Nathan Lothrop

Nathan Lothrop, MS
lothrop@email.arizona.edu

University of Arizona
Program Manager
COPH
Community, Environment & Policy Dept

I’d like to also add some advice to make sure I’m running appropriate modeling for a dataset. In a nutshell, I’m investigating what impacts modeled air pollution exposures at birth and age 6 have on respiratory outcomes in a group of about 1000 kids at these ages and then at ages from 8-32 years of age. In addition, I’d like to also get some ideas to visualize how participants have changed exposure quartiles between birth and age 6 years. We have info on other risk factors as well. I can’t share the data due to the owner not allowing it to go off their secure server. Would some fake but generally representative data help? I’m pasting in information below with questions in blue.


I will estimate the relationship between air pollution exposures in early life (at birth and age 6 years) and preschool wheezing phenotypes (i.e., nevertransient earlylate-onset, and persistent) (Martinez et al., 1995) using multinomial logistic regression. Results will be shown as adjusted odds ratios for an interquartile range increase in modeled annual average exposure for each pollutant. I will assess predictive capability of an unadjusted model (only air pollution exposure) and an adjusted model (adjusted model + covariates). To assess the importance of birth vs. age 6 pollutant exposure, I will compare health effects estimates for the pollutant exposure between the two model years.

Is this the appropriate approach to run 2 distinct models, one for each exposure, and compare them? Is there another that could incorporate exposures at both birth and age 6?

I will estimate the relationship between predicted air pollution exposures in early life and wheeze prevalence (i.e., no wheeze, infrequent wheeze (1-3 wheeze episodes), or frequent wheeze (>3 wheeze episodes) in the past year (Morgan et al., 2005)) at ages 8, 11, 13, 16, 18, 22, 24, 26, 29, and 32 years using general estimating equations with an unstructured covariance matrix to account for non-independent wheezing outcomes by participant over time (Ballinger, 2004; Morgan et al., 2005).

Is this the appropriate approach to run 2 distinct models, one for each exposure, and compare them? Is there another that could incorporate exposures at both birth and age 6?

SUMMARY

Who:

Client- Nathan Lothrop

Consulting Team: Lisa, Andy, Drew (author), Dave

Time- 9/30 3:30-4:30

Discussion: 

Clients Study- Nathan is studying how childrens lung function is affected after being exposed to various pollutants at birth. The study follows ~1200 children since 1980 and records data related to the 'wheezing phenotype' by asking the children if they have experienced wheezing, and the severity of this wheezing. This is done at age 3 and then again at age 6. The study also records the pollution exposure at age 6. Nathan is interested in how the categorization of the wheezing affects the childs lung function in the long term. Specifically the study lists the outcomes of wheezing as Never, Early Onset, Late, Persistent. Nathan's hypothesis is that kids who are exposed to polluted air at birth are more likely to be in the persistent wheezing category rather than those kids who were not exposed to pollution until age 6. 

The meeting- Nathan wants to know if multinomial logistic regression is appropriate to use in this study. He is not sure if multinomial logistic regression is the correct model to use, or if it is better to use multiple regression models for each category. Specifically Nathan wants to incorporate the exposure to pollution at birth and also at age 6. The consulting team came to the conclusion that running two models and comparing the coefficients is not the correct step and discussed other solutions. Nathan also wants visual data representations for all children in the data set and how they are grouped in the outcome categories. 

Next Steps: The consulting team does not have access to the data but will build some logistic models in R for Nathan to try out on his dataset and will also present some visuals on how participants changed exposure quartiles between birth and age 6.