Page 72 - Functional impairment and cues for rehabilitation of head and neck cancer patients -

70
Chapter 3
Statistical analysis
Analyses were performed using IBM® SPSS® Statistics 25.0. Participant characteristics, swallowing outcomes and results of the (first attempt of the) SPEAD-test were compared between the patients and healthy participants using the Mann-Whitney U test for continuous variables, the Fisher’s exact test for categorical outcomes or the linear-by-linear approximation of the Chi-square test for ordinal outcomes. The correlation between SPEAD-rate and age, gender, length, dental prosthesis and spoon use was evaluated by means of the Spearman correlation coefficient (ρ).
The single measures Intraclass Correlation Coefficient (ICC) from a two-way random model using a consistency definition was used to evaluate intra-rater and inter-rater reliability. A two-way mixed model was used to evaluate test-retest reliability. If the ICC was below 0.50, reliability was considered poor, between 0.50 and 0.75 moderate, between 0.75 and 0.90 good and above 0.90 excellent (32). If the ICC was > 0.75, the test was considered sufficiently reliable. Since no gold standard is available, construct validity was assessed using known groups validity, convergent validity and discriminant validity. Known groups validity was evaluated by assessing the difference in SPEAD-rate between patients and healthy participants. SPEAD-rate was expected to be lower in patients compared to healthy participants. The Mann-Whitney U test for continuous variables or chi-square test for categorical outcomes was used to test the difference. Also, r-type effect sizes were calculated (r = Z/√N). According to the guidelines of Cohen (33), the effect sizes (r) of 0.1 to 0.3 were considered small, 0.3 to 0.5 moderate, and above 0.5 large. All effect sizes were expected to be large (> 0.5). In addition, an effect size for healthy participants versus patients, adjusted for age, dental prosthesis, and spoon use, was calculated by means of a linear regression model. For this analysis, all variables were standardized by subtracting the mean and dividing by the standard deviation. Also, the difference in SPEAD-rate between participants with no, mild, moderate and severe SLP graded dysphagia was tested by means of the Kruskal-Wallis test as well as the difference in SPEAD-rate between participants with different DIGEST scores.
The construct validity of the test was assessed based on a number of hypotheses. First, a number of hypotheses were tested with regard to SPEAD-test correlations with subjective outcomes: we expected the SPEAD-rate to be lower with decreasing self-rated percentage of eating and drinking speed, decreasing self-rated percentage of swallow function, increasing SWAL-QOL total score, increasing SWAL-QOL eating duration subscore, and increasing degree of dysphagia rated by the SLP. Second, the following hypotheses were tested to evaluate correlations with objective outcomes: we expected the SPEAD-rate to be lower with decreasing FOIS, increasing DIGEST grade and decreasing maximal mouth opening. These hypotheses were tested by calculating the Spearman correlation coefficient. Correlation coefficients (ρ) < 0.3 were indicative of a weak correlation, from 0.3 to 0.6 of a fair correlation, from 0.6 to 0.8 of a moderately strong and ≥ 0.8 of a strong correlation (34). The SPEAD-rate was expected to be strongly related to perceived swallowing performance than to physical swallowing functions,