132 The number of WST failures increased over time to almost 12% three months after treatment, and the reason for failure changed from ‘not being able to drink the 100 mL of water’, to ‘coughing or choking post swallow’. Coughing or choking post swallow was found to have a specificity of up to 91.7% in predicting aspiration, and the WST is therefore a useful tool for early detection of swallowing dysfunction.21 Previous research found dysphagia and aspiration rates between 12% to 21%, similar to the results found in this study.18,22 Especially patients that received surgery with adjuvant treatment have a higher prevalence of dysphagia in comparison to patients that receive RT alone, as also seen in this research by the higher number of swallows.7 Besides WST failures, between 11% and 18% of patients had a WST score above the cut-off score (>2 standard deviations above the mean of healthy subjects), with the most problems 3 months after treatment. Previous research showed that the objective WST and subjective patient-reported outcomes measuring swallowing function have a low correlation and can therefore not be used interchangeably.23 A future study might aim at developing a prediction model with subjective questionnaires, to obtain individual risk scores for swallowing problems in patients with HNC, including a larger number of potential predictors. These predictors could then, apart from the predictors used in this study, also include a larger range of treatment modalities and normal tissue changes. This also makes it possible to study whether the factors found in this study are found with subjective outcome measures as well. Strengths and limitations Strengths of our study were the prospective study design, the use of the linear mixedeffects model checklist with recommendations for reporting multilevel data and analyses,24 and the use of an objective swallowing test with a high test-retest reliability.14 Limitations were the relatively low number of patients at follow-up, which limited the number of factors that could be explored, and the relative large drop-out and missing values. These missing data may have influenced the results, because it is unknown how these patients would have performed on the WST. Although linear mixed-effects model analysis is especially designed for repeated measurement analyses, and is better at handling missing values in comparison to other regression analyses,25 these regression models do not take into account the number of deaths as competing risk. Additionally, since the study group was relatively small, it was chosen to only look at interactions between timing of assessment and treatment, location, and tumor stage. Another limitation of this study were the significant differences found between treatment versus tumor stage and tumor site, as seen in Table 1. Patients receiving RT have an oropharynx, hypopharynx or larynx tumor, while patients receiving surgery most often have a tumor in the oral cavity. In addition, patients receiving CRT have larger tumors (stage III and IV), while patients receiving surgery most often have