In December, South Carolina's measles case reports appeared to be declining. A modeling tool inside the Centers for Disease Control and Prevention said the opposite, and the state kept its response staffing high through the holidays partly as a result. Within two weeks, cumulative confirmed cases more than doubled.
That sequence is the centerpiece of a CDC report published August 27 in the agency's Morbidity and Mortality Weekly Report, evaluating how a technique called nowcasting performed during the largest measles outbreak in the United States in roughly three decades. The outbreak ran from October 2025 through March 2026 and ended with 997 cases, a figure the report attributes to the South Carolina Department of Public Health.
The finding matters beyond epidemiology because it explains something families encounter constantly and rarely have a name for. Public health dashboards do not show what is happening now. They show what has been reported so far, and reported counts always lag reality by days. A dip on a chart can be a genuine improvement or just paperwork.
The Holiday Signal That Changed Staffing Decisions
Nowcasting corrects for that lag. The model, built on the EpiNow2 tool, adjusts incomplete reports using historical patterns in how long cases take to surface, then estimates both current case counts and the effective reproduction number, denoted as Rt. That figure represents the average number of secondary infections caused by a single infected person, and whether it sits above or below 1 indicates if an outbreak is growing or shrinking.
The first two model runs used line lists from mid- and late December, when provisional data accounted for 76.5 to 77.2 percent of the final count. Those provisional numbers indicated declining trends. The nowcasts indicated a likely increase, and the final daily and cumulative figures landed inside the model's 90 percent prediction intervals.
Partly because of those estimates, the South Carolina Department of Public Health maintained high staffing levels through the winter holiday period and began hiring additional staff. Cases then climbed rapidly, with the final cumulative count rising from 185 to 424 between December 23 and January 6.
Where the Model Failed and Why That Matters
The report is unusually candid about the model's worst stretch, which arrived precisely when accuracy mattered most.
On January 6, provisional data showed only 25.7 percent of what the final count would eventually show, reflecting a lower-than-expected reporting rate. The nowcasts for that date underestimated the final values and fell outside their own 90 percent prediction intervals. They still improved on the raw provisional counts and still correctly signaled that the outbreak was growing, but the numerical estimate was wrong.
Cases peaked on January 13 and 14. The model did not detect that turn in the January 16 or January 23 data, though it still improved estimates of total outbreak size by 10.5 to 11.3 percent compared with provisional counts. Only on January 27 did it estimate that the outbreak was likely decreasing.
The authors attribute the January failure to two things. Transmission increased rapidly at the end of December, likely due to changes in contact patterns around holiday gatherings, which the model did not account for. Data completeness then dropped because of backfilling and backlogging as investigators caught up. They also note that the analysis did not adjust for right truncation, a technical omission that can lead to underestimating cases during periods of growth, and that 36.7 percent of cases were identified through contact tracing before a rash appeared, which does not reflect the reporting delay the model is built to correct.
Reporting Lags Are the Hidden Variable in Every Outbreak Dashboard
The underlying delay in this outbreak was short by public health standards. The estimated median interval between rash onset and the date the state health department was notified was two to three days, and an average of 92.8 percent of cases were reported within 14 days of rash onset.
Even with that performance, the raw numbers repeatedly created the appearance of a decline that was not real. The authors credit South Carolina's contact tracing effort and its investment in disease surveillance for keeping reporting delays shorter and more consistent, which is what made the modeling workable.
For parents monitoring case counts in any state with an active outbreak, the practical implications are straightforward. A one-week dip in reported cases is not evidence that an outbreak has turned, and a state health department's own assessment carries more weight than a chart read cold. The same caution applies to the national measles case tracking that MedicalDaily has reported on weekly. Vaccination status, not case-count trends, remains the determinant of individual risk. Anyone unsure whether a child is up to date on MMR should confirm it with a pediatrician rather than infer safety from a graph.
Limits Families Should Keep in Mind
This report describes a single outbreak and does not establish that nowcasting will perform this way elsewhere. The authors state the estimates were sufficient to support situational awareness rather than to serve as precise forecasts, and they note that other nowcasting methods might perform better by incorporating epidemiologic changes or allowing faster shifts in Rt.
There is also no objective reference against which Rt trends can be validated, a limitation the authors identify. Contemporaneous estimates had to be compared visually against final data rather than tested quantitatively.
The tool required little additional effort from frontline staff, which is part of why it was usable. Because South Carolina supplied both rash-onset and public-health-report dates in its line lists, the model could produce estimates within minutes of receiving updated data and feed automated reports back to the state. CDC has made the code publicly available to jurisdictional and academic partners, which is the mechanism through which it could spread to other jurisdictions.
The report describes the first CDC implementation of nowcasting during an active measles outbreak. The technique had previously been applied to outbreaks retrospectively and to routine respiratory surveillance data. The agency frames it as a complement to provisional data rather than a replacement.
What happens next depends on state health departments. CDC has not announced that nowcasting will be applied to other active outbreaks, and no schedule has been published. MedicalDaily will report on deployments in other jurisdictions and any updated evaluations as they become available.
Key Questions Answered
What is nowcasting? A modeling method that corrects incomplete reported data for known reporting delays, producing real-time estimates of current case counts and transmission trends rather than waiting for reports to catch up.
What did the CDC report find? Nowcasts improved estimates of cumulative outbreak size compared with provisional counts and generally captured transmission trends, though accuracy declined in January when reporting completeness dropped unexpectedly.
Did the model get anything wrong? Yes. On January 6, it underestimated case counts, fell outside its own prediction intervals, and did not detect the mid-January peak until later in the month.
How long was the reporting delay? The estimated median interval between rash onset and public health notification was 2 to 3 days, with 92.8 percent of cases reported within 14 days.
Did this change anything on the ground? The state health department used the estimates to maintain staffing through the holidays and to begin hiring, and later to confirm that a late-January decline was real rather than a reporting artifact.
What should families take from this? A short-term dip in reported case counts is not reliable evidence that an outbreak is ending. Vaccination status determines individual risk, and MMR questions belong with a pediatrician.
Will other states use this? CDC has published the code for jurisdictional and academic partners, but no rollout to other outbreaks has been announced.
