Benefits and challenges of combining data from the CHEAR consortia

Abstract

Background/Aim: The constraints of sample size in environmental epidemiology studies can limit the detection of meaningful associations. Additionally, study-specific biases may arise from differences in data collection methods, exposure and outcome definitions, and confounder selection. Pooling individual level data across studies may reduce implicit study bias. Methods: Using the Children’s Health Exposure Analysis Resource (CHEAR), we assessed the benefits and potential challenges of pooling data by combining study participants (206 children ages 4-16 years) from the School Inner-City Asthma Intervention Study and the Denver Asthma Panel Study who submitted urine samples to CHEAR laboratories for phthalate metabolite measurement. Composite Asthma Severity Index (CASI) score was used to assess the relationship of phthalates, singularly and as a mixture, on asthma severity in each cohort and combined. All models controlled for age, sex, race, income, body mass index percentile, and creatinine with a random effect for study when appropriate to account for intra-study correlation. Results: For individual log2 transformed phthalate analyses conducted for each study alone, no significant associations with CASI score were observed after applying a false discovery rate (FDR) p-value. However, when the studies were combined, four phthalate metabolites had significant associations with CASI after FDR adjustment. One challenge of pooling data was the need to harmonize covariates - for example, income definition: yearly income compared to last month’s income. An additional challenge was the range in exposure distributions across cohorts. We utilized CHEAR quality control samples to evaluate potential systematic lab differences versus study-specific differences. Conclusion: Although pooling data poses several challenges, we were able to overcome some of them and identified associations in the pooled data that were null in the individual cohorts. The increased sample size may be more suitable for studying potential effect modifiers, such as sex, that could not be found in one cohort alone.