ObjectiveTo evaluate the application of improved Poisson regression model or negative binomial regression model in the early warning of the peak time of varicella incidence in Dalian, Liaoning province.

MethodsBased on the theory of Z-D phenomenon, the annual incidence data of varicella in Dalian from 2006 to 2014 were analyzed. Circular distribution method was used to calculate the incidence peak day and period. Considering over-dispersion might exist for data, Poisson regression model or negative binomial regression model were used to estimate the weekly baseline level of varicella. The sensitivity, detection time for early warning and false warning rate were used as evaluation indicators. Circular distribution method was used to select the appropriate warning threshold for reference standard to construct the early warning model of the peak time of varicella incidence in Dalian.

ResultsFrom 2006 to 2014, a total of 26 427 varicella cases were reported in Dalian with annual incidence rate of 46.366 /100 000. The results of circular distribution method showed that there was seasonal characteristic of varicella incidence. The total concentration trend r value was 0.195 (P＜0.010) between 2006 and 2014. The average annual incidence peak date was on 25 February, and the peak period lasted from 12 November to 10 June of the next year. Because the incidence data of varicella was over-dispersion, we established the baseline level of varicella by the improved negative binomial regression model. The Akaike information criterion (AIC) value was 418.854. Baseline level plus 20 cases in high incidence year and baseline level plus 10 cases in low incidence year were used as early warning threshold to accurately detect the time of varicella incidence. The sensitivity of early warning was 100%, the average detection time for early warning ranged from 1.50 to 4.67 weeks, and the false warning rate was 0%.

ConclusionThe early warning model of varicella incidence peak time in Dalian based on improved negative binomial regression model has high sensitivity and appropriate detection time, which can provide scientific evidence for the prevention and control of varicella and other infectious diseases.