Accepted Manuscript

Title: Criterion validity of the instrumented timed up and go

test: a partial least square regression study

Authors: Antonio Caronni, Irma Sterpi, Paola Antoniotti,

Evdoxia Aristidou, Fortunato Nicolaci, Michela Picardi,
Giuseppe Pintavalle, Valentina Redaelli, Gianluca Achille,
Luciana Sciumè, Massimo Corbo

PII: S0966-6362(18)30026-2
DOI: https://doi.org/10.1016/j.gaitpost.2018.01.015
Reference: GAIPOS 5929

To appear in: Gait & Posture

Received date: 13-8-2017

Revised date: 18-12-2017
Accepted date: 18-1-2018

Please cite this article as: Caronni Antonio, Sterpi Irma, Antoniotti Paola,
Aristidou Evdoxia, Nicolaci Fortunato, Picardi Michela, Pintavalle Giuseppe, Redaelli
Valentina, Achille Gianluca, Sciumè Luciana, Corbo Massimo.Criterion validity of the
instrumented timed up and go test: a partial least square regression study.Gait and
Posture https://doi.org/10.1016/j.gaitpost.2018.01.015

This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
Criterion validity of the instrumented Timed Up and
Go test: a partial least square regression study.

Antonio Caronni, Irma Sterpi, Paola Antoniotti, Evdoxia Aristidou, Fortunato Nicolaci,
Michela Picardi, Giuseppe Pintavalle, Valentina Redaelli, Gianluca Achille, Luciana
Sciumè, Massimo Corbo.

T
R IP
Authors’ affiliations. Department of Neurorehabilitation Sciences, Casa di Cura del
Policlinico, Via Dezza 48, 20144 Milano, Italy.

SC
Corresponding author. Antonio Caronni, a.caronni@ccppdezza.it

U
N
A
This research did not receive any specific grant from funding agencies in the public,
M

commercial, or not-for-profit sectors.

D
TE

HIGHLIGHTS
EP

 In this work we evaluate which ITUG test parameter is a valid balance

CC

measure.

 The Mini-BESTest (MB) scale was chosen as a criterion for balance.

A

 The PLSR evaluated the association between the ITUG variables and the MB

scale.

 The PLSR showed that turning parameters from the ITUG are good predictors

of the MB.
  Angular velocity and turning phase duration are proposed as valid balance

measures.

T
ABSTRACT

IP
The Timed Up and Go (TUG) test is a common mobility measure in rehabilitation.

R
With the instrumental TUG test (ITUG; i.e. the TUG measured by inertial

SC
measurement units, IMUs), several movement measures are newly available.

However, the clinical meaning of these new measures is not totally clear. Aim of the

U
current work is to evaluate the validity of different ITUG parameters as a measure of
N
balance.
A
M

Neurological patients (n=122; 52 females; 89 older than 65 years) completed the

TUG test with IMUs secured to their back. IMUs signals were used to split the TUG
D

test in five phases (sit-to-stand, walk1, turn1, walk2 and turn-and-sit) and twelve
TE

movement parameters were obtained. Experienced clinicians administered the Mini-

BESTest (MB) scale, a sound balance measure. The partial least square regression
EP

(PLSR) was used to explore the association between the ITUG variables and the MB
CC

measure.

A PLSR model with twelve ITUG variables had satisfactory fit parameters (RMSEP:
A

11%; R2: 0.41, 95% CI: 0.28-0.54; regression line: 1, 95% CI: 0.78-1.22). Three

ITUG variables (i.e. turn1 vertical angular velocity, turn1 duration and turn2 vertical

angular velocity) were found to be the most important predictors of the MB measure.

2
A PLSR model with the turning variables only had fit parameters comparable to that

of the twelve variables model.

Turning parameters from the TUG test are good predictors of the MB scale. The

mean angular velocity during turning and the duration of the turning phase are thus

proposed as a valid, ratio-level measures of balance in neurological patients.

T
IP
KEYWORDS

R
SC
1. Instrumental Timed Up and Go test

2. ITUG

3. Validity assessment U
N
4. Partial least square regression
A

5. Turning
M

6. Angular velocity
D
TE

INTRODUCTION
EP

The timed up and go (TUG) [1] test is widely used to assess mobility, both in elderly
CC

and young patients. It consists of rising from a chair, walking three meters, turning,

walking back and sitting on the chair. Traditionally, the only measure get from the
A

TUG is the time (seconds) the patient needs to complete the test. The clinical utility

of the TUG test total duration is widely recognized [2].

3
In the instrumented timed up and go (ITUG) test, subjects wear inertial measurement

units while completing the timed up and go (TUG) test [3,4]. By means of the inertial

sensors, the TUG test is split in its sub-phases (e.g. sit to stand, walk [5]) and

several movement measures are obtained (e.g. phase duration, angular velocity).

This abundance of new movement measures poses to the researcher a validity

T
problem. In simple terms, clinicians need to know the clinical meaning of the new

IP
movement measures. The relevance of this question is recognised and ongoing

research is highlighting some of the possible applications of the ITUG measures. For

R
SC
example, ITUG measures work better than the TUG total time in identifying elderly

people with cognitive impairment [6] and Parkinson disease [7].

U
Balance is the ability of a person not to fall while keeping a specified body position
N
(e.g. sitting, standing), during movement (e.g. gait) and in reaction to an external
A

perturbation (e.g. a push) [8]. Balance is a variable of paramount importance for the
M

correct assessment and treatment of patients with a motor impairment. The Mini-
D

BESTest (MB) is a Rasch consistent scale, which returns an interval measure of

TE

balance [9]. The MB scale is increasingly being used for balance evaluation and it is

actually considered a standardized measure of balance [10].

EP

The partial least square regression (PLSR) [11] couples dimension reduction with
CC

regression so that to predict a set of response (Y) variables from a set of predictor

(X) variables. In PLSR, latent variables (LVs) are computed as composites of the
A

original X variables and extracted so that to maximizes the covariance of the LVs

themselves with the Y variable. When a satisfactory model is found, variable

selection methods help to identify a subset of important predictors [12] and only the

most relevant predictors are kept in the final model.

4
The present work is aimed to evaluate the criterion validity of the ITUG parameters.

Criterion validity assessment requires the comparison of the measure under

assessment with other measures (i.e. the criteria) whose validity is already

recognised [13]. We assumed that the MB scale returns a proper measure of

balance and we explored the association between the ITUG variables and the MB

interval measure in a cohort of neurological patients with balance and gait

T
impairments. To this end, PLSR and variable selection methods were used to

IP
identify the ITUG variables that are the best predictors of the MB measure. The

R
ITUG parameters which are strong predictors of the MB measure are eventually

SC
proposed as indicators of the same latent variable measured by the MB scale (i.e.

balance).
U
N
A

METHODS
M
D

Participants selection.
TE

We conducted an observational, retrospective study in which data collected from 122

consecutive patients attending the rehabilitation clinic of Casa di Cura del Policlinico
EP

(CCP) in Milano was analysed (May 2016 – May 2017). In CCP, movement data
CC

(including the ITUG and the MB scale) are routinely collected at both patient’s

admission and discharge. All data used in the current work come from the admission
A

measurement.

We enrolled neurological patients with a wide range of balance impairment so that to

explore the whole MB balance continuum. Stroke, lower limbs polyneuropathy,

Parkinson’s disease and parkinsonisms were the most represented diagnoses

5
(Table 1). Patients were tested if able to walk and to move from sitting to standing

without touching assistance. Exclusion criteria were an acute medical condition, a

condition in addition to the neurological disease that causes a mobility impairment

per se (e.g. severe heart failure) and a TUG test >40 s.

When possible, patients were tested without gait aid. A gait aid was used, if the

T
measuring clinician judged that the risk of falling during the test was too high (20 and

IP
7 patients completed the TUG test with a walker with fixed front wheels or a cane,

R
respectively). Ankle foot orthoses were allowed as needed.

SC
All participants gave their written informed consent for using data for retrospective

studies. The current study was notified to the local ethical committee (Comitato Etico

U
Milano Area 2, Fondazione IRCCS Ca’ Granda Ospedale Maggiore Policlinico, n.
N
806_2017).
A
M

Motor assessment.
D
TE

Patients were tested in a corridor (2x6 m) close to the rehabilitation gym.

The conventional three-meters TUG test was performed [1,14]. Participants were
EP

asked to get out of the chair, walk three meters, turn around, walk back to the chair
CC

and sit down. An ordinary chair was used (seat height: 44 cm, no wheels, no

armrests, fixed back) and a traffic cone marked the turning point. Patients were
A

prompted by a go signal from the clinician. In each measuring session, the TUG test

was repeated five times each. Patients were instructed to use a comfortable and

safe walking speed.

6
The Mini-BESTest (MB) scale [9] consists of 14 items evaluating four different

aspects of balance. In the current work, the MB total score was converted in the MB

interval measure [15] and the latter has been inputted in the PLSR model (see

below). The MB interval measure ranges from 0% (poor balance) to 100% (proper

balance). Additional details on the MB scale can be found as supplementary data file

1.

T
IP
An experienced therapist performed the mobility assessment. Two physicians

R
completed the Functional Independence Measure (FIM) [16] scale.

SC
TUG test phases and parameters.
U
N
Participants completed the TUG test with a single trunk-worn sensor (mTUG medical
A
device; mHT - mHealth technologies, Bologna, Italy), attached at the level of the 2nd
M

– 3rd lumbar vertebra. The sensor was mounted so that its vertical, lateral and roll
D

axes were as parallel as possible to the vertical, lateral and roll axes of the room,

compatibly with the anatomical shape of the participant’s trunk (e.g. lumbar lordosis).
TE

We were as specific as possible in the sensor’s positioning.

EP

Inertial sensors signals were used to subdivide the TUG test into consecutive phases
CC

and for calculating the movement parameters. The beginning and end of each phase

was manually identified and the following consecutive phases were identified: sit to
A

stand, walk 1, turn 1, walk 2, turn 2, turn to sit (Figure 1).

TUG test parameters with a clear clinical value were calculated and included in the

PLSR model. The following 12 parameters were chosen:

1. sit to stand duration (STS s);

7
 2. walk 1 duration (W1 s);

3. turn 1 duration (T1 s);

4. walk 2 duration (W2 s);

5. turn and sit duration (T&S s);

6. mean amplitude of the Vert angular velocity in turn 1 (T1 °/s Vert);

7. mean amplitude of the Vert angular velocity in turn 2 (T2 °/s Vert);

T
8. mean amplitude of the AP angular velocity in turn 1 (T1 °/s AP);

IP
9. mean amplitude of the AP angular velocity in turn 2 (T2 °/s AP);

R
10. root mean square ratio (RMSR) of the Vert acceleration in walk 1 (W1 ratio

SC
Vert);

U
11. RMSR of the ML acceleration in walk 1 (W1 ratio ML);
N
12. RMSR of the AP acceleration in walk 1 (W1 ratio AP).
A
The supplementary data file 2 details the TUG test analysis.
M
D

Data and statistical analysis.

TE

Mean and standard deviation (SD) are used to summarise interval and ratio
EP

measures, while median and interquartile range (IQR) are used for ordinal

measures. Spearman’s rho was chosen for analysing the relationship between the
CC

ITUG variables and the MB measure and the relationship between the ITUG

measures themselves.
A

We performed a partial least squares regression (PLSR) to evaluate which ITUG

variables are associated with the MB measure. In the current work, the PLSR was

run in the following three steps. First, the optimum number of LVs was identified

8
using the root mean squared error of prediction (RMSEP). Second, the overall

quality of the PLSR model was assessed by the coefficient of multiple determination

(i.e. R2). Third, variables were selected using two filter methods [17]: the variable

importance on projection (VIP) and the partial least square regression coefficients

(RC). We considered variables with VIP > 1.21 important predictors of the MB

measure. RCs indicate the "direction" of the relationship between predictors and

T
response. Each of the 12 parameters included in the PLSR model is the mean value

IP
of the five TUG repetitions. Details on the PLSR are given as supplementary data file

R
3.

SC
A simple linear regression was calculated to predict the MB measure from each of

U
the turning variables (T1 °/s Vert; T2 °/s Vert; T1 s; T&S s), i.e. the most important
N
predictors of the MB measure according to the PLSR results. Type 1 error probability
A
was set at 5% (2-sided significance testing).
M

The Signal CED software (Signal version 4.07, Cambridge Electronic Design Ltd.
D

1997-2010) was used for manual analysis of the TUG test and for measuring the
TE

ITUG parameters (e.g. duration, mean angular velocity). The R software [18] and the

pls [19], plsVarSel [20], ggplot2 [21] and cowplot [22] packages were used for
EP

statistics and graphics.

CC

RESULTS
A

Table 1 reports patients’ demographics. The MB scale was well tailored to the

severity of balance impairment of our patients’ sample. Participants’ MB measure

spanned from 13 to 89% (no floor or ceiling effect was apparent) and the sample MB

9
average measure was 54.7% (SD: 13.9%), which is quite close to the scale middle

measure (i.e. 50%).

Table 2 reports the Spearman's rho of the correlation between the ITUG variables

and the MB measure and between the ITUG measures themselves. A moderate

correlation (0.5 ≤rho< 0.7) was found between seven (out of twelve) ITUG variables

T
and the MB measure, with turning parameters (e.g. turning vertical angular

IP
velocities) showing the strongest relationship.

R
SC
PLS regression of the ITUG parameters vs the MB measure.

U
Figure 2A shows the root mean squared error of prediction (RMSEP) versus the
N
model number of components. A one component model has the smallest RMSEP
A

(~11%) and therefore a single component is kept in the final model. The single
M

component model explains the 58.4% of the total variance.

D

The prediction plot (Figure 2B) shows the predicted MB values versus the measured
TE

values. A slight curvature is present only for extreme MB values. The model tends to

overestimate patients with very poor balance and underestimate patients with good
EP

balance. The R2 is 0.41 (95%CI: 0.28-0.54). The regression line has slope equal to 1
CC

(95%CI: 0.78-1.22).

Figure 3A shows the VIP values of the different predictors inputted in the model. VIP
A

values of three out of 12 variables (i.e. T1 °/s Vert, T2 °/s Vert and T1 s) are larger

than the 1.21 threshold. These variables are thus considered the most important

predictors of the MB measure. The regression coefficients of each predictor are

shown in Figure 3B . A positive relationship was found between angular velocity

10
during turning and the MB measure (i.e. higher the angular velocity in turning during

walking, higher the MB measure). A negative relationship was found between the

duration of the turn 1 phase and the MB measure. It is worth noting the negative

regression coefficient of turn and sit duration and its VIP value (1.12), which is just

below the 1.21 threshold.

T
A one-component PLSR model in which turning angular velocities and durations (T1

IP
°/s Vert; T2 °/s Vert; T1 s; T&S s) were chosen as predictors (i.e. those predictors

with the largest VIP) returned an RMSEP of about 10.5% and an R2 value equal to

R
SC
0.43 (95%CI: 0.30-0.56).

Figure 4 shows the relationship between each of the four major predictors and the

U
MB measure. A significant regression equation (p < 0.001) was found for T1 °/s Vert
N
(F(1,120) = 90.61; R2 = 0.43), T2 °/s Vert (F(1,120) = 89.07; R2 = 0.43), T1 s (F(1,120) =
A

69.54; R2 = 0.37) and T&S s (F(1,120) = 51.09; R2 = 0.29).

M
D

DISCUSSION
TE
EP

The main objective of the present study was to determine the criterion validity of the

ITUG parameters as a measure of balance in neurological patients. By using the

CC

PLS regression, we showed that the ITUG turning parameters are the best predictors

of balance as measured by the MB scale. For this reason, we propose the mean
A

angular velocity during turning and the duration of the turn phase as valid, ratio

measures of balance. In light of these findings, in what follows we want to draw

particular attention on the clinical importance of the evaluation of turning in the

neurological patient.

11
The Mini-BESTest scale as a balance criterion for validity

assessment.

In the MB scale [9] the patient’s ability to reach the upright stance and to complete

different tasks while standing (e.g. rising to toes, locomotion) is evaluated. The MB

T
IP
scale has several strengths. First, it is deeply rooted in a strong theoretical

framework, according to which sound balance corresponds to the integrity of

R
different motor functions (i.e. anticipatory postural adjustments, postural responses).

SC
Second, being Rasch consistent it returns an interval measure of balance [23,24].

U
Third, its clinical usefulness is high. It is able to identify fall history [25] and good
N
convergence has been shown between the MB scale and a number of measures
A
assessing balance related constructs [26]. The MB scale has been proposed as part
M

of a core outcome set [27], thus becoming a real criterion for the clinical assessment

of balance [10].
D
TE

Different definitions of validity have been proposed. Traditionally, a test is considered

valid if it “really measures what it purports to measure” [28]. The assessment of

EP

criterion validity requires the comparison of the new measure with other measures

(i.e. the criteria) whose validity is not questioned. A measure with high criterion
CC

validity agrees well with the measures of the gold standard.

A

Several methods have been used for validity assessment and correlation is probably

the earliest and still the predominant one [29]. In more recent times, the PLSR has

been proposed for validity evaluation [30]. In the current work we used the PLSR

because of three main reasons. ITUG variables are highly intercorrelated and, unlike

12
other regression techniques (e.g. multiple linear regression), the PLSR works well in

case of multicollinearities [31]. The PLSR works well even in case of a large number

of variables and a small sample size. Thanks to the variable selection methods [12],

the predictors most relevant for estimating the response variable are easily identified.

Additional details on the validity evaluation are given in supplementary data file 1.

T
IP
The clinical importance of turning.

R
SC
In the current work we showed that, among the ITUG parameters, turning

parameters (i.e. the mean angular velocity during turning and the duration of the turn

U
phase) are the best predictors of the MB measure. Based on these results, we
N
propose the turning measures as valid measures of balance.
A

The evaluation of turning (i.e. the change in the direction of movement) is a part of
M

the motor examination of paramount importance. Dysdiadochokinesis is a

D

neurological sign indicating the patient’s inability to make smooth, well-coordinated

TE

alternating movements, such as the pronation and supination of the forearm.

Moreover, it is a common clinical observation that turning during walking can make
EP

overt some gait unsteadiness that is not apparent during walking at even speed in a

straight line. Difficulty in turning is considered a risk factor for falls [32] and falls while
CC

turning are more likely to result in major trauma, such as an hip fracture [33].
A

Turning parameters were proposed as balance indicators also by other Authors.

Poor balance control was proposed to contribute to longer turning durations on the

TUG test [34] and significant correlations between clinical measures of balance and

13
turning speed have been shown [35]. In addition, the association between increased

turn duration and falls history has been shown in elderly [36] and stroke patients [37].

It is worth emphasising that turn 1 and turn 2 of the TUG test are quite different

tasks. In turn 1, subjects basically turn while walking. In turn 2, a few steps are

followed by a pivot turn (Figure 1) and some subjects complete the stand to sit phase

T
actually during turning (see Methods). We therefore propose to consider turn 1

IP
(rather than turn 2) as a genuine measure of turning during walking.

R
Instrumental measures have better metric properties than questionnaires and scales.

SC
In Cronbach's words, “the expansion of a mercury column does not have face validity

as an index of hotness. But it turns out that there is a statistical relation between

U
expansion and sensed temperature, observers employ the mercury method with
N
good interobserver agreement, the regularity of observed relations is increased by
A

using the thermometer...” [38]. We believe that the time has come for using
M

instrumental measures also for measuring movement-related constructs, leaving to

D

questionnaires the measure of “true latent variables” (e.g. health-related quality of

TE

life).
EP

Study limitations and its future developments.

CC

The R2 value of the PLSR models (~0.4) is commonly considered a low to moderate
A

coefficient. Accordingly, a good part of the total variance (~40%) is not explained by

the PLSR models. Moreover, the RMSEP value, which indicates the mean error in

the MB measure prediction, is ~10%. As an example, this means that the true MB

measure of a predicted MB measure equal to 75%, could actually range from 65 to

14
85%. The actual ability of the PLSR models to predict the exact MB measure from

the ITUG parameters is thus sub-optimal. However, the aim of the current work is not

to build a model to have a reliable estimation of the MB measure from the ITUG

parameters. Rather, we used the PLSR to show the significant association between

the ITUG parameters and a clinical measure of balance and to order the ITUG

parameters from the least to the most important for balance prediction (see VIP

T
values).

IP
After assessing validity, other metric properties of the turning parameters, such as

R
SC
the reliability and the minimal detectable change [39], remain to be evaluated.

Moreover, consistency of these results should be confirmed in homogeneous

U
population of patients (e.g. young, elderly) and in patients with specific diseases (e.g.
N
Parkinson’s disease, stroke).
A

For sake of precision, each trial of the TUG test was manually split into different
M

phases. This process of analysis obviously limits the clinical application of our
D

findings. Different algorithms for the automatic analysis of the TUG test are already
TE

available [40] and hybrid systems, in which the clinician supervises and adjusts the

automatic analysis, could be developed. This implementation is actually

EP

indispensable for a widespread distribution of the ITUG test.

CC

Conclusions.
A

In the current work we provide an understanding of some new movement measures

(i.e. the ITUG test parameters) relative to an accepted clinical measure (i.e. the MB

measure). The PLS regression showed that turning parameters from the TUG test

explains a good proportion of the variance of the Mini-BESTest scale. Therefore, the
15
mean angular velocity during turning and the duration of the turning phase are

proposed as valid, ratio-level measures of balance in neurological patients. A ratio-

level, instrumental measure of balance is expected to have a positive impact on the

clinical practice, for example by increasing the reliability of the balance assessment,

by simplifying data storage and transmission and their statistical processing.

T
IP
Contributors and authorship. All authors materially participated in the research or
article preparation. Antonio Caronni conceived and designed the study, contributed

R
to data acquisition, analysed and interpreted data, wrote the first version of the

SC
manuscript and updated it according to the other Authors’ suggestions. Irma Sterpi,
Luciana Sciumè and Massimo Corbo substantially contributed to the interpretation of

U
data and critically revised the manuscript. Paola Antoniotti, Evdoxia Aristidou,
N
Fortunato Nicolaci, Michela Picardi, Giuseppe Pintavalle, Valentina Redaelli and
Gianluca Achille collected data, participated to the interpretation of data and revised
A
the first version of the manuscript. All Authors gave their final approval of the
M

submitted manuscript.
D

CONFLICT OF INTEREST STATEMENT

TE
EP

None of the Authors has any financial or personal relationships with people or

organisations (e.g. employment, consultancies, stock ownership, honoraria, paid

CC

expert testimony, patent applications/registrations or grants) that could bias the

current work.
A

16
REFERENCES
[1] D. Podsiadlo, S. Richardson, The timed “Up & Go”: a test of basic functional
mobility for frail elderly persons., J. Am. Geriatr. Soc. 39 (1991) 142–148.
[2] E.R. Vieira, R.C. Palmer, P.H.M. Chaves, Prevention of falls in older people
living in the community, Bmj. 353 (2016) i1419.

T
[3] D. Vervoort, N. Vuillerme, N. Kosse, T. Hortobágyi, C.J.C. Lamoth, Multivariate

IP
Analyses and Classification of Inertial Sensor Data to Identify Aging Effects on
the Timed-Up-and-Go Test, PLoS One. 11 (2016) e0155984.

R
[4] A. Weiss, A. Mirelman, A.S. Buchman, D.A. Bennett, J.M. Hausdorff, Using a
body-fixed sensor to identify subclinical gait difficulties in older adults with

SC
IADL disability: maximizing the output of the timed up and go, PLoS One. 8
(2013) e68885.
[5] A. Weiss, T. Herman, M. Plotnik, M. Brozgol, N. Giladi, J.M. Hausdorff, An

U
instrumented timed up and go: the added value of an accelerometer for
identifying fall risk in idiopathic fallers, Physiol. Meas. 32 (2011) 2003.
N
[6] A. Mirelman, A. Weiss, A.S. Buchman, D.A. Bennett, N. Giladi, J.M. Hausdorff,
A
Association between performance on Timed Up and Go subtasks and mild
cognitive impairment: further insights into the links between cognitive and
M

motor function, J. Am. Geriatr. Soc. 62 (2014) 673–678.

[7] L. Palmerini, S. Mellone, G. Avanzolini, F. Valzania, L. Chiari, Quantification of
D

motor impairment in Parkinson’s disease using an instrumented timed up and

go test, IEEE Trans. Neural Syst. Rehabil. Eng. 21 (2013) 664–673.
TE

[8] A.S. Pollock, B.R. Durward, P.J. Rowe, J.P. Paul, What is balance?, Clin.
Rehabil. 14 (2000) 402–406.
EP

[9] F. Franchignoni, F. Horak, M. Godi, A. Nardone, A. Giordano, Using

psychometric techniques to improve the Balance Evaluation Systems Test: the
mini-BESTest, J. Rehabil. Med. Off. J. UEMS Eur. Board Phys. Rehabil. Med.
CC

42 (2010) 323.
[10] K.M. Sibley, T. Howe, S.E. Lamb, S.R. Lord, B.E. Maki, D.J. Rose, V. Scott, L.
Stathokostas, S.E. Straus, S.B. Jaglal, Recommendations for a core outcome
A

set for measuring standing balance in adult populations: a consensus-based

approach, PLoS One. 10 (2015) e0120568.
[11] A. Krishnan, L.J. Williams, A.R. McIntosh, H. Abdi, Partial Least Squares
(PLS) methods for neuroimaging: a tutorial and review, Neuroimage. 56 (2011)
455–475.
[12] M. Farrés, S. Platikanov, S. Tsakovski, R. Tauler, Comparison of the variable
importance in projection (VIP) and of the selectivity ratio (SR) methods for

17
 variable selection and interpretation, J. Chemom. 29 (2015) 528–536.
[13] H.C.W. De Vet, C.B. Terwee, L.B. Mokkink, D.L. Knol, Measurement in
medicine: a practical guide, Cambridge University Press, 2011.
[14] A. Caronni, C. Cattalini, A.M. Previtera, Balance and mobility assessment for
ruling-out the peripheral neuropathy of the lower limbs in older adults, Gait
Posture. 50 (2016) 109–115.
[15] F. Franchignoni, M. Godi, S. Guglielmetti, A. Nardone, A. Giordano, Enhancing
the usefulness of the Mini-BESTest for measuring dynamic balance: a Rasch
validation study., Eur. J. Phys. Rehabil. Med. 51 (2015) 429–437.

T
[16] R.A. Keith, C. V Granger, B.B. Hamilton, F.S. Sherwin, The functional

IP
independence measure: a new tool for rehabilitation., Adv. Clin. Rehabil. 1
(1987) 6–18. http://www.ncbi.nlm.nih.gov/pubmed/3503663 (accessed June

R
10, 2015).

SC
[17] T. Mehmood, K.H. Liland, L. Snipen, S. Sæbø, A review of variable selection
methods in partial least squares regression, Chemom. Intell. Lab. Syst. 118
(2012) 62–69.

U
[18] R Core Team, R: A Language and Environment for Statistical Computing,
(2016). https://www.r-project.org/.
N
[19] B.-H. Mevik, R. Wehrens, K.H. Liland, pls: Partial Least Squares and Principal
A
Component Regression, (2016). https://cran.r-project.org/package=pls.
[20] T. Mehmood, K.H. Liland, L. Snipen, S. Sæbø, A review of variable selection
M

methods in Partial Least Squares Regression, Chemom. Intell. Lab. Syst. 118
(2012) 62–69.
D

[21] H. Wickham, ggplot2: Elegant Graphics for Data Analysis, Springer-Verlag

New York, 2009. http://ggplot2.org.
TE

[22] C.O. Wilke, cowplot: Streamlined Plot Theme and Plot Annotations for
“ggplot2,” (2016). https://cran.r-project.org/package=cowplot.
EP

[23] A. Caronni, F. Zaina, S. Negrini, Improving the measurement of health-related

quality of life in adolescent with idiopathic scoliosis: the SRS-7, a Rasch-
developed short form of the SRS-22 questionnaire., Res. Dev. Disabil. 35
CC

(2014) 784–799. doi:10.1016/j.ridd.2014.01.020.

[24] A. Caronni, L. Sciumè, S. Donzelli, F. Zaina, S. Negrini, ISYQOL: a rasch-
consistent questionnaire for measuring health-related quality of life in
A

adolescents with spinal deformities, Spine J. (2017).

[25] A. Marques, S. Almeida, J. Carvalho, J. Cruz, A. Oliveira, C. Jácome,
Reliability, Validity, and Ability to Identify Fall Status of the Balance Evaluation
Systems Test, Mini--Balance Evaluation Systems Test, and Brief--Balance
Evaluation Systems Test in Older People Living in the Community, Arch. Phys.
Med. Rehabil. 97 (2016) 2166–2173.
[26] S. O’Hoski, K.M. Sibley, D. Brooks, M.K. Beauchamp, Construct validity of the

18
 BESTest, mini-BESTest and briefBESTest in adults aged 50 years and older,
Gait Posture. 42 (2015) 301–305.
[27] P.R. Williamson, D.G. Altman, J.M. Blazeby, M. Clarke, D. Devane, E. Gargon,
P. Tugwell, Developing core outcome sets for clinical trials: issues to consider,
Trials. 13 (2012) 132.
[28] T.L. Kelley, Interpretation of educational measurements., (1927).
[29] J.P. Guilford, New standards for test evaluation, Educ. Psychol. Meas. 6
(1946) 427–438.

T
[30] B.D. Zumbo, 3 Validity: Foundational Issues and Statistical Methodology,
Handb. Stat. 26 (2006) 45–79.

IP
[31] M.A.J. van Gerven, Z.C. Chao, T. Heskes, On the decoding of intracranial data
using sparse orthonormalized partial least squares, J. Neural Eng. 9 (2012)

R
26017.

SC
[32] F.-Y. Cheng, Y.-R. Yang, C.-J. Wang, Y.-R. Wu, S.-J. Cheng, H.-C. Wang, R.-
Y. Wang, Factors influencing turning and its relationship with falls in individuals
with Parkinson’s disease, PLoS One. 9 (2014) e93572.

U
[33] R.G. Cumming, R.J. Klineberg, Fall frequency and characteristics and the risk
N
of hip fractures, J. Am. Geriatr. Soc. 42 (1994) 774–778.
[34] T.B. Chen, L.-S. Chou, Impacts of muscle strength and balance control on sit-
A
to-walk and turn durations in the timed up and go test, Arch. Phys. Med.
Rehabil. (2017).
M

[35] L.A. King, M. Mancini, K. Priest, A. Salarian, F. Rodrigues-de-Paula, F. Horak,

Do clinical scales of balance reflect turning abnormalities in people with
D

Parkinson’s disease?, J. Neurol. Phys. Ther. 36 (2012) 25.

TE

[36] M. Mancini, H. Schlueter, M. El-Gohary, N. Mattek, C. Duncan, J. Kaye, F.B.

Horak, Continuous monitoring of turning mobility and its association to falls
and cognitive function: A pilot study, Journals Gerontol. Ser. A Biol. Sci. Med.
Sci. 71 (2016) 1102–1108.
EP

[37] K. Leigh Hollands, M.A. Hollands, D. Zietz, A. Miles Wing, C. Wright, P. Van
Vliet, Kinematics of turning 180 during the timed up and go in stroke survivors
CC

with and without falls history, Neurorehabil. Neural Repair. 24 (2010) 358–367.
[38] L.J. Cronbach, P.E. Meehl, Construct validity in psychological tests., Psychol.
Bull. 52 (1955) 281.
A

[39] A. Caronni, L. Sciumè, Is my patient actually getting better? Application of the

McNemar test for demonstrating the change at a single subject level, Disabil.
Rehabil. (2016) 1–7.
[40] A. Salarian, F.B. Horak, C. Zampieri, P. Carlson-Kuhta, J.G. Nutt, K. Aminian,
iTUG, a sensitive and reliable measure of mobility, IEEE Trans. Neural Syst.
Rehabil. Eng. 18 (2010) 303–310.

19
FIGURES LEGENDS
Figure 1. Accelerometers and gyroscopes recordings from a single
representative trial. Abbreviations as follows: acc, acceleration signal; gyro,

T
gyroscopes signal; Vert, vertical axis; ML, medial-lateral axis; AP, anterior-posterior

IP
axis. The lower graph shows the CUSUM of gyro.Vert. Vertical cursors split the TUG

R
test into six different phases. From cursor 1 to 2: sit-to-stand; from cursor 2 to 3:
walk1; from cursor 3 to 4: turn 1; from cursor 4 to 5: walk2; from cursor 5 to 6: turn;

SC
from cursor 6 to 7: turn to sit. The turn-and-sit phase is obtained by combining turn2
and turn-to-sit (from cursor 5 to 7). From cursor 1 to 7: total TUG duration.

U
Acceleration and angular velocity are expressed as g (1 g = 9.81 m/s2) and °/s,
N
respectively.
A
M

Figure 2. RMSEP and prediction plot of the PLS regression model. The RMSEP
was the lowest with a one component model (A). The prediction plot (B) shows the
D

predicted MB values versus the measured ones (R2: 0.41; regression line: 1).
TE

Figure 3. VIP values and regression coefficients of the twelve ITUG predictors
EP

inputted in the PLS regression model. VIP values (A) of three out of 12 variables
(i.e. turn 1 vertical angular velocity, turn 1 duration and turn 1 vertical angular
CC

velocity) are larger than the 1.21 threshold (horizontal dashed line). The regression
coefficients (B) for each of the 12 predictor. The “*” highlight those predictors with a
A

VIP larger than 1.21. “+” and “-” in the VIP plot report the sign of the regression
coefficient (positive and negative, respectively) for the three predictors above the
1.21 threshold.

20
Figure 4. Linear regression between each ITUG turning parameter and the MB
measure. For each regression, the corresponding R2 value is given in the plot.
Linear regression equations are as follows. (A): MB measure = 0.56 * (T1, Vert
angular velocity) + 28.9. (B): MB measure = 0.41 * (T2, Vert angular velocity) + 32.6.
(C): MB measure = -4.99 * (T1 duration) + 76.1. (D): MB measure = -3.59 * (turn-
and-sit duration) + 70.5.

T
R IP
SC
U
N
A
M
D
TE
EP
CC
A

21
 A
CC
EP
TE
D
M
A
N
U
SC
RIP

22
T
 A
CC
EP
TE
D
M
A
N
U
SC
RIP

23
T
 T
IP
R
SC
U
N
A
M
D
TE
EP
CC

Age (years) 69.0 (14.6)

Sex (M/F) 52/70
Diagnosis Stroke 45
A

LL-PNP 30
Park 26
Others 21
TUG test (s) 19.0 (7.6)
FIM scale score Motor 68 (25)
Cognitive 32 (7)
Total 97 (27)

24
Table 1. Patients’ demographics. Mean and standard deviation (SD) are used for
age and TUG test total duration. Median and interquartile range (IQR) are used for
the FIM scale motor, cognitive and total score. 89 out of 122 patients were older than
65 years. Abbreviations: LL-PNP: lower limbs polyneuropathy; Park, Parkinson’s
disease and parkinsonisms. The category Others includes myelopathy (9),
motoneurone disease (6), traumatic brain injury (3) and surgery because of brain

T
tumor (3).

R IP
SC
ML
W1 ratio

W1 ratio

W1 ratio
AP
MB meas

T1 s

T&S s
T1 °/s AP

T2 °/s AP

STS s

W1 s

W2 s

Vert
T1 °/s Vert

T2 °/s Vert

U
N
MB meas 1.00 0.65 0.66 0.30 0.44 -0.58 -0.60 -0.64 -0.58 -0.58 0.40 -0.45 0.06
A
M

T1 °/s Vert 1.00 0.93 0.43 0.62 -0.79 -0.90 -0.98 -0.84 -0.90 0.48 -0.59 0.14

T2 °/s Vert 1.00 0.43 0.66 -0.75 -0.86 -0.92 -0.79 -0.92 0.40 -0.54 0.19
D

T1 °/s AP 1.00 0.87 -0.38 -0.43 -0.47 -0.35 -0.43 0.18 -0.18 0.00
TE

T2 °/s AP 1.00 -0.54 -0.60 -0.64 -0.52 -0.69 0.24 -0.30 0.08
EP

STS s 1.00 0.76 0.77 0.73 0.74 -0.48 0.51 -0.06

CC

W1 s 1.00 0.89 0.95 0.85 -0.50 0.54 -0.06

T1 s 1.00 0.81 0.88 -0.48 0.57 -0.13

A

W2 s 1.00 0.78 -0.47 0.52 -0.06

T&S s 1.00 -0.35 0.49 -0.19

W1 ratio Vert 1.00 -0.76 -0.39

25
W1 ratio ML 1.00 -0.24

W1 ratio AP 1.00

Table 2. Spearman’s correlation between the ITUG variables and the MB

measure. The Spearman's rho is given. Turning parameters (e.g. T1 °/s Vert; T2 °/s
Vert; T1 s) showed the strongest correlation with the MB measure. Note the high

T
intercorrelation between ITUG variables (e.g. T1 °/s Vert vs W1 s).

R IP
SC
U
N
A
M
D
TE
EP
CC
A

26