Grémy F, Salmi L R
Université de Montpellier I, Faculté de Médecine, LUSIEEM, Hôpital Lapeyronie.
Bull Acad Natl Med. 1995 Feb;179(2):317-31; discussion 331-3.
This paper tries to review what is scientifically known about the predictive values of biological tests of HIV infection. The epidemiological situation for that infection is characterized by two facts: the very high values of sensitivity and specificity which are close to unity; the prevalence of seropositivity which is on average--at least in western countries--, very low (except for some small specific groups). Under those conditions, Negative Predictive Values are always very close to unity, and the percentage of false negative tests is extremely low. Things are quite different for Positive Predictive Value, which varies very rapidly with very small shifts or uncertainties about specificity and prevalence. In the case when prevalence is very low (general population screening) and at the same time specificity is not excellent (that means < 0.99 or even < 0.995), Positive Predictive Value is very poor and the proportion of false positive tests rather important. Indeed the analysis of scientific literature, using the method of "best synthesis evidence", reveals numerous discrepancies as to the value of specificity among different tests. Figures vary a lot from one study to another. It is not obvious which screening strategies are concerned by the results, which finally entail a strong statistical uncertainty. Finally, the figures published in the literature are given by high standard laboratories. One may fear the tests realized in routine laboratories are less reliable. As a conclusion, let us say that despite their very good quality, the biological tests, when used separately, should not be trusted without strong previous criticism when applied to samples of the general population. Any biological screening should be preceded by a clinical examination, including a precise inquiry, in order to detect people at risk, that means with a high prior probability. Clinical dialogue has moreover another great interest: it allows health consulting and education, and calls for personal responsibility for both seropositive and negative subjects. It is the best choice of method to reach a high preventive effectiveness.
本文旨在综述关于HIV感染生物学检测预测价值的科学认知。该感染的流行病学情况有两个特点：一是灵敏度和特异度的值非常高，接近1；二是血清阳性率平均而言——至少在西方国家——非常低（某些特定小群体除外）。在这些条件下，阴性预测值总是非常接近1，假阴性检测的比例极低。阳性预测值的情况则大不相同，它会随着特异度和患病率的极小变化或不确定性而迅速变化。当患病率非常低（一般人群筛查）且同时特异度不太理想（即<0.99甚至<0.995）时，阳性预测值非常低，假阳性检测的比例相当高。事实上，运用“最佳综合证据”方法对科学文献进行分析后发现，不同检测的特异度值存在诸多差异。不同研究中的数据差异很大。不清楚哪些筛查策略与这些结果相关，这最终导致了很大的统计不确定性。最后，文献中公布的数据是由高标准实验室提供的。人们可能担心常规实验室进行的检测可靠性较低。总之，尽管生物学检测质量很高，但单独使用时，应用于一般人群样本时，如果没有事先的严格批判，不应轻信。任何生物学筛查都应先进行临床检查，包括详细询问，以检测出有风险的人群，即具有高先验概率的人群。此外，临床对话还有另一个重要意义：它有助于健康咨询和教育，并促使血清阳性和阴性个体都承担个人责任。这是实现高预防效果的最佳方法选择。
