Objectives: The Puerto Rico Department of Health (PRDH) seeks to identify dengue epidemics with high specificity as soon as possible. Design: Development and prospective application of an early warning system for dengue epidemics using routine historical surveillance data. An intercept-only weekly negative binomial regression model was adjusted using historical data from probable and confirmed cases of dengue. A range of threshold definitions was explored using three percentiles estimated by the model of weekly counts of dengue cases. Environment: Dengue is endemic in Puerto Rico with an irregular occurrence of major epidemics that have a substantial impact on health burden and health systems. Data on probable and confirmed cases of dengue fever are routinely collected at all hospitals and private clinics.Participants: A total of 86,282 confirmed or probable cases of dengue virus were reported from January 1, 1986 to June 30, 2024, with an annual average of 2,212 cases (median: 1,533; range: 40-10,356) .Main and secondary outcome measures: The model was retrospectively adjusted to mimic the real-time detection of epidemics and evaluated based on the sensitivity and specificity of the detection of epidemics.Results: The 75th percentile threshold was the one that best aligned with historical epidemic classifications, balancing false alarms and missed detections. This model provides a robust method for defining thresholds, considering biased data, using all historical data and improving traditional methods such as endemic channels. Conclusions: In March 2024, the PRDH declared a public health emergency due to an early and off-season increase in cases that exceeded the epidemic alert threshold developed in this study. This real-time application highlights the value of these thresholds to support the detection of dengue epidemics and the public health response. Integrating thresholds with other tools and strategies can improve epidemic preparation and management.