The Latium acute stroke registry: data quality and validation of the NIH Stroke Scale in predicting in-hospital death

Abstract

Background and Purpose: The experimental phase of the stroke registry for the region of Lazio, completed in 2003, collected clinical data using an ad hoc survey on emergency room (ER) patients who presented with suspected stroke. We evaluated the accuracy and completeness of the data collected and assessed the validity of the National Institute of Health Stroke Scale (NIHSS) in predicting in-hospital mortality.

Methods and Results: Two different indicators were developed to evaluate the quality of the registry: the response (% ER patients surveyed with suspected stroke) and the coverage rates (% ER patients surveyed discharged with stroke). Their values by type of emergency service were respectively: first aid ER =59.3% and 45.5%; level one ER =39.7% and 27.6%; level two ER =17.2% e 12.0%. Stroke severity, age, type of emergency facility and time of arrival at the ER were associated with the accuracy and completeness of the registry. The in-hospital mortality predictive model showed good discriminative ability (area under the receiveroperating characteristic (ROC) curve (AUC)=0.79 (95%CI 0.75, 0.84)) and good calibration (Hosmer- Lemeshow’s goodness-of-fit (HL) test p=0.48). AUC and HL tests, estimated after applying the linear predictor to an external validation cohort, were 0.79 (95%CI 0.74, 0.84) and p=0.23 respectively.

Conclusions: Survey completeness and accuracy were associated with patient and facility characteristics, suggesting that it could be improved. The NIHSS calculated at ER admission is a good predictor of inhospital mortality.