Abstract
BACKGROUND Dysphagia is a common consequence of intracerebral hemorrhages (ICH). It can lead to enduring impairments of dietary intake and the requirement for feeding via percutaneous gastrostomy (PEG) tubes. However, variabilities in the course of swallowing recovery after ICH make it difficult to anticipate the need for PEG placement in an individual patient. A new tool called the GRAVo score was recently developed to predict PEG tube placement after an ICH but has not been externally validated. Our study aims were to externally validate the GRAVo score in a multicenter cohort and reexamine the role of race in predicting PEG placement, given the uncertain biological plausibility for this relationship observed in the derivation cohort. METHODS Patients for this analysis were selected from a previously completed multicenter, randomized, double-blind futility design clinical trial, the Intracerebral Hemorrhage Deferoxamine trial, and underwent a retrospective review of prospectively collected data. The GRAVo scores were computed by using previously established methods using the following variables: Glasgow Coma Scale ≤ 12 (2 points), race (1 point for Black), age > 50 years (2 points), and ICH volume > 30 mL (1 point). Association of GRAVo scores with PEG placement were examined by using logistic regression analysis after adjustment for exposure to deferoxamine. Model performance was estimated by using area under the receiving operating characteristic curve (AUROC). Subsequently, a second model was created by excluding scores for race, and the AUROC of both models were compared. RESULTS A total of 291 patients with complete data points served as the study cohort; 38 (13%) underwent PEG placement. The median GRAVo score for patients in the PEG and non-PEG groups were 4 (interquartile range 3-4) versus 2 (interquartile range 2-3), respectively (p < 0.0001). External validation of the GRAVo score yielded an AUROC of 0.7008 (95% confidence interval 0.6036-0.78); the model obtained without assignment of scores for the variable race yielded an AUROC of 0.6958 (95% confidence interval 0.6124-0.7891). The receiver operating characteristic curves from both models demonstrated close overlap. CONCLUSIONS The results of our external validation demonstrate the validity of GRAVo scores for predicting PEG tube placement after an ICH. However, its performance was more modest compared with that of the derivation cohort. Inclusion of the race variable had no measurable effect on model performance. Differences in patient characteristics between these cohorts may have influenced our results. These findings should be taken into consideration when using the GRAVo score to assist clinical decision making on PEG placement after an ICH.
