Paediatric and perinatal epidemiology最新文献

Background: Empirically evaluating the potential impact of recall bias on observed associations of prenatal medication exposure is crucial.

Objective: We sought to assess the effects of exposure misclassification on previous studies of the use of nonsteroidal anti-inflammatory drugs (NSAIDs) in early pregnancy and increased risk of amniotic band syndrome (ABS).

Methods: Using data from the National Birth Defects Prevention Study (NBDPS) on births from 1997 to 2011, we included 189 mothers of infants with ABS and 11,829 mothers of infants without congenital anomalies. We identified external studies of medication use during pregnancy to obtain validity parameters for a probabilistic bias analysis to adjust for exposure misclassification. Due to uncertainty about the transportability of these parameters, we conducted multidimensional bias analyses to explore combinations of values on the results.

Results: When we assumed higher specificity in cases or higher sensitivity in controls, misclassification-adjusted estimates suggested confounding-adjusted estimates were attenuated. However, in a few instances, when we assumed greater sensitivity in the cases than the controls (and Sp ≥ 0.9), the misclassification-adjusted estimates suggested upward bias in the confounding-adjusted estimates.

Conclusions: Results from our bias analysis highlighted that the magnitude of bias depended on the mechanism and the extent of misclassification. However, the parameters available from the validation studies were not directly applicable to our study. In the absence of reliable validation studies, considering mechanisms of bias and simulation studies to outline combinations of plausible scenarios to better inform conclusions on the effects of these medications on pregnancy outcomes remains important.

Background: In birth registries, incomplete recording of information leads to missing values. Multiple imputation (MI) by chained equations is a widely used method for analysing datasets with missing data. It is unknown whether using registry records from multiple pregnancies contributed by the same woman could potentially give more accurate values when resolving missing data.

Objectives: To investigate the relative performance of five methods to infer missing data on maternal characteristics using data from a medical birth registry, comparing longitudinal methods and MI with data from previous and future pregnancies.

Methods: We used data from the Medical Birth Registry of Norway (MBRN), selecting records among mothers with more than one pregnancy between 2004 and 2018. Longitudinal methods used reference pregnancies in three time directions: past, future and closest pregnancy record. MI was conducted with only index pregnancy records (single-pregnancy MI) and with both index and closest reference pregnancy records (multiple-pregnancy MI). Validity was assessed by comparing the actual values with inferred/imputed values. For continuous variables, we calculated the proportion of inferred values within predefined increments. For binary variables, we calculated five parameters: agreement rate, sensitivity, specificity, positive predictive value and negative predictive value.

Results: We included 578,670 pregnancies among 256,658 women. For continuous variables, the longitudinal methods showed the highest proportion within predefined increments, followed by multiple-pregnancy MI, and single-pregnancy MI showed the lowest value. For binary variables, longitudinal methods generally showed higher values among the five validity parameters than MI. Single-pregnancy MI had substantially lower agreement, while multiple-pregnancy MI performed similarly to longitudinal methods.

Conclusions: The longitudinal method outperformed MI in inferring missing data on maternal characteristics in a medical birth registry.

Background: In the United States, birthing parent-infant dyads may receive care from multiple healthcare systems. Linkage of an individual's electronic health records (EHR) across healthcare systems, in addition to birthing parent-infant linkage, may be necessary to obtain appropriate clinical data for perinatal health research.

Objectives: To develop a privacy-preserving process to link the health records of patients shared by two health systems for a perinatal health study, and to assess data enhancements associated with the linkage.

Methods: We included pregnant patients who received care from at least one of two healthcare systems based in Minnesota, USA and their infants born between December 2020 and September 2022 who had at least one well visit. We identified infants from one health system with birthing parents who potentially received care in the second health system based on the infant's delivery hospital. We implemented a one-way matching process using an algorithm to generate unique hash values for each record at each health system. Specifically, we used four hash ID rules based on six identifiers available in the EHR at both sites plus a consistent salt.

Results: One health system identified 3524 infants with birthing parents who potentially received care in the second system. The second system identified 39,321 infants delivered at the hospitals of interest during the study period. The algorithm matched 3406 (96.7%) infant records. After applying the study eligibility criteria, the birthing-parent records gained through hash matching increased the study population by 7.2% from 8100 to 8686. Overall, 13.6% of the study population had data from the second health system. Some demographic and pregnancy characteristics differed from those with data from the first system only.

Conclusions: The hash matching approach can increase study size, patient diversity, and data completeness in a privacy-preserving manner for perinatal health studies among patients that use multiple healthcare systems.

Background: Survey data are essential in paediatric epidemiology, providing valuable insights into child health outcomes. The potential outcomes framework has advanced causal inference using observational data. However, traditional design-based adjustments, especially sample weights, are often overlooked. This omission limits the ability to generalise findings to the broader population.

Objective: This study demonstrates three approaches for estimating the population average treatment effect (PATE) in a practical example, examining the impact of household second-hand smoke (SHS) exposure on blood pressure in school-aged children.

Methods: Using data from the National Health and Nutrition Examination Survey (NHANES) 2017-2020, we assessed the effect of household SHS exposure, a non-randomised treatment, on blood pressure in school-aged children. We applied estimators based on Inverse Probability of Treatment Weighting (IPTW), G-computation, Targeted Maximum Likelihood Estimation (TMLE), and regression adjustment. Models without adjustments were run for comparison. We examined point estimates and the efficiency of the estimates obtained from these methods.

Results: The largest differences were observed between the unadjusted regression models and the fully adjusted methods (IPTW, G-computation, and TMLE), which account for both confounding and survey weights. While the inclusion of the sample weights leads to wider confidence intervals for all methods, G-computation and TMLE showed comparatively narrower confidence intervals. Confidence intervals for the models not adjusted for sample weights were likely underestimated.

Conclusions: This study highlights the important role of sample weights in causal inference. Generalisability of the average treatment effect as estimated on data sampled using common survey designs to a defined population requires the use of sample weights. The estimators described provide a framework for incorporating sample weights, and their use in health research is recommended.

Background: Generally, studies in perinatal epidemiology restrict cohort entry to 20 weeks of gestation, but exposures and outcomes may occur earlier. This restriction may introduce left truncation bias.

Objectives: To examine the impact of left truncation bias when estimating the causal effect of abruption on perinatal mortality in the context of abnormal placentation, with spontaneous abortion (SAB) as a censoring event.

Methods: Through 80 Monte Carlo simulation scenarios based on realistic clinical assumptions, we estimated risk differences (RD), risk ratios (RR) and bias parameters for the abruption-perinatal mortality association.

Results: Censoring by SAB ranged from 5.6% to 7.6% across simulation setups. The risk of mortality was overestimated in observable (left-truncated) data at ≥ 20 weeks compared to an unobservable cohort starting follow-up at placental implantation (conception cohort). Underestimation of risks was stronger among abruption pregnancies. RDs for the abruption-mortality association were biased by +1% to +3% among conceptions with normal implantation and by +5% to +43% among abnormal placentation. Due to the disproportionate underestimation of mortality among nonabruption pregnancies, RRs were overestimated by 1.1 to 1.2-fold for normal implantations and by 1.1 to 8.4-fold for abnormal implantations.

Conclusions: The findings of this simulation study highlight the critical importance of placentation in successful pregnancy. Abnormal placentation has profound consequences for unsuccessful pregnancies, remarkably increasing the risks of early losses, placental abruption and other obstetrical complications. This study underscores that left truncation can bias the abruption-perinatal mortality association, differentially by whether the placentation was normal or abnormal. However, defining the causal question regarding the abruption-perinatal mortality association requires consideration of the target population, which may include all conceptions. In studies of these effects, outcome follow-up capability may introduce left truncation bias. We do not prescribe one analytic approach to account for left truncation, but rather, the approach should be guided by the causal question.

Background: Data on antihypertensive medication for non-severe gestational hypertension may suffer from immortal time and selection bias. Emulating target trials can prevent these biases by aligning follow-up with treatment initiation.

Objectives: We estimated the safety of antihypertensive medication initiation for the treatment of non-severe gestational hypertension on adverse birth outcomes in Botswana using sequential target trial emulation.

Methods: Data from the Tsepamo study (2014-2022), capturing birth outcomes at government delivery sites in Botswana, was used to examine antihypertensive medication initiation ≥ 24 weeks gestation for non-severe gestational hypertension (140-159 systolic or 90-109 diastolic blood pressure ≥ 20 weeks gestation without chronic hypertension). Sequential weekly target trial emulation compared initiation versus no initiation during 24-35 weeks' gestation on the risk of stillbirth and birth of infant small for gestational age (SGA), with secondary outcomes including very SGA, preterm birth, very preterm birth, neonatal death, and severe gestational hypertension. For each trial, eligible individuals were without chronic hypertension, had not previously initiated antihypertensive medication and had ≥ 2 non-severe blood pressure readings, at least one within 1 week of trial start. Log-binomial models estimated gestational week-specific and pooled risk ratios (RR) with 95% confidence intervals (CI) using bootstrapping.

Results: Of eligible individuals, there were 1676 antihypertensive initiator 'person-trials' and 5211 non-initiator 'person-trials'. In the pooled analysis, the adjusted RR for stillbirth and SGA comparing initiators to non-initiators was 0.92 (0.68, 1.19) and 1.09 (0.97, 1.23), respectively. The pooled adjusted RR for secondary outcomes were: very SGA, 1.05 (95% CI 0.88, 1.25); preterm birth, 1.09 (95% CI 0.96, 1.22); very preterm birth, 1.05 (95% CI 0.78, 1.47); neonatal death, 1.23 (95% CI 0.68, 2.24); severe gestational hypertension, 0.88 (95% CI 0.74, 1.07).

Conclusions: In this retrospective cohort study, antihypertensive medication initiation between 24 and 35 weeks' gestation for non-severe gestational hypertension was not associated with increased risk of adverse birth outcomes.