From a Hospitalization Count to a Transmission Signal

A corrected hospitalization or emergency-department count is still not an infection count. The important focus here is the chain of delays between infection, an observable clinical event, and the report that enters surveillance. Keeping those links separate lets readers understand what a nowcast contributes to a transmission estimate—and why a more timely signal is still a model-based inference rather than a direct reading of spread.

Several delays connect infection to a reported observation

A surveillance system rarely observes infection at the moment it happens. CDC uses a COVID-19 emergency-department example to make the sequence concrete. First comes infection. A delay may follow before symptoms appear. Among symptomatic cases, only a proportion may visit an emergency department, and there is another delay between symptom onset and the visit. The visit then has its own delay before the surveillance system receives and records the report.

These stages answer different questions. The infection-to-symptom interval describes biological timing. The symptom-to-visit interval describes when an illness becomes an observed encounter. The visit-to-report interval describes when information about that encounter becomes available to the surveillance system. The same word—“delay”—can refer to all three, but correcting one does not correct the others.

The distinction matters when a recent event-based curve is used to say something about current transmission. Suppose the newest emergency-department counts are low. A reporting delay may mean that some visits have not yet appeared. A nowcast can estimate the recent visit total after accounting for that incomplete reporting. But even a perfectly completed visit count would still occur after infection, and the relationship between visits and infections would still need to be modeled. A corrected observation is a better-timed input, not an infection timestamp recovered by definition.

The delay chain also explains why the date attached to a metric must be stated precisely. A hospitalization date, an ED-visit date, and a report date locate different stages of the process. A curve organized by report date describes when information entered the system. A curve organized by hospitalization or visit date is closer to the time of the observed event, but its newest dates are vulnerable to right truncation. A transmission analysis needs to know which of these dates it is using before interpreting a change.

The stages can also vary across settings. The source presents the infection-to-symptom-to-ED-to-report path as an explanatory example for COVID-19 and ED surveillance, not as a universal template for every pathogen or metric. Hospitalization data may have different observation and reporting behavior from ED data. A case series may be shaped by testing and notification processes that do not appear in a clinical-visit series. The principle is stable—delays accumulate between an unobserved infection and reported evidence—but the distributions must match the metric and context.

It helps to separate the chain into two timelines. The first runs from infection to an event that the surveillance system can observe. The second begins with that observed event and ends when its report is available.