In this study, we evaluated perinatal risk factors for offspring NDD using a high-precision model with random forest machine learning and SHAP variable importance analysis. The results showed that maternal age and low socio-economic status had the greatest influence on the development of NDD. In addition, maternal risk factors such as psychological problems, pregnancy complications such as PIH and GDM, maternal pre-pregnancy DM, FGR, small for gestational age (SGA), and male fetal risk factors were associated with NDD. Furthermore, the top-ranked important variables such as pre-gestational diabetes (DM)/hypertension (HTN), gestational diabetes mellitus (GDM), and pregnancy with hypertension (PIH) are highly similar to previous literature evaluating risk factors for NDD.17,18,19,20.
The DOHaD theory suggests that an uncertain environment in utero during early fetal development influences health risk factors in offspring in adulthood.7,8Based on this hypothesis, we will predict and identify groups of pregnancies at high risk for fetal health and evaluate preventive diagnosis, early intervention, and treatment for mothers.Five.
Age is a well-known risk factor for pregnancy complications. Both very young and older mothers at the time of birth can have adverse outcomes for their children, including low birth weight and neonatal mortality.21,22Gao et al. reported that in terms of NDD, younger and older maternal age at birth were associated with risk of ADHD and LD.6Our results show that age is one of the most important variables for the model (Table 3) and show a U-shaped pattern in Figures S1, S2, and S3, indicating that younger and older age are associated with risk of MDD, CDD, and NDD, while SES and age are the most relevant factors.
Maternal psychological status and substance use also influenced the offspring's NDD. Stress during pregnancy is also known to cause brain inflammation and affect fetal brain development.twenty threeIt is well known that an increase in stress-related corticosteroid hormones such as cortisol and corticosterone is a consequence of stress. Exposure of the fetus to high levels of cortisol leads to developmental delays and NDDs.FiveFurthermore, several researchers have reported that antidepressants, such as selective serotonin reuptake inhibitors (SSRIs), may affect the development of ASD depending on whether or not disorders of the serotonin system are involved in the pathophysiology of ASD.24,25In our results, maternal pre-pregnancy history of depression and anxiety and antidepressants are important risk factors for the development of NDD. In particular, anxiety and antidepressant use had a very high positive correlation with NDD in the SHAP value analysis. Moreover, these factors are the covariates most affected by other variables in the SHAP independence plot. Maternal genetic vulnerability also influences the neurodevelopment of offspring and is associated with pregnancy-related factors.26However, maternal genetic psychopathology was not defined in this study, and this limitation requires further investigation.
In this study, fetal risk factors such as SGA, FGR, and male gender were associated with the development of NDD. In general, FGR leads to SGA and brain remodeling, resulting in reduced limbic gray matter volume. Furthermore, regional volume expansion of the fronto-insular cortex, frontal lobe, and temporoparietal lobe influences the impairment of balanced neurodevelopment.27Furthermore, the male predominance in the incidence of NDDs is often emphasized.28In many species, including humans, women have been shown to generally have stronger immune responses and are better able to resist disease and infection than men.29Sex differences in neurodevelopment arise due to several neurological disorders caused by pathological reactive microglia in the central nervous system.30Quinn et al. reported from a large-scale study that there are gender differences in reading comprehension disabilities, which are largely due to male vulnerability rather than confirmation bias.31In our study, FGR and male gender were highly associated with the risk of NDD. Furthermore, in the dependency plot, FGR and male gender were strongly associated with each other, covariates as risk factors for NDD. In a large cohort study on the relationship between low birth weight and LD at age 11 years, Johnson et al. reported that low birth weight was associated with an increased risk of LD in boys but not in girls (OR = 4.32, 95% CI 1.55–12.04). Furthermore, these results depend on differences in SES.15In our study, SES, FGR, and male gender were highly ranked variables in the importance analysis, a result that reinforces the findings of previous studies.
A limitation of this study is that it is a retrospective analysis of an administrative database and relies on the accuracy and consistency of individuals coding the data. Therefore, it was not fully adjusted for severity or grade of NDD. Furthermore, due to limitations in extracting data on body mass index, adjustment for well-known risk factors such as pre-pregnancy obesity was not performed.17,32In this cohort, major problems of NDD such as ADHD and ASD are excluded, so the application of these models is limited. However, as mentioned above, these two problems affect genetic and heritable factors more than other NDDs, so confounding bias occurs when evaluating pre-conception or pregnancy risk factors. In this study, age and SES are the two factors that most affect the MDD, CDD, and NDD models. Therefore, the influence of other variables, respectively, may be underestimated. For this, further subgroup analysis adjusting for age and SES is required. FGR and preterm birth (PTB) are also known as major risk factors for NDD.33,34Notably, FGR is of early rather than late onset, with significant implications for a more severe pattern of neurodevelopmental outcome in offspring due to linkage to placental insufficiency.33,35Furthermore, PTB during early pregnancy may interact with FGR and be an important risk factor for NDD.34However, the original coding dataset did not specify the gestational age at diagnosis of FGR and PTB, which may have underestimated the impact of these factors. Despite the above limitations, this study had the advantage of evaluating the association between NDD and various pregnancy risk factors on a large-scale, nationwide scale using a machine learning model with high accuracy and validity compared with logistic regression with interactions and nonlinear terms. Based on unprecedented scale population-based data and the extraordinary performance of random forests, this study is expected to ensure high validity and reliability.
