Objectives: Early hemorrhage detection in intensive care units (ICUs) enables timely intervention and reduces the risk of irreversible outcomes. In this study, we aimed to develop a machine learning model to predict hemorrhage by learning the patterns of continuously changing, real-world clinical data.Methods: We used the Medical Information Mart for Intensive Care databases (MIMIC-III and MIMIC-IV). A recurrent neural network was used to predict severe hemorrhage in the ICU. We developed three machine learning models with an increasing number of input features and levels of complexity: model 1 (11 features), model 2 (18 features), and model 3 (27 features). MIMIC-III was used for model training, and MIMIC-IV was split for internal validation. Using the model with the highest performance, external verification was performed using data from a subgroup extracted from the eICU Collaborative Research Database.Results: We included 5,670 ICU admissions, with 3,150 in the training set and 2,520 in the internal test set. A positive correlation was found between model complexity and performance. As a measure of performance, three models developed with an increasing number of features showed area under the receiver operating characteristic (AUROC) curve values of 0.61–0.94 according to the range of input data. In the subgroup extracted from the eICU database for external validation, an AUROC value of 0.74 was observed.Conclusions: Machine learning models that rely on real clinical data can be used to predict patients at high risk of bleeding in the ICU.

As the healthcare environment is being digitalized and changed rapidly, research using medical big data is increasing. One of the most applicable data is electronic medical records which can provide a large amount of clinically practical meaning. Electronic medical data include patient’s demographic information, laboratory test results, imaging and biosignal data. In this article, we provide support for a wide variety of researchers in their efforts to use electronic medical record data accurately and usefully in their work. From the basic concept of the research using electronic medical records to challenging aspects like data integration between multiple institutions are described. Also, examples of each type of data are covered; structured such as numeric data and unstructured such as images, biosignals and narrative text. Using these kinds of electronic medical records, analyses are processed by data cleansing, transforming, and reducing in order. Many kinds of variables such as the exposure and outcome of interest, covariate and the research design can be chosen during the preprocessing. As many machine-learning-based studies as well as epidemiologic-based studies have been conducted using electronic medical records, various research frameworks have been proposed. However, data quality management and data standardization for multicenter data analysis are still remaining as challenging tasks.

Introduction: Human exposure to parabens is very common in daily life, and prenatal exposure to these chemicals is associated with poor birth outcomes. Therefore, the aim of this study was to investigate the effect of glutathione S-transferase (GST) polymorphisms on the association between prenatal exposure to parabens and birth outcomes. Methods: We conducted a multivariate analysis involving 177 subjects to determine the association between paraben concentrations and birth outcomes in mothers with GST mu 1 (GSTM1) and GST theta 1 (GSTT1) polymorphisms from 2017 to 2019. Furthermore, we determined the interactive effect between paraben levels and GSTM1/GSTT1 polymorphisms using regression analysis, in addition to a generalized linear model after stratifying GSTM1/GSTT1 genotype into three categories. Results: Methyl and propyl paraben concentrations were significantly and positively associated with birth weight (methyl, β = 116.525, 95% confidence interval (CI) = 22.460–210.590; propyl, β = 82.352, 95% CI = 9.147–155.557) in individuals with the GSTM1-null genotype. Moreover, the propyl paraben concentration was significantly associated with an increase in gestational age (β = 0.312, 95% CI = 0.085–0.539) in individuals with the GSTM1-null genotype. Conclusions: This study reported the association between prenatal paraben exposure and birth outcomes in individuals with GST polymorphisms. We found positive relationships of maternal exposure to methyl parabens with birth weight in both mothers with GSTM1 and GSTT1-null genotypes.

Bisphenols are endocrine disruptors that may be associated with altered fetal growth in humans, and they have similar biological functions to mimic hormones. In addition, aggregated chemicals showed an adverse effect although individual concentration was at a low level. However, most studies between bisphenols and birth outcomes have focused on the effect of individual bisphenol. Thus, we explored the associations of urinary bisphenol mixtures with birth outcomes. We conducted a prospective birth cohort study in South Korea. One hundred eighty mother-infant pairs were recruited from 2017 to 2019. Bisphenol A (BPA), bisphenol F (BPF), and bisphenol S (BPS) in one spot urine were analyzed using ultra-performance liquid chromatography–tandem mass spectrometry. We used two statistical approaches to examine potential associations of BPA, BPF, and BPS with birth weight and gestational age: (1) multivariable linear regression; (2) Bayesian kernel machine regression (BKMR). The geometric means of BPA, BPF, and BPS were 2.1, 0.2, and 0.1 μg/L, respectively. In stratified linear analyses by each median value, a higher BPF was positively associated with birth weight (g) (β = 125.5; 95% CI: 45.0 to 205.9). Mixture analyses using BKMR suggested an inverse association between bisphenol mixtures and birth weight. Our findings suggest that in utero bisphenol exposure may influence birth weight and that such relationships may differ considering non-linearity and the combined effect.

Introduction: Bisphenol F (BPF) and bisphenol S (BPS) are chemical substitutes for, and may have similar physiological effects to, bisphenol A (BPA). Bisphenols provoke endocrine disorders and are cytotoxic, oxidize hemoglobin, and induce morphological changes in human red blood cells (RBC). It is more sensitive to changes in the RBC number and hemoglobin (Hb) level during pregnancy. Therefore, we investigated the effects of bisphenols (BPs) and their substitute compounds on hemopoiesis and the serum biochemical parameters of pregnant women. Methods: The study population comprised 196 pregnant women from the MAKE cohort study, recruitment for which occurred from 2017 to 2019. We measured the levels of BPA, BPF and BPS in urine samples and collected data on socioeconomic, lifestyle, and environmental factors at visits to the hospital. The associations between the levels of the three BPs and biochemical parameters were analyzed by multiple linear regression. Results: The geometric mean urinary concentrations of specific gravity adjusted BPA, BPF, and BPS were 2.1, 0.2, and 0.1 mu g/L, respectively. There was a significant negative association between the urinary concentration of BPA and the Hb level (beta = -0.5, p = 0.02). After stratifying by the median concentrations of the three BPs, the maternal urinary BPA level had a significantly negative effect on the RBC count, HB level, and hematocrit in the high BPA concentration group (RBC, beta = 0.5, p = 0.001; Hb, beta = -1.4, p = 0.002; Hct, beta = - 5.0, p = 0.001). Conclusions: BPA has a harmful effect on hemato-biochemical changes that occur during pregnancy. Further studies should investigate the relation between widespread exposure to bisphenols and effects on human health.