OBJECTIVES: Claims data are often used to estimate the prevalence of chronic diseases, typically by dividing the number of patients with disease-related claims (e.g., ≥1 or ≥2 claims) by the number of studied individuals. Such estimates will have a downward bias because not all diseased patients will generate disease-related claims within their enrollment period. This downward bias can be substantial for underserved diseases that lack effective treatments. We explored whether an empirical Bayes estimator for the number of diseased individuals who do not generate claims could improve the accuracy of claims-based prevalence estimates. METHODS: As an example, we studied the prevalence of a rare dermatological condition without any FDA-approved therapies. After accounting for enrollment time, individuals in a large nation-wide claims database were identified as having 0, 1, 2, 3, etc., disease-related claims. These counts were modeled using a mixture of Poisson distributions, with an unknown mixing distribution. Empirical Bayes approaches, which are frequently used to estimate numbers of unobserved species in ecological experiments, were used to estimate the number of diseased individuals without claims, and to provide adjusted prevalence estimates. RESULTS: Out of over 4 million individuals with at least one year of continuous enrollment, n=2,026 had disease-related claims, comprised of n=1,422 with one claim, n=317 with two claims, n=134 with 3 claims, etc. The traditional method for estimating prevalence identified 4.9 cases per 10,000 persons.  After applying the empirical Bayes approach, the estimated prevalence increased to 7.9 cases per 10,000 persons, and became closer to published prevalence estimates based on non-claims data sources. CONCLUSIONS: In this example application, prevalence estimates based on claims data were increased by over 60% by using empirical Bayes approaches to account for large numbers of diseased individuals who did not generate claims. The increased prevalence estimates were more consistent with the published literature.