Abstract
We aim to predict the duration of hospitalization for acute stroke in Belgium by evaluating the external validity of the prolonged length of stay (PLOS) score and by formulating a new prediction score that may be better suited for the Belgian healthcare system. This single-center retrospective study is based on data collected prospectively from the departmental stroke registry. To validate the PLOS score, receiver operating characteristic curves were constructed and Hosmer-Lemeshow tests were implemented. Odds ratios were calculated by models of logistic regression, based on predictors of length of stay (LOS) with significance in univariate analyses, and were translated into a new risk score. C-statistics for prediction of LOS ≥7 days, LOS ≥14 days, and LOS ≥30 days using the PLOS score were in the range of 0.6-0.7. Thrombolytic therapy, mortality, and need for institutionalization had a notable negative influence on the discrimination of the PLOS score. Overall, the PLOS score performed better for prediction of LOS ≥14 days than for LOS ≥7 days and ≥30 days. The Belgian length of stay for stroke (BLOSS) score is proposed as a simplified prediction model based only on the NIHSS score and age. The PLOS score showed moderate value for prediction of hospitalization duration for acute stroke in this Belgian cohort. A prediction model based only on age and stroke severity may be a worthy alternative.