Dealing with missings in predicting determinants for musculoskeletal pain among Danish fishermen via conditional imputation and missing categories

Gabriele Berg-Beckhoff, Helle Østergaard, Jørgen Riis Jepsen

Publikation: Bidrag til tidsskriftKonferenceabstrakt i tidsskriftForskningpeer review

Resumé

Background
The aim of the analysis is to estimate the determinants of musculoskeletal
pain among Danish fishermen after several positive
structural changes for the physical work environment have been initiated.
The present analysis focuses on considering missings via
different methods such as conditional imputation and a ‘‘missing’’-
category in multiple regression analyses.
Health—exploring complexity: an interdisciplinary systems approach HEC2016 S23
123
Methods
A cross sectional survey in a random sample of Danish fishermen was
done in 2015 with application of the Nordic questionnaire for musculoskeletal
pain in nine different body regions (neck, shoulder,
elbow, hand, upper back, lower back, hip, knee, and foot). In total,
270 fishermen participated in the study (response rate 27 %). Determinants
for musculoskeletal pain were tested using a multiple linear
regression analysis with an overall pain score summing up all nine
pain locations. Missing answers were calculated out of the mean value
of the remaining answers of the relevant person (conditional imputation).
Cronbach alpha of 0.91 indicates a very good internal
consistency of the scale. Additionally, multinomial logistic regression
analyses considering relevant confounders were used to look at each
single pain site with missing as an additional outcome. In all analyses,
sideline occupations, work position, vessel type, education, and
duration at sea were considered as further predictors.
Results
The prevalence of pain was high for all musculoskeletal locations. In
the multiple linear regression analysis, workload was positively
associated with musculoskeletal pain (Beta: 0.05 (95 % CI 0.04;
0.05). Having a sideline occupation was negatively associated with
musculoskeletal pain (Beta: -0.15; 95 % CI -0.24; -0.03). A linear
regression model excluding all missing’s revealed similar results.
Multinomial regression models showed that workload was the only
consistent predictor for musculoskeletal pain, in particular regarding
upper and lower limb pain. Two additional predictors were found for
the nine different pain locations models; sideline occupation was
associated with less shoulder pain, and work duration of more than
30 days per year was a predictor for hip pain. An additional category
for missing values was considered in the multinomial regression
analysis to see if missing by itself had an effect on the outcome.
Overall, the odds ratios for missing categories were small and far
from being significant which suggest that missing values do not bias
the overall results. Both methods considering missings, the multiple
linear regression model with conditional imputation and the multinomial
logistic regression using missing categories in categorical
outcomes revealed similar results.
Conclusion
The work as a fisherman remains physically demanding, although this
is much less than previously. Fishermen still have a high prevalence
of musculoskeletal pain. Workload is the only and consistent predictor
of pain. Different forms of missing imputation revealed similar
results suggesting missings to occur unsystematically.
OriginalsprogEngelsk
Artikelnummer166
TidsskriftEuropean Journal of Epidemiology
Vol/bind31
Udgave nummerSuppl. 1
Sider (fra-til)S23-S24
Antal sider2
ISSN0393-2990
DOI
StatusUdgivet - 2016
BegivenhedEuropean Congress of Epidemiology HEC - University of Munich, Munich, Tyskland
Varighed: 28. aug. 20162. sep. 2016

Konference

KonferenceEuropean Congress of Epidemiology HEC
LokationUniversity of Munich
LandTyskland
ByMunich
Periode28/08/201602/09/2016

Citer dette

@article{5fbd8a95651e4fb7b653d9020ab4b78f,
title = "Dealing with missings in predicting determinants for musculoskeletal pain among Danish fishermen via conditional imputation and missing categories",
abstract = "BackgroundThe aim of the analysis is to estimate the determinants of musculoskeletalpain among Danish fishermen after several positivestructural changes for the physical work environment have been initiated.The present analysis focuses on considering missings viadifferent methods such as conditional imputation and a ‘‘missing’’-category in multiple regression analyses.Health—exploring complexity: an interdisciplinary systems approach HEC2016 S23123MethodsA cross sectional survey in a random sample of Danish fishermen wasdone in 2015 with application of the Nordic questionnaire for musculoskeletalpain in nine different body regions (neck, shoulder,elbow, hand, upper back, lower back, hip, knee, and foot). In total,270 fishermen participated in the study (response rate 27 {\%}). Determinantsfor musculoskeletal pain were tested using a multiple linearregression analysis with an overall pain score summing up all ninepain locations. Missing answers were calculated out of the mean valueof the remaining answers of the relevant person (conditional imputation).Cronbach alpha of 0.91 indicates a very good internalconsistency of the scale. Additionally, multinomial logistic regressionanalyses considering relevant confounders were used to look at eachsingle pain site with missing as an additional outcome. In all analyses,sideline occupations, work position, vessel type, education, andduration at sea were considered as further predictors.ResultsThe prevalence of pain was high for all musculoskeletal locations. Inthe multiple linear regression analysis, workload was positivelyassociated with musculoskeletal pain (Beta: 0.05 (95 {\%} CI 0.04;0.05). Having a sideline occupation was negatively associated withmusculoskeletal pain (Beta: -0.15; 95 {\%} CI -0.24; -0.03). A linearregression model excluding all missing’s revealed similar results.Multinomial regression models showed that workload was the onlyconsistent predictor for musculoskeletal pain, in particular regardingupper and lower limb pain. Two additional predictors were found forthe nine different pain locations models; sideline occupation wasassociated with less shoulder pain, and work duration of more than30 days per year was a predictor for hip pain. An additional categoryfor missing values was considered in the multinomial regressionanalysis to see if missing by itself had an effect on the outcome.Overall, the odds ratios for missing categories were small and farfrom being significant which suggest that missing values do not biasthe overall results. Both methods considering missings, the multiplelinear regression model with conditional imputation and the multinomiallogistic regression using missing categories in categoricaloutcomes revealed similar results.ConclusionThe work as a fisherman remains physically demanding, although thisis much less than previously. Fishermen still have a high prevalenceof musculoskeletal pain. Workload is the only and consistent predictorof pain. Different forms of missing imputation revealed similarresults suggesting missings to occur unsystematically.",
author = "Gabriele Berg-Beckhoff and Helle {\O}stergaard and Jepsen, {J{\o}rgen Riis}",
year = "2016",
doi = "10.1007/s10654-016-0183-1",
language = "English",
volume = "31",
pages = "S23--S24",
journal = "European Journal of Epidemiology",
issn = "0393-2990",
publisher = "Springer",
number = "Suppl. 1",

}

Dealing with missings in predicting determinants for musculoskeletal pain among Danish fishermen via conditional imputation and missing categories. / Berg-Beckhoff, Gabriele; Østergaard, Helle; Jepsen, Jørgen Riis.

I: European Journal of Epidemiology, Bind 31, Nr. Suppl. 1, 166, 2016, s. S23-S24.

Publikation: Bidrag til tidsskriftKonferenceabstrakt i tidsskriftForskningpeer review

TY - ABST

T1 - Dealing with missings in predicting determinants for musculoskeletal pain among Danish fishermen via conditional imputation and missing categories

AU - Berg-Beckhoff, Gabriele

AU - Østergaard, Helle

AU - Jepsen, Jørgen Riis

PY - 2016

Y1 - 2016

N2 - BackgroundThe aim of the analysis is to estimate the determinants of musculoskeletalpain among Danish fishermen after several positivestructural changes for the physical work environment have been initiated.The present analysis focuses on considering missings viadifferent methods such as conditional imputation and a ‘‘missing’’-category in multiple regression analyses.Health—exploring complexity: an interdisciplinary systems approach HEC2016 S23123MethodsA cross sectional survey in a random sample of Danish fishermen wasdone in 2015 with application of the Nordic questionnaire for musculoskeletalpain in nine different body regions (neck, shoulder,elbow, hand, upper back, lower back, hip, knee, and foot). In total,270 fishermen participated in the study (response rate 27 %). Determinantsfor musculoskeletal pain were tested using a multiple linearregression analysis with an overall pain score summing up all ninepain locations. Missing answers were calculated out of the mean valueof the remaining answers of the relevant person (conditional imputation).Cronbach alpha of 0.91 indicates a very good internalconsistency of the scale. Additionally, multinomial logistic regressionanalyses considering relevant confounders were used to look at eachsingle pain site with missing as an additional outcome. In all analyses,sideline occupations, work position, vessel type, education, andduration at sea were considered as further predictors.ResultsThe prevalence of pain was high for all musculoskeletal locations. Inthe multiple linear regression analysis, workload was positivelyassociated with musculoskeletal pain (Beta: 0.05 (95 % CI 0.04;0.05). Having a sideline occupation was negatively associated withmusculoskeletal pain (Beta: -0.15; 95 % CI -0.24; -0.03). A linearregression model excluding all missing’s revealed similar results.Multinomial regression models showed that workload was the onlyconsistent predictor for musculoskeletal pain, in particular regardingupper and lower limb pain. Two additional predictors were found forthe nine different pain locations models; sideline occupation wasassociated with less shoulder pain, and work duration of more than30 days per year was a predictor for hip pain. An additional categoryfor missing values was considered in the multinomial regressionanalysis to see if missing by itself had an effect on the outcome.Overall, the odds ratios for missing categories were small and farfrom being significant which suggest that missing values do not biasthe overall results. Both methods considering missings, the multiplelinear regression model with conditional imputation and the multinomiallogistic regression using missing categories in categoricaloutcomes revealed similar results.ConclusionThe work as a fisherman remains physically demanding, although thisis much less than previously. Fishermen still have a high prevalenceof musculoskeletal pain. Workload is the only and consistent predictorof pain. Different forms of missing imputation revealed similarresults suggesting missings to occur unsystematically.

AB - BackgroundThe aim of the analysis is to estimate the determinants of musculoskeletalpain among Danish fishermen after several positivestructural changes for the physical work environment have been initiated.The present analysis focuses on considering missings viadifferent methods such as conditional imputation and a ‘‘missing’’-category in multiple regression analyses.Health—exploring complexity: an interdisciplinary systems approach HEC2016 S23123MethodsA cross sectional survey in a random sample of Danish fishermen wasdone in 2015 with application of the Nordic questionnaire for musculoskeletalpain in nine different body regions (neck, shoulder,elbow, hand, upper back, lower back, hip, knee, and foot). In total,270 fishermen participated in the study (response rate 27 %). Determinantsfor musculoskeletal pain were tested using a multiple linearregression analysis with an overall pain score summing up all ninepain locations. Missing answers were calculated out of the mean valueof the remaining answers of the relevant person (conditional imputation).Cronbach alpha of 0.91 indicates a very good internalconsistency of the scale. Additionally, multinomial logistic regressionanalyses considering relevant confounders were used to look at eachsingle pain site with missing as an additional outcome. In all analyses,sideline occupations, work position, vessel type, education, andduration at sea were considered as further predictors.ResultsThe prevalence of pain was high for all musculoskeletal locations. Inthe multiple linear regression analysis, workload was positivelyassociated with musculoskeletal pain (Beta: 0.05 (95 % CI 0.04;0.05). Having a sideline occupation was negatively associated withmusculoskeletal pain (Beta: -0.15; 95 % CI -0.24; -0.03). A linearregression model excluding all missing’s revealed similar results.Multinomial regression models showed that workload was the onlyconsistent predictor for musculoskeletal pain, in particular regardingupper and lower limb pain. Two additional predictors were found forthe nine different pain locations models; sideline occupation wasassociated with less shoulder pain, and work duration of more than30 days per year was a predictor for hip pain. An additional categoryfor missing values was considered in the multinomial regressionanalysis to see if missing by itself had an effect on the outcome.Overall, the odds ratios for missing categories were small and farfrom being significant which suggest that missing values do not biasthe overall results. Both methods considering missings, the multiplelinear regression model with conditional imputation and the multinomiallogistic regression using missing categories in categoricaloutcomes revealed similar results.ConclusionThe work as a fisherman remains physically demanding, although thisis much less than previously. Fishermen still have a high prevalenceof musculoskeletal pain. Workload is the only and consistent predictorof pain. Different forms of missing imputation revealed similarresults suggesting missings to occur unsystematically.

U2 - 10.1007/s10654-016-0183-1

DO - 10.1007/s10654-016-0183-1

M3 - Conference abstract in journal

VL - 31

SP - S23-S24

JO - European Journal of Epidemiology

JF - European Journal of Epidemiology

SN - 0393-2990

IS - Suppl. 1

M1 - 166

ER -