Construction and verification of a risk factor prediction model for neonatal severe pneumonia

Abstract

Objective: To construct and validate a risk factor prediction model for neonatal severe pneumonia.

Methods: This study collected data from newborns diagnosed with pneumonia in Children's Hospital Affiliated to Zhengzhou University. A total of 652 newborns were included. Risk factors were identified using Least Absolute Selection and Shrinkage Operator (LASSO) regression and logistic regression analysis. The nomogram was used to construct a prediction model. The effectiveness of the model was evaluated using calibration curve, receiver operating characteristic (ROC) curve, and decision curve analysis (DCA).

Results: Out of 652 newborns, 186 (29%) were diagnosed with severe pneumonia. The patients were randomly divided into a training set (n = 554) and a testing set (n = 98) in a ratio of 85:15. A total of 30 indicators were analyzed. Respiratory rate (OR = 1.058, 95% CI: 1.035-1.081), weight (OR = 0.483, 95% CI: 0.340-0.686), C-reactive protein (CRP) (OR = 1.142, 95% CI: 1.028-1.268), neutrophil (NEU) (OR = 1.384, 95% CI: 1.232-1.555), hemoglobin (HGB) (OR = 0.989, 95% CI: 0.979-0.999), uric acid (UA) (OR = 1.006, 95% CI: 1.002-1.010), and blood urea nitrogen (BUN) (OR = 1.230, 95% CI: 1.058-1.431) were identified as independent risk factors for neonatal severe pneumonia. The calibration curve showed significant agreement. The area under the ROC curve (AUC) was 0.884 (95% CI: 0.852-0.916) for the training set, and 0.835 (95% CI: 0.747-0.922) for the testing set. DCA demonstrated good predictive properties.

Conclusion: The prediction model based on respiratory rate, weight, CRP, NEU, HGB, UA, and BUN has shown promising predictive value in distinguishing between mild to moderate pneumonia and severe pneumonia in neonates.

Keywords: neonatal; nomogram; predictive model; risk factor; severe pneumonia.

FIGURE 1

The flow diagram for developing and validating the prediction model for neonatal severe pneumonia.

FIGURE 2

Correlation bubble chart shows the Spearman correlation between parameters. Where one variable is plotted on the x-axis and the other on the y-axis for both severe pneumonia and mild to moderate pneumonia; antique blue for positive correlation and red for negative correlation.

FIGURE 3

Variable selection using the LASSO binary logistic regression model. (A) A coefficient profile was generated based on the logarithmic (lambda) sequence, and the optimal lambda value was used to produce non-zero coefficients. The colored lines represent different variables. As log (λ) gradually increases, the regression coefficient continuously converges until it becomes zero, thereby screening the variables. (B) The LASSO coefficient curves for the 30 variables were plotted against the log (λ) sequence. A vertical line was drawn at the value selected using 10-fold cross validation, which resulted in nine non-zero coefficients for the optimal λ. X-axis: Log (λ) value at selection point. Y-axis, Standardized coefficient magnitude.

FIGURE 4

A nomogram has been developed to predict the likelihood of severe pneumonia in neonates with pneumonia. To calculate total score and predicted probability of neonatal severe pneumonia, points from individual variables are added and a vertical line is drawn from the total points line at the bottom downward to determine the diagnosis rate. Example calculation: Estimate the risk for a neonatal severe pneumonia with: respiratory = 40 rate/min (Points: 10), weight = 3 kg (Points: 10), CRP = 150 mg/L (Points: 75), NEU = 10 × 109 cells/L (Points: 12.5), HGB = 0.1 g/L (Points: 10), UA = 250 mmol/L (Points: 3.75), and BUN = 3 mmol/L (Points: 2.5). Step 1, Sum the points (123.75); Step 2, Locate 123.75 on “Total Points” axis; Step 3, Read corresponding risk (>99.9%) on “diagnosis rate” axis.

FIGURE 5

Calibration, discrimination, and clinical application of nomogram prediction models in both the training and testing sets. (A) Calibration plot for the training set. (B) Calibration plot for the testing set. (C) ROC curves for the training set. The starred point (★) denotes the optimal threshold. (D) ROC curves for the testing set. The starred point (★) denotes the optimal threshold. (E) DCA curves for the training set. (F) DCA curves for the testing set.