Background and purpose: Classical robust optimization considers uncertainties in patient setup and particle range. However, anatomical changes occurring during the treatment are neglected. Our aim was to compare classical robust optimization (cRO) with anatomical robust optimization (aRO), to quantify the influence of anatomical variations during the treatment course, and to assess the need of adaptation. Materials and methods: Planning CT and weekly control CTs (cCTs) from 20 head and neck patients were analysed. Three intensity-modulated proton therapy (IMPT) plans were compared: conventional PTV-based plan; cRO, using solely the planning CT, and aRO, including additionally the first 2 cCTs in the optimization. Weekly and total cumulative doses, considering anatomical variations during the treatment, were calculated and compared with the nominal plans. Results: Nominal plans fulfilled clinical specifications for target coverage (D_98% ≥95% of prescribed dose). The PTV-based and cRO approaches were not sufficient to account for anatomical changes during the treatment in 10 and 5 patients, respectively, resulting in the need of plan adaptation. With the aRO approach, in all except one patient the target coverage was conserved, and no adaptations were necessary. Conclusion: In 25% of the investigated cases, classical robust optimization is not sufficient to account for anatomical changes during the treatment. Adding additional information of random anatomical variations in the optimization improves plan robustness.