Epidemiology最新文献

Background: A set of conditions sufficient to identify the average treatment effect (ATE) in observational data includes no measurement error, causal consistency, and conditional mean exchangeability with positivity. The average treatment effect in the treated (ATT) is identified under a subset of these conditions, specifically relaxing the symmetry of conditional exchangeability with positivity.

Methods: We reanalyzed data from the Northwest Germany Stroke Registry (2020-2021) to estimate the effect of tissue-type plasminogen activator (tPA) on inhospital mortality. We used inverse probability of treatment weighting for the ATE and standardized mortality ratio (SMR) weighting for the ATT. We also conducted 5000 simulations of 6000 patients, varying the prevalence of treatment indication. We generated homogeneous and heterogeneous treatment effects under two scenarios: (1) positivity holds for treated and untreated groups and (2) positivity only holds for the treated.

Results: Among 6000 patients, 20% received tPA, and 5% died. The inverse probability of treatment weighting risk ratio (ATE) was 1.70 (95% CI: 0.80, 3.64), while the SMR-weighted risk ratio (ATT) was 0.82 (95% CI: 0.59, 1.14). In simulations, ATT estimates of the risk ratio remained unbiased when we violated positivity for the untreated. However, ATE estimates showed increasing log-scale bias with increased nonpositivity, ranging from 0.2 to 1.1 for homogeneous effects and 0.2 to 0.9 for heterogeneous effects.

Conclusions: While ATE estimates suggested harm from tPA, ATT estimates suggest a protective effect. Simulations show that when one-sided positivity violations exist, epidemiologists can leverage weaker identification conditions to consistently estimate the ATT, even when estimates of the ATE are biased.

Background: Last menstrual period (LMP) and ultrasound are commonly used to estimate pregnancy length. Ovulation, which precedes fertilization by ≤24 hours, should give a more accurate estimate.

Methods: The Effects of Aspirin in Gestation and Reproduction (EAGeR) trial preconceptionally enrolled participants from four US medical centers from 2006 to 2012. Participants in our analyses delivered a singleton live birth, had prospectively recorded LMP, ovulation detected by a fertility monitor, and early first-trimester crown-rump length measurements. We estimated pregnancy length, preterm birth (<37 weeks) prevalence, and sex-specific size for gestational age by LMP, ultrasound, and ovulation. We report the sensitivity and specificity of LMP and ultrasound for detecting preterm birth compared with our gold standard, ovulation.

Results: In our analytic sample (n = 392), pregnancies were longest, preterm birth was least common (prevalence = 0.07, 95% confidence interval [CI]: 0.04, 0.10), and small for gestational age was most common when measured by LMP. Pregnancies were shortest, preterm birth was most common (prevalence = 0.10, 95% CI: 0.07, 0.13), and small for gestational age was least common when measured by ultrasound. The prevalence of preterm birth was 0.08 (95% CI: 0.06, 0.12) by ovulation. Using ovulation as the gold standard measure, LMP was less sensitive in detecting preterm birth (0.76, 95% CI: 0.61, 0.90) than ultrasound (0.94, 95% CI: 0.86, 1.00). The specificity of LMP was 1.00 (95% CI: 0.99, 1.00), and the specificity of ultrasound was 0.97 (95% CI: 0.96, 0.99).

Conclusion: While this study's pregnancy length information is the best-case scenario, we observed misclassification of outcomes that may inform future bias analyses.

G-computation is a useful estimation method that can be adapted to address various biases in epidemiology. However, these adaptations may not be obvious for some complex causal structures. This challenge is an example of the much wider issue of translating a causal diagram into a novel estimation strategy. To highlight these challenges, we consider two recent cases from the selection bias literature: treatment-induced selection and co-occurrence of biases that lack a joint adjustment set. For each case study, we show how g-computation can be adapted, describe how to implement that adaptation, show some general statistical properties, and illustrate the estimator using simulation. To simplify both the theoretical study and practical application of our estimators, we express the proposed g-computation estimators as stacked estimating equations. These examples illustrate how epidemiologists can translate identification results into a g-computation estimator and study the theoretical and finite-sample properties of a novel estimator.

Background: Counseling on the harms and benefits of a planned vaginal versus a planned repeat cesarean delivery often relies on observational studies using routinely collected (or administrative) data. However, the accuracy of planned (rather than actual) mode of delivery classifications in such data remains unknown. This study aimed to evaluate the validity of an administrative data-based algorithm to identify planned vaginal and planned cesarean deliveries among individuals with a previous cesarean.

Methods: An algorithm based on diagnostic and procedural codes was applied to records from the Nova Scotia Atlee Perinatal Database. Included were individuals with a previous cesarean eligible for a trial of labor between 2017 and 2019. We compared the classification of planned mode of delivery using the algorithm with that determined through review of a random sample of 200 medical charts. We estimated sensitivity, specificity, and predictive values with 95% confidence intervals (CIs).

Results: Based on the chart review, 80 deliveries (40%) were planned vaginal deliveries. The algorithm had an estimated sensitivity of 99% (95% CI: 93%, 100%), specificity of 96% (95% CI: 91%, 99%), positive predictive value of 94% (95% CI: 87%, 98%), and negative predictive value of 99% (95% CI: 95%, 100%) for identifying planned vaginal deliveries.

Conclusions: An algorithm based on routinely collected data accurately classified planned vaginal and planned cesarean deliveries among individuals with a previous cesarean. These findings suggest that studies using similar algorithms to inform counseling on planned mode of delivery in this population are minimally impacted by misclassification of this data.

Unmeasured confounding is an important obstacle when estimating causal effects from observational data. Ding and VanderWeele (EPIDEMIOLOGY 2016;27:368) derived bounds for causal effects, based on sensitivity parameters that quantify the maximal strength of unmeasured confounding. These bounds translate to the popular E-value metric, which quantifies the magnitude of unmeasured confounding required to "explain away" an observed association. While Ding and VanderWeele mainly focused on conditional (on measured confounders) causal effects, they also outlined how their method might be used for marginal causal effects. However, this requires specification of the sensitivity parameters at each level of the measured confounders, which is impractical in high-dimensional settings, and it yields overly conservative bounds that lack a natural E-value analog. In this article, we propose novel bounds for marginal causal effects based on Ding and VanderWeele's sensitivity parameters. The proposed bounds only require the analyst to specify the maximal values of the sensitivity parameters across all levels of the measured confounders, thus substantially reducing dimensionality. Furthermore, the proposed bounds are often narrower than Ding and VanderWeele's bounds, and they translate naturally into an E-value for marginal causation. We show how the proposed bounds can be estimated using standard regression techniques, and we illustrate through an application to publicly available data, with accompanying R code provided.