摘要：The aim was to create an algorithm to transform self-reported outcomes from a stroke register to the modified Rankin Scale (mRS). Two stroke registers were used: the Väststroke, a local register in Gothenburg, Sweden, and the Riksstroke, a Swedish national register. The reference variable, mRS (from Väststroke), was mapped with seven self-reported questions from Riksstroke. The transformation algorithm was created as a result of manual mapping performed by healthcare professionals. A supervised machine learning method—decision tree—was used to further evaluate the transformation algorithm. Of 1145 patients, 54% were male, the mean age was 71 y. The mRS grades 0, 1 and 2 could not be distinguished as a result of manual mapping or by using the decision tree analysis. Thus, these grades were merged. With manual mapping, 78% of the patients were correctly classified, and the level of agreement was almost perfect, weighted Kappa (Kw) was 0.81. With the decision tree, 80% of the patients were correctly classified, and substantial agreement was achieved, Kw = 0.67. The self-reported outcomes from a stroke register can be transformed to the mRS. A mRS algorithm based on manual mapping might be useful for researchers using self-reported questionnaire data.
其他摘要：Abstract The aim was to create an algorithm to transform self-reported outcomes from a stroke register to the modified Rankin Scale (mRS). Two stroke registers were used: the Väststroke, a local register in Gothenburg, Sweden, and the Riksstroke, a Swedish national register. The reference variable, mRS (from Väststroke), was mapped with seven self-reported questions from Riksstroke. The transformation algorithm was created as a result of manual mapping performed by healthcare professionals. A supervised machine learning method—decision tree—was used to further evaluate the transformation algorithm. Of 1145 patients, 54% were male, the mean age was 71 y. The mRS grades 0, 1 and 2 could not be distinguished as a result of manual mapping or by using the decision tree analysis. Thus, these grades were merged. With manual mapping, 78% of the patients were correctly classified, and the level of agreement was almost perfect, weighted Kappa (K w ) was 0.81. With the decision tree, 80% of the patients were correctly classified, and substantial agreement was achieved, K w = 0.67. The self-reported outcomes from a stroke register can be transformed to the mRS. A mRS algorithm based on manual mapping might be useful for researchers using self-reported questionnaire data.
