Comparison of Dynamic Imaging Grade of Swallowing Toxicity (DIGEST) estimates and segmentation-derived measures of pharyngeal residue severity

The Dynamic Imaging Grade of Swallowing Toxicity (DIGEST) grades swallowing efficiency based on visuoperceptual estimates of the percentage of bolus in the pharynx after the primary swallow (DIGEST-PR). We compared DIGEST-PR ratings with segmentation-based metrics of % residue to characterize agreement and patterns of discrepancy across approaches. We analyzed 188 initial swallows from videofluoroscopies with available DIGEST-PR criterion-reference scores. Blinded raters segmented bolus area on three frames: (A) before bolus transfer into the pharynx (preswallow residue); (B) one frame before upper esophageal sphincter (UES) opening; and (C) one frame after UES closure. Measures of total bolus area (B), new bolus area (B − A), total residue area (C) and new bolus residue area (C − A) were used to derive four metrics: Bolus Clearance Ratio (“BCR” = C/B); new-bolus-residue-as-%-new-bolus (“NBRPNB” = [(C − A)/(B − A)]); new-bolus-residue-as-%-total-bolus (“NBRPTB” = [(C − A)/B]); and total-residue-as-%-new-bolus (“TRPNB” = [C/(B − A)]). These metrics were mapped to DIGEST-PR severity categories (0%-9%, 10%-49%, 50%-90%, > 90%). Agreement with DIGEST-PR was assessed using prevalence- and bias-adjusted kappa. Generalized estimating equations evaluated the odds of discrepant segmentation-derived metrics relative to DIGEST-PR using ordinal logistic regression. The sample consisted of 20 patients who performed a total of 114 thin liquid, 38 pudding, and 36 cracker swallows. Severity was skewed toward lower severity ratings. Across all consistencies, the NBRPNB metric showed the strongest correspondence with DIGEST-PR. Discrepancies were greatest when metrics did not distinguish the new bolus from preswallow residue. Preswallow residue increased across successive swallowing tasks in the protocol, contributing to the largest disagreement for cracker stimuli, which also had the smallest new bolus areas. Segmentation-based measures that express residue as a percent of the new bolus show good concordance with DIGEST-PR ratings, particularly for thin liquids and lower levels of residue (DIGEST-PR < 50%). Further work is needed to evaluate agreement for thick liquid and solid stimuli and in cases with greater residue severity.