Analysis of birth defects surveillance in Urumqi from 2018 to 2023 and application of three kinds of model in prediction

Objective: This study aimed to assess the incidence and risk factors of perinatal birth defects in Urumqi (2018–2023), and compare the predictive accuracy of Joinpoint regression, Prophet, Grey Model (GM(1,1)), and a Bayesian-optimized hybrid model. Methods: Data were extracted from the Xinjiang Maternal and Child Health Cloud Platform. After quality control, we constructed the database in Excel, and analyses were performed using JMP 14.0 and R 4.4.1. Using population-based surveillance data from 36 midwifery institutions, we conducted trend analysis (Joinpoint regression) and time-series forecasting (Prophet, GM(1,1), and a Bayesian-weighted hybrid model). Model performance was evaluated by MAE, RMSE, MAPE, and R2. Results: The overall incidence of birth defects was 149.47 per 10,000, with a significant upward trend (χ2trend = 25.268, P < 0.001). Congenital heart disease (53.65 %) was the most prevalent defect. Higher incidence rates were observed in male infants, urban areas, and mothers aged≥35 years. The Grey Model showed the lowest prediction error (MAE = 21.8, MAPE = 15.86 %), while the Combined model achieved the highest R2(0.82) and lowest RMSE (29.34). Conclusion: The rising incidence of perinatal birth defects in Urumqi underscores the need for enhanced monitoring. Our findings advocate a tiered public health surveillance strategy: GM(1,1) for immediate-term (0–12 months), the hybrid model for medium-term (13–24 months), and Prophet for long-term (>25 months) planning, enabling resource prioritization in low-resource settings.