Temporal dynamics and forecasting of laboratory-confirmed human nontyphoidal Salmonella cases in Malaysia, 2018-2024: implications for surveillance.

OBJECTIVES

Nontyphoidal Salmonella (NTS) is a major cause of foodborne illness in Malaysia. Time-series modeling can characterize temporal patterns, although structural disruptions, nonlinear dynamics, and regional variability may affect model performance. This study aimed to characterize NTS seasonality in Malaysia, assess temporal stability, and explore model-based approaches for surveillance.

METHODS

Monthly laboratory-confirmed human NTS cases from national laboratory surveillance databases for 2018-2024 were analyzed using time-series models, including autoregressive integrated moving-average (ARIMA) models with Fourier terms and an exploratory hybrid ARIMA-long short-term memory (LSTM) model. Interrupted time-series analysis was used to assess structural stability. Anomalies were identified using 95% prediction intervals, and stratified modeling assessed regional heterogeneity.

RESULTS

NTS cases showed recurring annual seasonal patterns of moderate strength, with temporal variation over the study period. A significant decline occurred in March 2020 (β=-62.24 cases, p=0.001), followed by gradual recovery (β=2.71 cases/month, p<0.001), suggesting a marked coronavirus disease 2019 pandemic-related disruption and a subsequent positive post-intervention trend. The ARIMA(0,1,1)-Fourier (K=1) model outperformed conventional ARIMA, benchmark, and exponential smoothing models (root mean squared error, 42.9 vs. 51.37-108.79). The hybrid ARIMA-LSTM model provided modest additional improvement, particularly during high-case periods. Model-based anomaly detection identified deviations from expected patterns that coincided with documented outbreak-associated months. Regional heterogeneity in temporal dynamics suggested the need for surveillance strategies tailored to local epidemiological patterns.

CONCLUSION

Parsimonious models captured most temporal patterns in NTS cases, with only modest gains from more complex approaches. Context-specific modeling strategies, model-based anomaly detection, and tiered surveillance frameworks may support more proactive NTS surveillance in Malaysia.