论文部分内容阅读
Background Few diabetes prediction model have taken non-diabetes deaths as competing events, which may overestimate the risk of developing diabetes, especially in long-term follow-up for older population with a relatively high mortality.It remains unknown whether proportional subdistribution hazards model perform better to risk prediction for diabetes incidence than Cox proportional hazards model among the middle-aged and older adults.Methods Retrospective review of prospectively collected data by the Beijing Longitudinal Study of Aging (BLSA) between August 1992 and December 2012.Diabetes was diagnosed as a self-reported history of diabetes diagnosis, taking antidiabetic medicine, or having FPG ≥7.0 mmol/L (126mg/dl) at any of the periodic examinations.A questionnaire with good reliability and validity was designed to measure the risk factors, including dietary habits, lifestyle, psychological factors, cognitive function and physical conditions.Proportional subdistribution hazards model and Cox proportional hazards model were used to evaluate the risk of developing a first diabetes event.Receiver operating characteristic (ROC) curve, areas under the ROC curves (AUC), and calibration plots were used to evaluate the discrimination and calibration ability of the both methods.Results 244 people having diabetes were ruled out at baseline.During the 20 years follow-up, 144 cases of 1857 participants were documented for diabetes incidence with a median 10.9 (Interquartile range: 8.0-15.3) years follow-up period.The incidence density was 7.908/1000 person-years.Cumulative incidence function (CIF) of diabetes was 11.60% after adjusting for the competing risks of non-diabetes deaths.Area under the ROC curve was 0.74 (95% CI: 0.70-0.78) and 0.70 (95%CI: 0.66-0.75) in proportional subdistribution hazards model and Cox proportional hazards model, respectively.The difference value of AUCs was more than zero between proportional subdistribution hazards model and Cox proportional hazards model (Z=2.52, P=0.012).Sensitivity, specificity, and Youden index of the proportional subdistribution hazards model was 0.81, 0.52, and 0.67, and that of Cox proportional hazards model was 0.84, 0.42, and 0.63.In terms of discrimination and calibration, the proportional subdistribution hazards model is slightly superior to Cox proportional hazards model.Conclusion In summary, a multivariable proportional subdistribution hazards model was developed, after accounting for non-diabetes death, which performed better than Cox proportional hazards model to risk prediction for 20-year incident diabetes among middle-aged and older adults.