 23:433–441.

32.

SPSS. Advanced Statistics, Version 6.1. Chicago, IL: SPSS Inc., 1994.

33.

Hozo

, Djulbegovic

. When is diagnostic testing inappropriate or irrational? Acceptable regret approach. Med Decis Making, 2008; 28,4:540–553.

34.

Djulbegovic

, Hozo

, Schwartz

, McMasters

. Acceptable regret in medical decision making. Med Hypotheses, 1999; 53:253–259.

35.

Djulbegovic

, Frohlich

, Bennett

. Acting on imperfect evidence: How much regret are we ready to accept? J Clin Oncol, 2005; 23,28:6822–6825.

36.

Djulbegovic

, Hozo

. When should potentially false research findings be considered acceptable? PLoS Med, 2007; 4:e26.

37.

Tymstra

. The imperative character of medical technology and meaning of “anticipated decision regret.” Int J Technol Asses Health Care, 1989; 5:207–213.

38.

Tinetti

, Kumar

. The patient who falls. JAMA, 2010; 303:258–266.

39.

AGS Newsletter: American Geriatrics Society, British Geriatrics Society S/BGS Clinical Practice Guidelines: Prevention of Falls in Older Persons. www.americangeriatrics.org/education/cp_indexshtml. 2010 October 5.

40.

Weber

. The Methodology of the Social Sciences. Shils

, Finch

. New York: Free Press, 1949.