Estimating the true size of a measles outbreak is a long-standing challenge that limits the ability to understand the true impact^16,17. As of May 2026, the largest active measles outbreak in the US was ongoing on the Utah-Arizona border with more than 600 reported cases. The outbreak originated in a community with low vaccination coverage and limited engagement with healthcare systems and spread throughout Utah. Our findings demonstrate that in the modern era where the availability of PCR testing and electronic laboratory reporting might be expected to provide more complete reporting than passive surveillance of clinical cases, the true size of a measles outbreak is not captured by surveillance. Using two independent approaches based on complementary data sources, we estimate that the Utah-Arizona border outbreak was substantially larger than reported, with confirmed cases representing approximately 20.96%-32.5% of total infections. Importantly, while both approaches identify substantial under-ascertainment, they do not estimate the same underlying quantity. The phylodynamic analysis estimated between 1,798 and 2,051 infections linked to the Utah-Arizona border outbreak (both within and outside the originating community), depending on case inclusion criteria. Similarly, the agent-based model estimated a median of 2,226 to 2,791 infections within the affected community, depending on assumptions of interventions. These differences reflect the populations captured by each approach: the phylodynamic analysis includes all infections linked to the outbreak regardless of location, whereas the agent-based model is bounded by the affected community. Phylogenetic reconstruction indicates that transmission moved frequently between the affected community and surrounding areas in both directions (Figure 3), consistent with the broader scope of the phylodynamic estimate. Despite these differences, the credible interval from the phylodynamic analysis encompassed the estimates from the agent-based model, providing strong evidence for substantial under-ascertainment of infections during the outbreak. The convergence of estimates from two methods relying on independent data sources and distinct modeling assumptions strengthens confidence that the degree of under-ascertainment is not just an artifact of model assumptions.

For a case to be included in surveillance, a patient must have symptoms, seek healthcare, have the correct diagnosis made, and that diagnosis be reported to public health. In routine surveillance this limitation is well recognized but often difficult to quantify. During a large-scale outbreak, however, the degree of under-ascertainment has direct consequences for public health response: estimates of transmission intensity, assessments of intervention effectiveness, and projections of outbreak trajectory all depend on case counts that, if substantially incomplete, will mislead in proportion to their incompleteness. Under-ascertainment also complicates estimation of true disease burden, including rates of severe outcomes, hospitalization, and population-level immunity accrued through infection, each of which informs both the immediate response and longer-term surveillance priorities. A number of methods have been used in the past to examine the completeness of measles surveillance, including household surveys, capture-recapture analysis of hospitalized cases, and examination of administrative data¹⁷. All have found substantial under-estimation of the true burden. These studies however were conducted when the diagnosis of measles largely relied on a clinical case definition, and reporting was manual. In the current era, most confirmed cases are based on laboratory testing with molecular testing, or less commonly, serology. In many jurisdictions, including Arizona and Utah, positive tests for reportable diseases are automatically transmitted to public health via electronic laboratory reporting. These factors improve the completeness and accuracy of surveillance, however they still depend on individuals presenting for care and being tested. However, using two complementary methods, we found substantial under-ascertainment of measles in this outbreak. The magnitude of under-ascertainment likely varies by setting. Factors such as access to care, health care seeking behavior, testing practices, and the efficiency of laboratory reporting are probable drivers. With many large outbreaks of measles arising in tight knit communities, it is critical to better understand how these factors influence reporting in different settings.

The two approaches we used were not formally integrated and were instead applied independently to different data sources and indeed generated estimates for different underlying populations: the total number of cases related to the outbreak for the phylodynamic models and the total number of cases in the outbreak community for the agent-based model. Recent work has shown that integrating genomic and epidemiologic data can improve inference of outbreak dynamics; however, such approaches typically rely on a single integrated model^33–36. Such joint frameworks combine pathogen sequence data with epidemiologic observations to reconstruct transmission, estimate reproduction numbers, and infer unobserved infections. However, because these approaches rely on a single integrated model they cannot provide independent validation of findings. Our dual, independent approach draws on distinct data sources and assumptions and therefore allows for independent assessment of outbreak size. Estimates from multiple approaches can lend confidence to inference, particularly in settings where surveillance is believed to be incomplete.

This study has several limitations. The phylodynamic analysis relies on coalescent assumptions, and the sampling proportion is unknown; in a setting with substantial under-ascertainment, observed sequences may not be representative of all infections. Limited genetic diversity in measles viruses over short time scales also constrains phylogenetic resolution. The estimates presented in this study should not be used to determine rates of hospitalization or severe sequelae, since we know that this community provided home treatments to many individuals who would have been hospitalized in other communities.

Similarly, the agent-based model depends on assumptions regarding contact structure, transmission dynamics, and vaccination coverage. Contact patterns were based on age structure and school-based mixing and may not fully capture behavioral heterogeneity, particularly in this unique population. Transmission was calibrated to a fixed basic reproduction number; however, sensitivity analyses across a range of plausible R₀ values (5-17) yielded similar estimates of outbreak size, suggesting that findings are robust to uncertainty in this parameter. Isolation and quarantine were applied uniformly throughout the simulation upon case detection, and the model did not incorporate post-exposure prophylaxis. Case detection was represented as a fixed probability, which may not capture variability in healthcare-seeking behavior or testing practices across individuals and over the course of the outbreak. The model also assumes no change in behavior over the course of the outbreak, an assumption that may not hold given the substantial public attention this outbreak received. Finally, in the absence of a gold standard for true outbreak size, direct validation of either approach is impossible.

The analytical approaches applied here are pathogen agnostic and are useful in settings where surveillance completeness is uncertain, such as influenza and mpox. However, the magnitude of under-ascertainment in this outbreak reflects characteristics specific to this community, including low vaccination coverage, large household sizes, and limited engagement with healthcare systems, and may not be representative of other settings. In settings with higher vaccination coverage or more diffuse contact patterns, both the magnitude and timing of transmission may differ.

By combining genomic and mechanistic modeling approaches, we conclude that the Utah-Arizona border outbreak was considerably larger than reported. These findings suggest that under-ascertainment may be substantial even in the modern setting with laboratory testing, electronic laboratory reporting, and effective public health infrastructure. Although this outbreak occurred in a close-knit community with limited engagement with external institutions, the mechanisms driving underreporting are unlikely to be unique to it. Vaccine skepticism, distrust of public health authorities, and limited engagement with healthcare systems are increasingly common features of populations with low vaccination coverage in the United States, and the same factors that lead individuals to decline vaccination may also reduce their likelihood of seeking care, accepting testing, or otherwise being captured by passive surveillance when illness occurs. The result is a structural bias in which outbreaks become less visible precisely in the communities where they are most likely to occur.

As pockets of low coverage continue to grow, the need for accurate assessment of the impact will increase. Complementary approaches such as sero-surveillance, wastewater monitoring, or community-based case finding may be needed to accurately characterize burden and inform outbreak response in these settings.