Incorporating Clinical Considerations into Statistical Analyses of Markers: A Quiet Revolution in How We Think About Data
[摘要] I can say 2 things about the biostatistical methods I learned in college for the analysis of markers and diagnostic tests. First, they closely reflect the methods actually used in the literature. Second, they are completely useless. So although my professors were completely right to teach me about the current state of the art, they left me completely unprepared for conducting informative analyses.The basic problem shared by many of the biostatistical methods used in marker research is that they relate numbers to other numbers, rather than to anything that we might care about in the real world. A correlation coefficient, for instance, tells us the average difference in y for a 1SD increase in x , where x and y are first divided by their SDs. If this is not sufficiently abstract for you, remember that the standard deviation is the square root of the average of the squared difference between each value of a variable and the mean of that variable. Sensitivity tells you the probability that a patient with disease has a positive test result, which assumes that you know whether the patient has disease and are trying to guess whether he or she will test positive or not. An area under the curve (AUC) tells you the probability that for a randomly selected pair of patients, 1 with disease and 1 without, the patient with disease has a higher test score. Is a doctor really ever asked to guess which of 2 patients actually has a disease?The abstract nature of many statistical metrics becomes particularly problematic when we try to interpret the results of diagnostic studies. Take, for instance, …
[发布日期] [发布机构]
[效力级别] [学科分类] 过敏症与临床免疫学
[关键词] [时效性]