Cooper Jackie A, Miller George J, Humphries Steve E
Division of Cardiovascular Genetics, Department of Medicine, British Heart Foundation Laboratories, Rayne Building, Royal Free and University College Medical School, London, UK.
Atherosclerosis. 2005 Jul;181(1):93-100. doi: 10.1016/j.atherosclerosis.2004.12.026.
We have compared the predictive value of the PROCAM and Framingham risk algorithms in healthy UK men from the Second Northwick Park Heart Study (NPHS-II) (50-64 years at entry), followed for a median of 10.8 years for coronary heart disease (CHD) events. For PROCAM, the area under the receiver operating characteristic (ROC) curve was 0.63 (95% CI, 0.59-0.67), and not significantly different (p = 0.46) from the Framingham score, 0.62 (0.58-0.66). Sensitivities for a 5% false-positive rate (DR(5)) were 13.8 and 12.4%, respectively. Calibration analysis for PROCAM gave a ratio of observed to expected events of 0.46 (Hosmer-Lemeshow test, p < 0.0001) and 0.47 for Framingham (p < 0.0001). Using measures taken at 5 years of high-density lipoprotein cholesterol and (estimated) low-density lipoprotein cholesterol levels increased the ROC by only 1%. An NPHS-II risk algorithm, developed using a 50% random subset, and including age, triglyceride, total cholesterol, smoking status, and systolic blood pressure at recruitment, gave an ROC of 0.64 (0.58-0.70) with a DR(5) of 10.7% when applied to the second half of the data. Adding family history and diabetes increased the DR(5) to 18.4% (p = 0.28). Adding lipoprotein(a) >26.3 mg/dL (relative risk 1.6, 1.1-2.4) gave a DR(5) of 15.5% (p = 0.55), while adding fibrinogen levels (relative risk for 1S.D. increase = 1.5, 1.1-2.0) had essentially no additional impact (DR(5) = 16.9%, p = 0.95). Thus, the PROCAM algorithm is marginally better as a risk predictor in UK men than the Framingham score, but both significantly overestimate risk in UK men. The algorithm based on NPHS-II data performs similarly to those for PROCAM and Framingham with respect to discrimination, but gave an improved ratio of observed to expected events of 0.80 (p = 0.01), although no score had a high sensitivity. Any novel factors added to these algorithms will need to have a major impact on risk to increase sensitivity above that given by classical risk factors.
我们比较了PROCAM风险算法和弗雷明汉风险算法对来自第二次诺斯威克公园心脏研究（NPHS-II）的健康英国男性（入组时年龄为50 - 64岁）的预测价值，这些男性随访冠心病（CHD）事件的中位时间为10.8年。对于PROCAM，受试者工作特征（ROC）曲线下面积为0.63（95%CI，0.59 - 0.67），与弗雷明汉评分0.62（0.58 - 0.66）无显著差异（p = 0.46）。5%假阳性率（DR(5)）时的敏感度分别为13.8%和12.4%。PROCAM的校准分析得出观察到的事件与预期事件的比率为0.46（Hosmer-Lemeshow检验，p < 0.0001），弗雷明汉为0.47（p < 0.0001）。使用5年时测量的高密度脂蛋白胆固醇和（估计的）低密度脂蛋白胆固醇水平，ROC仅增加了1%。使用50%随机子集开发的NPHS-II风险算法，包括入组时的年龄、甘油三酯、总胆固醇、吸烟状况和收缩压，应用于另一半数据时，ROC为0.64（0.58 - 0.70），DR(5)为10.7%。加入家族史和糖尿病后DR(5)增加到18.4%（p = 0.28）。加入脂蛋白(a)>26.3mg/dL（相对风险1.6，1.1 - 2.4）时DR(5)为15.5%（p = 0.55），而加入纤维蛋白原水平（1个标准差增加的相对风险 = 1.5，1.1 - 2.0）基本没有额外影响（DR(5) = 16.9%，p = 0.95）。因此，在英国男性中，PROCAM算法作为风险预测指标略优于弗雷明汉评分，但两者都显著高估了英国男性的风险。基于NPHS-II数据的算法在区分能力方面与PROCAM和弗雷明汉算法相似，但观察到的事件与预期事件的比率提高到了0.80（p = 0.01），尽管没有评分具有高敏感度。添加到这些算法中的任何新因素都需要对风险有重大影响，才能将敏感度提高到超过经典风险因素所提供的水平。
