Article Text


0079 Examining multiple exposure pathways of beryllium using mixed and structural equation moding techniques
  1. Jenna Armstrong1,
  2. Abbas Virji1,
  3. Mary Davis2,
  4. Gregory Day1,
  5. Aleksandr Stefaniak1,
  6. Marcia Stanton1,
  7. David Deubner3,
  8. Christine Schuler1
  1. 1National Institute of Occupational Safety and Health, Morgantown, WV, USA
  2. 2Tufts University, Boston, MA, USA
  3. 3Materion Brush, Inc., Elmore, OH, USA


Objectives Inhalation beryllium exposures are associated with sensitisation, however dermal exposures are also important. In a previous study, we identified strong correlations between dermal-air, dermal- surface, and air- surface measurements. The aim of this study was to investigate workplace factors associated with exposures using mixed-effects models and structural equation modelling (SEM).

Method Beryllium was measured in personal air, on gloves, and on surfaces at three manufacturing facilities. Predictor variables included substance and activity emission potential (REACH classification), dilution, segregation, PPE, personal behaviour, and work shift.

Results The mixed model described 57 and 59% of total variance for air and dermal, respectively. The total variance explained by the SEM model for air and dermal was 0.51 and 0.48% respectively. In both models activity and substance emission potential, surface contamination, dilution, and personal behaviour were significant predictors of air concentrations (p ≤ 0.05); and surface contamination and air concentrations were significant predictors of dermal loading on cotton gloves (p ≤ 0.05). However, work shift and personal behaviour were predictive of dermal loading in the SEM (p ≤ 0.03), but not in the mixed model. In addition, the SEM reported a parameter estimate for air concentration as a predictor of dermal loading that was an order of magnitude higher than in the mixed model.

Conclusions Although SEM requires relatively large sample sizes, it is useful for modelling multiple, correlated dependent variables. In addition, full-information maximum likelihood (FIML) methods can be used in SEM to include missing predictor variable data. Although we found both models to be useful, SEM has the potential to illustrate indirect pathways of outcome variables.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.