BMJ No 7115 Volume 315
Papers Saturday 25 October 1997
Development and evaluation of evidence based risk assessment tool (STRATIFY) to predict which elderly inpatients will fall: case-control and cohort studies
D Oliver, M Britton, P Seed, F C Martin, A H Hopper
Abstract
Objectives: To identify clinical characteristics of elderly inpatients that predict their chance of falling (phase 1) and to use these characteristics to derive a risk assessment tool and to evaluate its power in predicting falls (phases 2 and 3).Design: Phase 1: a prospective case-control study. Phases 2 and 3: prospective evaluations of the derived risk assessment tool in predicting falls in two cohorts.
Setting: Elderly care units of St Thomas's Hospital (phase 1 and 2) and Kent and Canterbury Hospital (phase 3).
Subjects: Elderly hospital inpatients (aged 65 years and over): 116 cases and 116 controls in phase 1, 217 patients in phase 2, and 331 in phase 3.
Main outcome measures: 21 separate clinical characteristics were assessed in phase 1, including the abbreviated mental test score, modified Barthel index, a transfer and mobility score obtained by combining the transfer and mobility sections of the Barthel index, and several nursing judgments.
Results: In phase 1 five factors were independently associated with a higher risk of falls: fall as a presenting complaint (odds ratio 4.64 (95% confidence interval 2.59 to 8.33); a transfer and mobility score of 3 or 4 (2.10 (1.22 to 3.61)); and primary nurses' judgment that a patient was agitated (20.9 (9.62 to 45.62)), needed frequent toileting (2.48 (1.08 to 5.70)), and was visually impaired (3.56 (1.26 to 10.05)). A risk assessment score (range 0-5) was derived by scoring one point for each of these five factors. In phases 2 and 3 a risk assessment score g2 was used to define high risk: the sensitivity and specificity of the score to predict falls during the following week was 93% and 88% respectively in phase 2 and 92% and 68% respectively in phase 3.
Conclusion: This simple risk assessment tool predicted with clinically useful sensitivity and specificity a high percentage of falls among elderly hospital inpatients.
Introduction
Falls are common among elderly hospital inpatients.(1-2) For the patient, consequences may include fracture,(3-4) fear of falling,(5) anxiety and depression,(6) and loss of confidence,(7) all of which lead to greater disability. Falls by inpatients are associated with increased duration of stay in hospital and a greater chance of unplanned readmission or of discharge to residential or nursing home care.(8)
Successful rehabilitation to minimise long term disability of elderly people requires that staff aim to reduce patients' dependency and to increase their autonomy during recovery from acute illness when it is associated with disability. The occurrence of some falls is an unwelcome but probably inevitable consequence of encouraging patients to regain mobility early after acute illness. None the less, there may be simple measures that could reduce the incidence of falls(2)(9) without the need for physical restraints, sedation, excessive supervision, or other measures that undermine a patient's dignity and independence.
A strategy which has proved successful in the prevention of pressure sores(10) is to select patients at high risk and target prevention strategies. Various clinical characteristics (over 400 in total on systematic review(11)) have been shown to be associated with an increased incidence of falls occurring at home or outdoors. Examples include use of particular drugs, muscle weakness, unstable gait, postural hypotension, and poor visual acuity.(12-14) Some of these characteristics may not be evident early in an inpatient episode or may require specialised equipment or diagnostic skills. They are therefore unsuitable for use in a routine clinical assessment applicable to large numbers of hospital inpatients. Moreover, most factors predictive of falls among elderly people in the community may not apply to hospital inpatients, where recovery from acute illness that is associated with changing mobility is more common.
Studies to identify risk factors for elderly inpatients falling have shown that a few readily assessable risk factors may predict a large proportion of these falls.(2)(15-17) Since the occurrence of falls depends on patient characteristics (case mix) and institutional characteristics such as clinical and nursing practice,(2-3) risk factors may be specific to particular hospital units. Most of the presently available information is from the United States and often from nursing homes whose populations are more clinically stable. There is little evidence from British acute hospital wards for elderly patients.
We report a three phase investigation. In phase 1 we conducted a
case-control study to discover which risk factors were significantly
associated with falls occurring in the Elderly Care Unit at St
Thomas's H Settings Phase 3 was conducted at the Kent and Canterbury Hospital, a 500 bed
district general hospital with two acute and four rehabilitation wards
for elderly patients. The service has an age related admission policy
(over 75 years). On the acute wards the patients' mean age is 83
years, their mean length of stay is seven days, and there were 3,000
admissions in 1995. On the four wards in the separate rehabilitation
unit patients' mean age is 83 years and their mean length of stay is
14 days.
Phase 1: case-control study Within 48 hours of each fall the patient's primary nurse was
interviewed and the case notes reviewed. For each fall and each
control, 21 separate pieces of information were recorded (see box).
These included the abbreviated mental test score,(18)
modified Barthel index,(19) and a transfer and mobility
score (range 0-6) obtained by combining the transfer and mobility
sections of the Barthel index (each with ranges 0-3). Several nursing
judgments were also obtained. No attempt was made to standardise the
formation of these judgments as the aim was to produce an easily
useable risk assessment tool.
Characteristics (putative risk factors) listed for patients
who fall and controls (phase 1) Phase 2: investigation of risk assessment tool in local cohort
(St Thomas's Hospital) Phase 3: investigation of risk assessment tool in remote cohort
(Kent and Canterbury Hospital) Statistical analysis Phases 2 and 3 - We determined the specificity and
sensitivity of STRATIFY scores 2 and over and 3 and over (0.5) in predicting falls
in the following week from the proportions of fallers and non-fallers
correctly identified. Exact confidence intervals for the proportions
were calculated. Each week was treated as a separate datum, and no
adjustment was made for repeated measures on the same patient.
STRATIFY risk assessment tool Do you think the patient is (questions 2-5)
2 Agitated?
3 Visually impaired to the
extent that everyday function is affected?
4 In need of especially frequent toileting?
5 Transfer and mobility score of 3 or 4?
Total score In selecting only five variables for the final risk assessment score
(see box), we bore several considerations in mind. We chose only
variables that showed significant differences in univariate analysis.
Two variables that gave significant results in multiple regression were
not used. At the time of a patient's admission the nurses felt that
they were less able to assess instability of gait than to use the
objective transfer and mobility score. Treatment with antiarrhythmic
drugs was not used as some ward nurses might not easily be able to
assess which drugs were in this category.
Phase 2: local validation study
Phase 3: remote validation study
The initial case-control study showed that seven of the
putative risk factors for falling were significantly more prevalent in
the patients who fell than in the controls. Five of these factors were
used to construct the risk assessment tool. These were the factors that
were readily assessable by ward nurses based on their day to day
observation of patients and could be performed shortly after admission
to hospital. This conferred the advantage of generating a pragmatic
risk assessment tool, taking about one minute per patient per week and
requiring no formal measurements, additional training, or equipment. In
the validation study at St Thomas's Hospital a score of 2 as a
definition of "high risk" identified 93% of falls that occurred in
the following week. At any one time only five or six patients on a 26
bed ward had a score of 2 or more. This could allow targeting of
strategies to prevent falls on a small group of ward patients without
"missing" many future fallers. In the validation study at Kent and
Canterbury Hospital a risk score of 2 or more identified 70% of
patients who subsequently fell, with a high (98%) negative predictive
value, so again few future fallers were not
identified.
At least three factors might explain the reduced power of the risk
assessment tool at Kent and Canterbury Hospital. Firstly, the
predictive power of risk factors is likely to be specific to one unit
or patient group. Secondly, the risk assessment tool was completed by
the nurses themselves, rather than by interview of the nurses, which
might have reduced consistency of assessment. Thirdly, nurses aware of
the predictive power of the score demonstrated elsewhere might have
altered their care of "high risk" patients, thus preventing some
falls (the Hawthorne effect(20) ).
In choosing the appropriate "cut off" score that defines high risk,
there is a trade off between a score that confers high sensitivity or
high specificity. It may be that a prospective validation is necessary
in any hospital unit before use of the risk assessment tool since case
mix, ward design, and nursing philosophy and skills vary widely. A high
cut off score that gave high specificity would lose sensitivity,
thereby missing many patients who would fall. However, a low score,
with high sensitivity, might define more than half the ward patients as
high risk, which would be of no practical benefit.
There are examples of unit based programmes to prevent falls in
Ireland(21) and the United States(2)(15)(22-24)
which successfully extended standard nursing practice to prevent falls
in inpatients. A similar programme may be effective in Britain.
STRATIFY may be applicable to many acute hospital elderly units.
Further study is needed to determine whether the falls of inpatients
identified as high risk can be prevented by a targeted
intervention.
Department of Elderly
Care (Division of Medicine), Department of
Statistics (Division of Public Health Sciences),
We thank C O'Connor, D Sturdy, and S Rohatgi.
Funding: None.
(Accepted 24 June 1997)
References
1 Rubenstein L, Josephson K, Robbins A. Falls in the nursing
home. Ann Intern Med 1994;121:442-51.
2 Morse K. Preventing inpatient falls. London:
Sage Publications, 1996.
3 De Vicenzo D, Watkins S. Accidental falls in a
rehabilitation setting. Reh
4 Baker S, Harvey A. Fall injuries in the elderly. Clin
Geriatr Med 1985;1:501-12.
5 Murphy J, Isaacs B. The post-fall syndrome. A study of 36
elderly inpatients. Gerontology 1982;28:265-70.
6 Vetter N, Ford D. Anxiety and depression scores in elderly
fallers. Int J Geriatr Psychiatry 1989;4:158-63.
7 Liddle J, Gilleard C. The emotional consequences of falls for
patients and their families. Age Ageing 1984;23(suppl
4):17.
8 Bates D, Pruess K, Souney P, Platt R. Serious falls in
hospitalised patients; correlates and resource utilisation. Am J
Med 1995;99:137-43.
9 Werner P, Koroknay V, Braun J. Individualised care
alternatives in the process of removing physical restraints in the
nursing home. J Am Geriatr Soc 1994;42:321-5.
10 Waterlow J A. A risk assessment card. Nurs Times
1985;81:48, 49-55.
11 Preventing falls and subsequent injury in older people.
Effective Health Care Bull 1996;2(4):1-15.
12 Hindmarsh J J, Estes E J. Falls in older persons. Causes and
interventions. Arch Intern Med 1989;149:2217-22.
13 Kapoor W N. Diagnostic evaluation of syncope. Am J
Med 1991;90:91-106.
14 Nevitt M C, Cummings S R. Risk factors for recurrent non-syncopal
falls. JAMA 1989;261:2663-8.
15 Schmid N. Reducing patient falls; a research-based
comprehensive falls prevention programme. Mil Med
1990:155:202-7.
16 Morse J, Tylko S. Characteristics of the fall prone patient.
Gerontologist 1987;27:516-22.
17 Gluck T, Rai G, Wientjes H. An evaluation of risk factors for
falls in an acute and rehabilitation elderly care ward.
Gerontology 1996;42(2):104-7.
18 Hodkinson H M. Evaluation of a mental test score for assessment
of mental impairment in the elderly. Age Ageing
1972;1:233-8.
19 Collin C, Wade D T, Davies S, Horne V. The Barthel ADL index: a
reliability study. Int Disability Stud 1988;10:61-9.
20 Streiner D L, Normal G R, Blum H M. PDQ
Epidemiology. Toronto: B C Decker, 1989:33-8.
21 Cannard G. Falling trend. Nurs Times
1996;92(2):36-7.
22 Brady R, Chester F, Pierce L, Salter J, Schreck R, Radziewicz
R. Geriatric falls; prevention strategies for the staff. Metro Health
falls prevention programme. J Gerontol Nurs
1993;19(9):26-32.
23 Hendrich A. An effective unit-based fall prevention plan.
J Nurs Qual Assur 1988;3(1):28-36.
24 Morton D. Five years of fewer falls. Am J Nurs
1989;6:204-5.
Subjects and methods
Phases 1 and 2 were conducted at St Thomas's Hospital in London,
an inner city, 700 bed, teaching hospital. The Elderly Care Unit has 96
beds on four wards, admitting patients with a minimum age of 65 years
on the basis of need. One of the four wards is dedicated to stroke
rehabilitation, and the others admit acutely ill and usually disabled
patients with a mean age of 79.5 years and a mean length of stay of 13
days. There were 1,800 admissions in 1995.
From the adverse events recorded in ward incident books, we
identified all falls occurring on the four elderly care wards at St
Thomas's Hospital over a three month period. A fall was defined as
"an incident in which a patient suddenly and involuntarily came to
rest upon the ground or surface lower than their original station."
Each fall was regarded as a new incident, and thus patients who fell
several times were the subject of several data sets. For each fall, a
control patient was selected who was a patient in the next bed and who
had not yet fallen during his or her admission. Fallers and control
patients were not matched for age or sex since these variables were
among the risk factors being investigated. If control patients
subsequently fell, new data were collected to reflect their status on
this occasion as fallers Thus, one patient could contribute two or more
sets of data.
Age Current Barthel index (range 0-20) Current transfer and mobility score (0-6), by combining the transfer
and mobility sections of the Barthel index* Most recent documented abbreviated mental test score (0-10) Use of walking aid? (Yes/No) Catheter or drip in situ? (Yes/No) Was a fall a presenting complaint on admission to hospital or at any
previous time during current admission? (Yes/No) Current medication:
Antidepressants? (Yes/No)
Diuretics?
(Yes/No)
Vasodilators? (Yes/No)
Antiarrhythmics? (Yes/No)
Antiparkinsonian drugs? (Yes/No)
Sedative or hypnotics?
(Yes/No)
Opiates? (Yes/No)
Drugs in two or more of the above
categories? (Yes/No)Nurses' judgments about patient's current clinical
characteristics:
Agitated? (Yes/No)
In need of
especially frequent toileting? (Yes/No)
Has visual impairment which
seems to affect functioning on the ward? (Yes/No)
Has hearing
impairment which seems to affect functioning on the ward? (Yes/No)
Has language impairment which seems to affect functioning on the ward?
(Yes/No)
Has an unstable gait? (Yes/No)*Transfer score: 0=unable, 1=major help
needed (one or two people, physical aids), 2=minor help
(verbal or
physical), 3=independent.
Mobility score: 0=immobile, 1=independent
with aid of wheelchair, 2=walks with help of one person, 3=independent
The five risk factors found to be significantly and
independently associated with falling (see table 1) were tested on a
cohort of patients in the same unit. A si
We obtained a STRATIFY score for all inpatients on the elderly
care wards on admission (within 24 hours) and weekly over an eight week
period. The risk assessment was completed by ward nurses themselves,
who were asked not to instigate specific measures on the basis of the
score obtained. Falls were documented in the same way as for phase
2.
Phase 1 - To compare the fallers and controls we used
Student's t tests for continuously distributed
data, Wilcoxon's non-parametric test for categorical scales
(abbreviated mental test score, Barthel index, and transfer and
mobility scores), and logistic regression for categorical data. To
select variables for construction of the risk assessment tool, we
calculated odds ratios for all differences.
Results
Phase 1: case-control study
One hundred and sixteen falls were recorded. Only seven risk
factors were found to be significantly and independently more prevalent
among fallers than controls (table 1). These were agitation, need for
frequent toileting, unstable gait, visual impairment (all as judged by
the patient's primary nurse), fall as a presenting complaint, and a
transfer and mobility score of 3 or 4. There was no significant
difference between fallers and controls in the median transfer and
mobility score, but the distributions were significantly different (see
figure). A transfer and mobility score of 3 or 4 (that is, the patient
could stand with minimal or no assistance but had limited mobility with
a walking aid) was significantly associated with risk of falling
compared with higher or lower scores (chi2=24.4, P<0.001)
Transfer and mobility scores of elderly inpatients who subsequently fell and of control patients in phase 1 of study
1 Did the patient present to hospital with a fall or has he or
she fallen on the ward since admission?
(Yes=1, No=0)
(Yes=1, No=0)
(Yes=1, No=0)
(Yes=1, No=0)
(Yes=1, No=0)
Table 1 - Prevalence of putative risk factors in elderly
inpatients who fell and control patients
(phase 1). Values are
numbers
(percentages) unless stated otherwise
Characteristic
Controls
(n=116Cases
(n=116)
Odds ratio
(95% confidence interval) Mean
(SD)
age
(years)82.3
(7.4)84.6
(7.0)2.28
(0.42 to 4.14)
Median
(interquartile range) abbreviated mental test
score 7
(4-10) 6
(4-9)-0.5
(-1.5 to 0) Median
(interquartile range) Barthel index 13
(7-17) 10.5
(6.5-14)-2
(-3.5
to 0) Median
(interquartile range) transfer and mobility
score* 4
(2-5.5) 3
(2-4.5)-0.5
(-1 to 0) Transfer and
mobility score of 3 or 4† 34
(29)54
(46)2.10
(1.22 to 3.61) Agitation† 9
(8)74
(64)20.9
(9.62 to 45.62) Fall
as presenting complaint† 23
(20)62
(53)4.64
(2.59 to 8.33) Visual
impairment† 5
(4)16
(14)3.55
(1.26 to 10.05) Frequent
toileting† 9
(8)20
(17)2.48
(1.08 to 5.70) Walking aid 64
(55)71
(61)0.80
(0.61 to 1.04)
Hearing impairment 15
(13)21
(18)1.49
(0.72 to 3.60) UnsTable
gait 40
(34)90
(78)6.58
(3.68 to 11.75) Catheter or drip in situ 11
(9)7
(6)0.61
(0.23 to 1.64)
Drugs: Opiates 11
(9)18
(16)1.75
(0.79 to 3.90)
Sedatives 7
(6)16
(14)2.49
(0.98 to 6.31)
Anticonvulsants 7
(6)12
(10)1.80
(0.68 to 4.74)
Vasodilators 30
(26)23
(20)0.71
(0.38 to 1.31) Diuretics 16
(14)20
(17)1.30
(0.64 to 2.66)
Antidepressants 25
(22)14
(12)0.50
(2.45 to 1.02)
Antiarrhythmics 2
(2)31
(27)20.79
(4.84 to 89.27)
Antiparkinsonian 13
(11)4
(3)0.28
(0.89 to 0.90) Drugs in 2 or more of above
classes 28
(24)31
(27)1.15
(0.63 to 2.07)
*Transfer and mobility score (range 0-6) obtained by combining
the transfer and mobility sections of the Barthel index
(each with
ranges 0-3).
†Characteristics used in the final risk assessment
(STRATIFY)
score. Although differences for antiarrhythmics and unsTable gait were
significant, these factors were not used for methodological reasons
(see main text for details).
The risk assessment tool was completed 395 times on 217 patients,
among whom 71 falls occurred in the week after an assessment (table 2).
A risk score of 2 or more was calculated for 66 (93%) of the 71
fallers and for only 40 (12%) of the 324 non-fallers. Thus, in
correctly predicting a fall in the week after assessment, a risk score
of 2 or more had a sensitivity of 93% and a specificity of 88% (table
3). A score of three or more identified 49 (69%) of the fallers and 12
(4%) of the non-fallers - that is, a sensitivity of 69% and
specificity 96%.
Table 2 - Distribution of risk assessment scores in
case-control study (phase 1) local validation study (phase 2) and
remote validation study (phase 3)
Risk assessment
score Case-control
Local
validation
Remote validation
Controls
(n=116)Falls
(n=116)Non-falls
(n=324)Falls
(n=71)Non-falls
(n=363)Falls
(n=79)
0 58 13 181 1 134 3
1 41 25 103 4 114 3
2 12 38 28 17 70 30
3 5 35 10 34 36 33
4 0 5 2 13 5 5
5 0 0 0 2 4 5
A total of 446 risk assessments were completed on 331 patients,
who sustained 79 falls (table 2). In predicting falls in the following
week, a risk score of 2 or more had a sensitivity of 92% and a
specificity of 68%, and a score of three or more had a sensitivity of
54% and specificity of 88% (table 3).
Table 3 - Usefulness of risk assessment scores of 2 and over
and 3 and over in predicting falls among elderly inpatients in local and
remote validation cohorts (phases 2 and 3). Values are percentages
(95% confidence intervals)
Local validation cohort
Remote validation cohort
Score 2 and over Score
3 and over Score 2 and over Score 3 and over
Sensitivity 93.0
(84.3 to
97.7)69.0
(56.9 to
79.5)92.4
(84.2 to
97.2)54.4
(42.8 to 65.7)
Specificity 87.7
(83.6 to
91.0)96.3
(93.6 to
98.1)68.3
(63.3 to
73.1)87.6
(83.8 to 90.8) Positive predictive
value* 62.3
(52.3 to 71.5)80.3
(68.2 to 89.4)38.8
(31.8 to
46.2)48.4
(38.1 to 59.8) Negative predictive
value† 98.3
(96.0 to 99.4)93.4
(90.2 to 95.8)97.6
(94.9 to
99.1)89.8
(86.2 to 92.8)
*Positive predictive value = No of falls with score
n and over/No of all scores n and over.
†Negative predictive value = No of falls with score
<n/No of all scores <n.Discussion
Key Points
Falls among elderly inpatients are common and result in
morbidity, loss of independence, and higher healthcare costs Preventive strategies to reduce falls require identification
of high risk patients In a case-control study investigating 21 possible risk factors
we identified five factors significantly associated with falls We used these five factors to construct a risk assessment
tool, which showed high sensitivity and specificity in predicting falls
when tested in two elderly care units This simple assessment tool could be used in hospital to
target prevention programmes to patients at high risk of falling
United Medical and Dental Schools,
St
Thomas's Hospital,
London SE1 7EH
D
Oliver, senior registrar
M
Britton, senior registrar
F C
Martin, consultant physician
A H
Hopper, consultant physician
United Medical and
Dental Schools,
St Thomas's Hospital
P
Seed, lecturer
Conflict of interest: None.
Home | Current contents | Past issues |
Classified ads | Career Focus | Feedback
Collections |
About this site | About the BMJ | BMA | Medline