Identifying and Adjusting for Length-Of-Pregnancy Bias in Estimating the Effects of Air Pollution in Birth Outcome Studies

Abstract

Listen

Translate article

Exposure to ambient air pollution during pregnancy can have consequences on the outcome of a pregnancy and on the health of the newborn child. Epidemiological studies have found associations between air pollutants and preterm birth. Many of these studies look at exposures during different periods of gestation (e.g., first trimester, month before birth) and also include an exposure measure for the entire gestation period. But when characterizing exposures over the full duration of the pregnancy, simple averaging of exposures is not sufficient protection against potential biases arising from the differing lengths of the exposure period that are intrinsic to any assessment of possible effects of exposure on preterm versus full-term births. For example, in our study of ozone exposure and preterm birth, when we calculated average concentration over the full gestation period, we found the following percentages of preterm births from the first through fourth exposure quartiles: 10.4%, 5.0%, 5.9%, and 10.0%. These results look like a strong U-shaped dose-response. However, we will show that this pattern is an artifact of a methodological bias resulting from inadequate accounting for length of gestation. We identify a mechanism that can cause this pattern when accounting for different lengths of gestation only via averaging exposures over time. We illustrate the problem with our own data and with published results from other studies. We identify methods that avoid the length-of-pregnancy bias and emphasize the need to use survival analysis methods when dealing with time-dependent variables and outcomes.