Department of Emergency Medicine, Department of Cellular and Integrative Physiology, Indiana University School of Medicine, Indianapolis, IN.
Department of Emergency Medicine, Department of Cellular and Integrative Physiology, Indiana University School of Medicine, Indianapolis, IN.
Ann Emerg Med. 2014 Mar;63(3):275-80. doi: 10.1016/j.annemergmed.2013.08.023. Epub 2013 Sep 23.
Pretest probability helps guide diagnostic testing for patients with suspected acute coronary syndrome and pulmonary embolism. Pretest probability derived from the clinician's unstructured gestalt estimate is easier and more readily available than methods that require computation. We compare the diagnostic accuracy of physician gestalt estimate for the pretest probability of acute coronary syndrome and pulmonary embolism with a validated, computerized method.
This was a secondary analysis of a prospectively collected, multicenter study. Patients (N=840) had chest pain, dyspnea, nondiagnostic ECGs, and no obvious diagnosis. Clinician gestalt pretest probability for both acute coronary syndrome and pulmonary embolism was assessed by visual analog scale and from the method of attribute matching using a Web-based computer program. Patients were followed for outcomes at 90 days.
Clinicians had significantly higher estimates than attribute matching for both acute coronary syndrome (17% versus 4%; P<.001, paired t test) and pulmonary embolism (12% versus 6%; P<.001). The 2 methods had poor correlation for both acute coronary syndrome (r(2)=0.15) and pulmonary embolism (r(2)=0.06). Areas under the receiver operating characteristic curve were lower for clinician estimate compared with the computerized method for acute coronary syndrome: 0.64 (95% confidence interval [CI] 0.51 to 0.77) for clinician gestalt versus 0.78 (95% CI 0.71 to 0.85) for attribute matching. For pulmonary embolism, these values were 0.81 (95% CI 0.79 to 0.92) for clinician gestalt and 0.84 (95% CI 0.76 to 0.93) for attribute matching.
Compared with a validated machine-based method, clinicians consistently overestimated pretest probability but on receiver operating curve analysis were as accurate for pulmonary embolism but not acute coronary syndrome.
在疑似急性冠脉综合征和肺栓塞患者中，预测概率有助于指导诊断性检查。基于临床医生的非结构化整体评估的预测概率比需要计算的方法更容易且更容易获得。我们比较了医生整体评估对急性冠脉综合征和肺栓塞的预测概率的诊断准确性，以及一种经过验证的计算机化方法。
这是一项前瞻性收集的多中心研究的二次分析。患者（N=840）有胸痛、呼吸困难、非诊断性心电图和没有明显诊断。临床医生通过视觉模拟量表和基于网络的计算机程序使用属性匹配方法对急性冠脉综合征和肺栓塞的整体预测概率进行评估。对患者进行 90 天的随访以评估结局。
临床医生对急性冠脉综合征（17%比 4%；P<.001，配对 t 检验）和肺栓塞（12%比 6%；P<.001）的估计值明显高于属性匹配。两种方法在急性冠脉综合征（r(2)=0.15）和肺栓塞（r(2)=0.06）方面相关性均较差。与计算机化方法相比，临床医生的评估方法对急性冠脉综合征的受试者工作特征曲线下面积较低：临床医生的整体评估为 0.64（95%置信区间 [CI] 0.51 至 0.77），而属性匹配为 0.78（95% CI 0.71 至 0.85）。对于肺栓塞，这些值分别为 0.81（95% CI 0.79 至 0.92）和 0.84（95% CI 0.76 至 0.93）。
与经过验证的基于机器的方法相比，临床医生始终高估了预测概率，但在接收者操作曲线分析中，对于肺栓塞的准确性与属性匹配相当，但对于急性冠脉综合征则不然。
