Robust and highly sensitive outcomes are crucial for small trials in rare diseases. Combining different outcome types might improve sensitivity to identify disease severity and progression, yet innovative methodologies are scarce. Here we develop an Item Response Theory framework that allows integrated modeling of both continuous and categorical outcomes (ccIRT). With degenerative ataxias, a group of rare neurological coordination diseases, as a showcase, we developed a ccIRT model integrating two ataxia outcome types: a clinician-reported outcome (Scale for the Assessment and Rating of Ataxia; SARA; categorical data) and digital-motor outcomes for gait and limb coordination (continuous data). The ccIRT model leveraged data from 331 assessments from a natural history study for spastic ataxias. The model describes SARA items and digital-motor outcomes data as functions of a common underlying ataxia severity construct, evaluating 9 gait and 17 limb coordination digital-motor measures for their ability to add to SARA in estimating individual ataxia severity levels. Based on our proposed workflow for assessing digital-motor outcomes in ccIRT models, the final model selected three digital gait and three limb coordination measures, reducing average uncertainty in ataxia severity estimates by 49% (10% SD) compared to the SARA-only IRT model. Trial simulations showed a 49% and 61% reduction in sample sizes needed to detect disease-modifying effects in two genotypes. Overall, our ccIRT framework for combining multiple outcome domains, even of different variable types, facilitates a more precise estimation of disease severity and a higher power, which is particularly relevant for rare diseases with inherently small and short trials.