Diagnostic Models of Neonatal Respiratory Distress Syndrome and Congenital Pneumonia: A Retrospective Cohort Study

Background: The differential diagnosis of respiratory distress syndrome (RDS) and congenital pneumonia (CP) in newborns remains a complex clinical challenge due to the similarity in their clinical manifestations and their potential to coexist. Objective: We aimed to determine differential diagnostic predictors of RDS and CP in newborns by using mathematical modeling and machine learning methods. Methods: A retrospective cohort study was conducted; de-identified medical records of 244 newborns (97 with RDS and 143 with CP) were collected to assess clinical, anamnestic, laboratory, and instrumental data by applying multiple regression analysis, ROC analysis, logistic regression models, and Random Forest. Results: Patients with CP presented with a more severe condition at admission (57.1% vs. 23.3%; p = 0.023), required mechanical ventilation (MV) more frequently (22.4% vs. 8.2%; p = 0.004), and were more often transferred to the intensive care unit (ICU) (77.3% vs. 55.7%; p = 0.001). They further had lower hemoglobin levels (151 ± 28 g/L vs. 164 ± 31 g/L; p = 0.001) and red blood cell counts (p = 0.021). Regression analysis demonstrated that the severity of the condition and the presence of cerebral ischemia were dependent on hemoglobin levels in the case of CP, while gestational age played a determining role in RDS. The machine learning models achieved an accuracy of 0.69 and an area under the curve (AUC) of 0.82 (Random Forest). The key predictors for differential diagnosis of RDS were low gestational age, hyperbilirubinemia, and congenital heart defects, while for CP, they were hemoglobin < 151 g/L, lymphocytes < 4.8 × 10³/μL, oxygen saturation < 90-91%, and cerebral ischemia. Conclusions: The use of mathematical modeling methods made it possible to identify prognostically significant predictors for the differential diagnosis of RDS and CP. The resulting models are best viewed as proof-of-concept tools for hypothesis generation and future research, as external validation is necessary before they can be considered for clinical use.