Читать книгу Diabetic Retinopathy and Cardiovascular Disease - Группа авторов - Страница 20

Risk scores are generally developed using data from prospective observational cohort studies. The predictive models calculate a score from a series of weighted risk factors. The weights are regression coefficients or hazard ratios that account for how important each risk factor is in predicting cardiovascular disease. There are a number of stages in selecting the variables to be incorporated into a model, with various statistical methodologies employed. Potentially relevant risk factor variables are selected for investigation on the basis of clinical knowledge and existing epidemiological evidence. The associations between these variables and cardiovascular outcomes are modelled, initially in univariate analyses. Various models are then built, adding and removing variables to see whether they benefit the model in terms of goodness of fit (i.e., whether they improve the correlation between risk predicted by the model and risk observed in the study cohort). Variables that improve the model’s predictive capabilities are generally retained. Other factors may also contribute to the decision to retain or exclude variables from the final predictive model, including cost and ease of measurement and potential impact on patient motivation for lifestyle modification for example [32].