Bias from restricting to live births when estimating effects of prescription drug use on pregnancy complications: A simulation

The Methodological Quality of Observational Studies Examining the Risk of Pregnancy Drug Use on Congenital Malformations Needs Substantial Improvement: A Cross-Sectional Survey

BACKGROUND AND OBJECTIVE

An increasing number of observational studies have investigated the risk of using drugs during pregnancy on congenital malformations. However, the credibility of the causal relationships drawn from these studies remains uncertain. This study aims to evaluate the potential methodological issues in existing observational studies.

METHODS

We used a stepwise approach to investigate this issue. First, we identified observational studies published in 2020 that examined the risk of congenital malformations associated with medication use during pregnancy. We assessed the methodological characteristics for establishing causality, including study design, confounding control, and sensitivity analysis, and compared them between "core clinical journals" and "general journals." For studies reporting an increased risk of congenital malformations in core clinical journals, we searched for subsequent studies addressing the same research question published between January 2021 and May 2023 to assess the consistency of the literature.

RESULTS

A total of 40 eligible studies were published in 2020, primarily focused on the safety of vitamin B12 and folic acid (n = 4), antidepressants (n = 4), and others (n = 32). Our findings suggest that only two (5.00%) studies used causal models to guide the identification of confounding, and only eight (20.00%) studies assessed the potential dose-response relationship. In all, 15 (37.50%) studies used propensity score analysis strategy to achieve "mimic-randomization." In addition, 22 studies (55.00%) performed sensitivity analyses, while 10 (45.45%) showed inconsistency with the primary outcome. Furthermore, 5 studies reported positive outcomes, whereas only 1 out of 11 studies demonstrated a positive correlation between drug usage during pregnancy and major malformations in subsequent studies.

CONCLUSION

A significant portion of the studies has failed to sufficiently consider the essential methodological characteristics required to improve the credibility of causal inferences. The increased risk of congenital malformations documented in core clinical journal was not adequately replicated in subsequent studies.

Detecting and Dating Early Non-live Pregnancy Outcomes: Generation of a Novel Pregnancy Algorithm From Norwegian Linked Health Registries

PURPOSE

Pregnancies ending before gestational week 12 are common but not notified to the Medical Birth Registry of Norway. Our goal was to develop an algorithm that more completely detects and dates all possible pregnancy outcomes (i.e., miscarriages, elective terminations, ectopic pregnancies, molar pregnancies, stillbirths, and live births) by using diagnostic codes from primary and secondary care registries to complement information from the birth registry.

METHODS

We used nationwide linked registry data between 2008 and 2018 in a hierarchical manner: We developed the UiO pregnancy algorithm to arrive at unique pregnancy outcomes, considering codes within 56 days as the same event. To estimate the gestational age of pregnancy outcomes identified in the primary and secondary care registries, we inferred the median gestational age of pregnancy markers (45 ICD-10 codes and 9 ICPC-2 codes) from pregnancies registered in the medical birth registry. When no pregnancy markers were available, we assigned outcome-specific gestational age estimates. The performance of the algorithm was assessed by blinded clinicians.

RESULTS

Using only the medical birth registry, we identified 649 703 pregnancies, including 1369 (0.2%) miscarriages and 3058 (0.5%) elective terminations. With the new algorithm, we detected 859 449 pregnancies, including 642 712 live-births (74.8%), 112 257 miscarriages (13.1%), 94 664 elective terminations (11.0%), 6429 ectopic pregnancies (0.7%), 2564 stillbirths (0.3%), and 823 molar pregnancies (0.1%). The median gestational age was 10+1 weeks (IQR 10+0-12+2) for miscarriages and 8+0 weeks (IQR 8+0-9+6) for elective terminations. Gestational age could be inferred using pregnancy markers for 66.3% of miscarriages and 47.2% of elective terminations.

CONCLUSION

The UiO pregnancy algorithm improved the detection and dating of early non-live pregnancy outcomes that would have gone unnoticed if relying solely on the medical birth registry information.

Bias in the association between advanced maternal age and stillbirth using left truncated data

Restriction to analysis of births that survive past a specified gestational age (typically 20 weeks gestation) leads to biased exposure-outcome associations. This bias occurs when the cause of restriction (early pregnancy loss) is influenced by both the exposure and unmeasured factors that also affect the outcome. The aim of this study is to estimate the magnitude of bias resulting from left truncated data in the association between advanced maternal age and stillbirth. We simulated data for the causal pathway under a collider-stratification mechanism. Simulation parameters were based on an observed birth cohort from Western Australia and a range of plausible values for the prevalence of early pregnancy loss, unmeasured factor U and the odds ratios for the selection effects. Selection effects included the effects of maternal age on early pregnancy loss, U on early pregnancy loss, and U on stillbirth. We compared the simulation scenarios to the observed birth cohort that was truncated to pregnancies that survived beyond 20 gestational weeks. We found evidence of marginal downward bias, which was most prominent for women aged 40 + years. Overall, we conclude that the magnitude of bias due to left truncation is minimal in the association between advanced maternal age and stillbirth.

Longitudinal Methods for Modeling Exposures in Pharmacoepidemiologic Studies in Pregnancy

In many perinatal pharmacoepidemiologic studies, exposure to a medication is classified as "ever exposed" versus "never exposed" within each trimester or even over the entire pregnancy. This approach is often far from real-world exposure patterns, may lead to exposure misclassification, and does not to incorporate important aspects such as dosage, timing of exposure, and treatment duration. Alternative exposure modeling methods can better summarize complex, individual-level medication use trajectories or time-varying exposures from information on medication dosage, gestational timing of use, and frequency of use. We provide an overview of commonly used methods for more refined definitions of real-world exposure to medication use during pregnancy, focusing on the major strengths and limitations of the techniques, including the potential for method-specific biases. Unsupervised clustering methods, including k-means clustering, group-based trajectory models, and hierarchical cluster analysis, are of interest because they enable visual examination of medication use trajectories over time in pregnancy and complex individual-level exposures, as well as providing insight into comedication and drug-switching patterns. Analytical techniques for time-varying exposure methods, such as extended Cox models and Robins' generalized methods, are useful tools when medication exposure is not static during pregnancy. We propose that where appropriate, combining unsupervised clustering techniques with causal modeling approaches may be a powerful approach to understanding medication safety in pregnancy, and this framework can also be applied in other areas of epidemiology.

Quantifying the influence of bias in reproductive and perinatal epidemiology through simulation

PURPOSE

The application of simulated data in epidemiological studies enables the illustration and quantification of the magnitude of various types of bias commonly found in observational studies. This was a review of the application of simulation methods to the quantification of bias in reproductive and perinatal epidemiology and an assessment of value gained.

METHODS

A search of published studies available in English was conducted in August 2020 using PubMed, Medline, Embase, CINAHL, and Scopus. A gray literature search of Google and Google Scholar, and a hand search using the reference lists of included studies was undertaken.

RESULTS

Thirty-nine papers were included in this study, covering information (n = 14), selection (n = 14), confounding (n = 9), protection (n = 1), and attenuation bias (n = 1). The methods of simulating data and reporting of results varied, with more recent studies including causal diagrams. Few studies included code for replication.

CONCLUSIONS

Although there has been an increasing application of simulation in reproductive and perinatal epidemiology since 2015, overall this remains an underexplored area. Further efforts are required to increase knowledge of how the application of simulation can quantify the influence of bias, including improved design, analysis and reporting. This will improve causal interpretation in reproductive and perinatal studies.

Environmental hazards, social inequality, and fetal loss: Implications of live-birth bias for estimation of disparities in birth outcomes

Restricting to live births can induce bias in studies of pregnancy and developmental outcomes, but whether this live-birth bias results in underestimating disparities is unknown. Bias may arise from collider stratification due to an unmeasured common cause of fetal loss and the outcome of interest, or depletion of susceptibles, where exposure differentially causes fetal loss among those with underlying susceptibility.

METHODS

We conducted a simulation study to examine the magnitude of live-birth bias in a population parameterized to resemble one year of conceptions in California (N = 625,000). We simulated exposure to a non-time-varying environmental hazard, risk of spontaneous abortion, and time to live birth using 1000 Monte Carlo simulations. Our outcome of interest was preterm birth. We included a social vulnerability factor to represent social disadvantage, and estimated overall risk differences for exposure and preterm birth using linear probability models and stratified by the social vulnerability factor. We calculated how often confidence intervals included the true point estimate (CI coverage probabilities) to illustrate whether effect estimates differed qualitatively from the truth.

RESULTS

Depletion of susceptibles resulted in a larger magnitude of bias compared with collider stratification, with larger bias among the socially vulnerable group. Coverage probabilities were not adversely affected by bias due to collider stratification. Depletion of susceptibles reduced coverage, especially among the socially vulnerable (coverage among socially vulnerable = 46%, coverage among nonsocially vulnerable = 91% in the most extreme scenario).

CONCLUSIONS

In simulations, hazardous environmental exposures induced live-birth bias and the bias was larger for socially vulnerable women.

Ondansetron use in early pregnancy and the risk of late pregnancy outcomes

BACKGROUND

The effects of ondansetron, used off-label to treat nausea and vomiting during pregnancy, on common pregnancy complications are understudied. Modest effects of a commonly used drug could result in adverse events for large numbers of pregnant women. Therefore, our objective was to compare the risk of stillbirth, preterm birth, gestational hypertensive disorders, small for gestational age, and differences in birth weight between women prescribed ondansetron and women prescribed alternative antiemetics in early pregnancy.

METHODS

A cohort of pregnant women receiving a prescription for ondansetron or comparator antiemetics (metoclopramide or promethazine) during the first 20 weeks of pregnancy was identified using electronic health record data from a health care system in North Carolina, USA. Confounding by multiple covariates was controlled using stabilized inverse probability of treatment weights. Weighted hazard ratios (HR) and 95% confidence intervals (CI) accounted for competing events.

RESULTS

We identified 2677 eligible pregnancies with antiemetic orders, 66% for ondansetron. The small number of stillbirths (n = 15) resulted in an imprecise estimate of the association with ondansetron (HR = 1.60; 95%CI 0.51, 4.97). No association was observed for preterm birth (HR = 0.90; 95%CI 0.67, 1.20) or gestational hypertensive disorders (HR = 0.87; 95%CI 0.68, 1.12). We observed an association with small for gestational age (HR = 1.37; 95%CI 0.98, 1.90), however mean birth weight among term births was similar between groups.

CONCLUSIONS

Our results do not suggest that ondansetron increases the risk of preterm birth or gestational hypertensive disorders. The weak association observed between ondansetron use and small for gestational age warrants further investigation.

Quantification of selection bias in studies of risk factors for birth defects among livebirths

BACKGROUND

Risk factors for birth defects are frequently investigated using data limited to liveborn infants. By conditioning on survival, results of such studies may be distorted by selection bias, also described as "livebirth bias." However, the implications of livebirth bias on risk estimation remain poorly understood.

OBJECTIVES

We sought to quantify livebirth bias and to investigate the conditions under which it arose.

METHODS

We used data on 3994 birth defects cases and 11 829 controls enrolled in the National Birth Defects Prevention Study to compare odds ratio (OR) estimates of the relationship between three established risk factors (antiepileptic drug use, smoking, and multifetal pregnancy) and four birth defects (anencephaly, spina bifida, omphalocele, and cleft palate) when restricted to livebirths as compared to among livebirths, stillbirths, and elective terminations. Exposures and birth defects represented varying strengths of association with livebirth; all controls were liveborn. We performed a quantitative bias analysis to evaluate the sensitivity of our results to excluding terminated and stillborn controls.

RESULTS

Cases ranged from 33% liveborn (anencephaly) to 99% (cleft palate). Smoking and multifetal pregnancy were associated with livebirth among anencephaly (crude OR [cOR] 0.61 and cOR 3.15, respectively) and omphalocele cases (cOR 2.22 and cOR 5.22, respectively). For analyses of the association between exposures and birth defects, restricting to livebirths produced negligible differences in estimates except for anencephaly and multifetal pregnancy, which was twofold higher among livebirths (adjusted OR [aOR] 4.93) as among all pregnancy outcomes (aOR 2.44). Within tested scenarios, bias analyses suggested that results were not sensitive to the restriction to liveborn controls.

CONCLUSIONS

Selection bias was generally limited except for high mortality defects in the context of exposures strongly associated with livebirth. Findings indicate that substantial livebirth bias is unlikely to affect studies of risk factors for most birth defects.

Generation of a pregnancy cohort for medicine utilisation and medicine safety studies in New Zealand

PURPOSE

The aim of this study was to use national health databases to assemble a pregnancy cohort for undertaking medicine utilisation and safety studies in New Zealand.

METHOD

Pregnancies conceived between January 2005 and March 2015 were identified in the National Maternity Collection, the National Minimum Dataset, the Mortality Collection, and the Laboratory Claims Collection. Pregnancy start and end dates were calculated and used in conjunction with the National Health Index number to merge the records from the four collections to create the New Zealand Pregnancy Cohort. Records of live born and stillborn infants identified in the National Maternity Collection and the Mortality Collection that were linkable with a cohort member formed the baby cohort.

RESULTS

The cohort consists of 941 468 pregnancies to 491 272 women. One-third of the pregnancies, predominantly early pregnancy losses and terminations, were not found in the National Maternity Collection. Records of 632 090 live born or stillborn infants are linked with 623 099 pregnancies.

CONCLUSIONS

The New Zealand Pregnancy Cohort is a comprehensive collection of virtually all pregnancies which ended in a live or stillbirth and many, though not all, which ended as early pregnancy losses or terminations in New Zealand over the past decade, and better represents the pregnant population than a cohort generated from the National Maternity Collection alone would do. This cohort will be valuable for investigating patterns of medicine use during pregnancy in New Zealand and developing a fuller understanding of potential impacts of foetal exposure in early pregnancy.

The curse of the perinatal epidemiologist: inferring causation amidst selection

Purpose of review

Human reproduction is a common process and one that unfolds over a relatively short time, but pregnancy and birth processes are challenging to study. Selection occurs at every step of this process (e.g., infertility, early pregnancy loss, and stillbirth), adding substantial bias to estimated exposure-outcome associations. Here we focus on selection in perinatal epidemiology, specifically, how it affects research question formulation, feasible study designs, and interpretation of results.

Recent findings

Approaches have recently been proposed to address selection issues in perinatal epidemiology. One such approach is the ongoing pregnancies denominator for gestation-stratified analyses of infant outcomes. Similarly, bias resulting from left truncation has recently been termed "live birth bias," and a proposed solution is to control for common causes of selection variables (e.g., fecundity, fetal loss) and birth outcomes. However, these approaches have theoretical shortcomings, conflicting with the foundational epidemiologic concept of populations at risk for a given outcome.

Summary

We engage with epidemiologic theory and employ thought experiments to demonstrate the problems of using denominators that include units not "at risk" of the outcome. Fundamental (and commonsense) concerns of outcome definition and analysis (e.g., ensuring that all study participants are at risk for the outcome) should take precedence in formulating questions and analysis approach, as should choosing questions that stakeholders care about. Selection and resulting biases in human reproductive processes complicate estimation of unbiased exposure- outcome associations, but we should not focus solely (or even mostly) on minimizing such biases.