Observational studies of the effects of medication use during pregnancy are threatened by mechanisms of systematic bias which may impact the interpretation of effect estimates. Importantly, these biases may independently or jointly threaten validity, resulting in observed associations that may be incorrect in both direction and magnitude. Methods of quantitatively assessing and addressing these biases are available, and under the correct assumptions, can provide a more accurate understanding and interpretation of effect measures. The goals of this dissertation were to examine previous studies on associations of commonly used medications during pregnancy, adverse pregnancy and fetal outcomes and apply analytic methods to address validity concerns. We used data from the National Birth Defects Prevention Study (NBDPS) for all three studies, which was the largest, population-based case-control study of congenital anomalies from 1997–2011.
In the first study, we evaluated whether the previously reported increased risk of orofacial clefts associated with ondansetron in NBDPS may be explained by selection bias arising from differential participation. We used study records available on participants and non-participants to calculate inverse probability of participation weights (IPWs) to adjust for differential participation. The unadjusted odds ratio (OR) for ondansetron use and cleft palate was 1.5 (95% CI 1.0–2.0). After adjusting for age, education, study year and location, and periconceptional folic acid use, the estimate was 1.6 (95% 1.1–2.1) and the participation-weighted OR was 1.4 (95% CI 1.0–2.0). When we adjusted for confounding using the same covariates as the confounding-only model and selection bias, the OR was 1.6 (95% CI 1.3–2.2). Our estimates suggested limited evidence of selection bias from differential participation in the association between ondansetron use in the first trimester and cleft palate reported in NBDPS.
In the second study, we assessed and quantified the presence of immortal time bias in a study on the use of decongestants in late pregnancy and preterm delivery, comparing time-fixed to time-varying analyses. We observed that results from a time-fixed approach (aHR = 0.99, 95% CI 0.75, 1.31) for our time-dependent outcome resulted in downward bias compared to results from the time-varying approach (aHR = 1.09, 95% CI 0.82, 1.44). However, we did not observe the same reductions in risk of preterm delivery associated with use of decongestants in the when using a time-fixed approach as previously reported in the literature. Overall, we found that both time-fixed and time-varying approaches suggested that use of decongestants in second and/or third trimester of pregnancy did not confer a protective effect for preterm delivery.
In the third study, we conducted probabilistic and multidimensional bias analyses to address differential and nondifferential exposure misclassification for the association between periconceptional use of non-aspirin non-steroidal anti-inflammatory drugs (NSAIDs) and amniotic band syndrome. Under the assumption of differential misclassification, assuming better classification in the cases, the bias-adjusted estimates were compatible with either bias away or towards the null. When we assumed greater specificity in the cases, the bias-adjusted estimates suggested bias away from the null in the unadjusted estimates. If we assumed greater sensitivity in the cases, the adjusted estimates suggested bias away when specificity was high (> 0.9) or towards the null as specificity decreased. Results suggested substantial bias towards the null when we assumed nondifferential exposure misclassification, particularly if sensitivity and specificity were low (0.3 and 0.8 respectively).
All three studies highlight the importance of examining and quantifying the effect of proposed mechanisms of systematic bias on associations in observational studies otherwise we may be led astray by intuitions. These analyses also underline the critical importance of explicitly stating assumptions since all results are conditional on assumptions being correct. These methods (and others) can be used to quantitively assess important, potential sources of systematic error to ultimately improve the rigor of observational studies and our ability to draw conclusions.
| Identifer | oai:union.ndltd.org:bu.edu/oai:open.bu.edu:2144/46628 |
| Date | 25 August 2023 |
| Creators | Adrien, Nedghie Julie Claire Angel |
| Contributors | Parker Kelleher, Samantha E. |
| Source Sets | Boston University |
| Language | en_US |
| Detected Language | English |
| Type | Thesis/Dissertation |
| Rights | Attribution-NonCommercial-NoDerivatives 4.0 International, http://creativecommons.org/licenses/by-nc-nd/4.0/ |
Page generated in 0.1429 seconds