"Even as raw data suggested a measles outbreak was waning, CDC’s advanced ‘nowcasting’ model accurately predicted continued transmission, prompting state officials to maintain critical response staffing through the holidays and ultimately averting a potentially more severe public health crisis."
In late December, public health officials in South Carolina were presented with a seemingly positive trend: the raw numbers of confirmed measles cases indicated an outbreak that was winding down. However, a sophisticated modeling tool developed by the Centers for Disease Control and Prevention (CDC) offered a starkly different assessment, predicting that the outbreak was, in fact, still on the rise. This discrepancy led state officials to maintain robust staffing levels for their response efforts throughout the holiday season, a decision that proved prescient. Just two weeks later, the cumulative number of confirmed measles cases had more than doubled, underscoring the critical role of advanced data analytics in navigating public health emergencies.
This pivotal sequence of events is detailed in a new CDC report published this week, marking a significant milestone as the agency’s first deployment of real-time modeling during an active measles outbreak. The technique, known as "nowcasting," is designed to address a fundamental challenge faced by all health departments during rapidly evolving outbreaks: discerning whether current case numbers represent a true picture or merely an incomplete snapshot due to reporting delays. The implications of this distinction extend far beyond a single state, impacting the timeliness and effectiveness of public health interventions for communities nationwide.
Reporting delays are an inherent and universal challenge in disease surveillance. These delays systematically create the illusion that growing outbreaks are shrinking, potentially leading to delayed or inadequate responses. For families and communities, the downstream consequences can be profound, determining whether vital resources like clinics, contact tracing teams, and vaccination outreach campaigns are deployed at the critical moment or weeks too late. The South Carolina experience offers a compelling case study in how advanced modeling can mitigate these risks.
The Holiday Signal That Changed Staffing Decisions
The measles outbreak in South Carolina, which spanned from October 2025 through March 2026, ultimately resulted in 997 confirmed cases. This figure represents the largest measles outbreak in the United States in approximately three decades, according to data attributed to the South Carolina Department of Public Health. During the course of the outbreak, the state diligently transmitted case line lists to the CDC on a roughly bi-weekly basis. These lists included crucial data points such as the date each patient developed a rash and the date public health authorities were officially notified.
The CDC’s analysis, detailed in the report, revealed that the initial nowcasts, generated from data collected on December 19 and December 23, were produced at a time when the provisional data represented between 76.5% and 77.2% of the final case counts. At face value, these reported numbers were trending downwards, suggesting a declining outbreak. However, the nowcasting model provided a contradictory estimate, indicating a likely increasing trend. Crucially, the model’s prediction intervals – a range within which the true final counts were expected to fall – encompassed the actual final case numbers, a testament to its predictive power.
This sophisticated modeling insight played a direct role in the state health department’s decision-making. Primarily due to the model’s projections, the department chose to maintain high staffing levels throughout the holiday period and initiated plans to hire additional personnel. This proactive approach proved vital, as cumulative cases surged from 185 on December 23 to 424 by January 6. The nowcasting model’s ability to generate these estimates within minutes of receiving updated data, feeding automated reports back to state officials, enabled a swift and informed response.
Reporting Delays Are the Problem Being Solved
It is essential to differentiate nowcasting from forecasting. While forecasting aims to predict future events, nowcasting focuses on estimating the present situation using incomplete data, specifically by correcting for historical reporting delays. This distinction, though seemingly academic, carries significant weight in public health. An incomplete dataset pointing in the wrong direction, as occurred in South Carolina, can lead to critical misinterpretations of an outbreak’s trajectory.
In the South Carolina outbreak, the estimated median gap between the onset of a rash and public health notification was between two to three days. On average, 92.8% of cases were reported within 14 days, which are considered unusually robust reporting times. The report commends the state’s diligent contact-tracing efforts and its investment in surveillance systems for these positive outcomes. Notably, more than a third of all cases (362) were identified through contact tracing even before a rash had appeared, demonstrating the effectiveness of proactive public health measures.
The nowcasting model also estimated the effective reproduction number (R_t), which represents the average number of secondary infections generated by each infected individual. An R_t value above 1 signifies an expanding epidemic, while a value below 1 indicates a declining trend. The CDC utilized the open-source package EpiNow2 for its analysis, and the agency has made the underlying code publicly available to jurisdictional and academic partners. This accessibility means that other health departments can adopt and implement this powerful modeling technique without the need for extensive in-house development.
Where the Model Broke Down
The CDC report is commendably direct in acknowledging the limitations and instances where the model did not perform optimally. This section offers valuable insights into the complexities of real-time disease surveillance. When transmission rates accelerated rapidly around the holiday period, the nowcasting model struggled to keep pace with the surge.
On January 6, the completeness of provisional data dropped sharply, with reports reflecting only 25.7% of the final case count. This significant decline in reporting completeness was attributed to a combination of backfilling of older cases and processing backlogs. Consequently, the nowcast underestimated the true number of cases and fell outside its own 90% prediction interval. The outbreak subsequently peaked on January 13 and 14, and the model failed to accurately detect this turning point in the data from January 16 and January 23.
Despite these limitations, even during these challenging weeks, the nowcasts provided improved estimates of the total outbreak size by 10.5% to 11.3% compared to raw provisional counts. Furthermore, they correctly indicated the general direction of transmission. By January 27, the model was able to estimate that the outbreak was likely decreasing. After January 23, its estimates once again fell within their prediction intervals and accurately reflected the true case counts. This later accuracy allowed state officials to confidently conclude that the observed decline was genuine and not merely a reporting artifact, enabling them to scale down response efforts through February and March.
The authors of the report candidly list three key limitations. Firstly, the model did not explicitly account for the impact of holiday gatherings, which likely drove the surge in transmission. Secondly, it did not adjust for "right truncation," a phenomenon that can lead to underestimation of cases during periods of rapid growth. The report suggests that other epidemiological modeling methods might offer improved performance by incorporating dynamic changes in transmission patterns. Lastly, there is no objective external reference point against which to definitively validate reproduction number estimates, meaning their accuracy was primarily assessed through visual comparison with observed trends.
Application Beyond One Outbreak
The report’s conclusion is measured rather than overtly triumphant. The success of nowcasting in this instance was contingent on several factors: the consistent collection of two critical date fields (rash onset and notification date), and the relatively fast and stable reporting rates. The authors emphasize that in settings where disease surveillance is slower or more erratic, the reliability of this same method would be diminished.
This conditionality points to an important accountability aspect. While the nowcasting technique itself is relatively low-cost to implement once developed, its effectiveness is entirely dependent on the robustness of state and local surveillance capacity, which varies significantly across the country. Health departments facing budget cuts or staffing reductions may collect essential data fields less consistently, thereby degrading the performance of such modeling tools.
For individual households, the fundamental advice regarding measles remains unchanged. Two doses of the measles, mumps, and rubella (MMR) vaccine continue to offer approximately 97% protection against the virus. The CDC’s measles guidance and recommended vaccination schedules are also unchanged. Parents in areas experiencing active transmission are strongly advised to confirm their children’s vaccination status. Furthermore, individuals experiencing symptoms such as fever and rash should call healthcare providers in advance rather than walking into clinics, to minimize potential exposure to others. This caution extends to interpreting public health dashboards and data reports, including those tracking national measles cases.
The ultimate impact and widespread adoption of this nowcasting methodology remain to be seen. Key unanswered questions include whether other states will readily adopt the technique, whether the CDC will deploy it proactively in future outbreaks, and how its performance will be assessed in settings with less complete reporting. The CDC’s decision to publish the underlying code represents a crucial and necessary first step towards broader implementation and further research.
