BRIEF METHODOLOGICAL REPORTS
A Comparative Study of the Use of Four Fall Risk Assessment
Tools on Acute Medical Wards
Michael Vassallo, FRCP, PhD,Ã Rachel Stockdale, MRCP (UK), w Jagdish C. Sharma, FRCP,Ã
Roger Briggs, FRCP,z and Stephen Allen, FRCP§
OBJECTIVES: To compare the effectiveness of four falls Tinetti tools (log rank P 5.41) did not demonstrate this
risk assessment tools (STRATIFY, Downton, Tullamore, characteristic.
and Tinetti) by using them simultaneously in the same CONCLUSION: Significant differences were identiﬁed
environment. in the performance and complexity between the four risk
DESIGN: Prospective, open, observational study. assessment tools studied. The STRATIFY tool was the
SETTING: Two acute medical wards admitting predomi- shortest and easiest to complete and had the highest pre-
nantly older patients. dictive value but the lowest sensitivity. J Am Geriatr Soc 53:
PARTICIPANTS: One hundred thirty-ﬁve patients, 86 1034–1038, 2005.
female, mean age Æ standard deviation 83.8 Æ 8.01 (range Key words: falls; risk assessment; comparative study
MEASUREMENTS: A single clinician prospectively com-
pleted the four falls risk assessment tools. The extent of
completion and time to complete each tool was recorded.
Patients were followed until discharge, noting the occur-
rence of falls. The sensitivity, speciﬁcity, negative predictive
B ecause there is some evidence that falls in hospital can
be reduced,1–3 it is important to identify high-risk pa-
tients likely to beneﬁt from expensive multidisciplinary in-
accuracy, positive predictive accuracy, and total predictive
terventions.4 Several fall risk factors have been identiﬁed,
accuracy were calculated.
and some of them have been compiled into fall risk assess-
RESULTS: The number of patients that the STRATIFY ment tools.5,6 Such tools are based on the premise that the
correctly identiﬁed (n 5 90) was significantly higher than higher the number of risk factors, the higher the risk of
the Downton (n 5 46; Po.001), Tullamore (n 5 66; falling.7
P 5.005), or Tinetti (n 5 52; Po.001) tools, but the Although a number of tools have been used to identify
STRATIFY had the poorest sensitivity (68.2%). The
fall risk in hospitalized patients,8 not all have been validat-
STRATIFY was also the only tool that could be fully com-
ed.9 Several of those that have been validated have high
pleted in all patients (n 5 135), compared with the Down-
accuracy, but when tested outside the specific setting in
ton (n 5 130; P 5.06), Tullamore (n 5 130; P 5.06), and
which they were originally validated, the predictive accu-
Tinetti (n 5 17; Po.001). The time required to complete
racy is not reproduced.10 This may have occurred because
the STRATIFY tool (average 3.85 minutes) was signiﬁcant-
ly less than for the Downton (6.34 minutes; Po.001), of different patient and staff characteristics, as well as op-
Tinetti (7.4 minutes; Po.001), and Tullamore (6.25 min- erational differences between the environments. There is
utes; Po.001). The Kaplan-Meier test showed that the considerable overlap between the characteristics used to
STRATIFY (log rank P 5.001) and Tullamore tools (log compile the tools, raising the question of whether fall risk
rank Po.001) were effective at predicting falls over the ﬁrst assessment tools are indeed any different. A comparative
week of admission. The Downton (log rank P 5.46) and study of four assessment tools was therefore conducted to
determine whether there are any differences in performance
of the tools in the same environment. The chosen fall risk
tools were the STRATIFY,10 the Tullamore,11 a Tinetti-
From the ÃKings Mill Hospital, Sutton in Ashﬁeld, United Kingdom; wRoyal based assessment,11 and the Downton.12 They were chosen
Bournemouth Hospital, Bournemouth, United Kingdom; zSouthampton
General Hospital, Southampton, United Kingdom; and §University of because staff were familiar with their use, they had previ-
Bournemouth, Poole, United Kingdom. ously been used on elderly people in various hospital set-
Address correspondence to Dr. Michael Vassallo, Royal Bournemouth tings to predict the risk of falls,1,8,10,11,13 and they are still
Hospital, Castle Lane East, Bournemouth, BH7 7DW, United Kingdom. being used widely in many hospitals.
E-mail: firstname.lastname@example.org By studying four fall risk assessment tools simulta-
DOI: 10.1111/j.1532-5415.2005.53316.x neously with the same investigators, under the same ward
JAGS 53:1034–1038, 2005
r 2005 by the American Geriatrics Society 0002-8614/05/$15.00
JAGS JUNE 2005–VOL. 53, NO. 6 A COMPARISON OF FOUR FALLS RISK ASSESSMENT TOOLS 1035
conditions, it was hoped to determine whether there are real presenting with a fall or having a fall on the ward, the
differences in effectiveness between fall risk assessment presence of agitation, visual impairment, need for frequent
tools and whether more complex tools are any better than toileting, and impaired ability to transfer and walk. Scores
simple tools at identifying patients who fall on medical of 2 or more were considered to be high risk.
wards. The Tinetti fall risk index is based on number of chron-
ic disabilities.7 The higher the number of chronic disabil-
ities, the higher the likelihood of having recurrent falls. The
METHODS nine risk factors included in the fall risk index are mobility
The study was conducted in two medical wards admitting score, morale score, mental status score, distance vision,
predominantly elderly patients. They were admitted for hearing, postural blood pressure drop, back examination,
treatment of a wide range of medical conditions. Approval medications on admission, and admission activity of daily
was obtained from the North Notthinghamshire ethics living score. To simplify the tool, the Geriatric Depression
committee. One hundred thirty-ﬁve consecutive patients Scale score, Get Up and Go test, and Abbreviated Mental
were studied. None declined to participate. On admission, Test score were used instead of the Philadelphia Morale
patients had medical and nursing assessments. A single cli- score, gait and balance assessment, and Mini-Mental State
nician prospectively conducted the medical assessment, Examination score, respectively. These modiﬁcations were
which consisted of measurements of vision, depression, validated for use16 (unpublished data). The subjects’ fall
mobility, and a medication review. The nursing assessment risk score was the number of index risk factors present.
included information on agitation, need of frequent toilet- Scores of 0 to 3 were considered low, 4 to 6 was medium,
ing, activities of daily living, and postural blood pressure. and 7 to 9 was high risk. For the purposes of analysis, me-
The clinician completing the medical assessment completed dium and high risk were considered together.
the risk assessment tool. Fall prevention measures were not The Tullamore tool11 assesses sex, age, gait, sensory
dependent on the score obtained, but measures were taken deficits, falls history, medication, medical history and mo-
to try to correct any individual fall risk factors identiﬁed bility under various subheadings. Patients are classiﬁed as
using the various tools. For example, postural hypotension low (score 3–8), medium (score 9–12) or high (score 13).
would be treated even though the patient might be at low Medium- and high-risk scores were considered together.
overall risk of falls. For all the tools, the cutoff point from low to higher
Information was collected on patients’ age, sex, history risk was that suggested by the respective authors. Patients
of falls, and medications on admission. Patients had a were followed up to the point of discharge from the ward.
physical examination noting the presence of impaired vi- Nursing staff kept a record of falls as they occurred on the
sion, hearing loss, lower limb abnormalities, gait distur- wards in a falls diary. The clinician completing the tool was
bance, back extension, postural hypotension, and blinded to the occurrence of falls that occurred after tool
confusion. Patients were deemed to have impaired vision completion. Patients who fell at least once were classiﬁed as
if they were registered blind or partially sighted or were fallers. Other outcomes assessed were the number of risk
unable to see less than 6/60 on a Snellen chart using glasses, assessments that could be completed in their entirety on
if appropriate. Hearing impairment was deﬁned as the in- initial assessment and how long it took to complete the falls
ability to follow a conversation with or without using a risk assessment. The time was calculated by estimating the
hearing aid. A limb was considered abnormal if there was time taken for each individual assessment. The total for the
any evidence of weakness (Medical Research Council cri- tool was then the total of all the individual components
teria grade 4/5 or less), neuropathy, amputation, joint ab- required to complete the tool.
normality excluding minor osteoarthritic changes, or any
condition judged to interfere with normal gait such as cell-
ulitis or a deep vein thrombosis. A patient’s gait was as-
sessed using the Get Up and Go Test.14 On this basis, STATISTICS
patients were classiﬁed into four groups: normal, safe (with The sensitivity, speciﬁcity, and total predictive accuracy of
or without using aids), unsafe (with or without using aids), the tools were calculated. Sensitivity was deﬁned as the total
and unable, if the patient was bedridden. Back extension number of fallers correctly identiﬁed as high risk. Speciﬁcity
was studied after testing the patient’s mobility and was re- was deﬁned as total number of nonfallers correctly deﬁned
corded with the patient standing as being able or unable to as low risk. The total predictive accuracy was the total
perform the maneuver. Patients were considered to be con- number of patients correctly identiﬁed expressed as a per-
fused if they scored less than 7 of 10 on the Hodkinson centage. The positive predictive value was deﬁned as the
Abbreviated Mental Test score.15 number of high-risk patients who went on to fall whereas
The Downton Fall risk tool12 was compiled based on a the negative predictive value was the number of low-risk
history of falls, medications (tranquilizers/sedatives, diu- patients who did not fall. Results were expressed as a per-
retics, antihypertensives excluding diuretics, anti- centage. Fishers exact probability test was used to compare
parkinsonian drugs, and antidepressants), sensory deficits the accuracy of the various risk tools by comparing the
(visual impairment, hearing impairment), limb abnormal- numbers of patients correctly identiﬁed by the various tools
ities (such as hemiparesis), confusion, and unsafe gait (with to the best performing tool. Data was collected on 135 pa-
or without aids). Each of these factors scored a point; scores tients. Assessment items that could not be completed were
of 3 or above identify patients at risk. identiﬁed and recorded. Because we aimed to evaluate the
STRATIFY consists of ﬁve factors, each found to be practical utility of each of the tools, all items were included
independently associated with falling.10 These factors are in calculating the total score. Incomplete items received a
1036 VASSALLO ET AL. JUNE 2005–VOL. 53, NO. 6 JAGS
Table 1. Characteristics of Tools When Identifying Fallers
Downton STRATIFY (Medium/high risk) (Medium/high risk)
Characteristic (n 5 135) (n 5 135) (n 5 135) (n 5 135)
Sensitivity 81.8 68.2 90.9 77.3
Speciﬁcity 24.7 66.4 40.7 30.9
Positive predictive value 17.5 28.3 22.9 17.9
Negative predictive value 87.5 91.5 95.8 87.5
Patients correctly identiﬁed, nÃ 46 90 66 52
Total predictive accuracy, % 34.1 66.6 48.8 38.5
STRATIFY vs Downton, Po. 001; Tullamore, P 5.005; Tinetti, Po.001.
score of zero so that the total score for each tool was derived pleted in their entirety for all patients (Table 2). The
from the number of positively scored items. STRATIFY performed significantly better than the Tinetti
The Kaplan-Meier hazard statistic was used to assess (Po.001), which could be completed for only 17 patients.
the likelihood of falls in high- and low-risk patients for each Not all the Tinetti items could be completed because of
of the tools for the ﬁrst week of patient stay using all 135 inability to perform postural blood pressure measures
patients. Significance was expressed using the Log rank test. (n 5 75) or the Geriatric Depression Scale (n 5 62) because
The number of falls was censored in daily time intervals for of severe cognitive function, although it was still possible to
the ﬁrst week in both the high- and low-risk category for classify 70% of patients as medium- to high-risk for falls
each of the tools using the Tinetti tool. Completion of the STRATIFY did not
differ from the Downton (P 5.06) and Tullamore tools
(P 5.06). The Downton and Tullamore were not completed
RESULTS in ﬁve subjects because of an inability to complete a cog-
One hundred thirty-ﬁve patients were studied: 86 female, nitive assessment. It was still possible to classify the re-
mean age Æ standard deviation 83.8 Æ 8.01 (range 56– spective patients into a high-risk category. The time
100). Almost all patients had an acute illness of varying required to ﬁll in the STRATIFY was significantly less than
severity (e.g., respiratory tract infection, heart failure, that for the Downton (Po.001), Tinetti (Po.001), and
stroke, urinary tract infection) on a background of chron- Tullamore (Po.001).
ic disease (e.g., arthritis or dementia). The mean length of The predictive value of the tools to identify fallers over
stay was 14.6 Æ 7.5 days 7.5 (range 6–22 days). Twenty- the ﬁrst week was analyzed using the Kaplan-Meier hazard
two fallers, of whom six had recurrent falls, contributing 29 test. The STRATIFY (log rank P 5.001) and Tullamore (log
falls in total, were identiﬁed. The performance of the var- rank Po.001) were able to identify fallers from the time of
ious tools is shown in Table 1. The STRATIFY had the admission and throughout the ﬁrst week of patient stay
highest total predictive accuracy but the lowest sensitivity. (Figure 1). The Downton (log rank P 5.46) and Tinetti (log
The number of patients that the STRATIFY correctly iden- rank P 5.42) did not demonstrate a similar ability.
tiﬁed (number of high-risk patients who fell and low-risk
patients who did not fall) was significantly higher than the
Downton (Po.001), Tullamore (P 5.005), and Tinetti
(Po.001). In view of the low sensitivity, a separate anal- DISCUSSION
ysis with 1 or above as the cutoff point for high risk was It is well recognized that the performance of any given fall
performed. This change gave a sensitivity of 86% and risk assessment tool varies when used in different settings.17
speciﬁcity of 25%. The number of correctly identiﬁed pa- This may result from differences in patient, staffing, and
tients was 48 (35%). This was significantly inferior to the environmental characteristics. By studying and comparing
Tullamore (P 5.03), which now had the highest number of the performance of the various tools under the same
correctly identiﬁed patients. conditions in the same environment, this study attempted
It was possible to complete the STRATIFY tool for all to identify any differences in the effectiveness of these tools.
patients evaluated. None of the other tools could be com- Although there is considerable overlap between the
Table 2. Number of Completed Risk Assessment Tools and Time to Complete
Downton STRATIFY (Medium/high risk) (Medium/high risk)
Other Score Characteristic (n 5 135) (n 5 135) (n 5 135) (n 5 135)
Time to complete, minutes, mean Æ standard deviation 6.34 Æ 2.62 3.85 Æ 1.67 6.25 Æ 2.56 7.40 Æ 3.88
Number fully completedw 130 135 130 17
STRATIFY vs Downton, Po.001; Tinetti, Po.001; Tullamore, Po.001.
STRATIFY vs Tinetti, Po.001.
JAGS JUNE 2005–VOL. 53, NO. 6 A COMPARISON OF FOUR FALLS RISK ASSESSMENT TOOLS 1037
0.4 0.4 High
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
Time (Days) Time (Days)
Log Rank P = 0.46 Log Rank P = 0.001
0.4 High 0.4 High
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
Time (Days) Time (Days)
Log Rank P < 0.001 Log Rank P = 0.42
Figure 1. Cumulative hazard to ﬁrst fall for all the fall risk assessment tools.
characteristics compiling the various tools, this study iden- Tinetti resulted from the missing data. Despite this, it was
tiﬁed significant differences in their performance. still possible to categorize the majority of patients into the
The total predictive accuracy of the various tools was medium- to high-risk group. The analyses included all pa-
low, the highest being the STRATIFY at 66.6%. This is tients, regardless of whether the tools were completed in
principally because of low speciﬁcity. Speciﬁcity refers to their entirety because it was desired to study the utility of it
patients who do not fall having been correctly identiﬁed as in real life, where one needs to decide on the outcome of a
at low risk of falling. However, the hallmark of effective fall risk assessment regardless of whether a tool is completed.
prevention measures and high-quality care is that high-risk This is important because fall risk assessment needs to be
patients are prevented from falling. A low speciﬁcity is ex- accurate, simple, and not time consuming to be implement-
pected in an environment in which patients are prevented ed effectively on wards without adding a considerable bur-
from falling. In addition, not all patients identiﬁed as being den on already hard-pressed staff.
at high risk will fall even if left on their own, further re- This study has a number of limitations. The fall risk
ducing a tool’s total predictive accuracy. The ideal tool, assessment was done only once; therefore a change in pa-
with high sensitivity and speciﬁcity, is difﬁcult to develop, tient condition could explain the low predictive value of the
and the most important measure of a fall risk assessment tools. The results obtained may not be reproducible on
tool is arguably its sensitivity. A possible way of improving other units because of differing staff and patient character-
fall risk assessment is to focus on items in existing tools that istics, including a different sex mix. Another limitation is
improved their sensitivity, such as a history of falls, confu- that it did not include all available fall risk assessment tools.
sion, and an unsafe gait. The STRATIFY, despite having the Some, such as the Morse Fall Scale,18 are widely used in the
highest total predictive accuracy, had the lowest sensitivity United States, but significant differences were identiﬁed
because it failed to identify the highest number of fallers as between the four risk assessment tools studied, with
being at high risk. This is an important weakness of the tool. STRATIFY having the best predictive accuracy but the
Changing the cutoff point of the tool to 1 improved its lowest sensitivity.
sensitivity by increasing the number of fallers identiﬁed as
high risk but reduced the predictive accuracy.
The tools differed in complexity, as reﬂected in the sig- REFERENCES
nificant differences in the time required to ﬁll them out and 1. Vassallo M, Vignaraja R, Sharma JC et al. The effect of changing practice on
the inability to complete some of the tools in their entirety. fall prevention in a rehabilitative hospital. The Hospital Injury Prevention
It is likely that the low accuracy of some tools like the (HIP) Study. J Am Geriatr Soc 2004;52:335–339.
1038 VASSALLO ET AL. JUNE 2005–VOL. 53, NO. 6 JAGS
2. Haines TP, Bennell KL, Osborne RH et al. Effectiveness of targeted falls pre- 10. Oliver D, Britton M, Seed P et al. Development and evaluation of
vention programme in subacute hospital setting: Randomised controlled trial. evidence based risk assessment tool (STRATIFY) to predict which elderly
BMJ 2004;328:676–679. inpatients will fall: Case-control and cohort studies. BMJ 1997;315:
3. Healey F, Monro A, Cockram A et al. Using targeted risk factor reduction to 1049–1053.
prevent falls in older in-patients: A randomised controlled trial. Age Ageing 11. Cunnington AL, Dhaylan B, Stott DJ. Predicting falls in long term careFare
2004;33:390–395. the Cannard and Tinetti tools useful? Age Ageing 2003;32(Suppl. 1):16.
4. Capezuti E. Building the science of falls-prevention research. J Am Geriatr Soc 12. Downton JH. Falls in the Elderly. London: Edward Arnold, 1993.
2004;52:461–462. 13. Nyberg L, Gustafson Y. Using the Downton index to predict those prone to
5. American Geriatrics Society, British Geriatrics Society, and American Acad- falls in stroke rehabilitation. Stroke 1996;27:1821–1824.
emy of Orthopaedic Surgeons Panel on Falls Prevention. Guideline for the 14. Mathias S, Nayak USL, Isaccs B. Balance in elderly patients. The ‘get up and
prevention of falls in older persons. J Am Geriatr Soc 2001;49:664–672. go’ test. Arch Phys Med Rehabil 1986;67:387–389.
6. Chu LW, Pei CK, Chiu A. Risk factors for falls in hospitalised older medical 15. Qureshi KN, Hodkinson HM. Evaluation of a ten question mental test in the
patients. J Gerontol A Biol Sci Med Sci 1999;54A:M38–M43. institutionalised elderly. Age Ageing 1974;3:152–157.
7. Tinetti ME, Williams TF, Mayewski R. Fall risk index for elderly patients 16. Vassallo M. The prevention of falls in hospital. PhD Thesis. University of
based on number of chronic disabilities. Am J Med 1986;90:429–434. Southampton, 2002.
8. Oliver D, Daly F, Martin FC et al. Risk factors and risk assessment tools for 17. Perell KL, Nelson A, Goldman RL et al. Fall risk assessment measures: An
falls in hospital in-patients: A systemic review. Age Ageing 2004;33:122–130. analytic review. J Gerontol A Biol Sci Med Sci 2001;56A:M761–M766.
9. Wyatt JC, Altman DG. Prognostic models: Clinically useful or quickly for- 18. Morse JM, Morse RM, Tylko SJ. Development of a scale to identify the
gotten? BMJ 1995;311:339–341. fall-prone patient. Can J Ageing 1989;8:366–377.