Abstract Body: Post-stroke dysphagia (PSD) is a common complication following stroke. Predicting the recovery from PSD remains a major challenge. The Predictive Swallow Score (PRESS) model, derived and validated in Switzerland, sought to predict recovery from PSD after an acute ischemic stroke (AIS). We sought to perform the first external validation of the PRESS model.

A two-center retrospective cohort study of 149 dysphagic patients with functional oral intake scale (FOIS) ≤4 following AIS was conducted. Demographic data and clinical co-variates (age, NIHSS, ANY2 score, frontal opercular stroke, FOIS) were collected to predict the recovery from dysphagia at post-stroke day 7. The primary outcome was impaired swallow at day 7 (FOIS ≤ 4).

The mean age was 72.1 years (±16.0) and 79 (53%) were females. Median (Q1-Q3) NIHSS at presentation was 14 (7-20), ANY2 score 5 (3-6), FOIS at presentation 1 (1-4). 66 patients had frontal-opercular involvement of stroke. Median FOIS at 7-day follow-up from stroke was 4 (1-5), with 71 (44%) being classified as impaired. The Hosmer-Lemeshow test rejected the null hypothesis that the PRESS model was a good fit to the validation data (p=<0.0001, x²=55.735, df=5). Calibration curve analysis revealed that the PRESS model overpredicted the likelihood of persistent dysphagia. The C-statistic of the PRESS model on the validation cohort was 0.74 (95% CI: 0.65-0.81). In a secondary analysis, impaired subjects (FOIS ≤ 4) at day-7 had a higher median (Q1-Q3) admission NIHSS (15 [7.5-21] vs 12 [6-17], p=0.02), ANY2 score (6 [4-6] vs 4 [3-5], p<0.0001), PRESS score (8 [6-9] vs 5 [2-8], p<0.00001), and higher frequency of frontal opercular stroke location (64% vs 46%, p=0.03). However, age (p=0.17) and sex (p=0.07) were associated with persistent dysphagia at day 7.

This first external validation study of the PRESS model did not support its use as a tool to predict recovery from PSD, potentially owing to regional variation in the predictive power of certain co-variates such as stroke severity, stroke location, and age. The study is limited by its retrospective design. A larger prospective cohort study is needed to evaluate the PRESS model prior to clinical adoption.