Predicting CCHF incidence and its related factors using time-series analysis in the southeast of Iran: comparison of SARIMA and Markov switching models.

Crimean-Congo haemorrhagic fever (CCHF) is endemic in the southeast of Iran. This study aimed to predict the incidence of CCHF and its related factors and explore the possibility of developing an empirical forecast system using time-series analysis of 13 years' data. Data from 2000 to 2012 were obtained from the Health Centre of Zahedan University of Medical Sciences, Climate Organization and the Veterinary Organization in the southeast of Iran. Seasonal autoregressive integrated moving average (SARIMA) and Markov switching models (MSM) were performed to examine the potential related factors of CCHF outbreaks. These models showed that the mean temperature (°C), accumulated rainfall (mm), maximum relative humidity (%) and legal livestock importation from Pakistan (LIP) were significantly correlated with monthly incidence of CCHF in different lags (P < 0·05). The modelling fitness was checked with data from 2013. Model assessments indicated that the MSM had better predictive ability than the SARIMA model [MSM: root mean square error (RMSE) 0·625, Akaike's Information Criterion (AIC) 266·33; SARIMA: RMSE 0·725, AIC 278·8]. This study shows the potential of climate indicators and LIP as predictive factors in modelling the occurrence of CCHF. Our results suggest that MSM provides more information on outbreak detection and can be a better predictive model compared to a SARIMA model for evaluation of the relationship between explanatory variables and the incidence of CCHF.