诊断性试验的设计与评价研究生-3节课

1、诊断性试验的分诊断性试验的分析与评价析与评价四川大学华西医院四川大学华西医院实验医学科实验医学科秦秦 莉莉2一、定义一、定义v对疾病进行诊断的试验方法，即称为诊对疾病进行诊断的试验方法，即称为诊断性试验。断性试验。诊断性试验不仅包括实验室检查，还包括病诊断性试验不仅包括实验室检查，还包括病史、体检结果、影象学检查、各种公认的诊史、体检结果、影象学检查、各种公认的诊断标准等。断标准等。3诊断性试验的应用诊断性试验的应用1.1.诊断疾病诊断疾病2.2.筛检无症状病人筛检无症状病人3.3.疾病的随访疾病的随访4.4.判断疾病的严重性判断疾病的严重性5.5.估计疾病的临床过程极其预后估计疾病的临床过程

2、极其预后6.6.估计对治疗的反应估计对治疗的反应7.7.测定目前对治疗的实际反应测定目前对治疗的实际反应4二、评价诊断性试验的条件二、评价诊断性试验的条件v1.1.标准诊断：标准诊断： 疾病的诊断，必须有标准诊断（即金标准疾病的诊断，必须有标准诊断（即金标准, , gold standardgold standard，或参考标准，或参考标准，reference reference standardstandard）。标准诊断是目前公认的诊断方）。标准诊断是目前公认的诊断方法，如：活检、手术、尸检、特殊检查或长法，如：活检、手术、尸检、特殊检查或长期随访的结果期随访的结果5二、评价诊断性试验的条

3、件二、评价诊断性试验的条件v2.2.诊断方法的对比：诊断方法的对比：评价新的诊断性试验，必须与标准诊断方法进行评价新的诊断性试验，必须与标准诊断方法进行比较。比较。新的诊断性试验，应该具备方法更为简便、更为新的诊断性试验，应该具备方法更为简便、更为可靠或者减少危险、减少创伤、节约费用等优点，可靠或者减少危险、减少创伤、节约费用等优点，这样的诊断性试验才具有推广意义。这样的诊断性试验才具有推广意义。6二、评价诊断性试验的条件二、评价诊断性试验的条件v3.3.列出四格表列出四格表标准诊断标准诊断有病有病无病无病诊断性试验诊断性试验+ +a a( (真阳性真阳性) )b b( (假阳性假阳性) )-

4、 -c c( (假阴性假阴性) )d d( (真阴性真阴性) )7所评价的资料应能列出四格表，方法如下所评价的资料应能列出四格表，方法如下用标准诊断方法，诊断的病例数为用标准诊断方法，诊断的病例数为 a+ca+c在有病的受试者中，诊断性试验阳性者为在有病的受试者中，诊断性试验阳性者为 a a，阴性者为阴性者为 c c用标准诊断方法，判断无该病的例数为用标准诊断方法，判断无该病的例数为 b+db+d无该病的受试者中，诊断性试验阳性例数为无该病的受试者中，诊断性试验阳性例数为 b b，阴性例数为，阴性例数为 d d 从评价的资料中，不能绘制四格表的诊断性试从评价的资料中，不能绘制四格表的诊断性试验

5、，无法进行评价验，无法进行评价8三、评价诊断性试验的常用指标三、评价诊断性试验的常用指标v1.1.评价指标评价指标1)1)敏感度（敏感度（sensitivitysensitivity）：经金标准确诊有病的）：经金标准确诊有病的人中，诊断性试验阳性者所占的比例。人中，诊断性试验阳性者所占的比例。 SEN = a /SEN = a /（ a+ca+c）2)2)特异度（特异度（specificityspecificity）：经金标准诊断确定为）：经金标准诊断确定为无该病的人中，诊断性试验阴性所占的比例。无该病的人中，诊断性试验阴性所占的比例。 SPE = d /SPE = d /（b+db+d）93

6、)3)准确性（准确性（accuracyaccuracy）：经诊断性试验检查后）：经诊断性试验检查后真阳性与真阴性占总例数的比例。真阳性与真阴性占总例数的比例。 ACC =ACC =（a+da+d）/ /（a+b+c+da+b+c+d）4)4)阳性预测值（阳性预测值（positive predictive positive predictive valuevalue）：诊断性试验阳性的人中有病的人）：诊断性试验阳性的人中有病的人数所占的比例。（即诊断性试验阳性时，患数所占的比例。（即诊断性试验阳性时，患病的可能性病的可能性, ,即阳性结果的即阳性结果的验后概率验后概率） +PV = a /+PV

7、 = a /（a+ba+b）5)5)阴性预测值（阴性预测值（negative predictive negative predictive valuevalue）：诊断性试验阴性的人中，无该病）：诊断性试验阴性的人中，无该病的人数所占的比例。（即诊断性试验阴性时，的人数所占的比例。（即诊断性试验阴性时，不患该病的可能性）不患该病的可能性） -PV = d /-PV = d /（c+dc+d）106)6)患病率（患病率（prevalenceprevalence）：）： PREV =PREV =（a+ca+c）/ /（a+b+c+da+b+c+d）7)7)阳性似然比（阳性似然比（positive

8、likelihood ratiopositive likelihood ratio）：有）：有病者诊断性试验阳性的概率与无病者试验阳性的病者诊断性试验阳性的概率与无病者试验阳性的概率之比。概率之比。 +LR = a /+LR = a /（a+ca+c） / b / / b /（b+db+d） = =SenSen/ /（1-Spe1-Spe）8)8)阴性似然比（阴性似然比（negative likelihood rationegative likelihood ratio）：有）：有病者试验阴性的概率和无病者试验阴性的概率之病者试验阴性的概率和无病者试验阴性的概率之比。比。-LR = c /-L

9、R = c /（a+ca+c） / d / / d /（b+db+d） = =（1-Sen1-Sen）/ Spe/ Spe11LRLR：有病者得出某一试验结果的概率与无病者得出该试验结果的：有病者得出某一试验结果的概率与无病者得出该试验结果的概率之比。表示一个诊断试验结果出现在有病者和出现在无病概率之比。表示一个诊断试验结果出现在有病者和出现在无病者的可能性比值大小，代表了一个诊断性试验区分有病和无病者的可能性比值大小，代表了一个诊断性试验区分有病和无病的能力大小。的能力大小。LRLR越大，患病可能性越大越大，患病可能性越大12举例举例An Oxfordshire (England) grou

10、p of clinical investigators invited general practitioners in their area “to refer patients with suspected heart failure to our clinic.” Once there, these 126 patients underwent independent, blind BNP measurements and echocardiography. The first set of results from that study is shown in Table13Perfo

11、rmance of B-type Natriuretic Peptide 18 pg/mLPerformance of B-type Natriuretic Peptide 18 pg/mL As a As a diagnostic test for left ventricular dysfunctiondiagnostic test for left ventricular dysfunction141.You can calculate the proportion of patients with LVD who also have elevated BNP. （敏感度）（敏感度）a/

12、(a + c) = 35/40 = 0.88, or 88%“positivity in the presence of the target disorder” is Sensitivity.2.You can calculate the proportion of patients who are free of LVD who also have normal BNP. （特异度）（特异度）d/(b + d) = 29/86 = 0.34, or 34% “negativity in the absence of the target disorder” is Specificity.1

13、53.You can calculate the proportion of patients with elevated BNP who also have LVD. a/(a + b) = 35/92 = 0.38, or 38% “presence of the target disorder among positives” is Positive Predictive Value (PPV).Another term to express this value is the Post-test Likelihood given a Positive Test Result（阳性结果验

14、后概率）（阳性结果验后概率）4.You can calculate the proportion of patients with normal BNP who also are free of LVD. d/(c + d) = 29/34 = 0.85, or 85% “absence of the target disorder among negatives” as Negative Predictive Value (NPV).Clinicians more commonly think in terms of the post-test likelihood given a nega

15、tive result.（阴性结果验后概率）（阴性结果验后概率） 165.You can calculate the proportion of patients with LVD before you even measure their BNP. (a + c)/(a + b + c + d) = 40/126 = 0.32, or 32%By convention, we refer to that “pre-test probability of the target disorder” in the total population at risk (not considering

16、any additional diagnostic information) as PrevalencePrevalence, because it describes the prevailing rate of the target disorder in the patients who are undergoing the diagnostic test. 17验前概率、验后概率、患病率验前概率、验后概率、患病率v验前概率：患者没有进行诊断性试验时的患验前概率：患者没有进行诊断性试验时的患病概率病概率v验后概率：根据患者诊断性试验结果，结合验后概率：根据患者诊断性试验结果，结合验前概率

17、推测的患病概率验前概率推测的患病概率v患病率患病率= =验前概率验前概率186.You can calculate the odds that a patient has LVD before you ever measure their BNP. Pre-test odds（验前比）（验前比）= Pre-test Probability (100% Pre-test Probability) = 32%/(100% 32%)= 32%/68%= 0.47By convention, we refer to this as Pre-test Odds.And you can convert a

18、n odds back into a probability. Probability=odds/(odds + 1) = 0.47/1.47 = 0.32, or 32%197.You can calculate the likelihood that an elevated BNP is found in patients with, as opposed to patients without, LVD. a/(a + c)/b/(b + d) = Sensitivity/(100% Specificity)= 88%/(100% 34%)= 88%/66%= 1.3By convent

19、ion, we refer to that as a Likelihood Ratio of a positive test (some prefer to call it a Positive Likelihood Ratio).208.You can calculate the likelihood that a normal BNP is found in patients with, as opposed to patients without, LVD. c / (a + c) / d / (b + d) = (100% Sensitivity)/Specificity= (100%

20、 88%)/34%= 12%/34%= 0.4.By convention, we refer to that as a Likelihood Ratio of a negative test (some prefer to call it a Negative Likelihood Ratio).219.You can discover that if you multiply the Pre-test Odds（验前比）（验前比）from the population studied by the LR of a positive test result and convert the r

21、esulting Post-test Odds（验后比）（验后比） back to a probability, it is identical to the +PV. 验后比验后比 = 验前比验前比似然比（似然比（Pre-test odds x LR） = 0.47 x 1.3= 0.61and验后概率验后概率 = 验后比验后比/(1+验后比验后比) =0.61/1.61 = 0.38, or 38%(the same as you calculated in no. 3 above).2210. 阴性似然比阴性似然比(-LR)验前比得到阴性结果的验后比，可用公验前比得到阴性结果的验后比，可

22、用公式将此验后比转化为验后概率，即阴性结果的验后概率。式将此验后比转化为验后概率，即阴性结果的验后概率。Pre-test odds（-LR） = 0.47 x 0.4 = 0.19and0.19/1.19 = 0.15, or 15% (阴性结果验后概率）阴性结果验后概率）and100% 15% = 85%（阴性预测值）（阴性预测值）(the same as you calculated in no. 4 above).阴性结果的验后概率阴性结果的验后概率=100% (-PV)23You may have noticed that we havent introduced the terms

23、“true-positive rate” and “false-positive rate.” This is because weve found inconsistencies in their construction. Sure, the obvious numerator in a “false-positive” rate is cell b of Table 81, but what should we use for its denominator? Weve encountered three different denominators. Some folks insert

24、 (b+d) for its denominator, creating a number equal to (1-specificity); others use (a+b), creating a number equal to (1-PPV); and weve even encountered folks using (a+b+c+d) for its denominator, telling us the percentage of false-positive results in the entire study population. These are ambiguous t

25、erms and we wont use them here.24真阳性率、真阴性率、假阳性率、假阴性率真阳性率、真阴性率、假阳性率、假阴性率尽管临流学家不提倡用上述说法，但很多尽管临流学家不提倡用上述说法，但很多文献仍然使用上述说法，目前公认的定义如文献仍然使用上述说法，目前公认的定义如下：下：真阳性率真阳性率= =敏感度敏感度真阴性率真阴性率= =特异度特异度假阳性率假阳性率=1=1特异度特异度假阴性率假阴性率=1=1敏感度敏感度25vIf you apply these rules-of-thumb to BNP and LVD, youd conclude that this diag

26、nostic test had a good sensitivity (88%) and -PV (85%), but that its poor specificity (34%) dragged down its +PV (38%) and its +LR (1.3), and led to an -LR (0.4) that was almost as useless as the +LR. In fact, its +PV or post-test probability (38%) was only slightly higher than its pre-test probabil

27、ity or prevalence (34%). And thats the way it was reported. These investigators concluded, “introducing routine measurement (of BNP) would be unlikely to improve the diagnosis of symptomatic (LVD) in the community.”26vHowever, their report also documented the effect of two other cut-points for BNP.

28、This led both to a counter claim on the usefulness of BNP in the subsequent letters to the editor and to an opportunity for us to describe some alternative ways of presenting information about the accuracy of a diagnostic test. When we applied a higher cut-point for a positive BNP test (75 rather th

29、an 18 in the original report) we could construct the following table27SenSen=26/40=0.65=26/40=0.65SpeSpe=75/86=0.87=75/86=0.87+LR=5.08+LR=5.0828vMultilevel Likelihood Ratios（多水平似然比）（多水平似然比）Because the authors of the BNP study presented their results for two other cutoffs (10 pg/mL and 76 pg/mL), you

30、 can divide their test results into three groups ( 75). Although you cant any longer describe these results with binary measures like sensitivity and specificity, you can make great use of “multilevel” LRs. That is, you can describe, for any level of the test result, the likelihood that that level w

31、ould be observed in a patient with, as opposed to one without, the target disorder. 2930You can easily apply the LR for a test result to any prevalence (pre-test odds) of the target disorder. Suppose a patient has a pre-test probability of 50% (a pre-test odds of 1:1). You can simply multiply that p

32、atients pre-test odds (say, 1:1) by the LR for that patients test result (say, 80 pg/mL, with an LR of 5.1). This generates a post-test odds of 5.1, which you can convert into a post-test probability by solving 5.1/(1+5.1). This yields a posttest probability of 84%, which is much higher than you wou

33、ld generate with the cutoff of 10 pg per mL. For the latter case, you multiply 11.3 and get a post-test probability of LVD of only 1.3/2.3=0.56 or 56%31v2.2.诊断性试验指标的临床意义诊断性试验指标的临床意义稳定的指标：敏感性、特异性、稳定的指标：敏感性、特异性、LRLR、LRLR（是最重要的指标）（是最重要的指标） 相对稳定的指标：准确性相对稳定的指标：准确性不稳定的指标：阳性预测值、阴性预测值、不稳定的指标：阳性预测值、阴性预测值、患病率

34、患病率 32不稳定指标及其影响因素不稳定指标及其影响因素 现举例说明不稳定指标及其影响因素：某地运动现举例说明不稳定指标及其影响因素：某地运动员有胸前区疼痛史者例，分别作运动心员有胸前区疼痛史者例，分别作运动心电图及冠状动脉造影，结果如下：电图及冠状动脉造影，结果如下： 冠状动脉狭窄冠状动脉狭窄75%75%（金标准）（金标准） 是是 否否运动心电图运动心电图 + 55+ 55（a a） 7 7（b b） 6262 - 49 - 49（c c） 8484（d d） 133133 104 91 195 104 91 19533vSEN=a/(a+c)=55/104=53%SEN=a/(a+c)=5

35、5/104=53%vSPE=d/(b+d)=84/91=92%SPE=d/(b+d)=84/91=92%vACC=(a+d)/(a+b+c+d)=55+84/195=71%ACC=(a+d)/(a+b+c+d)=55+84/195=71%v+PV=a/(a+b)=55/62=89%+PV=a/(a+b)=55/62=89%v-PV=d/(c+d)=84/133=63%-PV=d/(c+d)=84/133=63%vPREVPREV（冠状动脉狭窄）（冠状动脉狭窄）=(a+c)/(a+b+c+d)=104/195=53%=(a+c)/(a+b+c+d)=104/195=53%v+LR=SEN/(1-

36、SPE)=0.53/+LR=SEN/(1-SPE)=0.53/（1-0.921-0.92）=6.6=6.6v-LR=(1-SEN)/SPE=(1-0.53)/0.92=0.51-LR=(1-SEN)/SPE=(1-0.53)/0.92=0.51v阳性率阳性率=(a+b)/(a+b+c+d)=62/195=31%=(a+b)/(a+b+c+d)=62/195=31%34如果扩大检查范围，将该地全体运动员均如果扩大检查范围，将该地全体运动员均作上述检查，结果如下：作上述检查，结果如下： 冠状动脉狭窄冠状动脉狭窄 是是 否否运动心电图运动心电图 55(a) 42(b) 55(a) 42(b) 979

37、7 49(c)49(c)478(d)478(d)527527 104104520520 624624 35vSEN=55/104=53%SEN=55/104=53%（不变）（不变）vSPE=478/520=92%SPE=478/520=92%（不变）（不变）vACC=(55+478)/624=85%ACC=(55+478)/624=85%（增加，个百（增加，个百分点）分点）v+PV=55/97=57%+PV=55/97=57%（下降，个百分点）（下降，个百分点）v-PV=478/527=91%-PV=478/527=91%（增加，个百分点）（增加，个百分点）vPREV=104/624=17%P

38、REV=104/624=17%（原为）（原为）v+LR=0.53/+LR=0.53/（1-0.921-0.92）=6.6=6.6（不变）（不变）v-LR=-LR=（1-0.531-0.53）/0.92=0.51/0.92=0.51（不变）（不变）v阳性率阳性率=97/624=15%=97/624=15%（下降，个百分点）（下降，个百分点）36结果解释结果解释PREV 的下降是由于扩大了检查范围，被的下降是由于扩大了检查范围，被检人群中，患病者例数减少所致。检人群中，患病者例数减少所致。随着随着PREV的下降，、的下降，、，而，而、阳性率、阳性率。SEN、SPE、LR、LR、稳定不变。、稳定不变

39、。v在评价诊断性试验中，一般不用阳性率，因阳在评价诊断性试验中，一般不用阳性率，因阳性病例数并未说明是真阳性或假阳性。性病例数并未说明是真阳性或假阳性。37 从以上组数据可以看出从以上组数据可以看出： v当患病率变化后，当患病率变化后，SENSEN、SPESPE、LRLR及及-LR-LR都很稳定，都很稳定，两组数字相同，而两组数字相同，而PVPV随随PREVPREV不同有很大变化，不同有很大变化，PVPV随患病率增高而增加。因此，随患病率增高而增加。因此，PVPV不能看做试验本身不能看做试验本身的特性。敏感度越高，则假阴性越低，假阴性率等的特性。敏感度越高，则假阴性越低，假阴性率等于漏诊率。因

40、此，高敏感度的试验，用于临床诊断于漏诊率。因此，高敏感度的试验，用于临床诊断时漏诊率低。通常用高敏感度试验，阴性结果排除时漏诊率低。通常用高敏感度试验，阴性结果排除诊断，又称为诊断，又称为SnNoutSnNout。v高敏感度试验用于：高敏感度试验用于：疾病漏诊可能造成严重后果；疾病漏诊可能造成严重后果；用于排除疾病；用于排除疾病；用于筛选无症状且发病率又比较低的疾病。用于筛选无症状且发病率又比较低的疾病。38v特异性越高，则假阳性率越低，假阳性率等于误特异性越高，则假阳性率越低，假阳性率等于误诊率。因此，特异性高的试验，用于临床时误诊诊率。因此，特异性高的试验，用于临床时误诊机会少。高特异性试

41、验，用于肯定诊断、确诊疾机会少。高特异性试验，用于肯定诊断、确诊疾病。当试验结果阳性时，临床确诊价值最大。病。当试验结果阳性时，临床确诊价值最大。v用高特异性试验，阳性结果肯定诊断，又称为用高特异性试验，阳性结果肯定诊断，又称为SpPinsSpPins。v特异性高的试验适用于：特异性高的试验适用于：肯定疾病诊断；肯定疾病诊断；凡假阳性结果会导致病人精神负担，或不当防凡假阳性结果会导致病人精神负担，或不当防治措施会给病人带来严重危害。治措施会给病人带来严重危害。39四、诊断性试验的样本大小的计算四、诊断性试验的样本大小的计算v计算时，先查阅文献或作预试估计计算时，先查阅文献或作预试估计SENSE

42、N、SPESPEv病例组（）：用该试验敏感性估计病例组（）：用该试验敏感性估计= =（）（）v对照组（）：用该试验特异性估计对照组（）：用该试验特异性估计 P P= =（）（）v用估计总体率的样本公式分别计算用估计总体率的样本公式分别计算vm ma a2 2（）（）2 2vm ma a2 2（）（）2 2显著性水平显著性水平a a取取. . . .（双侧检验）（双侧检验）诊断性试验的允许误差诊断性试验的允许误差一般定在一般定在0.05-0.100.05-0.1040五、似然比的临床应用五、似然比的临床应用 v可用于临床计算患病的概率，便于更准确地对患者作出可用于临床计算患病的概率，便于更准确地

43、对患者作出诊断。诊断。 例如：对怀疑急性心肌梗塞患者，作肌酸磷酸激酶例如：对怀疑急性心肌梗塞患者，作肌酸磷酸激酶（CPKCPK）测定，根据其结果可计算似然比。爱丁堡皇家）测定，根据其结果可计算似然比。爱丁堡皇家医院将怀疑心肌梗死者医院将怀疑心肌梗死者360360例收入病房例收入病房, ,检测检测CPKCPK，由一，由一位不知位不知CPKCPK结果的医生根据心电图和尸检结果判断有心结果的医生根据心电图和尸检结果判断有心肌梗死者肌梗死者230230例，无心梗者例，无心梗者130130例，测定值如下：例，测定值如下： 2-39 40-79 80-119 120-159 160-199 200-239

44、 240-279 280-319 320-359 360-399 400-439 440-4792-39 40-79 80-119 120-159 160-199 200-239 240-279 280-319 320-359 360-399 400-439 440-479 =480 =480AMI AMI + 2 13 30 30 21 19 18 13 19 15 7 8 35+ 2 13 30 30 21 19 18 13 19 15 7 8 35 - 88 26 8 5 0 1 1 1 0 0 0 0 0 - 88 26 8 5 0 1 1 1 0 0 0 0 0 41如将如将CPK

45、80CPK 80单位作为诊断心肌梗死的临界值，列出四格单位作为诊断心肌梗死的临界值，列出四格表表 急性心肌梗塞急性心肌梗塞 是是 否否 80u 215 1680u 215 16 CPK CPK 80u 15 11480u 15 114 230 130 230 130 SEN=a/ SEN=a/（a+ca+c）=215/230=0.93=215/230=0.93 SPE=d/ SPE=d/（b+db+d）=114/130=0.88=114/130=0.88 +LR=SEN/ +LR=SEN/（1-SPE1-SPE）=0.93/=0.93/（1-0.881-0.88） =7.75=7.7542再一

46、步分析，则可计算不同水平的似然比：再一步分析，则可计算不同水平的似然比：CPK AMICPK AMI（+ +） AMIAMI（- -） LRLR n n 比例比例 n n 比例比例 280 97 97/230=0.42 1 1/130=0.01 0.42/0.01=42 280 97 97/230=0.42 1 1/130=0.01 0.42/0.01=42 80279 118 118/230=0.51 15 15/130=0.12 0.51/0.12=4.2 80279 118 118/230=0.51 15 15/130=0.12 0.51/0.12=4.2 4079 13 13/230=

47、0.06 26 26/130=0.02 0.06/0.02=0.34079 13 13/230=0.06 26 26/130=0.02 0.06/0.02=0.3139 2 2/230=0.01 88 88/130=0.67 0.01/0.67=0.01139 2 2/230=0.01 88 88/130=0.67 0.01/0.67=0.01合计合计 230 130230 13043v似然比的应用：似然比的应用：验前比（验前比（Pretest OddsPretest Odds）= =验前概率验前概率/ /（1-1-验前概率）验前概率）验后比（验后比（Post-test OddsPost-te

48、st Odds）= =验前比验前比似似然比然比验后概率（验后概率（Post-test ProbabilityPost-test Probability）= =验验后比后比/ /（1+1+验后比）验后比）44例：某患者活动后即感胸前区疼痛，在医院检查例：某患者活动后即感胸前区疼痛，在医院检查CPKCPK为为7272单位，试问患单位，试问患AMIAMI的可能性有多大？的可能性有多大？解：（解：（1 1） 估计患估计患AMIAMI的可能性有的可能性有50%50%（试验前概（试验前概率）率） （2 2） 按前表中按前表中CPKCPK为为72U72U的的 LR=0.30LR=0.30 （3 3） 计算：

49、计算： 验前比验前比=0.50/=0.50/（1-0.501-0.50）=1=1 验后比验后比=1=10.30=0.300.30=0.30 验后概率验后概率=0.3/=0.3/（1+0.31+0.3）=0.23=0.23 答：该病例患答：该病例患AMIAMI的机会只有的机会只有23%23%。4546部份常见病、诊断性试验结果的阳性似然比部份常见病、诊断性试验结果的阳性似然比病名病名金标准金标准诊断性试验诊断性试验阳性似然比阳性似然比冠心病冠心病冠状动脉造影，狭窄冠状动脉造影，狭窄75%75%典型心绞痛发作典型心绞痛发作115115冠心病冠心病冠状动脉狭窄（血管造冠状动脉狭窄（血管造影）影）不典

50、型心绞痛，有阳性病史不典型心绞痛，有阳性病史1414心肌梗塞心肌梗塞心电图或尸检心电图或尸检肌酸激酶肌酸激酶80u80u7.757.75深静脉血栓形成深静脉血栓形成静脉造影静脉造影深静脉血栓形成症状（疼痛、皮深静脉血栓形成症状（疼痛、皮肤颜色改变、局部发热、压痛、肤颜色改变、局部发热、压痛、周径增大周径增大3cm3cm）全部体征伴周径）全部体征伴周径增大增大2.62.6深静脉血栓形成深静脉血栓形成静脉造影静脉造影以上体征以上体征4 4项，且无周径改变项，且无周径改变0.150.15深静脉血栓形成深静脉血栓形成静脉彩色多普勒静脉彩色多普勒血浆血浆D-dimerD-dimer1292 ng1292

51、 ng/ml/ml2.02.03.13.1冠心病冠心病冠状动脉狭窄（血管造冠状动脉狭窄（血管造影）影）心电图运动试验心电图运动试验 STST下抑下抑2.5mm2.5mm393922.4922.4911111.51.991.51.994.24.21.01.491.01.492.12.10.05-0.990.05-0.990.920.9247六、六、ROCROC曲线曲线v“Receiver” or “Response” Operating Characteristic (ROC) curves，受试者工作特性曲线，受试者工作特性曲线，a helpful way of distinguishing

52、real signals for false noises in the early days of radar.vIf you plot the SEN versus (1-SPE) or “false alarms” that result from selecting different cutoffs for the diagnostic test results, you generate a useful picture of the tests accuracy that is called an “ROC curve.” ROC curves nicely display th

53、e trade-offs of using one or more cutoffs for the test. 4849vAn ROC curve has some useful properties:It illustrates the performance of a dichotomous diagnostic test when you select different cut-points to distinguish “normal” from “abnormal” results.It demonstrates the fact that any increase in sens

54、itivity will be accompanied by a decrease in specificity, and vice versa. The closer the curve gets to the upper left corner of the display, the more the overall accuracy of the test. That is, choosing the point labelled “BNP 76” correctly identifies 26 affected and 75 normal patients out of the tot

55、al of 126, or 80% overall accuracy. However, choosing the point labelled “BNP18 ” correctly identifies 35 affected but only 29 normal patients from the total of 126, which is only 51% overall accuracy. 50The closer the curve comes to the 45-degree diagonal of the ROC space, the less accurate the tes

56、t. At 45 degrees, the test adds no diagnostic information at all. Getting a bit fancier, the slope of the tangent at a cut-point gives the LR for that value of the test. Notice how much steeper the tangent is for the cutoff of 76 than it is for the cutoff of 18. The area under the curve provides an

57、overall measure of a tests accuracy. This property can help decide which of two competing tests for the same target disorder is the better one. 51ROCROC：描述诊断性试验优劣描述诊断性试验优劣确定确定cut-offcut-off值值比较不同诊断性试验比较不同诊断性试验52七、提高诊断性试验效率的方法：七、提高诊断性试验效率的方法：联合试验联合试验1.1.平行试验：同时做几个试验，只要有一个阳性，即平行试验：同时做几个试验，只要有一个阳性，即可认为

58、有患病证据。平行试验提高了可认为有患病证据。平行试验提高了敏感度和阴性敏感度和阴性预测值预测值，但降低了特异度及阳性预测值。，但降低了特异度及阳性预测值。SenSen=Sen1 + Sen2 - Sen1 X Sen2=Sen1 + Sen2 - Sen1 X Sen2SpeSpe=Spe1 X Spe2=Spe1 X Spe2验后比验后比= =验前比验前比 X LR1 X LR2X LR1 X LR253联合试验联合试验2.2.序列试验：依次相继的试验，要所有的试验阳性序列试验：依次相继的试验，要所有的试验阳性才能做出诊断。序列试验提高了才能做出诊断。序列试验提高了特异度及阳性预特异度及阳性

59、预测值测值。但降低了敏感度及阴性预测值。但降低了敏感度及阴性预测值。例如：诊断心肌梗死的例如：诊断心肌梗死的CPKCPK、ASTAST、LDHLDH，没有一，没有一种试验是很特异的，如采用序列试验，即三项均种试验是很特异的，如采用序列试验，即三项均阳性才能诊断，这样可提高诊断心肌梗死的特异阳性才能诊断，这样可提高诊断心肌梗死的特异度。度。SEN = SEN1 X SEN2SEN = SEN1 X SEN2SPE = SPE1 +SPE2 - SPE1 X SPE2SPE = SPE1 +SPE2 - SPE1 X SPE254七、诊断性试验的评价原则七、诊断性试验的评价原则v新的诊断性试验用于

60、临床之前或杂志上有关新的诊断性试验用于临床之前或杂志上有关诊断性试验的结论，均须经过科学的评价。诊断性试验的结论，均须经过科学的评价。1.1.是否采用盲目法将诊断性试验与标准诊断法是否采用盲目法将诊断性试验与标准诊断法（金标准）作过对比研究？（金标准）作过对比研究？J诊断性试验必须与金标准比较，才能确定是诊断性试验必须与金标准比较，才能确定是否可靠。否可靠。J盲法对比，更为科学。盲法对比，更为科学。J列出四格表进行分析对比，计算各项指标，列出四格表进行分析对比，计算各项指标，根据根据ACCACC、SENSEN、SPESPE确定诊断性试验有无应用确定诊断性试验有无应用价值。价值。552.2.被检