Forecasting the future burden of major diseases in Kazakhstan using global burden of disease and time series models (ARIMA, Prophet, LSTM), 2019-2032

Background: Over the last years, it has become a widespread practice to use the global burden of disease (GBD) metrics for anticipating the disease patterns all over the world. Disability-adjusted life years (DALYs) is the main indicator to use for quantifying the losses in terms of health caused by disease. This study projects Kazakhstan’s disease burden from 2019 to 2032 by applying and comparing four forecasting approaches: GBD projections, auto-regressive integrated moving average (ARIMA), Prophet, and long short-term memory (LSTM) neural networks. Methods: DALY data for the top 10 disease categories in Kazakhstan were modeled using Python 3.11 with statistical and machine learning libraries. Each model was trained and validated for short- and medium-term forecasts, with performance compared across trajectory trends and disease ranking stability. Results: GBD and LSTM models projected stable rankings among the top 10 DALY contributors through 2032, with only malignant neoplasms of the colon and rectum showing a decline, while ARIMA and Prophet exhibited greater temporal fluctuations, predicting a drop in lower respiratory infections. Across all models, noncommunicable diseases, particularly, cardiovascular and metabolic disorders, remain dominant drivers of Kazakhstan’s future health burden. Conclusions: Deep learning (LSTM) and GBD approaches yielded smoother, more robust long-term predictions, whereas ARIMA and Prophet captured short-term variability more sensitively, highlighting the benefit of integrating statistical and AI-based paradigms for comprehensive national health forecasting and policy design.