Publication:

Integrating statistical and machine-learning models to determine tuberculosis mortality rates in Malaysia

Background Tuberculosis (TB) mortality remains an important indicator of disease burden and health-system performance, reflecting the effectiveness of detection, treatment, and prevention efforts. In Malaysia, TB mortality rates based on temporal patterns remain insufficiently characterized, limiting understanding of mortality burden and forecasting performance; addressing this gap may strengthen public health planning. Thus, this study aimed to determine TB mortality rates and evaluate predicted mortality in 2024 alongside characterize temporal patterns from 2014 to 2024 using autoregressive integrated moving average (ARIMA), long short-term memory (LSTM), convolutional neural network (CNN), and hybrid CNN–LSTM models.

Methods This time-series study analyzed monthly TB mortality data from 2014 to 2024 obtained from the National Tuberculosis Registry. The dataset comprised 132 monthly observations, with 80% used for model training and 20% for testing. Four forecasting models were applied: ARIMA, LSTM, CNN, and hybrid CNN–LSTM. Relative error (RE%) was used to evaluate deviation between observed and predicted mortality values.

Results In 2024, observed monthly TB mortality rates ranged from 0.42 to 0.79 per 100,000 population, while age-standardized mortality rates ranged from 0.43 to 0.80 per 100,000 population. The hybrid CNN–LSTM model provided the best representation of Malaysia’s TB mortality pattern, capturing the upward trend, mid-year peaks, early-year troughs, and cyclical fluctuations.

Conclusion Observed TB mortality rates in Malaysia were slightly higher than ranges reported in other international studies. The hybrid CNN–LSTM model produced the most accurate estimates, suggesting that hybrid deep-learning methods can strengthen TB mortality monitoring and support timely public health decision-making.