In late December, the raw measles numbers coming out of South Carolina looked like an outbreak that was winding down. A CDC model told state officials the opposite, and the state kept its response staffed through the holidays. Two weeks later, the cumulative number of confirmed cases had more than doubled.
That sequence is the substance of a new CDC report published this week, and it describes the first time the agency has run this kind of real-time modeling during an active measles outbreak. The technique is called nowcasting, and it exists to answer a question every health department faces during a fast-moving outbreak: are today's numbers real, or just incomplete?
The distinction matters far beyond one state. Reporting delays are universal in disease surveillance, and they systematically make growing outbreaks appear to be shrinking. For families, the downstream effect is whether a health department scales up clinics, contact tracing and vaccination outreach at the right moment or two weeks too late.
The Holiday Signal That Changed Staffing Decisions
The outbreak ran from October 2025 through March 2026 and produced 997 confirmed cases, the largest US measles outbreak in about three decades, according to figures the report attributes to the South Carolina Department of Public Health. The state sent case line lists to CDC roughly twice weekly, including the date each patient developed a rash and the date public health was notified.
The CDC analysis reports that the first two nowcasts, run on line lists from December 19 and December 23, were produced at a time when provisional data contained 76.5 to 77.2 percent of what the final counts would show. The reported numbers were pointing downward. The model instead estimated a likely increasing trend, and its prediction intervals contained the true final counts.
Partly because of those estimates, the state health department maintained high staffing through the holiday period and began hiring additional staff. Cumulative cases then climbed from 185 on December 23 to 424 by January 6. The model produced its estimates within minutes of receiving each updated line list, feeding automated reports back to state officials.
Reporting Delays Are the Problem Being Solved
Nowcasting is not forecasting. Forecasting predicts the future; nowcasting estimates the present from incomplete data by correcting for how long reports historically take to arrive. The distinction sounds academic until an incomplete count points the wrong direction, which is precisely what happened here.
In South Carolina the estimated median gap between rash onset and public health notification ran two to three days, and an average of 92.8 percent of cases were reported within 14 days. Those are unusually good numbers, and the report credits the state's contact-tracing effort and its investment in surveillance. More than a third of cases, 362, were identified through contact tracing before a rash even appeared.
The model also estimated the effective reproduction number, the average number of secondary infections each infected person generates. Above 1 means growth; below 1 means decline. The tool used was the open-source package EpiNow2, and the CDC has made its code publicly available to jurisdictional and academic partners, meaning other health departments can adopt the method without having to build it from scratch.
Where the Model Broke Down
The report is unusually direct about failure, and that section is the one worth reading closely. When transmission accelerated rapidly around the holidays, the model did not keep up.
On January 6, provisional data held only 25.7 percent of what the final count would become, a collapse in reporting completeness driven by backfilling and backlogs. The nowcast underestimated the true figure and fell outside its own 90 percent prediction interval. Cases peaked on January 13 and 14, and the model failed to detect that turn in the January 16 and January 23 data.
Even during those weeks, the nowcasts improved estimates of total outbreak size by 10.5 to 11.3 percent compared with raw provisional counts, and correctly indicated the general direction. By January 27 the model estimated the outbreak was likely decreasing, and after January 23 its estimates again captured true counts within their intervals. That later accuracy allowed the state to conclude the decline was genuine rather than a reporting artifact, and to scale down its response through February and March.
The authors list three limitations plainly. The model did not account for the holiday-gathering patterns that drove the surge, nor did it adjust for right truncation, which can lead to underestimating cases during growth. Other methods might perform better by incorporating epidemiologic change. And there is no objective reference against which to validate reproduction number estimates, so accuracy there was assessed by visual comparison.
Application Beyond One Outbreak
The report's own conclusion is conditional rather than triumphant. Nowcasting worked here because two date fields were consistently collected, and because reporting was fast and stable. Where surveillance is slower or more erratic, the same method would be less reliable.
That condition is the accountability angle. The technique costs little once built, but it depends entirely on state and local surveillance capacity that varies widely across the country. Health departments facing staffing reductions collect the same fields less consistently, and the tool degrades as a result.
For households, none of this changes what to do about measles. Two doses of MMR remain about 97 percent effective, and CDC measles guidance and vaccination schedules are unchanged. Parents in areas with active transmission should confirm their children's vaccination status, and anyone with fever plus a rash should call ahead rather than walk into a clinic waiting room. The same caution about reading dashboards applies to the national measles case-trackingMedicalDaily reports each week.
What remains unknown is whether other states will adopt the method, whether the CDC will deploy it in future outbreaks, and how it performs in settings where reporting is less complete. The agency has published the code, which is the necessary first step.
Key Questions Answered
What is nowcasting? A statistical method that estimates what is happening right now by correcting incomplete case reports for known reporting delays.
How is it different from forecasting? Forecasting predicts future case counts. Nowcasting estimates present conditions from data that has not finished arriving.
What did it accomplish in South Carolina? It signaled rising transmission in late December while raw counts suggested decline, supporting a decision to keep response staffing high through the holidays.
Did the model ever get it wrong? Yes. During January, when reporting completeness dropped sharply, it underestimated cases and missed the outbreak peak.
How large was the outbreak? 997 confirmed cases from October 2025 through March 2026, the largest US measles outbreak in roughly 30 years.
Can other health departments use this? CDC has published the code publicly, though reliability depends on consistent collection of onset and report dates.
Does this change what parents should do about measles? No. Two doses of MMR remain the recommended protection, and current guidance is unchanged.
