Predicting human plasma concentrations of persistent organic pollutants from dietary intake and socio-demographic information in the Norwegian Women and Cancer study

Abstract

Background: Concentrations of persistent organic pollutants (POPs) in humans are influenced by a large number of factors including birth year, reproductive history and diet. Accordingly, information on dietary habits and socio-demographic variables may predict plasma concentrations of POPs, thus enabling studies on health effects in large epidemiological studies, without performing time consuming and expensive chemical analyses on entire cohorts.

Aims: To develop and evaluate statistical models for predicting concentrations of POPs in participants of the Norwegian Women and Cancer (NOWAC) study, using questionnaire information and measured plasma POP concentrations.

Materials and methods: Information on estimated dietary intakes and socio-demographic variables from four different questionnaires (in 1991, 1994, 2004 and 2005) were obtained from participants in the NOWAC study. We measured POP concentrations in a total of 367 blood samples from 2005 and built multivariable linear regression models for p,p'-DDE, PCB-118, -138, -153, -180 and summed PCB concentrations in one subsample (N = 259) and evaluated the models in another subsample (N = 108). Measured and predicted values were compared using correlation coefficients and inter-method agreement was evaluated using weighted Cohen's κ for tertile categorization.

Results: Median POP concentrations in the population ranged from 13 ng/g lipid to 162 ng/g lipid (lowest for PCB-118 and highest for p,p'-DDE). Common predictors for all POPs were birth year, breastfeeding and the weight-related variables (BMI or weight change), whereas influential dietary variables differed and were of varying importance. The predicted plasma concentrations were significantly correlated with the measured values (r_s = 0.24, 0.33, 0.41, 0.50, 0.56, and 0.54 for p,p'-DDE, PCB-118, -138, 153, -180 and summed PCBs, respectively). Tertiles of predicted plasma concentrations displayed significant, but varying agreement with measured concentrations (Weighted Cohen's κ = 0.19, 0.22, 0.33, 0.42, 0.45, and 0.50 respectively).

Conclusion: Predicted plasma concentrations of certain PCBs showed good precision (Kw > 0.4) when compared to measured concentrations. Thus, the models can be used to classify NOWAC participants into high, medium and low PCB exposure groups.

Keywords: Diet; Exposure; Lifestyle; Persistent organic pollutants; Regression models.