Influenza Rt Categories Track Emergency Department Trends, but Only Partly
CDC’s influenza reproduction-number framework is built around a deceptively simple question: is transmission accelerating, holding steady, or easing? In this preprint, investigators tested whether those weekly Rt epidemic trend categories actually mirror what clinicians see next in the emergency department, where rising influenza activity eventually becomes visible through worsening respiratory illness and syndromic surveillance signals.
The answer was mixed. Across the 2024/2025 and 2025/2026 influenza seasons in the United States, alignment between the categorical Rt signal and observed emergency department trends was only moderate overall, and it shifted materially depending on how change in ED activity was defined. The best correspondence came when investigators used a 5% relative change threshold, where growing and declining categories aligned with the next week’s ED trends 61% of the time. Under the tighter 0.5% difference threshold, alignment fell to 38.2%.
Low-level noise in surveillance data can obscure direction before it obscures magnitude.
How the Surveillance Comparison Was Built
The analysis used the posterior probability that Rt exceeded 1, expressed as P(Rt>1), and translated that probability into five epidemic trend categories: growing, likely growing, not changing, likely declining, and declining. Those categories were then compared with emergency department activity captured through the National Syndromic Surveillance Program. the comparison was made against ED trends in the following week, matching the expected lag between infection, symptom onset, and care seeking.
Three different definitions were used to decide whether ED activity had truly changed enough to count as an observed trend. One required at least a 0.1% difference between consecutive weeks, another required at least 0.5%, and the third used a 5% relative change. That design matters because a direction-based metric and a magnitude-based metric do not answer the same surveillance question. A category can be directionally correct while still failing to capture a small but measurable shift in clinical burden, especially when influenza circulation is low or fluctuating near a turning point.
Threshold Choice Shaped the Apparent Performance
The strongest alignment emerged under the 5% relative change threshold, where the framework was most successful at matching clear week-to-week movement in ED activity. At the other end of the spectrum, alignment was weakest when the threshold for observed change was only 0.5%. The pattern suggests the Rt categories behave better when the bedside signal is large enough to distinguish from routine surveillance variation, and worse when investigators ask them to anticipate very small changes in ED counts.
Calibration followed the same general pattern. Performance was best under the 5% relative change threshold, reinforcing the idea that the Rt classification system is most dependable when compared with more substantial movement in care-seeking data. In practical surveillance terms, the categories appear better suited to identifying clinically meaningful swings in influenza burden than to flagging marginal week-to-week drift.
Misalignment Clustered Around Inflection Points
The weakest agreement did not occur randomly. Misalignment concentrated around epidemic inflection points and during periods of low influenza activity, when transmission is shifting direction and small model errors can push a week into the wrong category. Those are exactly the intervals when the underlying epidemic curve is hardest to classify in real time. A short delay in case accumulation, a local wobble in health-seeking behavior, or ordinary statistical uncertainty can move a week from “not changing” to “likely declining” or from “likely growing” to “growing” without a corresponding shift in emergency department burden.
This is not a failure unique to influenza surveillance. Directional metrics are inherently vulnerable when the true signal is close to the decision boundary. Rt estimation converts noisy, incomplete observations into a real-time forecast of transmission direction, but the surveillance output still depends on how sharply the epidemic curve is changing. When circulation is sparse, the model has less information and the ED data themselves may be more erratic, creating the kind of mismatch that becomes most visible near peaks and troughs.
Why Direction and Magnitude Do Not Always Agree
The categories used here reflect confidence in the direction of change rather than its size. That distinction is central to the interpretation of the results. A week labeled “growing” means the probability distribution favors expansion, not necessarily that influenza activity is climbing quickly enough to produce an obvious step-up in emergency care. Conversely, a week classified as “declining” may still carry a substantial caseload if the epidemic is easing from a high plateau rather than collapsing from peak transmission.
From a public health operations perspective, that mismatch is understandable. Emergency departments respond to the cumulative consequences of infection, not to the abstract probability that Rt has moved above or below 1. A transmission signal can turn before the clinical burden does, and a clinical burden can remain elevated even after transmission begins to soften. The study therefore exposes a structural limitation in surveillance translation: a metric built for direction may not cleanly predict a metric built for magnitude.
A More Size-Based Definition Could Improve Surveillance Usefulness
The investigators conclude that defining epidemic trend categories based on Rt magnitude could improve correspondence with observed change indicators. In other words, a framework that incorporates how far Rt is from 1, not just whether it is above or below that line with some confidence, may better match the real-world size of influenza shifts seen in emergency departments. That approach would not remove uncertainty, but it could make the categories more operationally meaningful for real-time monitoring.
For public health teams, the implication is straightforward: the current Rt categories are useful, but they are not interchangeable with ED trend data. They work best when the epidemic is moving enough for the surveillance system to detect a clear change and less well when activity is flat, low, or pivoting. During those transitional weeks, even a good model can disagree with observed care patterns because the two signals are measuring different things on different timelines.
The study remains a preprint and has not undergone peer review. It also relies on aggregated surveillance data rather than patient-level clinical records, which limits the ability to probe why specific weeks diverged. The next step is not simply more modeling, but tighter alignment between the statistical definition of epidemic trend and the magnitude of the clinical signal public health officials are trying to track.