Statistical boosting is a powerful tool that has become increasingly more popular in recent years. It works very well in the high-dimensional data setting, and it has many beneficial properties, e.g., shrinkage, automated (implicit) variable selection, and interpretable statistical models. Boosting has especially good potentials within the field of biomedicine, where the high-dimensional data setting is typical. However, its use in biomedical applications has so far been somewhat limited. A factor that holds boosting back is the lack of uncertainty measures, e.g., confidence intervals for the parameter estimates, which is often required by medical doctors. In this thesis, we focus on this issue. We consider generalized additive models and how likelihood-based boosting, which uses Fischer scoring to compute the iterative updates, is applied to estimate the underlying model structure. An advantage of the likelihood-based methodology is the possibility of deriving an approximate hat matrix, which is exact in the Gaussian case. Using it, we can compute approximate confidence intervals for the expected response and pointwise confidence bands for the effect functions. We will elaborate on an existing method and generalize it. In particular, we will develop the necessary results to apply it to likelihood-based boosting with penalized stumps as base learners. We conduct an extensive simulation study where we compare the coverage of these approximate uncertainty measures with empirical counterparts. The empirical confidence bands and intervals are derived by bootstrap, which is the current standard practice in the field of boosting to obtain uncertainty measures.