Hofer T, Weissfeld J
Division of General Internal Medicine, University of Michigan, Ann Arbor.
Med Decis Making. 1994 Oct-Dec;14(4):357-68. doi: 10.1177/0272989X9401400406.
To determine whether the complex strategy of lipid measurements for the detection of patients with high blood cholesterol levels proposed by the first Expert Panel of the National Cholesterol Education Panel (NCEP) could be simplified without significant loss of accuracy.
Decision-analysis-based model of competing case detection strategies as compared with the NCEP strategy. A Markov model was used to estimate numbers of people treated over ten years as a result of the different classification strategies.
Conditional probabilities for the decision trees were derived from cholesterol distributions in national population-based surveys. Parameters for the Markov model were from published major epidemiologic studies and clinical trials.
Misclassification to treatment vs non-treatment as a continuous function of the distribution of true low-density lipoprotein (LDL).
A simplified strategy was designed that screens high-risk persons with two LDL measurements and low-risk people with one cholesterol measurement followed by two LDL measurements if the initial value is high. This algorithm requires 37% fewer measurements to classify a population. The overall accuracy of classification to treatment based on the NCEP I cutoff points is high, with a positive predictive value of 95% and a negative predictive value of 87% (relative to 97% and 80%, respectively, for the NCEP I protocol). This strategy is very similar to published NCEP II guidelines. A strategy that recommends an LDL determination for everyone, as a recent NIH consensus panel suggested, also significantly reduces the number of measurements required by 48%. The positive predictive value is 93%, vs 97% for the NCEP I protocol. The negative predictive value is 92%, vs 80% for the NCEP I.
The complex measurement strategy initially proposed in the NCEP I guidelines did not improve accuracy of classification over the simpler and more convenient strategies that the authors evaluated and that have been substantially adopted in the NCEP II guidelines.
确定美国国家胆固醇教育计划（NCEP）首个专家小组提出的用于检测高血胆固醇水平患者的复杂脂质测量策略是否可以简化而不会显著降低准确性。
与NCEP策略相比，基于决策分析的竞争病例检测策略模型。使用马尔可夫模型估计由于不同分类策略在十年内接受治疗的人数。
决策树的条件概率来自基于全国人群调查的胆固醇分布。马尔可夫模型的参数来自已发表的主要流行病学研究和临床试验。
将误分类为治疗组与非治疗组作为真实低密度脂蛋白（LDL）分布的连续函数。
设计了一种简化策略，即对高危人群进行两次LDL测量，对低危人群进行一次胆固醇测量，如果初始值高则随后进行两次LDL测量。该算法对人群进行分类所需的测量次数减少了37%。基于NCEP I临界值进行治疗分类的总体准确性较高，阳性预测值为95%，阴性预测值为87%（相对于NCEP I方案分别为97%和80%）。该策略与已发表的NCEP II指南非常相似。正如最近美国国立卫生研究院共识小组所建议的，一种建议对每个人都进行LDL测定的策略也显著减少了所需测量次数的48%。阳性预测值为93%，而NCEP I方案为97%。阴性预测值为92%，而NCEP I为80%。
NCEP I指南最初提出的复杂测量策略在分类准确性方面并没有比作者评估的更简单、更方便的策略有所提高，而这些策略已在NCEP II指南中大量采用。
