台湾35~74岁男性体检者骨质疏松5年发病风险预测模型

目的 构建台湾35~74岁男性体检者骨质疏松（OP）5年发病风险（个体化）预测模型。方法 选择1999~2005年首次参加台湾美兆健检的35~74岁男性体检者7801人，去除基线患OP者505人后余下7296人，将4个体检中心分为建模队列（台北中心，n=3844），余下3中心为验证队列（n=3452）。按建模队列5年后是否发生OP为因变量，以该队列基线指标为自变量在进行单变量分析后，建立多元逐步Logistic回归模型，以ROC曲线下的面积（AUC）为判定预测模型拟合优度的主要指标，用验证队列对模型的外部效度进行评估。模型建立后，将人群的预测风险概率正态化后再转化为可实际应用操作的4个风险等级。结果 基线4个体检中心OP患病率范围为4.77%~7.88%。去除基线患者后，全部受检者5年发病率为2.34%（171/7296），4个中心OP发病率范围为：1.52%~4.89%。多变量Logistic回归构建的预测模型包括年龄、日常工作性质、腰围、体重和血肌酐水平5个指标。建模队列建立的预测模型的ROC曲线下面积（AUC）约为0.728 (95%CI：0.675-0.772) ，验证队列外部效度验证结果为0.698(0.633-0.762)。将建模队列划分为4个风险等级后，显示中危（占11.9%）和高危（占1.1%）的个体5年内发生OP的危险分别比一般人群高2.4倍和8.1倍。结论 利用台湾美兆健检纵向数据资料建立的OP 5年个体风险预测模型效应与信度均较高，且纳入的预测变量和建立的风险等级评价标准简单实用，因此对于今后无论个体进行自身OP风险评价还是社区工作者对社群人群进行OP监测均具有较大的应用价值。

Risk predicting model of 5-year osteoporosis occurrence in 35-74-year-old males in Taiwan

Objective To build a risk predicting model of 5-year osteoporosis (OP) occurrence in 35-74-year-old males in Taiwan. Methods A total of 7081 individuals who received health check-up in Taiwan MJ Health Administration Organization for the first time from 1999 to 2005 were selected. After excluding 505 individuals who were OP patients at baseline, 7296 subjects were enrolled. Four health check-up centers were selected as the modeling cohort (Taipei center, n=3844), and the left 3 centers were selected as the testing cohort (n=3452). Multivariate logistic regression model was established based on an univariate model with variables that predicted OP incidence in 5 years and with variables that appeared on baseline. We evaluated model predictability by the area under the receiver-operating characteristic (ROC) curve (AUC) and testified its diagnostic property on the testing sample. Once final model was defined, we next established rules to characterize 4 different degrees of the risk based on cut points of these probabilities after transforming into normal distribution by log-transformation. Results At baseline, the range of OP prevalence was 4.77-7.88% in 4 check-up centers. After excluding 505 OP individuals at baseline, the incidence of OP was 2.34% (171/7296). The range of incidence was 1.52-4.89% in 4 check-up centers. Final multivariable logistic regression model included 5 risk factors: age, routine work status, waist circumference, weight, and serum creatinine. The AUC of the model consisted of modeling cohort was 0.728 (95% CI, 0.675-0.772). The AUC of testing cohorts was 0.698 (95% CI, 0.633-0.762). After labeling 4 risk degrees in modeling cohort, the risks of developing OP for subjects in moderate risk degree (11.9% of all subjects) and high risk degree (1.1% of all subjects) were 2.4 times and 8.1 times higher than those of general population, respectively. Conclusion The predictability and reliability of the model of estimated risks on developing OP within 5 years, which is based on Taiwan MJ longitudinal check-up population database, are good and satisfied. The predictive variables involved and the evaluation criterions to establish the risk degrees are simple and practicable. It will be helpful both for individual to assess the risk on developing OP and for community workers to survey the development of OP in community population.