Early prediction of gait ability in patients with hip fracture
Environ Health Prev Med
Early prediction of gait ability in patients with hip fracture
Eiki Tsushima 0 1
Ryukichi Hada 0 1
Manabu Iwata 0 1
Hitoshi Tsushima 0 1
0 R. Hada Medical Informatics, University Hospital, Hirosaki University , Hirosaki, Aomori , Japan
1 E. Tsushima (&) M. Iwata H. Tsushima Graduate School of Health Sciences, Hirosaki University , 66-1 Hon-cho, Hirosaki, Aomori 036-8564 , Japan
Objective Many elderly patients with hip fracture (HF) present with gait deficits. As such, an HF both indirectly and directly increases the number of elderly people requiring care, making it a major medical and economic problem in an aging society. To facilitate the treatment of HF and attempt to resolve the consequences, we have attempted to derive an equation that would predict gait ability. The prediction equation was developed by multivariate analysis using standard evaluation methods, with inclusion of guaranteed objectivity where possible. We attached greater importance to the prediction of gait ability early in the period of hospitalization, since this allows for early determination of an efficient therapeutic strategy. Methods The subjects were 54 HF patients (six men, 48 women; mean age: 78.0 ± 8.4 years) admitted to general hospitals in Hirosaki, Aomori prefecture, between 1998 and 2007. All were aged 60 years or older and were able to walk immediately before injury; physical therapy was initiated for all individuals during hospitalization. Evaluation items related to physical function, psychological function, and complications that may affect gait were evaluated; these included the manual muscle test, motor age test, Katz's index, dementia (HDS-R), consciousness disturbance, among others.
Assessment of functional state; Gait on discharge; Hip fracture; Physical therapy; Prediction of gait ability
Results Based on data for 35 patients who could gait at
discharge and 19 patients who could not, a model including
MAT, HDS-R, and the New York Heart Association
classification of cardiac function scores (P \ 0.001) was
obtained using multiple logistic regression analysis
(discriminant hitting ratio: 94.4%).
Conclusions The effectiveness of the derived model
suggests that both physical and psychological functions
should be considered for gait prediction.
Many elderly patients with hip fracture (HF) present with
gait deficits. As such, an HF both indirectly and directly
increases the number of elderly people requiring care,
making this condition a major medical and economic
problem in an aging society. Major factors that affect the
prognosis of HF patients include dementia [
], extensor muscle strength of the knee joint [
], such as cerebral stroke, heart disease,
respiratory and circulatory disorders, and arthropathy.
Improvements in surgical methods now enable the early
initiation of rehabilitation, including physical therapy (PT).
The probable duration of therapy and hospitalization can be
estimated based on the above factors, and this allows the
therapy to be performed efficiently. Choong et al. 
reported that discussion of the treatment process and goals
with the patient, the patient’s family, and the medical staff
led to shortening of the hospitalization period. In addition,
early exercise with weight-bearing on the affected leg [
and shortening of the hospitalization period [
effective approaches for functional recovery, as well as
being of benefit financially.
While it has been accepted that the early prediction of
the prognosis for physical function (gait ability) is likely to
be of benefit for therapy, achieving such a prediction has
two associated problems. The first is the difficulty in
establishing concrete judgment criteria based on a cutoff
point because the extent to which each of the above factors
influences the prediction is unclear. Moreover, the degree
to which each factor influences the prediction may vary
depending on the specific combination of factors;
consequently, only a rough prediction of gait ability can be
obtained based on experience. The other problem is the
lack of objective evaluation-based prediction criteria;
information-sharing among physical therapists in this field
would be greatly improved if gait ability could be predicted
using an objective evaluation method with concrete indices
of motor ability.
To resolve these problems, we have attempted to derive
an equation for predicting gait ability, based on an
investigation of the prognosis of gait at discharge, which is an
important index of motor ability. The prediction equation
was developed by multivariate analysis using standard
evaluation methods, with the inclusion of guaranteed
objectivity where possible. The time of evaluation was set
to the day of initiation of weight-bearing exercise in HF
patients; we attached greater importance to the prediction
of gait ability early in the period of hospitalization, since
this allows for early determination of an efficient
The subjects were HF patients admitted to the general
wards of district hospitals in Hirosaki, Aomori prefecture,
between 1998 and 2007. The average duration of
hospitalization for the HF patients was around 2 months.
Patients aged 60 years or older who were able to gait
immediately before injury and for whom PT was initiated
during hospitalization were selected for the study. Patients
who were admitted to hospital within 1 year, patients in
whom PT was suspended or discontinued due to other
diseases, and patients who died before discharge were
excluded. Among those patients who satisfied these
conditions, 54 patients (six men and 48 women) were
randomly selected for inclusion in the study. Patients
admitted to a single facility were selected to ensure
uniformity in the procedures performed by the surgeon and the
physical therapist in charge as well as consistency in the
inpatient environment and the therapeutic policy of the
staff. The items measured were essential to being able to
evaluate and treat the functional states and were not
harmful to the patients. The nature and purpose of these
measurements were explained to the patients and their
families, and consent was obtained before the evaluation
was performed. Institutional ethics committee approval for
the study was not obtained because such a committee had
not yet been established at this time this study was
initiated. All of the procedures were conducted in accordance
with the Helsinki Declarations of 1975.
The mean age of the subjects was 78.0 ± 8.4 years
(range 60–90 years), and the mean duration of
hospitalization was 54.0 ± 18.2 days (range 24–134 days). All
subjects underwent surgery: c-locking nail, cannulated
cancellous screw, multiple pinning, and captured hip screw
were used for internal fixation in 36, two, one, and two
patients, respectively. Thirteen patients underwent femoral
head prosthetic replacement (Omnifit-UHR; Howmedica
Osteonics, Allendale, NJ).
Physical therapy procedure
Physical therapy at the bedside was initiated 1 day after
surgery and mainly involved range of motion exercise of
the hip joint on the treated side and exercise for muscle
strengthening. Weight-bearing gait training was initiated
3–5 days after surgery in patients who received
osteosynthesis. and 14–16 days after surgery in patients who
received femoral head prosthetic replacement, depending
on the physical condition of the patient after surgery.
Postponement of initiation of weight-bearing exercise due
to insufficient fixation of the surgical region was not
necessary in any of the study subjects. The physical therapist
selected the appropriate assistive devices for training for
standing and gait, such that the patients could utilize the
exercise to their maximum ability, and appropriate changes
to the assistive device were made when the gait ability of
the patient improved. One of the authors (E. Tsushima) was
the physical therapist in charge of PT for all subjects.
With reference to several previous reports [
related to physical function, psychological function, and
complications that may affect gait were evaluated
(Table 1). Standard items that can be readily measured
without a specific instrument were selected based on a
consideration of general clinical applicability. Physical
function was evaluated based on muscle strength, motor
ability, and activities of daily living, and psychological
function was evaluated based on intelligence and
Japan coma scale: Consciousness was judged as follows: 0 (clear consciousness), 1 (consciousness unclear), 2 (disorientation), 3 (the patient
cannot give their own name and birth date), 10 (the patient opens their eyes in response to calling), and 20 (the patient opens their eyes in
response to a loud voice or to shaking of his/her body)
Night delirium: Patients in whom night delirium was noted were judged (?) and patients with no delirium were judged (–). The judgment was
made based on information obtained from nurses, with only a single episode of delirium required for a judgment of (?) to be made
Influence of heart disease: Evaluation was made using the classification of cardiac function established by the New York heart association
(NYHA). Patients with no heart disease and class-I and class-II heart diseases were judged as ‘none’, and patients with class-III and class-IV
heart diseases were judged as ‘mild’ and ‘severe’, respectively
Influence of respiratory disorder: The Hugh-Jones classification was used. Patients with no respiratory disorder and grades I–IV respiratory
disorders were judged as (–), and patients with grade V respiratory disorder were judged (?)
Influence of cerebral stroke: The legs were staged by Brunnstrom staging as I (severest paralysis) to VI (almost no paralysis)
Influence of other bone and joint disorders: Complications of rheumatoid arthritis, gonarthrosis, and compression fracture were classified into
two categories, ‘mild’ and ‘severe’, corresponding to the degrees of functional disturbance and pain
For the evaluation of muscle strength, the hip joint
flexor muscle strength and the knee joint extensor muscle
strength on the affected side were evaluated based on the
manual muscle test [
] (MMT), which is a simple and
widely used test. For the evaluation of motor ability, leg
motor function was evaluated by using the motor age test
(MAT) reported by Johnson et al. [
] (Table 2). Activities
of daily living were evaluated using the Katz’s index (KI)
]. When a test item could not be performed because the
patient who reported pain refused to undergo the test or
could not understand the content, the item was judged to be
impossible to determine and was not included in the score.
For the evaluation of intelligence, the presence (in this
case, also the severity) or absence of a decrease in
intelligence was evaluated using the standard method in Japan,
which is a revised version of Hasegawa’s dementia scale
]. For the evaluation of the level of
consciousness disturbance, the Japan coma scale (JCS) was
used. When night delirium (henceforth referred to as
delirium) was noted once or more during the period
between admission and the evaluation day, a judgment of
‘yes’ was made regarding delirium.
For complications, the New York Heart Association
classification of cardiac function [
] (NYHA) was used to
evaluate heart disease, and the Hugh–Jones classification
] was used for respiratory disease. The Brunnstrom
] of the legs was used for evaluating the degree of
paralysis in patients with cerebral stroke. This evaluation
method has a high reliability. Therefore, anyone can easily
take the evaluation method. For patients with orthopedic
disorders other than HF, subjective severity was evaluated
using motor function as the index, because no
corresponding standard evaluation method was available.
Physical and psychological functions were evaluated on
the day weight-bearing exercise was initiated in a
rehabilitation room; for example, standing up from a chair and
gait training. Physical therapy was continued, and MAT
and KI scores were re-evaluated 1 week after the initial
evaluation. The MMT and HDS-R scores were not
re-evaluated because no major changes were noted after
1 week. Age, the number of days after admission at the
time of initiation of bedside PT, and the number of days
after initiation of bedside PT that exercise in a
rehabilitation room was started were also included in the analysis.
A single examiner (E. Tsushima) performed all evaluations
to increase examiner reliability.
The patients were divided into the walking group, which
consisted of 35 patients who could gait at discharge, and
the non-walking group, consisting of 19 patients who could
not do so. When the patient could gait independently in the
ward without attention, the patient was included in the
walking group, regardless of whether an assistive device
was required. This judgment was made based on the
unanimous agreement of the physician, the physical
therapist, and the nurses in charge of the patient.
We used multiple logistic regression (MLR) analysis [
to prepare a prediction equation to differentiate between
the walking and non-walking groups. As prior measures,
bivariate statistical analysis was performed and significant
items identified. For quantitative data, a two-sample t test
was applied. When the variance of the two samples was
unequal in a Levene test, a Welch test was performed. For
categorical data, a Mann–Whitney test and a X2 test for
independence (exact test when a category with an expected
value of 5 or lower was present) were applied for items
with many categories and few categories, respectively.
Discriminant analysis using the stepwise method
(pin = 0.25) was performed for a selection of items, with
classification of the walking and non-walking groups as
dependent variables and all other items as independent
Discriminant analysis is a parametric method, and it
has limitations, including the assumption of a
multivariate normal distribution and the necessity of continuity of
data scales. Since most data did not show a normal
distribution or the data scale was categorical, MLR
analysis was selected because of its wide applicability.
The MLR analysis was performed using significant items
(identified as described above) as independent variables
and the walking (category number = 1) and non-walking
(category number = 0) groups as dependent variables.
To select the independent variables for MLR, we
constructed an optimal model based on the likelihood
ratiobased forward selection method satisfying the following
conditions: (1) independent variables are lower than the
significance level; (2) a -2 maximal likelihood (-2ML)
reflects a significant decrease; (3) the increase in the
model X2 value is significant; (4) the discriminant hitting
ratio is high. We noted that construction of a model
based on these statistical values may lead to illogical
results without ensuring consistency with medical
common sense [
]. Thus, the investigation was repeated
with inclusion of additional items that were likely to
have an effect on prognosis based on medical and
empirical knowledge. A linear relationship between the
logit and changes in a given value was confirmed based
on the log odds-ratio for the independent variables
finally selected by MLR.
The 95% confidence interval (95% CI) of the odds ratio
(OR) of each item was calculated from the coefficient and
standard error. To analyze the accuracy of the prediction
equation, we calculated the predicted value using
probability, and a cross table of the measured and predicted
values was prepared based on values of p [ 0.5 for the
walking group and p \ 0.5 for the non-walking groups,
respectively, to obtain the discriminant hitting ratio.
We used the SPSS ver.12.01 J software package (SPSS,
Chicago, IL) for the above statistical analyses, and the
significance level was set to P = 0.05 or P = 0.01.
The descriptive statistics for the evaluation items are
shown in Table 3. Since all patients were of Brunnstrom
stage III or IV (moderate paralysis), the classification was
altered such that it was based on two categories: with (?)
and without (–) paralysis. For orthopedic disorders other
than HF, the patients had either no disorders or mild
disorders, and thus were categorized as with (?) or
without (–) orthopedic disorders.
The patients underwent periodic X-ray examination
during hospitalization, and dislocation or deformity of
the fractured region was confirmed. In one patient (an
89-year-old female in the walking group), the fractured
region was dislocated, and the patient underwent
gait training without weight-bearing for 2 weeks during
hospitalization. In a number of other patients, PT was
suspended for 2–3 days due to poor physical
condition, but general therapy proceeded without major
Walking group (n = 35)
Non-walking group (n = 19)
Mean ± SD
Mean ± SD Median Range t
a MMT, Manual muscle test; HDS–R, Hasegawa’s dementia scale; MAT, motor age test; PT, physical therapy
b KI, Katz’s index; JCS, Japan coma scale
c NYHA, New York heart Association Classification of cardiac function; Hugh-Jones, Hugh-Jones classification for respiratory disease;
Brunnstrom stage, evaluation of paralysis in patients with cerebral stroke
The walking group included 35 patients, of whom ten
used a cane and 25 used a walker. The non-walking group
comprised 19 patients, of whom nine required assistance
for all movement. Differences between the walking and
non-walking groups were noted for all items related to
physical and psychological functions and in the NYHA and
age (Table 3). Significant correlations were noted among
the MMT, MAT, KI and HDS-R scores (Table 4).
Discriminant analysis by the stepwise method was applied
using KI, HDS-R, MAT, delirium, and NYHA, which were
selected as items showing significant differences between
the two groups. In MLR analysis using these items, only
the MAT and HDS-R were significant. However, based on
empirical knowledge, the influence of complications could
not be ruled out. Thus, items associated with complications
were added one by one to the initial MLR model. Addition
of the NYHA caused a significant decrease of -2ML
(difference: 7.701), an increase in the model X2 value
(difference: 7.741), and improvement of the discriminant
hitting ratio (payoff rate: 3.7%). Thus, this model was
designated as the final MLR model (MLR model 1:
Table 5). Based on the equation shown in Table 5, we
calculated a value for P. The overall discriminant hitting
ratio calculated from the cross table of the measured and
predicted values, and the hitting ratios of the walking and
non-walking groups were all 94.4%.
Since the weight-bearing exercise was initiated early
after surgery, patients may have been unable to utilize their
full motor ability. Thus, of the variables selected for the
MLR model 1, MAT was replaced with a value measured
1 week later (MAT-1w), and the analysis was repeated.
However, the evaluation could not be made after 1 week in
four patients because of their poor physical condition and a
delay in measurements caused by holiday periods. Thus,
95% Confidence interval of odds ratio
-2ML (-2 maximal likelihood) = 16.71; Model X2 = 53.34 (P \ 0.001)
Prediction equation: p = 1/[1 ? exp(-1 9 Score)]
Score = 1.08 9 MAT(points) ? 0.31 9 HDS-R(points) - 4.11 9 NYHA (category values of none = 0, mild = 1, severe = 2) - 12.9
a If the odds ratio [ 1, the probability categorized into walking group is higher as the point is larger. If the odds ratio \ 1, the probability
categorized into non-walking group is higher as the point is larger
-2ML (-2 maximal likelihood) = 12.54; Model X2 = 53.87 (P \ 0.001)
Prediction equation: p = 1/[1 ? exp(-1 9 Score)]
Score = 1.26 9 MAT-1w(points) ? 0.43 9 HDS-R(points) - 6.28 9 NYHA(category values of none = 0, mild = 1, severe = 2) - 17.1
a MAT-1w is motor age test that evaluated after 1 week from initiation of weight-bearing exercise in a rehabilitation room
b If the odds ratio is [ 1, the probability categorized into walking group is higher as the point is larger; if the odds ratio \ 1, the probability
categorized into non-walking group is higher as the point is larger
these patients were excluded from the analysis, and a total
of 50 patients (mean age 77.9 ± 8.4 years old; walking
group, n = 31; non-walking group, n = 19) were
analyzed. The results are shown in Table 6 (MLR model 2).
Comparison of -2ML showed that the goodness of fit was
higher in MLR model 2 than in MLR model 1, and the
discriminant hitting ratio of MLR model 2 was 96.0%, with
the hitting ratios of the walking and non-walking groups
being 96.8 and 94.7%, respectively.
Significant differences were noted in physical and
psychological function between the walking and non-walking
groups, which is consistent with the results of several
earlier studies. Thus, the prediction of gait ability may be
possible by evaluating one of these items. However,
considering that the determination of gait ability in elderly HF
patients is not simple, prediction based on a single item
may lead to an incorrect judgment; thus, a combination of
two or more items may increase the accuracy of prediction.
Hence, construction of a model of prediction from multiple
items using a multivariate analytical method may be
suitable for predicting gait ability. In support of this, the
prediction accuracy of our MLR analysis was high.
Items selected by MLR, such as MAT and HDS-R, were
highly correlated with other measures of physical and
psychological functions, demonstrating the representative
nature of these items. Thus, the influence of both physical
and psychological functions needs to be considered for gait
prediction. Johnson et al. [
] first described the MAT
based on an evaluation table with reference to the typical
motor development process between 0 and 72 months
(6 years) after birth. However, this test is now frequently
used in the field of rehabilitation. The decline of motor
function with age is markedly influenced by neurology.
Such a decline does not necessarily regress to the
childhood development process, but it may be represented based
on standard and objective indices using the MAT, and an
overall evaluation including balance ability and muscle
strength may also be possible. In addition, no specific
instrument is needed for the evaluation, and the procedure
is simple. The HDS-R is a method of evaluating
intelligence that was designed in Japan, and its usefulness has
also been confirmed in other countries [
Although the NYHA was not significant in the MLR, we
added it to the prediction equation because it has been
reported to influence prognosis [
]. Since the NYHA
score was not significant in the initial model, its inclusion
was not strictly necessary, but it may become significant in
a patient population comprising an increased number of
subjects with the complication of heart disease; the number
of patients with heart disease in the current study was
The patients may not have been able to use their full
motor ability on the day of initiation of weight-bearing
exercise because they were elderly and post-surgery
interval was short. Hence, MAT in MLR model 1 was replaced
with MAT-1w. In MLR model 2, the OR of MAT-1w was
improved, and the accuracy of prediction increased. It is
understandable that the use of the value after the 1-week
period increased the accuracy of prediction, but MLR
model 1 was also significant and had a high hitting ratio.
The precision of the prediction was quite high with both
models. Therefore, MLR model 1 may be fully applicable in
terms of early prediction of gait ability, making these
findings useful for planning a PT strategy. Since the model
is capable of predicting the appropriate target strategy at an
early stage, assessments of the requirement for continuation
of PT and the utilization of social resources upon early
discharge may possibly be made more easily.
Prolongation of the injury-associated recumbency
period (the period between admission and PT) increases the
possibility of abasia, but there was no significant difference
in the duration of this period between the groups, and no
patients were recumbent for a prolonged period after
injury. Moreover, the patients for whom PT was suspended
for a prolonged period were excluded from the study. Thus,
construction of multiple models that take these conditions
into account will be necessary to further extend the work
presented here. In addition, changes due to differences in
rehabilitation or the ward system among facilities and
developments in surgical therapy will also need to be
accounted for in future models. Nonetheless, the model that
we have constructed should prove useful for prediction of
gait ability in hip-fracture patients. Accumulation of data
regarding the items selected for the prediction equation at
each institution and application of the model using these
data may lead to the construction of a model with even
1. Bitsch M , Foss N , Kristensen B , Kehlet H . Pathogenesis of and management strategies for postoperative delirium after hip fracture: a review . Acta Orthop Scand . 2004 ; 75 : 378 - 89 .
2. Tanaka J , Tokimura F , Seki N. Outcomes of hip fracture surgery in patients aged C 90 years . Orthopedics. 2003 ; 26 : 55 - 8 .
3. Marcantonio ER , Flacker JM , Michaels M , Resnick NM . Delirium is independently associated with poor functional recovery after hip fracture . J Am Geriatr Soc . 2000 ; 48 : 618 - 24 .
4. Duppils GS , Wikblad K. Delirium : behavioural changes before and during the prodromal phase . J Clin Nurs . 2004 ; 13 : 609 - 16 .
5. Binder EF , Brown M , Sinacore DR , Steger-May K , Yarasheski KE , Schechtman KB . Effects of extended outpatient rehabilitation after hip fracture: a randomized controlled trial . JAMA . 2004 ; 292 : 837 - 46 .
6. Masuda T , Miura N , Ishii S , Hibino Y , Beppu M . New preoperative evaluation system of the physical findings of aged patients with femoral neck fracture . J Orthop Sci . 2004 ; 9 : 434 - 9 .
7. Zanocchi M , Maero B , Francisetti F , Giona E , Nicola E , Margolicci A , et al. Multidimensional assessment and risk factors for prolonged hospitalization in the elderly . Aging Clin Exp Res . 2003 ; 15 : 305 - 9 .
8. Choong PF , Langford AK , Dowsey MM , Santamaria NM. Clinical pathway for fractured neck of femur: a prospective, controlled study . Med J Aust. 2000 ; 172 : 423 - 6 .
9. Jarnlo GB , Ceder L , Thorngren KG . Early rehabilitation at home of elderly patients with hip fractures and consumption of resources in primary care . Scand J Prim Health Care . 1984 ; 2 : 105 - 12 .
10. Crotty M , Whitehead CH , Gray S , Finucane PM . Early discharge and home rehabilitation after hip fracture achieves functional improvements: a randomized controlled trial . Clin Rehabil . 2002 ; 16 : 406 - 13 .
11. Lucille D , Catherine W. Muscle testing, 5th edn . Philadelphia: WB Saunders; 1986 .
12. Johnson MK , Zukc FN , Wingate K. The motor age test . J Bone Joint Surg . 1951 ; 33 -A: 698 - 707 .
13. Katz S , Ford AB , Moskowitz RW , Jackson BA , Jaffe MW . Studies of illness in the aged: the index of ADL, a standardized measure of biological and psychosocial function . JAMA . 1963 ; 185 : 914 - 9 .
14. Kim KW , Lee DY , Jhoo JH , Youn JC , Suh YJ , Jun YH , et al. Diagnostic accuracy of mini-mental status examination and revised Hasegawa dementia scale for Alzheimer's disease . Dement Geriatr Cogn Disord . 2005 ; 19 : 324 - 30 .
15. Committee on Medical Rating of Physical Impairment . Guides to the evaluation of permanent impairment; the cardiovascular system . JAMA . 1960 ; 172 : 1049 - 60 .
16. Hugh-Jones S . A simple standard exercise test and its use for measuring exertion distress . Br Med J. 1952 ; 4749 : 65 - 71 .
17. Brunnstrom S. Movement therapy in hemiplegia . New York: Harper & Row; 1970 .
18. David WH , Stanley L. Applied logistic regression . New York: John Wiley & Sons; 1989 .
19. Daniels L , Worthingham C . Muscle testing: technqiues of manual examination . Philadelphia: W.B. Saunders ; 1980 .