Using epidemiological data from February 2020 to September 2022, the study evaluates two- and five-day forecasts of positive SARS-CoV-2 tests, hospital admissions and intensive-care admissions. Penalized-spline generalized additive models with negative-binomial outcomes account for overdispersion. Forecast quality varies across epidemic phases and outcomes: ICU admissions are predicted more reliably than hospital admissions or positive cases, and abrupt trend changes produce the largest errors. The findings emphasize the importance of timely, reliable health-information systems. First published online on 7 August 2025; included in the December 2025 issue.