Practical prediction model for the risk of 2-year mortality of individuals in the general population

Abstract

This study proposed to validate a prediction model and risk-stratification tool of 2-year mortality rates of individuals in the general population suitable for office practice use. A risk indicator (R) derived from data in the literature was based on only 6 variables: to calculate R for an individual, starting with 0, for each year of age above 60, add 0.14; for a male, add 0.9; for diabetes mellitus, add 0.7; for albuminuria >30 mg/g of creatinine, add 0.7; for stage ≥3 chronic kidney disease (CKD), add 0.9; for cardiovascular disease (CVD), add 1.4; or for both CKD and CVD, add 1.7. We developed a univariate logistic regression model predicting 2-year individual mortality rates. The National Health and Nutrition Examination Survey (NHANES) data set (1999–2004 with deaths through 2006) was used as the target for validation. These 12,515 subjects had a mean age of 48.9±18.1 years, 48% males, 9.5% diabetes, 11.7% albuminuria, 6.8% CVD, 5.4% CKD, and 2.8% both CKD and CVD. Using the risk indicator R alone to predict mortality demonstrated good performance with area under the receiver operating characteristic (ROC) curve of 0.84. Dividing subjects into low-risk (R=0–1.0), low intermediate risk (R>1.0–3.0), high intermediate risk (R>3.0–5.0) or high-risk (R>5.0) categories predicted 2-year mortality rates of 0.52%, 1.44%, 5.19% and 15.24%, respectively, by the prediction model compared with actual mortality rates of 0.29%, 2.48%, 5.13% and 13.40%, respectively. We have validated a model of risk stratification using easily identified clinical characteristics to predict 2-year mortality rates of individuals in the general population. The model demonstrated performance adequate for its potential use for clinical practice and research decisions.