Factors associated with adverse long-term outcomes in ischemic stroke: development of a prediction model and its validation


We sought to develop and validate the long-term outcomes in ischemic stroke based on demographic, clinical, and echocardiographic variables. Univariable and multivariable logistic regression modeling was performed in 325 consecutive ischemic stroke subjects (initial cohort). Prediction rules were developed and applied in similar 1305 patients (validation cohort). Only age, renal failure, and aortic root sclerosis were associated with increased all-cause mortality on multivariable logistic regression analyses. Based on the hazard ratios, 2 points were given for each decade of age over 40 years old, 6 for renal failure, and 3 for aortic root sclerosis. Long-term mortality was compared between three groups: low risk (0–5 points), moderate risk (6–10 points), and high risk (more than 11 points). In the initial cohort, Kaplan–Meier mean survival estimates were 59·8 ± 1·4 months in low risk and 37·4 ± 3·3 months high-risk groups (P < 0·001). In the validation cohort, Kaplan–Meier mean survival estimates were 53 ± 0·8 months in low risk and 33·1 ± 2·1 months high-risk groups (P < 0·001) (Fig. 1).

Figure 1.

Kaplan-Meier survival estimates among patients in the validation cohort.

A high prevalence of stroke along with the heterogeneity in its clinical outcome has generated interest in correctly identifying poor prognostic factors and in the development of prognostic models. Accurate early prediction of long-term prognosis may enable healthcare providers optimize rehabilitation process, secondary prevention care, and potentially improve resource allocation [1]. Predictors have been previously identified to have a univariable relationship with the adverse outcome. However when multivariable analysis were employed the relationship of these predictors to the adverse outcomes have not been well established [2]. Our study was aimed at developing a simple, accurate prediction model, incorporating only readily available variables, which retained predictive power in a multivariable logistic regression analysis.

We conclude that during the index admission, a simple predictive score, which includes age, aortic root sclerosis, and renal function, can easily distinguish the patients with poor prognosis. Prospective validation has confirmed this finding.


The authors would like to acknowledge the contributions of Drs. Amar Shah, Bulibek Batyrjan, Du le, Houman Nourkeyhani, Jaspreet Arora, and Leigh Cagino.