RESUMO
<p><b>OBJECTIVE</b>To provide methods and alert thresholds which are scientific, sensitive, specific and practical for Early Warning System in Public Health Surveillance.</p><p><b>METHODS</b>Alert data was based on historical infectious diseases reports. Control chart was used to detect outbreaks or epidemics. An epidemic was defined by consulting Specialists. After calculating sensitivity, specificity, positive predictive value and describing receiver-operating characteristic curve (ROC), the optimal model and thresholds were chosen.</p><p><b>RESULTS</b>At 80 percentile, the sensitivities and the specificities of epidemic haemorragia fever, hepatitis A, dysentery, epidemic cerebrospinal meningitis and malaria were over 90%, and there was a high efficacy of early warning. At 90 percentile, the sensitivities and the specificities of tuberculosis and measles were over 85%, and there was a high efficacy of early warning also.</p><p><b>CONCLUSION</b>Control chart based on five years was chose as a essential method in early warning system. The alert threshold for epidemic haemorragia fever, hepatitis A, dysentery, epidemic cerebrospinal meningitis and malaria was 80 percentile. The alert threshold for tuberculosis and measles was 90 percentile.</p>