Establishment of a risk prediction model for hydrocephalus complicated by neonatal bacterial meningitis

Abstract

Background: Hydrocephalus is a severe complication of neonatal bacterial meningitis (NBM), threatening the health and quality of life of neonates, affecting the outcome of nervous system development, and leading to neurological sequelae, such as movement disorders, hearing impairment, mental retardation, and epilepsy. Improvement in prognosis is closely related to early identification and active treatment.

Objective: To find the independent risk factors of NBM complicated with hydrocephalus, to construct the related risk prediction model and validate it, in order to provide help for clinicians to identify the children with high risk of hydrocephalus at an early stage, to guide clinical decision-making and improve prognosis.

Methods: 528 children with NBM hospitalized in Kunming Children's Hospital from January 2019 to December 2022 were selected. After excluding 46 patients with incomplete medical records And 1 death case, 481 patients remained. They were randomly divided into a training set (n = 337) and a validation set (n = 144) (the division ratio was 7:3) by using the split function in R language. The basic information, cerebrospinal fluid routine biochemistry, blood routine, blood culture, imaging findings, and other indicators of the children were collected. Determination of whether hydrocephalus was complicated based on the child's brain magnetic resonance imaging or CT. LASSO regression was used to screen independent risk factors for NBM complicated by hydrocephalus, And independent risk factors were Analyzed by using multivariate logistic regression. The risk prediction model for NBM complicated by hydrocephalus was constructed by using the analysis results, and a nomogram was created. The model was internally validated based on the cases in the training and internal validation sets. A total of 132 children with NBM who were hospitalized at Peking University First Hospital from January 2006 to December 2021 were included in the study. After excluding 2 cases with incomplete medical records, the remaining 130 cases were used as external validation cases to externally validate the model.

Results: Twenty predictive variables were screened out by LASSO regression analysis, including NBM type, BW, age of onset, pregnancy complications, gestational age, birth asphyxia, umbilical cord, amniotic fluid, maximum body temperature, vomiting, convulsions, anterior fontanel, blood culture, PLT, peak value of WBC, peak value of N, peak value of PLT, CSF multinucleated percentage peak, lowest value of CSF glucose, and intracranial hemorrhage. The results of multifactorial Logistic regression analysis after oversampling showed that the significant risk factors were intracranial hemorrhage (OR = 6.922, P < 0.001), anterior fontanel (OR = 8.002, P < 0.001), lowest value of CSF glucose (OR = 0.416, P < 0.001), gestational week (OR = 0.870, P = 0.0088), maternal pregnancy complications (OR = 0.284, P = 0.0118), convulsions (OR = 2.906, P = 0.0178), amniotic fluid (OR = 2.417, P = 0.0263), and CSF multinucleated percentage peak (OR = 1.011, P = 0.0350). There was no statistically significant difference between convulsions, maternal pregnancy complications and CSF multinucleated percentage peak in binary logistic regression. Therefore, a nomogram risk prediction model was created with the remaining five predictive variables. The area under the ROC curve (AUC) of the training set after weighting was 0.925 (95%CI = 0.899-0.951), the internal validation set was 0.894 (95%CI = 0.829-0.959), And the external validation set was 0.758 (95%CI = 0.677-0.839); the goodness-of-fit test showed that the training set P = 0.431, internal validation set P = 0.224, and external validation set P = 0.176. Decision curve analysis (DCA) showed that the net benefit of the model was higher than the net benefit at the extremes in a large range of thresholds in the training set, internal validation set, and external validation set.

Conclusion: The Nomogram risk prediction model established in this study, which includes five indicators of the lowest CSF glucose level, combined intracranial hemorrhage, anterior fontanel, gestational week, and amniotic fluid, can early predict the risk of NBM complicating hydrocephalus.

Keywords: Complications; Hydrocephalus; Neonatal bacterial meningitis; Nomogram; Risk factors.

Fig. 1

Schematic diagram of inclusion and exclusion process for NBM cases

Fig. 2

Distribution of Propensity Score Weights in Internal and External Validation Sets

Fig. 3

Distribution Histogram of Weight Values in Internal and External Validation Sets

Fig. 4

Selection of predictors of prostate biopsy using the LASSO binary logistic regression model. Note: A represents the coefficient curve of 32 variables, and B represents the process of screening the optimal λ

Fig. 5

Nomogram predicting NBM complicated with hydrocephalus

Fig. 6

ROC curve of the nomogram for predicting NBM complicated with hydrocephalus

Fig. 7

Calibration curve of the nomogram for predicting NBM complicated with hydrocephalus

Fig. 8

Decision curve analysis of the nomogram predicting NBM complicated with hydrocephalus. Note: In the figure, the abscissa represents the threshold probability, and the ordinate represents the net benefit value. The black line indicates the net benefit when no participants are predicted, while the red line represents the net benefit when all participants are predicted. The area between the black line and the red line in the model curve represents the clinical utility of the model. Training set (a), internal test set (b), and external test set (c)