A Familiar Stroke Prognosis Model Finds External Validation
Prognosis can be a trepidacious component of the post-stroke conversation. Practitioners are often hesitant to provide concrete answers to patients and their families, leading to frustration on both sides. Multiple stroke prognosis scores have been created over the years and focus on NIHSS, age, and non-lacunar sub-type as the main variables. These scores are limited by lack of external validity, use of variables not universally recorded in the stroke registry, and limitation of use for either ischemic or hemorrhagic stroke.
This British study demonstrates external validity of a model that uses these common variables to predict 30 day ischemic/hemorrhagic stroke mortality. Internal data was derived from the Sentinel Stroke National Audit Programme (January-June 2013) and external validation was done using the South London Stroke Register (2005-2012).
2 models were ultimately created, A and B. The variables used were:
age (<60, 60-69, 70-79, 80-89, 90 or greater), Afib, ischemic or hemorrhagic stroke subtype, and NIHSS. The 2 models differed in that A used the full NIHSS while B used only the level of consciousness component.
Model A had a c-statistic of 0.86 (95% CI 0.85-0.88) and 0.87 (0.84-0.89) for internal and external validation respectively. For Model B, the corresponding c-statistics were 0.82 (0.81-0.83) and 0.86 (0.83-0.89)
The disadvantage of this model is that its c-statistic predictive value is not significantly better than prior models, like the iScore. Other limitations include this model being internally and externally validated in a UK (albeit ethnically diverse) population. It is also unknown whether these patients received IV-tPA, clot extraction, or aggressive measures for their bleeds, all of which would certainly skew outcomes.
The advantage of this model is that it only uses a few key variables and may be used for ischemic or hemorrhagic stroke. NIHSS’s are often incomplete in the medical record and level of consciousness alone appears to be a good surrogate. The population from the Sentinel program was unselected thereby making it more robust. Future stroke studies may now consider this model given its external validation and permission of cases mixing.