Impact of birth weight to placental weight ratio and other perinatal risk factors on left ventricular dimensions in newborns: a prospective cohort analysis

Study design and setting

This single-center prospective cohort study investigated the impact of key parameters, including placental factors, on echocardiographic LV dimensions and morphology in newborns. The primary objective of the study was to evaluate the relationship between the BW/PW ratio and different LV dimensions in newborns while considering other neonatal, maternal, and placental factors. Echo evaluations were conducted within 48–72 h after delivery and before hospital discharge. The study was conducted at NewYork-Presbyterian Brooklyn Methodist Hospital’s nursery ward and neonatal intensive care unit (NICU) from 2014 to 2018, adhering to the Helsinki Declaration and approved by the hospital’s Institutional Review Board (IRB). Data security was maintained by appointing distinct identification numbers to each participant and keeping the data securely in a protected repository. Reporting followed the strengthening the reporting of observational studies in epidemiology (STROBE) guidelines [14].

Eligibility criteria

This study included neonates who underwent echocardiographic evaluations at the hospital mostly for murmur evaluation as part of routine screening. Neonates below the 10^th percentile or above the 90^th percentile on the Fenton growth chart were selected for routine screening echocardiography [15]. Neonates with major cardiac pathology, significant congenital abnormalities, hypoxic respiratory failure, severe sepsis/shock, low 5 min APGAR score, genetic diagnosis, perinatal depression, or born before the 35^th week of pregnancy were excluded. Significant cardiac pathologies, as exclusion criteria in our study, encompass any structural anomalies detected via echocardiography, irrespective of the clinical presentation. This includes, but is not limited to, congenital heart diseases such as septal defects, complex cyanotic heart conditions, and outflow tract obstructions, as well as valvular disorders like severe stenosis or regurgitation. Symptomatic newborns with conditions such as persistent pulmonary hypertension of the newborn and cardiac dysfunctions that lead to clinical instability were also excluded from the study. Conversely, neonates with trivial cardiac findings, which do not impact hemodynamics – like a patent ductus arteriosus that presents with minimal shunting or a patent foramen ovale measuring less than 3 mm – were not excluded and therefore participated in the study. These minor conditions are typically not associated with significant clinical consequences and are often considered normal variants in neonates. Parental consent was not obtained as echocardiography scans were part of routine screening protocols for newborns. Following echocardiography testing and discharge, no further follow-up was done for the included neonates.

Predictor variables

We used neonatal, maternal, and placental factors as predictor variables. The included population was divided into three categories based on their BW/PW ratio using reference ranges provided by the study conducted by Flatley et al. (2022) [16]. These reference ranges were adjusted for GA, sex, and parity. The three groups were categorized as follows: small BW/PW (less than the 10^th percentile), normal BW/PW (between the 10^th and 90^th percentiles), and large BW/PW (more than the 90^th percentile). Utilizing the same percentiles cutoff, we categorized the population into three size groups: SGA, appropriate for gestational age (AGA), and large for gestational age (LGA) based on Fenton charts [15].

Neonatal variables included sex, GA, birth weight, height, head circumference, chest circumference, size category, ponderal index, APGAR scores at 1 and 5 min, blood pressure (BP) measurements (diastolic, systolic, and mean PB), and NICU admissions. Maternal covariates included gravidity, parity, maternal age, body mass index (BMI), mode of delivery, gestational diabetes mellitus (GDM) diagnosis and management, ethnicity, and preeclampsia. GDM diagnosis followed the criteria set by the American Diabetes Association (ADA) collected from medical records at the time of delivery [17], and its management was divided into diet/lifestyle, medication, or insulin. The diagnosis of preeclampsia in our study was extracted from medical records at the time of delivery, which was primarily made using the criteria established by the American Congress of Obstetricians and Gynecologists (ACOG) during the study period, while BP readings and BMI (using height and weight) were extracted at the time of delivery by our institution nurses. Placental factors considered were PW, BW/PW ratio, BW/PW category, placental dimensions (largest and smallest), and placental thickness. Relevant maternal medical history was obtained from electronic medical records, while neonatal and placental information was collected after birth.

Outcome variables

The dependent variables included inter-ventricular septal thickness during diastole (IVSd) and systole (IVSs), left ventricular internal dimension during diastole (LVIDd) and systole (LVIDs), left ventricular posterior wall thickness at end of diastole (LVPWd) and systole (LVPWs), shortening fraction (FS), left ventricular mass (LVmass), left ventricular mass to volume ratio (LVmass/vol), and asymmetrical septal hypertrophy (ASH). The only binary variable, ASH, was derived from the IVSd to LVPWd ratio, while all other variables were continuous and either measured directly or estimated using echocardiography. LV morphology was evaluated using the 2D method for structural assessment and the M-mode method for functional evaluation. Echocardiography was conducted by a single board-certified cardiologist. In cases where the cardiologist is not available, the echo study is conducted by a technician and is subsequently re-reviewed and interpreted by the cardiologist. Most LV dimensions measures were conducted using the short-axis view, while some were done using the parasternal long-axis view. All echo LV dimensions measured were done in one single cycle. A Philips 5500 ECHO (Philips, Andover, MA, USA) device was used for examining the heart by echocardiography.

Data analysis

We utilized multiple imputations to handle missing data in two sub-datasets – one for numerical variables and another for categorical variables. For numerical variables, we normalized the data using a logarithm transformation and then utilized a technique called multiple imputations with principal component analysis (MIPCA) with PCA to fill in missing values, creating a complete dataset [18]. For categorical variables, we employed MIPCA with multiple correspondence analysis (MCA) [18, 19]. This approach has advantages in managing large datasets with missing values, accurately estimating missing values, reducing dataset complexity, and effectively analyzing categorical data [19].

Then, we compared the characteristics of three BW/PW categories at baseline using different statistical tests. For categorical variables, the Chi-squared test was used, while for continuous variables, a one-way analysis of variance (ANOVA) was employed [20]. For statistical modeling, a genetic algorithm was used for model selection based on the lowest Akaike information criterion (AIC) [21, 22]. The genetic algorithm iteratively searches for the optimal variable combination based on fitness criteria, inspired by natural selection and genetics. Afterward, a backward stepwise algorithm based on the F-test was applied to refine the model, retaining only significant variables at a 5 % significance level [23]. The backward stepwise algorithm starts with all the variables selected by the genetic algorithm and removes non-significant ones based on the F-test [23].

Baseline characteristics were analyzed and presented in tables with frequencies, percentages, mean, and standard deviations. Regression model results were also presented in tables, showing estimates, standard errors, 95 % confidence intervals (CI), and p-values for each predictor variable. Statistical significance was determined using a p-value threshold of less than 0.05. The model fit evaluation included R-squared (R²) and log-likelihood. Statistical analyses for Chi-squared and one-way ANOVA tests were conducted using Stata software (version 16.0, StataCorp LLC, College Station, TX). The missMDA package was used for imputing missing values, and the glmulti package was utilized for model selection using the genetic algorithm in RStudio Build 446 [18, 19, 22].

Selection and inclusion process

Figure 1 presents an overview of the selection and inclusion process used in our study. Within the study period, a total of 908 neonates were admitted to the hospital. After excluding 81 neonates based on the selection criteria (reasons outlined in Figure 1), we included 827 neonates in our final analysis. Among these neonates, 16 were classified in the small BW/PW group (with BW/PW ratio less than the 10^th percentile), 488 were in the large BW/PW group (with BW/PW ratio above the 90^th percentile), and the remaining 323 neonates were categorized in the normal BW/PW group.

Figure 1:

Overview of the selection process of neonates according to BW/PW ratio groups in our study.

Baseline characteristics

Table 1 presents the baseline characteristics of the neonates categorized based on their BW/PW ratio groups. The comparison of categorical variables among the groups revealed no statistically significant differences in sex, birth weight category, APGAR scores at 1 and 5 min, NICU admission, mode of delivery, gravidity, parity, preeclampsia, GDM status, GDM management, and ethnicity. Maternal insulin use for GDM was documented in 37 cases, while medication use was found in 66 cases (65 using glyburide and one using metformin). Notably, there were no significant variations in the distribution of these variables across the different BW/PW ratio groups.

Table 2 presents the differences in baseline characteristics between the BW/PW ratio groups in continuous variables. Neonates in the small BW/PW ratio group had significantly lower birth weight, birth height, head circumference, chest circumference, and higher PW compared to those in the normal and large BW/PW ratio groups as shown in Table 2. Conversely, GA was significantly lower in the large BW/PW ratio group compared to the other groups. However, there were no significant differences in the ponderal index, maternal age, maternal BMI, maternal BP (systolic, diastolic, mean), placental diameter 1 (largest diameter), placental diameter 2 (smallest diameter), and placental thickness among the BW/PW ratio groups. Table 3 presents the differences in LV parameters between the BW/PW ratio groups. Upon analysis, no significant differences were observed in most of the LV parameters among the groups, including IVSd, IVSs, LVIDd, LVIDs, LVPWd, LVPWs, FS, LVmass, and LVmass/vol. The presence of ASH showed no significant difference among the BW/PW ratio groups as well (p=0.664).

Regression results

Table 4 presents the regression analysis results, revealing significant associations between neonatal and maternal factors with various LV parameters. The IVSd model shows a strong fit with an R² of 0.341 and an adjusted R² of 0.333, explaining approximately 34.1 % of its variance. Its log-likelihood is −974.455, indicating a good model fit. IVSs also has a robust fit (R²=0.302, adjusted R²=0.295), while LVIDd has a slightly lower but still significant R² of 0.266. In contrast, LVIDs and FS exhibit weaker fits with R² values of 0.154 and 0.050, respectively. Notably, LVmass has the highest R² at 0.406, indicating a strong relationship with the model, while ASH, though lacking an R² value, shows an association with the model through a log-likelihood of −475.797.

GA displayed strong correlations with multiple LV dimensions, showing significant relationships with IVSd (p=0.011), IVSs (p<0.001), LVIDd (p<0.001), LVIDs (p<0.001), LVPWd (p<0.001), LVPWs (p<0.001), LVmass (p<0.001), and LVmass/volume ratio (p=0.047). Neonatal size factors, including size category, BW, ponderal index, and chest circumference, also influenced LV parameters. Additionally, male sex, NICU admission, and APGAR scores were associated with specific LV dimensions.

Among maternal factors, primiparity showed associations with decreased IVSd (estimate=−0.15, p=0.016), LVmass (estimate=−0.62, p=0.005), and LVmass/volume ratio (estimate=−1.84, p=0.032). GDM exhibited a significant increase in IVSs (p=0.010), while insulin use was positively correlated with IVSd (p<0.001). Maternal BMI was linked to decreased LVID during diastole (p=0.002) and systole (p=0.005), whereas maternal age was associated with increased LVPWd (p=0.036). Furthermore, White and Hispanic ethnicities related to increased ASH.

Among the placental factors analyzed, placental thickness and smallest diameter exhibited significant associations with increased LV dimensions. Placental thickness was linked to substantial increases in IVSd (p=0.002), IVSs (p=0.001), LVPWd (p=0.003), LVPWs (p<0.001), LVmass (p=0.017), and LVmass/volume (p=0.011). Similarly, the shortest diameter of the placenta displayed significant positive correlations with IVSd (p<0.001), IVSs (p<0.001), LVIDd (p<0.001), LVIDs (p=0.016), LVPWd (p<0.001), LVPWs (p<0.001), LVmass (p<0.001), and LVmass/volume (p<0.001). Moreover, the BW/PW ratio was associated with an increase in FS, with an estimated coefficient of 0.29 (95 % CI 0.03–0.55, p=0.027). Additionally, PW showed a significant correlation with IVSs (p=0.019), while the longest diameter of the placenta was associated with a decrease in LVIDd (estimate =−0.07, p=0.039), LVmass (estimate =−0.11, p=0.024), and LVmass/volume (estimate =−0.55, p=0.005).

Strength and limitations

This study exhibits several notable strengths that contribute to its scientific merit. Firstly, it adopts a prospective cohort design, allowing for the exploration of associations and reducing the potential for bias inherent in retrospective studies. The inclusion of neonates with echocardiographic evaluations conducted primarily for murmur evaluation as part of routine screening adds clinical relevance to the findings, making them more generalizable to clinical settings. Comprehensive consideration of neonatal, maternal, and placental factors provides a holistic understanding. Meticulous handling of missing data with MIPCA, coupled with genetic algorithm-based model selection followed by a backward stepwise algorithm, ensures comprehensive exploration and statistical validity in assessing predictor variables’ impact on LV parameters.

Despite the valuable findings of this study, it is important to acknowledge its limitations. The single-center nature of the study may limit generalizability of the results to other populations [37]. Additionally, the use of routine screening echocardiography may introduce selection bias, as neonates with major cardiac pathologies were excluded, potentially affecting the representation of certain LV dimensions in the study. The imputation technique used to handle missing data may introduce imprecision in the results, as using MIPCA for missing data has limitations. Efforts to address missing data through transformations and MIPCA may introduce biases. PCA for imputing missing data assumes a multivariate normal distribution, potentially misaligned with real-world datasets. It also assumes missing values occur at random, introducing bias, and may result in information loss or noise. Furthermore, due to the study’s observational design, causality between the BW/PW ratio and LV dimensions cannot be established, and other confounding factors may influence the results.

In addition, the study has other limitations. The lack of long-term follow-up of participants after hospital discharge makes it challenging to assess the long-term outcomes of LV function and overall cardiovascular risk for SGA infants. Moreover, the incorporation of neonates undergoing echocardiography for murmurs, predominantly functional in nature, introduces a potential confounding factor, including the likelihood of detecting minor congenital heart defects. This complicates the derivation of clear conclusions and hinders generalizability to broader populations, thereby challenging the clinical applicability and utility of the findings. Ideally, the study should be conducted on a cohort of unselected and healthy children to enhance the robustness and relevance of the results. Therefore, caution must be taken when interpreting the findings of this study. The study design was not blinded, which could introduce bias in measurements and interpretation of the results. Additionally, important confounding variables like maternal smoking, a known risk factor for low birth weight and cardiac dysfunction, were not included in the analysis [38]. Another missing risk factor and a potential limitation of our study is the lack of records regarding pre-existing maternal type one and two diabetes, which is well-known to influence both neonatal size and increase the risk of CHDs. Moreover, our collected data lacks information on the timing of the diagnosis (preterm/term onset) and the severity of preeclampsia, potentially introducing bias into our results. Preeclampsia is a significant influencer of cardiac remodeling during development as it imposes strain on the placenta and, consequently, affects the size of the baby. This, in turn, can place a burden on the developing heart. Understanding the specific timing and severity of preeclampsia is crucial for a more comprehensive analysis of its impact on cardiac health in neonates, which was missing in our study. We also did not have records of other medications used during pregnancy. Furthermore, the reliance solely on echocardiography to evaluate LV function has inherent limitations [39], and advanced imaging techniques for assessing LV morphology were not employed in the study. Lastly, it is worth noting that a single cardiologist conducted the echocardiographic evaluations, which may impact interobserver variability.