Noise is the signal: variability in surveillance data provide early indication of epidemic phase

Predicting epidemic surges is challenging and often relies on prior seasons, outbreaks in other regions, and advanced modeling of surveillance data. Using foundational principles of disease growth and spread, including trend analysis and overdispersion, we developed a surge detection algorithm for emerging pathogens that requires no historical baselines or seasonal assumptions. We used it to evaluate temporal variability (coefficient of variation, CV) and spatial clustering (global Moran's I) as early warning indicators across three parallel SARS-CoV-2 surveillance streams: wastewater, clinical cases, and hospitalizations. All metrics showed consistent variability patterns across epidemic phases that predicted surge onset and decline approximately one month in advance. For weekly pre-surge detection, sensitivity and positive predictive value reached 75% and 54%. CV additionally signaled surge termination, with up to 70% sensitivity and 59% positive predictive value. When evaluated per surge, by requiring only a single correct signal within each phase, CV detected 86 to 98% of onsets and 73 to 85% of terminations. These signatures held across surveillance streams and geographic scales, suggesting they reflect system-level transmission dynamics rather than stream-specific artifacts, consistent with overdispersed transmission producing high variability before and after peak. Requiring only basic quantitative methods, our framework is implementable by practitioners without specialized training in dynamical systems.