Professional Documents
Culture Documents
Lins Et Al 2021
Lins Et Al 2021
Lins Et Al 2021
Applied Ergonomics
journal homepage: http://www.elsevier.com/locate/apergo
A R T I C L E I N F O A B S T R A C T
Keywords: The Ovako Working posture Assessment System (OWAS) is a commonly used observational assessment method
OWAS for determining the risk of work-related musculoskeletal disorders. OWAS claims to be suitable in the application
Working postures for untrained persons but there is not enough evidence for this assumption. In this paper, inter-rater (inter-
Risk assessment
observer) reliability (agreement) is examined down to the level of individual postures and categories. For this
Reliability analysis
purpose, the postures of 20 volunteers have been observed by 3 varying human raters in a laboratory setting and
the inter-rater agreement against reference values was determined. A high agreement of over 98% (κ = 0.98) was
found for the postures of the arms but lower agreements were found for posture classification of the legs
(66 − 97%, κ = 0.85) and the upper body (80 − 96%, κ = 0.85). No significant difference was found between
raters with and without intense prior training in physical therapy. Consequently, the results confirm the general
reliability of the OWAS method especially for raters with non-specialized background but suggests weaknesses in
the reliable detection of a few particular postures.
* Corresponding author. Fraunhofer Institute for Digital Media Technology IDMT, Division Hearing, Speech and Audio Technology, Marie-Curie-Straße 2, 26129,
Oldenburg, Germany.
E-mail addresses: christian.lins@idmt.fraunhofer.de (C. Lins), sebastian.fudickar@uol.de (S. Fudickar), andreas.hein@uol.de (A. Hein).
https://doi.org/10.1016/j.apergo.2021.103357
Received 29 July 2019; Received in revised form 1 October 2020; Accepted 4 January 2021
Available online 30 January 2021
0003-6870/© 2021 Elsevier Ltd. All rights reserved.
C. Lins et al. Applied Ergonomics 93 (2021) 103357
The reliability of assessment methods for the analysis of postures has individual categories include the postures shown in Table 1. In each of
already been investigated in a number of studies. Kazmierczak et al. the three categories, the rater (observer) selects the partial posture that
(2006) have investigated the inter- and intra-rater reliability for most closely corresponds to the actual posture of the subject. Each
assessing working postures. However, their study was based on video partial posture is assigned a corresponding numerical code. Together,
recordings and did not use any well-known pen-and-paper method such the codes of back, arms, and legs form a three-digit code that describes
as OWAS. The study by Dartt et al. (2009) also used video recordings, the posture. For this OWAS code there are 72 possible combinations (if
here to determine the inter- and intra-rater reliability in the assessment one excludes leg position 7).
of upper limb postures using Multimedia Video Task Analysis (MVTA) —
a software that supports the manual assessment via RULA and OWAS. 2.2. Study design
Bao et al. (2009) also used video recordings but assessed the reliability
of the observations not on the basis of a specific method, but with regard In the study 20 participants were separately observed by three
to the flexion and abduction angles of the individual upper limbs. Trask human raters while posing in each of the 72 (excluding lower limbs
et al. (2017) investigated (again using video recordings) the influence of posture 7 ‘walking’) OWAS postures (see Table 1). Two groups of raters
partly or fully visible limbs on the reliability of the observations. Rhén were selected, the first group being students of the graduating class of a
and Forsman (2020) used video clips that were presented to professional physiotherapy school, the second group being employees
OHS-ergonomists to assess the inter- and intra-rater reliability for the and students of the University of Oldenburg without such a previous
OCRA checklist method. Oliv et al. (2019) have determined the education (see Fig. 1).
inter-rater reliability for both the total score as well as individual items The three raters were equipped with pens and structured sheets (see
of the Quick Exposure Check (QEC) assessment method. Appendix Fig. 5) and viewed each participant from one viewing-angle
Karhu et al. (1977) have given inter- and intra-rater agreement (see Fig. 2). The raters sat next to each other at a distance of around
values (23–99% and 70–100% respectively) for OWAS but for the whole 1 m among each other, and were requested to remain seated, to ensure
body assessment. De Bruijn et al. have already investigated inter- and the independent evaluation of the raters. Due to the resulting position-
intra-rater reliability of OWAS in more detail (De Bruijn et al., 1998). offset, the viewing angle from the outer rater seats on the participants
Their study also examined the differentiability of individual OWAS was not frontal, but slightly shifted. However, raters were allowed to
categories (back, arms, legs), i.e. how confident the raters could shift their viewing angle while remaining seated to aid visual perception
distinguish between the individual categories. Pictures of people in
different postures were presented to the raters in this study for evalua Table 1
tion. However, the different postures belonging to a specific category OWAS categories and postures.
(back, arms, legs) are assigned to different action classes, i.e. their
Category Code Posture Description
associated risk is varying. Therefore, a statement about the possibility of
differentiating the individual categories is not sufficient. It is crucial to Back 1xx Straight
know the degree to which raters can distinguish between the individual
postures of a category. Since this has not yet been investigated, in this 2xx Bent
paper we present the results of a controlled study that examined
inter-rater reliability down to the level of individual postures of real
persons and not only pictures of them. 3xx Twisted
observation, i.e. not video recordings, this has not yet been shown, so we
have also investigated this in this study. A part of our raters were stu
x2x One arm at or above shoulder level
dents of a physiotherapist school in their last year of training, so that one
can assume a solid education in physical therapy and knowledge of the
musculoskeletal system. x3x Both arms at or above shoulder level
Based on the above, this paper examines the following two research
questions:
Legs xx1 Sitting
1. How strong is the inter-rater agreement for the different postures of
the OWAS assessment method?
2. How do OWAS assessments differ from observations with and xx2 Standing on both straight legs
2
C. Lins et al. Applied Ergonomics 93 (2021) 103357
Fig. 3. Screenshot of the study software. The text below the figures are a tex
tual explanation of the part posture (in German).
3
C. Lins et al. Applied Ergonomics 93 (2021) 103357
the random agreement pc has been removed from the overall agreement Table 3
po . Here, the kappa calculation is used for the agreement between the Confusion matrix for reference against rater votes for the back postures.
human raters and the ground truth reference. Reference
Cohen’s kappa is defined for two raters, so the generalization Fleiss’
1 2 3 4
kappa for n raters and N participants is used as well (Fleiss, 1971).
Consider the following notations from Fleiss’ work: Rater 1 (Straight) 96% 1% (13) 2% (25) 0% (4)
(1049)
2 (Bent) 1% (16) 93% 0% (5) 6% (60)
1 ∑N
P= Pi (2) (999)
N i=1 3 (Twisted) 1% (8) 1% (11) 86% 12%
(893) (125)
Pi in Equation (2) is the fraction of agreement of all rater pairs (here: 4 (Bent & 0% (1) 6% (63) 14% 80%
Twisted) (153) (891)
rater 1 and 2, rater 1 and 3, rater 2 and 3) regarding participant i. So P is
the average agreement over all participants.
∑
k
Table 4
Pe = p2j (3)
j=1
Confusion matrix for reference against rater vote for arms postures.
Reference
pj in Equation (3) is the fraction of all ratings to category j. Pe is the mean 1 2 3
agreement that can be expected by random ratings.
Rater 1 (Arms below shoulders) 98% 0% (7) 1% (16)
(1428)
P − Pe
κF = (4) 2 (One arm above shoulder) 0% (3) 98% 1% (19)
1 − Pe (1412)
3 (Both arms above 0% (7) 1% (9) 99%
In Equation (4) Fleiss’ kappa κF is the normalized agreement between
shoulder) (1417)
n raters minus the random agreement.
In this paper, Fleiss’ kappa is used to compare the consistency of the
ratings between the three raters observing each posture, and Cohen’s
Table 5
kappa to compare the consistency of the posture ratings made by the
Confusion matrix for reference against rater votes for legs postures. 1: Sitting, 2:
human raters with the known reference posture.
Standing on straight legs, 3: Standing on one straight leg, 4: Standing on both
Therefore, in the second step, the individual ratings of the raters
knees bent, 5: Standing on one knee bent, 6: Kneeling on one or both knees.
were treated as single ratings when compared to the reference posture.
Reference
The postures, which the study software selects at random and then
displays to the participant modelling the posture, serve as reference. 1 2 3 4 5 6
When comparing the observations against a reference that is inde Rater 1 98% 1% (4) 0% (3) 1% (4) 0% (2) 0% (0)
pendently interpreted by the individual modeling the posture, a misin (707)
terpretation of the instructions may cause additional errors or a bias. To 2 0% (2) 94% 1% (5) 4% (31) 1% (5) 0% (0)
(693)
cancel out this reference interpretation bias, a majority vote was
3 0% (4) 2% (19) 66% 1% (8) 30% 1% (5)
calculated, i.e. when two of three raters share the same vote there is high (648) (291)
probability that the third rater’s observation is incorrect. Cohen’s kappa 4 0% (2) 1% (6) 0% (3) 94% 3% (19) 1% (8)
was used to assess this agreement, which is intended for two raters (in (669)
our case the agreement between the majority vote for observations and 5 0% (0) 0% (1) 13% 1% (3) 81% 4% (20)
(61) (384)
the reference values). 6 0% (1) 0% (0) 0% (0) 1% (6) 1% (9) 97%
With regard to the second research question, the two rater type (681)
groups were considered separately and Fleiss’ kappa values were
calculated for each group.
3. Results
2.3.2. Confusion matrices
A confusion matrix visualizes the error (therefore sometimes called In total, 32 raters participated (8 of 32 trained in physical therapy) in
error matrix) that observations can have in comparison with a reference. the study. As a result, we have a total of 4320 individual posture ob
Here the possible categories are listed in the rows and columns, where servations (20 participants times 72 possible OWAS postures times 3
the rows refer to recognized categories of the raters and the columns raters). The individual observations at the level of the three categories
represent the “true” categories of the reference. Therefore, the diagonal back, arms, and legs were compared. Tables 3–5 shows the confusion
elements of the matrix contain the true positives, i.e. the correct matches matrix for the reference against rater votes for the back, arms, and legs
between raters and reference, whereas the remaining elements contain postures.
errors, i.e. no matches, of the raters (see Table 2).
3.1. Rater votes compared with reference
For the back postures (Table 3) the results show a high agreement of
Table 2
Example confusion matrix for reference against raters for two possible
96% and 93% for the OWAS codes 1 and 2 and significantly1 lower
categories. agreements for codes 3 (86%) and 4 (80%). The kappa for the back
postures is κ = 0.85. The differences between all four category ratings
Reference
on the diagonal are significant (p = 0.02).
1 2 The overall agreement for the arms postures (Table 4) is greater 98%
Rater 1 Rater is correct Rater is wrong (sees 1 but 2 is
correct)
2 Rater is wrong (sees 2 but 1 is Rater is correct 1 2
χ -test of OWAS codes 1/2 against 3/4 ratings with α = 0.05 results in p =
correct)
0.472.
4
C. Lins et al. Applied Ergonomics 93 (2021) 103357
(κ = 0.98) with only few not significant disagreements. The differences Table 8
between the three diagonal values are not significant (p = 0.98). Confusion matrix for reference against rater (PT-trained) votes for arms
The results for the legs postures (Table 5) are mixed: there was a high postures.
agreement for sitting posture (98%), standing on one straight leg (94%), Reference
standing on both knees bent (94%), and kneeling (97%). Least agree 1 2 3
ment was for the standing on one straight leg (66%) which was often
Rater 1 (Arms below shoulders) 98% (570) 1% (3) 1% (8)
confused with standing on one knee bent (81%). The differences between
2 (One arm above shoulder) 0% (2) 98% (558) 2% (9)
all six category ratings are significant (p = 7.92⋅10− 14 ). 3 (Both arms above shoulder) 1% (3) 1% (3) 99% (571)
5
C. Lins et al. Applied Ergonomics 93 (2021) 103357
6
C. Lins et al. Applied Ergonomics 93 (2021) 103357
Table 15
Confusion matrix for 2-rater vs. 3rd-rater votes for legs postures. 1: Sitting, 2: Standing on straight legs, 3: Standing on one straight leg, 4: Standing on both knees bent,
5: Standing on one knee bent, 6: Kneeling on one or both knees.
Two raters (as reference)
1 2 3 4 5 6
Table 16
Agreements summarized for 2-raters vs. 3rd rater (see Section 3.3) and all raters
vs. reference (see Section 3.1) agreements together with χ 2 -test results.
Back Arms Legs
5. Conclusion and usefulness of OWAS as simple and easy to use assessment method for
postures. This is also shown by several ergonomics studies carried out
In the conducted study, the inter-rater (inter-observer) agreement of with the help of the OWAS method and its wide usage throughout the
the OWAS method was examined down to the level of the individual community (Diego-Mas and Alcaide-Marzal, 2014; Hellig et al., 2018;
postures. With a few exceptions, a very high agreement was found for White and Lee Kirby, 2003; Li and Lee, 1999; Mattila et al., 1993). Thus,
the individual postures. Comparatively low agreements were found for the OWAS method remains a practical and easy-to-use assessment
the back (“twisted” and “bent and twisted”) as well as for the lower method for postures in work environments.
limbs (“load on a straight leg” and “load on bent leg”). This might be
caused by confusion in terms of perception among raters. The elusive
postures cannot be summarized either, because in the literature different Declaration of competing interest
action classes are given for them (Mattila and Vilkki, 1999). For the
practical application of OWAS, these results show that OWAS users have The authors declare that they have no known competing financial
to pay special attention to these elusive postures of back and legs. interests or personal relationships that could have appeared to influence
Training instructions should take this into account. the work reported in this paper.
The influence of prior training in physical therapy on the inter-rater
agreement was also investigated. No significant variations of the inter- Acknowledgements
rater agreement among both considered rater groups (with and
without prior training in physical therapy) have been found, even This work was supported by the funding initiative Niedersächsisches
though non-trained raters performed with slightly lower agreements. Vorab of the Volkswagen Foundation and the Ministry of Science and
This supports the claim of OWAS to be practical usable for everyone Culture of the Lower Saxony State (MWK) as a part of the Interdisci
without the requirement of specific training. plinary Research Centre on Critical Systems Engineering for Socio-
Although our study indicates that it is difficult to differentiate be Technical Systems II.
tween distinct partial postures (leg postures standing on one straight leg The authors would like to thank both the editor and the anonymous
and standing on one leg bent), this does not reduce the overall suitability reviewers for their valuable comments and suggestions.
7
C. Lins et al. Applied Ergonomics 93 (2021) 103357
Appendix
Fig. 5 shows the checklist that was used throughout the study to manually assess the postures of the participants. One row is for one posture.
Fig. 5. OWAS checklist for up to 26 posture samples (see rows). Posture images are from (Diego-Mas and Alcaide-Marzal, 2014).
8
C. Lins et al. Applied Ergonomics 93 (2021) 103357
Hellig, T., Mertens, A., Brandl, C., 2018. The interaction effect of working postures on Plantard, P., Shum, H.P., Le Pierres, A.-S., Multon, F., 2016. Validation of an ergonomic
muscle activity and subjective discomfort during static working postures and its assessment method using Kinect data in real workplace conditions. Appl. Ergon. 1–8.
correlation with OWAS. Int. J. Ind. Ergon. https://doi.org/10.1016/j. https://doi.org/10.1016/j.apergo.2016.10.015.
ergon.2018.06.006. Punnett, L., Wegman, D.H., 2004. Work-related musculoskeletal disorders: the
Lins, C., Eichelberg, M., Rölker-Denker, L., Hein, A., 2015. SIRKA: Sensoranzug zur epidemiologic evidence and the debate. J. Electromyogr. Kinesiol. 14 (1), 13–23.
individuellen Rückmeldung körperlicher Aktivität, Dokumentationsband zur 55. https://doi.org/10.1016/j.jelekin.2003.09.015.
DGAUM–Jahrestagung, pp. 301–303. https://doi.org/10.13140/RG.2.1.4269.5128. Rhén, I.-M., Forsman, M., 2020. Inter- and intra-rater reliability of the ocra checklist
Lins, C., Fudickar, S., Gerka, A., Hein, A., 2018. A wearable vibrotactile interface for method in video-recorded manual work tasks. Appl. Ergon. 84, 103025. https://doi.
unfavorable posture awareness warning. In: Proceedings of 4th International org/10.1016/j.apergo.2019.103025. http://www.sciencedirect.com/science/artic
Conference on Information and Communication Technologies for Aging Well and E- le/pii/S0003687019302339.
Health (ICT4AWE). INSTICC, Funchal. https://doi.org/10.5220/ Schaub, K., Caragnano, G., Britzke, B., Bruder, R., 2013. The European Assembly
0006734901780183. Worksheet. Theor. Issues Ergon. Sci. 14 (6), 616–639. https://doi.org/10.1080/
Lowe, B.D., Dempsey, P.G., Jones, E.M., 2019. Ergonomics assessment methods used by 1463922X.2012.678283.
ergonomics professionals. Appl. Ergon. 81 (June), 102882. https://doi.org/10.1016/ Takala, E.-P., Pehkonen, I., Forsman, M., Hansson, G.-Å., Mathiassen, S.E., Neumann, W.
j.apergo.2019.102882. P., Sjøgaard, G., Veiersted, K.B., Westgaard, R.H., Winkel, J., 2010. Systematic
Matsui, H., Maeda, A., Tsuji, H., Naruse, Y., 1997. Risk Indicators of Low Back Pain evaluation of observational methods assessing biomechanical exposures at work.
Among Workers in Japan. Association of Familial and Physical Factors with Low Scand. J. Work. Environ. Health 36 (1), 3–24. https://doi.org/10.5271/sjweh.2876.
Back Pain. https://doi.org/10.1097/00007632-199706010-00015. http://www.nc Trask, C., Erik, S., Rostami, M., Heiden, M., 2017. Observer variability in posture
bi.nlm.nih.gov/pubmed/9201863. assessment from video recordings : the effect of partly visible periods. Appl. Ergon.
Mattila, M., Karwowski, W., Vilkki, M., 1993. Analysis of working postures in hammering 60, 275–281. https://doi.org/10.1016/j.apergo.2016.12.009.
tasks on building construction sites using the computerized OWAS method. Appl. van der Beek, A.J., Mathiassen, S.E., Windhorst, J., Burdorf, A., 2005. An evaluation of
Ergon. 24 https://doi.org/10.1016/0003-6870(93)90172-6. methods assessing the physical demands of manual lifting in scaffolding. Appl.
Mattila, M., Vilkki, M., 1999. OWAS methods. In: The Occupational Ergonomics Ergon. 36 (2), 213–222. https://doi.org/10.1016/j.apergo.2004.10.012. http://
Handbook. CRC Press LLC, pp. 447–459. www.sciencedirect.com/science/article/pii/S0003687004001474.
Li, K.W., Lee, C.L., 1999. Postural analysis of four jobs on two building construction sites: Weir, P.L., Andrews, D.M., Wyk, P.M.V., Callaghan, J.P., 2011. The influence of training
an experience of using the OWAS method in Taiwan. J. Occup. Health. https://doi. on decision times and errors associated with classifying trunk postures using video-
org/10.1539/joh.41.183. based posture assessment methods. Ergonomics 54 (2), 197–205. https://doi.org/
Oliv, S., Gustafsson, E., Baloch, A.N., Hagberg, M., Sandén, H., 2019. The Quick Exposure 10.1080/00140139.2010.547603.
Check (QEC) Inter-rater reliability in total score and individual items. Appl. Ergon.
76, 32–37. https://doi.org/10.1016/j.apergo.2018.11.005. February 2018.