Röhrig Bernd
Gesundheitswesen. 2023 Jun;85(6):578-594. doi: 10.1055/a-1937-9516. Epub 2023 Feb 27.
Many diagnostic tests are currently being performed around the world to detect SARS-CoV-2 infection. Positive and negative test results are not one hundred percent accurate, but have far-reaching consequences. There are false positives (test positive, uninfected) and false negatives (test negative, infected). A positive/negative result does not necessarily mean that the test subject is actually infected/non-infected. This article has two objectives: 1. to explain the most important characteristics of diagnostic tests with binary outcome 2. to point out problems and phenomena of interpretation of diagnostic tests, on the basis of different scenarios.
Presentation of the basic concepts of the quality of a diagnostic test (sensitivity, specificity) and pre-test probability (prevalence of test group). Calculation (including formulas) of further important quantities.
In the basic scenario, sensitivity is 100%, specificity 98.8%, and pre-test probability of 1.0% (10 infected persons per 1,000 tested). For 1,000 diagnostic tests, the statistical mean is 22 positive cases, 10 of which are true-positive. The positive predictive probability is 45.7%. The prevalence calculated from this (22/1,000 tests) overestimates the actual prevalence (10/1,000 tests) by a factor of 2.2. All cases with a negative test outcome are true negative. The prevalence has a strong influence on the positive and negative predictive value. This phenomenon occurs even with otherwise very good test values of sensitivity and specificity. At a prevalence of only 5 infected persons per 10,000 (0.05%), the positive predictive probability drops to 4.0%. Lower specificity amplifies this effect, especially with small numbers of infected persons.
If the sensitivity or specificity is below 100%, diagnostic tests are always error-prone. If the prevalence of infected persons is low, a large number of false positive results are to be expected - even if the test is of good quality with a high sensitivity and especially a high specificity. This is accompanied by low positive predictive values, i. e. positive tested persons are not infected. A false positive test result in the first test can be clarified by carrying out a second test.
目前全球正在进行许多诊断测试以检测新型冠状病毒感染。阳性和阴性测试结果并非百分之百准确，但却有着深远影响。存在假阳性（测试呈阳性但未感染）和假阴性（测试呈阴性但已感染）情况。阳性/阴性结果并不一定意味着受测者实际已感染/未感染。本文有两个目标：1. 解释二元结果诊断测试的最重要特征；2. 根据不同场景指出诊断测试解释中的问题和现象。
介绍诊断测试质量的基本概念（敏感性、特异性）和预测试概率（测试组患病率）。计算（包括公式）其他重要量值。
在基本场景中，敏感性为100%，特异性为98.8%，预测试概率为1.0%（每1000名受测者中有10名感染者）。对于1000次诊断测试，统计平均值为22例阳性病例，其中10例为真阳性。阳性预测概率为45.7%。据此计算出的患病率（22/1000次测试）高估了实际患病率（10/1000次测试）2.2倍。所有测试结果为阴性的病例均为真阴性。患病率对阳性和阴性预测值有很大影响。即使敏感性和特异性测试值在其他方面非常好，这种现象也会出现。当每10000人中只有5名感染者（0.05%）时，阳性预测概率降至4.0%。较低的特异性会放大这种影响，尤其是在感染者数量较少时。
如果敏感性或特异性低于100%，诊断测试总是容易出错。如果感染者患病率较低，即使测试质量良好，敏感性高尤其是特异性高，也会出现大量假阳性结果。这伴随着较低的阳性预测值，即测试呈阳性的人未被感染。首次测试中的假阳性结果可通过进行第二次测试来澄清。
