Development of a flexible self-calculation scoring model to determine stroke occurrence

Abstract

Stroke has become a significant threat to global public health, the ideal solution to which is primary prevention. Identification and management of determinants of stroke among various variables in different datasets are essential steps for its primary prevention. This study aimed to develop a flexible scoring model, which can easily modify different datasets. The public dataset containing 41,931 cases with 643 occurrences of stroke was randomly divided into training, validation, and test datasets comprising 25,158 (60%), 8,386 (20%), and 8,387 (20%) cases, respectively. Three continuous variables (age, body mass index, and average glucose level) and seven categorical variables (heart disease, hypertension, sex, married/smoking/work/residence status) in the dataset were converted using the weight of evidence method. The significant variables among 10 transformed variables were selected using multivariable logistic regression analyses. The scoring model for stroke occurrence was developed in the training and validation datasets, and performance was evaluated in the test dataset. Age, average glucose level, heart disease, and hypertension were significant variables of stroke occurrence. The scoring model was easily calculated using four determinants and indicates that the stroke occurrence ranged from 0.04 to 12.50%. The performance of the scoring model on the test dataset was similar to that on the validation dataset. This novel point scoring model is flexible enough to modify various datasets and can be used for determinant identification. Furthermore, its simplicity allows individuals to manage determinants by self-calculating stroke occurrence. Our model contributes to primary prevention using determinant identification and management.