CONTEXT: A statistical model that accurately predicts human forced expiratory volume in one second (FEV1) response to ozone exposure has been identified and proposed as the foundation for future risk assessments for ambient ozone. We believe that the assumptions about intra-subject variability in the published model can be improved and hypothesize that more realistic assumptions will improve the fit of the model and the accuracy of risk assessments based on the model. OBJECTIVE: Identify alternate assumptions about intra-subject variability and compare goodness-of-fit for models with various variability structures. MATERIALS AND METHODS: Models were fit to an existing data set using a statistical program for fitting nonlinear mixed models. Goodness-of-fit was assessed using Akaike's Information Criteria (AIC) and visual examination of graphical figures showing observed and predicted values. RESULTS: The AIC indicated that a model that assumed intra-subject variability was related to the magnitude of individual response fit the data better than a model that assumes intra-subject variability is constant across individuals and exposures (the original model). This finding was consistent with the variability of observed responses for filtered air exposures and for exposures predicted to be below the threshold for response. CONCLUSION: An ozone exposure-response model that assumes intra-subject variability increases with individual mean FEV1 response appears to fit the data better than one that assumes constant variability.
CONTEXT: A statistical model that accurately predicts human forced expiratory volume in one second (FEV1) response to ozone exposure has been identified and proposed as the foundation for future risk assessments for ambient ozone. We believe that the assumptions about intra-subject variability in the published model can be improved and hypothesize that more realistic assumptions will improve the fit of the model and the accuracy of risk assessments based on the model. OBJECTIVE: Identify alternate assumptions about intra-subject variability and compare goodness-of-fit for models with various variability structures. MATERIALS AND METHODS: Models were fit to an existing data set using a statistical program for fitting nonlinear mixed models. Goodness-of-fit was assessed using Akaike's Information Criteria (AIC) and visual examination of graphical figures showing observed and predicted values. RESULTS: The AIC indicated that a model that assumed intra-subject variability was related to the magnitude of individual response fit the data better than a model that assumes intra-subject variability is constant across individuals and exposures (the original model). This finding was consistent with the variability of observed responses for filtered air exposures and for exposures predicted to be below the threshold for response. CONCLUSION: An ozone exposure-response model that assumes intra-subject variability increases with individual mean FEV1 response appears to fit the data better than one that assumes constant variability.