125 7 appropriate covariance structure to fit the data.17 To account for within-patient correlations, a random patient factor was added, and a random intercept was used to account for the different entry levels of patients. The fixed-effect factors tumor site, treatment modality, tumor stage, timing of assessment, sex, and age, as well as 2-way interactions of the factors tumor site, treatment modality, and tumor stage during the assessment period were assessed using the AR(1) method (first-order autoregressive covariance pattern) for parameter estimation. Tumor site consisted of 3 levels: oral cavity, oropharynx, or larynx and hypopharynx. Treatment modality consisted of 4 levels: RT, CRT, surgery, or surgery followed by post-operative (C)RT. Tumor stage consisted of 4 levels (stage 1 to 4), timing of assessment consisted of 5 levels (M0, M3, M6, M12, and M24), sex consisted of 2 levels (male or female), and age was defined as a continuous variable. The model included a stepwise backward selection of factors, in which factors that were not significant at a p<0.10 level were removed, beginning with the interactions. A hierarchical structure was maintained, meaning that if an interaction was included in the model, the main effects were also represented in the model. Risk factors were reported as estimated unstandardized regression coefficients with 95% confidence intervals (CI) and p-values. Swallowing dysfunction (a score above the cut-off value of 8 number of swallows) was used to create a Receiver Operating Characteristic (ROC) curve, to help facilitate the use of the linear mixed–effects model in identifying factors associated with swallowing problems in patients with HNC. The coefficients of the significant covariates, together with the value of the intercept of the mixed model analysis, were combined into a formula for the estimated number of swallows. The intercept is the value of the estimated number of swallows when all coefficients remain zero. Addition of the coefficients will lead to an increase or decrease of the estimated number of swallows. For each time point, the formula was filled with average variable values for significant coefficients, as calculated by a restricted maximum likelihood approach (REML). Model assumptions were verified by plotting the residuals versus the fitted values. All analyses were performed using Statistical Package for the Social Sciences (SPSS) version 25 (Chicago, IL). A p-value <0.10 was considered statistically significant. Results Of 135 patients that met the inclusion criteria, 128 were included and 115 performed baseline measurements. During the study period with 2 years follow-up, 25 patients were deceased, and 24 patients dropped out. In addition, five measurements at M24 could not be performed because of the COVID-19 situation and were indicated as missing. The flow diagram of the study is shown in Figure 1. Personal and clinical characteristics of the study