December 2, 2025Open Access

Comparative effectiveness analysis of univariate time-series forecasting models for disease mortality rates in the global burden of disease database: a case study of global hypertensive heart disease among women of childbearing age

Key Points

The LSTM model demonstrated superior predictive accuracy on the validation set, reducing the Mean Absolute Error significantly.
Mean Absolute Error metrics showed LSTM outperformed ARIMA, with lower error rates indicating a robust forecasting approach.
Evaluation involved univariate forecasting models using the global burden of disease database spanning 1990-2021 data.
Implications suggest enhanced resource allocation for hypertensive heart disease management based on predictive insights.

Abstract

Objective The mortality rate of hypertensive heart disease (HHD) among women of childbearing age (WCBA) worldwide is continuously increasing. Accurate prediction of the mortality rate of HHD among WCBA globally plays a crucial role in evaluating the effectiveness of intervention measures and predicting future disease trends. To date, there has been few systematic comparative evaluations of prediction methods for epidemiological indicators in the field of disease burden. The purpose of this study was to systematically compare the performance of univariate prediction models in the global burden of disease (GBD) database. Method Global mortality data on HHD in WCBA (1990–2021) were split into training and validation sets. We implemented and compared four models: AutoRegressive Integrated Moving Average (ARIMA), Prophet, eXtreme Gradient Boosting (XGBoost), and Long Short-Term Memory (LSTM). Model performance was assessed using Mean Squared Error (MSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and the Diebold-Mariano (DM) test for statistical significance. Results The LSTM model demonstrated superior predictive accuracy on the validation set, with the lowest error rates across all metrics (MSE: 0.00021; MAE: 0.00872; MAPE: 0.662%). All the other models demonstrated statistically significant superiority over ARIMA (MSE: 0.03645; DM test p 0.05 for all metrics). According to the DM test, both Prophet and LSTM demonstrated high predictive accuracy ( p = 0.8762 for DM test based on MSE; p = 0.4292 for DM test based on MAE; p = 0.4303 for DM test based on MAPE). The LSTM model predicted that the mortality rate will exhibit an initial decline followed by a stabilization trend from 2022 to 2030, while the Prophet model predicted that the mortality rate will continue to rise. Conclusion This study provided the first systematic comparison of univariate forecasting models for HHD mortality in WCBA using GBD data. A key finding was that both LSTM and Prophet performed exceptionally well statistically, LSTM achieves superior predictive capability via its gated mechanisms and state memory, while Prophet enhances interpretability through its additive model structure. This study therefore provides practical guidance for health authorities to select appropriate models based on actual needs to support improved resource planning for HHD.

Ask AI

Helpful

Bookmark

View Full Paper