3. Statistical considerations
According to the penetration aspiration scale (PAS), dysphagia was classified four degrees: no dysphagia (PAS score = 1 to 2), mild dysphagia (PAS score = 3 to 4), moderate dysphagia (PAS score = 5 to 6), and severe dysphagia (PAS score =7 to 8), corresponded to minimal、low、moderate and high risks of aspiration.14We compared the hierarchical entries of MAPS (Table.1) and PAS, and also classified MAPS to four degrees: no dysphagia (MPAS score = 1) with a minimal risk of aspiration, mild dysphagia (MPAS score = 2) with a low risk of aspiration, moderate dysphagia (MPAS score = 3 to 4) with a moderate risk of aspiration, and severe dysphagia (MPAS score = 5 to 6) with a high risk of aspiration. The highest score achieved in solid、semisolid or liquid swallowing trials was taken as the final score.
Dysphagia severity of modified GUSS was graded to four degrees that were the same as GUSS. 0 to 9 points were rated as severe dysphagia with a high risk of aspiration, 10 to 14 points as moderate dysphagia with a moderate risk of aspiration, 15 to 19 points as mild dysphagia with a low risk of aspiration, and 20 points as no dysphagia.6 Reliability for modified GUSS was calculated by Spearman’s rank correlation coefficient and Kappa statistics and the proportion of overall agreement (P0) as a raw agreement index. A rs or κ coefficient between 0.4 and 0.8 was rated substantial, and values > 0.8 were considered excellent. Sensitivity and specificity, as well as positive and negative predictive values, were determined by comparing the results of modified GUSS with the results of VFSS. Spearman’s rank correlation coefficient was also used to predict the correlation between results of modified GUSS and days from oral feeding to gastric tube extubation. Test validity of the values of rs or κ coefficient was the same as above. Wilcoxon signed rank tests was used to distinguish the risks of aspiration caused by solid、semisolid、and liquid food. Statistical analyses were performed with SPSS 17.0, and descriptive analyses were performed according to data characteristics.