<!DOCTYPE article PUBLIC "-//NLM//DTD JATS (Z39.96) Journal Publishing DTD v1.1 20151215//EN" "https://jats.nlm.nih.gov/publishing/1.1/JATS-journalpublishing1.dtd">
<article xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" article-type="research-article" dtd-version="1.0" specific-use="sps-1.5" xml:lang="en">
  <front>
    <journal-meta>
      <journal-id journal-id-type="nlm-ta">Arq Bras Cardiol</journal-id>
      <journal-id journal-id-type="publisher-id">abc</journal-id>
      <journal-title-group>
        <journal-title>Arquivos Brasileiros de Cardiologia</journal-title>
        <abbrev-journal-title abbrev-type="publisher">Arq. Bras. Cardiol.</abbrev-journal-title>
      </journal-title-group>
      <issn pub-type="ppub">0066-782X</issn>
      <issn pub-type="epub">1678-4170</issn>
      <publisher>
        <publisher-name>Sociedade Brasileira de Cardiologia - SBC</publisher-name>
      </publisher>
    </journal-meta>
    <article-meta>
      <article-id pub-id-type="publisher-id" specific-use="scielo-v3">3MrkTJmcXTvJRGcPJ86ckQG</article-id>
      <article-id pub-id-type="publisher-id" specific-use="scielo-v2">S0066-782X2017000400304</article-id>
      <article-id pub-id-type="doi">10.5935/abc.20170037</article-id>
      <article-categories>
        <subj-group subj-group-type="heading">
          <subject>Original Articles</subject>
          <subj-group>
            <subject>Acute Coronary Artery Disease</subject>
          </subj-group>
        </subj-group>
      </article-categories>
      <title-group>
        <article-title>A Multivariate Model for Prediction of Obstructive Coronary Disease in Patients with Acute Chest Pain: Development and Validation</article-title>
      </title-group>
      <contrib-group>
        <contrib contrib-type="author">
          <name>
            <surname>Correia</surname>
            <given-names>Luis Cl&#225;udio Lemos</given-names>
          </name>
          <xref ref-type="aff" rid="aff1">1</xref>
          <xref ref-type="aff" rid="aff2">2</xref>
          <xref ref-type="corresp" rid="c1"/>
        </contrib>
        <contrib contrib-type="author">
          <name>
            <surname>Cerqueira</surname>
            <given-names>Maur&#237;cio</given-names>
          </name>
          <xref ref-type="aff" rid="aff1">1</xref>
        </contrib>
        <contrib contrib-type="author">
          <name>
            <surname>Carvalhal</surname>
            <given-names>Manuela</given-names>
          </name>
          <xref ref-type="aff" rid="aff1">1</xref>
        </contrib>
        <contrib contrib-type="author">
          <name>
            <surname>Ferreira</surname>
            <given-names>Felipe</given-names>
          </name>
          <xref ref-type="aff" rid="aff1">1</xref>
        </contrib>
        <contrib contrib-type="author">
          <name>
            <surname>Garcia</surname>
            <given-names>Guilherme</given-names>
          </name>
          <xref ref-type="aff" rid="aff2">2</xref>
        </contrib>
        <contrib contrib-type="author">
          <name>
            <surname>Silva</surname>
            <given-names>Andr&#233; Barcelos da</given-names>
          </name>
          <xref ref-type="aff" rid="aff1">1</xref>
        </contrib>
        <contrib contrib-type="author">
          <name>
            <surname>S&#225;</surname>
            <given-names>Nicole de</given-names>
          </name>
          <xref ref-type="aff" rid="aff1">1</xref>
        </contrib>
        <contrib contrib-type="author">
          <name>
            <surname>Lopes</surname>
            <given-names>Fernanda</given-names>
          </name>
          <xref ref-type="aff" rid="aff1">1</xref>
        </contrib>
        <contrib contrib-type="author">
          <name>
            <surname>Barcelos</surname>
            <given-names>Ana Clara</given-names>
          </name>
          <xref ref-type="aff" rid="aff1">1</xref>
        </contrib>
        <contrib contrib-type="author">
          <name>
            <surname>Noya-Rabelo</surname>
            <given-names>M&#225;rcia</given-names>
          </name>
          <xref ref-type="aff" rid="aff1">1</xref>
          <xref ref-type="aff" rid="aff2">2</xref>
        </contrib>
      </contrib-group>
      <aff id="aff1">
        <label>1</label>
        <institution content-type="orgname">Escola Bahiana de Medicina e Sa&#250;de P&#250;blica</institution>
        <addr-line>
          <named-content content-type="city">Salvador</named-content>
          <named-content content-type="state">BA</named-content>
        </addr-line>
        <country country="BR">Brazil</country>
        <institution content-type="original">Escola Bahiana de Medicina e Sa&#250;de P&#250;blica; Salvador, BA - Brazil</institution>
        <institution content-type="normalized">Escola Bahiana de Medicina e Sa&#250;de P&#250;blica</institution>
      </aff>
      <aff id="aff2">
        <label>2</label>
        <institution content-type="orgname">Hospital S&#227;o Rafael</institution>
        <addr-line>
          <named-content content-type="city">Salvador</named-content>
          <named-content content-type="state">BA</named-content>
        </addr-line>
        <country country="BR">Brazil</country>
        <institution content-type="original">Hospital S&#227;o Rafael, Salvador, BA - Brazil</institution>
      </aff>
      <author-notes>
        <corresp id="c1"><bold>Mailing Address: Lu&#237;s Cl&#225;udio Lemos Correia</bold>, Av. Princesa Leopoldina, 19/402. Postal Code: 40150-080, Gra&#231;a, Salvador, BA - Brazil. E-mail: <email>lccorreia@cardiol.br</email>; <email>lccorreia@terra.com.br</email>
				</corresp>
        <fn fn-type="con">
          <p>
            <bold>Author contributions</bold>
          </p>
          <p>Conception and design of the research: Correia LCL, Cerqueira M, Carvalhal M, Ferreira F, Garcia G, Silva AB, S&#225; N, Lopes F, Barcelos AC, Noya-Rabelo M; Acquisition of data: Cerqueira M, Carvalhal M, Ferreira F, Silva AB, S&#225; N, Lopes F, Barcelos AC, Noya-Rabelo M; Analysis and interpretation of the data: Correia LCL, Cerqueira M, Garcia G, Silva AB, S&#225; N, Lopes F, Barcelos AC, Noya-Rabelo M; Statistical analysis: Correia LCL, Cerqueira M, Ferreira F, Garcia G, Noya-Rabelo M; Writing of the manuscript: Correia LCL, Cerqueira M, Carvalhal M, Ferreira F, Silva AB, S&#225; N, Lopes F, Barcelos AC, Noya-Rabelo M; Critical revision of the manuscript for intellectual contente: Correia LCL, Cerqueira M, Carvalhal M, Garcia G, Silva AB, S&#225; N, Lopes F, Barcelos AC, Noya-Rabelo M.</p>
        </fn>
        <fn fn-type="conflict">
          <p>
            <bold>Potential Conflict of Interest</bold>
          </p>
          <p>No potential conflict of interest relevant to this article was reported.</p>
        </fn>
      </author-notes>
      <pub-date pub-type="epub-ppub">
        <month>04</month>
        <year>2017</year>
      </pub-date>
      <volume>108</volume>
      <issue>4</issue>
      <fpage>304</fpage>
      <lpage>314</lpage>
      <history>
        <date date-type="received">
          <day>14</day>
          <month>06</month>
          <year>2016</year>
        </date>
        <date date-type="rev-recd">
          <day>08</day>
          <month>08</month>
          <year>2016</year>
        </date>
        <date date-type="accepted">
          <day>04</day>
          <month>11</month>
          <year>2016</year>
        </date>
      </history>
      <permissions>
        <license license-type="open-access" xlink:href="http://creativecommons.org/licenses/by/4.0/" xml:lang="en">
          <license-p>This is an Open Access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.</license-p>
        </license>
      </permissions>
      <abstract>
        <title>Abstract</title>
        <sec>
          <title>Background:</title>
          <p>Currently, there is no validated multivariate model to predict probability of obstructive coronary disease in patients with acute chest pain.</p>
        </sec>
        <sec>
          <title>Objective:</title>
          <p>To develop and validate a multivariate model to predict coronary artery disease (CAD) based on variables assessed at admission to the coronary care unit (CCU) due to acute chest pain.</p>
        </sec>
        <sec>
          <title>Methods:</title>
          <p>A total of 470 patients were studied, 370 utilized as the derivation sample and the subsequent 100 patients as the validation sample. As the reference standard, angiography was required to rule in CAD (stenosis &#8805; 70%), while either angiography or a negative noninvasive test could be used to rule it out. As predictors, 13 baseline variables related to medical history, 14 characteristics of chest discomfort, and eight variables from physical examination or laboratory tests were tested.</p>
        </sec>
        <sec>
          <title>Results:</title>
          <p>The prevalence of CAD was 48%. By logistic regression, six variables remained independent predictors of CAD: age, male gender, relief with nitrate, signs of heart failure, positive electrocardiogram, and troponin. The area under the curve (AUC) of this final model was 0.80 (95% confidence interval [95%CI] = 0.75 - 0.84) in the derivation sample and 0.86 (95%CI = 0.79 - 0.93) in the validation sample. Hosmer-Lemeshow's test indicated good calibration in both samples (p = 0.98 and p = 0.23, respectively). Compared with a basic model containing electrocardiogram and troponin, the full model provided an AUC increment of 0.07 in both derivation (p = 0.0002) and validation (p = 0.039) samples. Integrated discrimination improvement was 0.09 in both derivation (p &lt; 0.001) and validation (p &lt; 0.0015) samples.</p>
        </sec>
        <sec>
          <title>Conclusion:</title>
          <p>A multivariate model was derived and validated as an accurate tool for estimating the pretest probability of CAD in patients with acute chest pain.</p>
        </sec>
      </abstract>
      <kwd-group xml:lang="en">
        <title>Keywords:</title>
        <kwd>CoronaryArtery Disease</kwd>
        <kwd>Methods</kwd>
        <kwd>Chest Pain</kwd>
        <kwd>Models Statistical</kwd>
        <kwd>Coronary Angiography</kwd>
        <kwd>Troponin</kwd>
        <kwd>Electrocardiography</kwd>
      </kwd-group>
    </article-meta>
  </front>
  <body>
    <sec sec-type="intro">
      <title>Introduction</title>
      <p>Acute chest pain is one of the most common reasons for emergency department visits. Since it may represent a clinical manifestation of cardiac ischemia, patient discharge is normally conditioned to a negative test for obstructive coronary artery disease (CAD).<xref ref-type="bibr" rid="B1">1</xref> However, the efficiency of this defensive strategy is challenged by a low yield of cardiac tests, since only a portion of patients ends up having obstructive CAD and a smaller part will need revascularization.<xref ref-type="bibr" rid="B2">2</xref> In addition, routine testing is not supported by evidence of beneficial effect<xref ref-type="bibr" rid="B3">3</xref> and may have unintentional consequences: overdiagnosis and overtreatment of coronary disease not causally related to symptoms, prolonged hospital stay, unnecessary invasive procedures due to false-positive test results, and increased medical expenses.<xref ref-type="bibr" rid="B4">4</xref>
			</p>
      <p>Therefore, a more rational approach is to indicate additional tests on the basis of pretest probability. Traditionally, this pretest evaluation is restricted to electrocardiogram and necrosis markers. However, the use of a multivariate model has the potential to improve accuracy and provide a more continuous range of probabilities. In order to develop and validate a multivariate model to predict CAD based on variables assessed at admission to the coronary care unit, 370 consecutive patients were studied. Thirty-five variables were tested as candidate predictors of obstructive CAD in order to generate a final model that was further validated in a subsequent sample of 100 patients.</p>
    </sec>
    <sec sec-type="methods">
      <title>Methods</title>
      <sec>
        <title>Sample selection</title>
        <p>During a period of 30 consecutive months, all patients admitted to the coronary care unit of our hospital were included in the study. Admission took place whenever medical judgment recognized any chance of a coronary etiology, regardless of electrocardiogram or troponin. The only exclusion criterion was the patient's decline to participate. As defined <italic>a priori,</italic> the first 370 patients were utilized as the derivation sample and the next 100 patients as the validation sample. The study was approved by an institutional review committee, and all the subjects gave informed consent to participate.</p>
      </sec>
      <sec>
        <title>Predictors of obstructive CAD</title>
        <p>At baseline admission, three sets of variables were recorded as candidates for prediction of obstructive CAD. The first comprised 13 variables related to medical history, such as age, gender, previous history of CAD, risk factors for CAD, and comorbidities; the second set included 14 characteristics of chest discomfort; and the third set was composed of eight variables related to either physical examination or basic admission tests, including physical and radiologic signs of left heart failure, ischemic electrocardiographic changes (T wave inversion &#8805; 1 mm or dynamic ST deviation &#8805; 0.5 mm), positive troponin (&gt; 99th percentile of the general population; Ortho-Clinical Diagnostics, Rochester, NY, USA), N-terminal pro-B-type natriuretic peptide (NT-proBNP, enzyme-linked fluorescent assay, Biom&#233;rieux, France), high-sensitivity C-reactive protein (CRP; nephelometry, Dade-Behring, USA), white cell count, plasma glucose, and hemoglobin. Laboratory tests were performed in plasma material collected at presentation to the emergency room. The medical history and chest pain characteristics were recorded by three investigators (M.C., A.M.C., and R.B.) trained to interview the patients in a systematic form, in order to decrease bias and improve reproducibility. Radiologic signs of ventricular failure and electrocardiogram were all interpreted by the same senior investigator (L.C.).</p>
      </sec>
      <sec>
        <title>Outcomes definition</title>
        <p>The primary outcome to be predicted by the model was a diagnosis of obstructive CAD, defined by subsequent tests performed during hospital stay. The outcome data were collected by three investigators (M.C., A.M.C., and R.B.) and adjudicated by a fourth investigator (L.C.). For diagnostic evaluation, the patients underwent invasive coronary angiography or a provocative noninvasive test (perfusion magnetic resonance imaging, nuclear single-photon emission computed tomography or stress-echocardiography with dobutamine), at the discretion of the assistant cardiologist. In the case of a positive noninvasive test, the patients had angiography for confirmation. Based on this diagnostic algorithm, obstructive CAD was defined as a &#8805; 70% stenosis on angiography. A normal noninvasive test (ischemic defect size &lt; 5% of the left ventricular myocardium) indicated the absence of obstructive CAD and no further test was required. Regardless of coronary tests, the patients were classified as presenting no obstructive CAD if one of the following dominant diagnoses was confirmed by image: pericarditis, pulmonary embolism, aortic dissection or pneumonia. Secondarily, the model was tested for the prediction of death within 30 days of admission.</p>
      </sec>
      <sec>
        <title>Statistical analysis</title>
        <p>The statistical analysis is depicted in <xref ref-type="fig" rid="f1">Figure 1</xref>. The initial sample of 370 consecutive patients was utilized for the derivation of the model. First, univariate associations between obstructive CAD and baseline characteristics were tested by unpaired Student's <italic>t</italic> test for numeric variables and Pearson's chi-square test for categorical variables. Numeric variables not normally distributed were expressed as median and interquartile range and compared by nonparametric Mann-Whitney's test. Second, variables with a p value &lt; 0.10 in the univariate analysis were included in the multivariate logistic regression analysis for prediction of obstructive CAD.</p>
        <p>
          <fig id="f1">
            <label>Figure 1</label>
            <caption>
              <title>Flowchart of the statistical analysis.</title>
            </caption>
            <graphic xlink:href="1678-4170-abc-108-04-0304-70037-gf01.jpg"/></fig>
        </p>
        <p>Multivariate models were developed by the stepwise method, forcing all selected variables into the regression and eliminating the least significant at each step, according to Wald's statistical test. Initially, three intermediate models were built, according to the type of predictive variables (medical history, chest pain characteristics or physical examination/laboratory tests). Independent predictors (p &lt; 0.05) in each intermediate model were included as covariates in the final model. This final model was built hierarchically, with the order of variable imputation defined by clinical reasoning. The improvement of the model at each step was described by the decrease in -2Log likelihood.</p>
        <p>Discrimination was evaluated by the area under the receiver operating characteristic (ROC) curve (AUC), while calibration was assessed by Hosmer-Lemeshow's test and correlation between predictive and observed prevalence of disease according to deciles of prediction. The incremental value of the full model in relation to the most basic model was evaluated by comparing the two AUCs by DeLong's test. In addition, <italic>integrated discrimination improvement</italic> by the full model was described according to Pencina's method.<xref ref-type="bibr" rid="B5">5</xref>
				</p>
        <p>Subsequently, 100 consecutive patients served as the validation sample. In this sample, discrimination of CAD was tested by the AUC. Since calibration analysis by deciles would not be appropriate in a sample of 100 patients, observed CAD prevalence was compared among tertiles of CAD prediction. The incremental value of the full model in relation to the most basic model was evaluated by comparing the two AUC by DeLong's test. <italic>Integrated discrimination improvement</italic> by the full model was also described in this sample.</p>
        <p>In a sensitivity analysis, the full sample of 470 patients was used to test whether the performance of the model changed according to the presence or absence of electrocardiographic or troponin changes. For this analysis, an interaction term was tested by logistic regression. The full sample was also used to test the prognostic value of the model. The AUC for 30-day mortality prediction was described and compared with the GRACE score<xref ref-type="bibr" rid="B6">6</xref> as a proxy of a model specifically created for a prognostic purpose. DeLong's test was used to compare the AUCs.</p>
        <p>Statistical significance was defined as alpha &lt; 0.05. For numerical variables with normal distribution, mean and standard deviation was used, while a non-normal distribution implied in the use of median and interquartile range. SPSS, version 21.0, was the software used for statistical analysis.</p>
      </sec>
      <sec>
        <title>Acute chest pain score</title>
        <p>In order to generate a score for CAD prediction, points were attributed to each positive variable, proportional to their regression coefficients in the final model. The prevalence of obstructive CAD was described according to score's deciles. Alternatively, the final regression formula was used to create a logistic calculator, provided as an Excel spreadsheet (electronic file) or application for smartphones (to be available in the near future).</p>
      </sec>
      <sec>
        <title>Sample size determination</title>
        <p>As described above, two consecutive samples of patients were selected: the derivation set and the validation set. For the derivation set, the sample size was planned to allow inclusion of at least 10 covariates in the logistic regression model. The calculation was based on the following assumptions: 30% prevalence of obstructive CAD and the need for 10 events for each covariate in the logistic regression model.<xref ref-type="bibr" rid="B7">7</xref> Therefore, a minimum of 300 patients would be required, and as a safety precaution, we planned to include a total of 370 individuals. The validation sample was set to test the discriminatory accuracy by the ROC curve analysis. Based on the assumption of an AUC of 0.70, to provide 90% power to reject the null hypothesis of an AUC equal 0.50, under the alpha of 5%, a minimum of 85 patients was required. Therefore, we planned to include 100 patients in the validation set.</p>
      </sec>
    </sec>
    <sec sec-type="results">
      <title>Results</title>
      <sec>
        <title>Sample population for model derivation</title>
        <p>In total, 370 patients were studied, aged 60 &#177; 16 years, 57% males, 33% with a previous history of coronary disease. The median time elapsed between the onset of symptoms and first clinical evaluation in the hospital was 4 hours (interquartile range = 1.8 - 13 hours). At presentation, 52% of the patients had ischemic changes on the electrocardiogram, and 48% had positive troponin. Further investigation according to study protocol identified obstructive CAD in 176 patients, a prevalence of 48%. All cases had diagnostic confirmation by invasive coronary angiography. Regarding the 194 patients without CAD, 74 were classified by a negative angiography, 105 by a negative noninvasive test and 15 had another dominant diagnosis (four with pulmonary embolism, two with aortic dissection, seven with pericarditis, and two with pneumonia).</p>
      </sec>
      <sec>
        <title>Predictors of obstructive CAD</title>
        <p>Among the 13 variables related to medical history, only four were associated with obstructive CAD: older age, higher prevalence of male gender, previous history of CAD, and a trend towards more diabetes (<xref ref-type="table" rid="t1">Table 1</xref>). When these four variables were included in the logistic regression, age and male gender remained statistically significant (Intermediate Model 1) (<xref ref-type="table" rid="t2">Table 2</xref>).</p>
        <table-wrap id="t1">
          <label>Table 1</label>
          <caption>
            <title>Comparison of medical history, chest pain characteristics, and laboratory tests between patients with and without obstructive coronary artery disease</title>
          </caption>
          <table frame="hsides" rules="groups">
            <colgroup>
              <col width="25%"/>
              <col width="25%"/>
              <col width="25%"/>
              <col width="25%"/>
            </colgroup>
            <thead>
              <tr style="border-bottom-width:thin;border-bottom-style:solid">
                <th align="left">
								</th>
                <th align="center" colspan="2">Obstructive Coronary Disease</th>
                <th align="center">p Value</th>
              </tr>
              <tr>
                <th align="left">
								</th>
                <th align="center">Yes (n = 176)</th>
                <th align="center">No (n = 194)</th>
                <th align="center">
								</th>
              </tr>
            </thead>
            <tbody>
              <tr>
                <td align="left" colspan="4">
                  <bold>Medical History</bold>
                </td>
              </tr>
              <tr>
                <td align="left">Age (years)</td>
                <td align="center">63 &#177; 14</td>
                <td align="center">57 &#177; 16</td>
                <td align="center">&lt; 0.001</td>
              </tr>
              <tr>
                <td align="left">Male gender</td>
                <td align="center">121 (69%)</td>
                <td align="center">90 (46%)</td>
                <td align="center">&lt; 0.001</td>
              </tr>
              <tr>
                <td align="left">Body mass index (kg/m2)</td>
                <td align="center">28 &#177; 4.8</td>
                <td align="center">28 &#177; 5.9</td>
                <td align="center">0.61</td>
              </tr>
              <tr>
                <td align="left">History of CAD</td>
                <td align="center">68 (39%)</td>
                <td align="center">55 (28%)</td>
                <td align="center">0.03</td>
              </tr>
              <tr>
                <td align="left">Diabetes</td>
                <td align="center">62 (36%)</td>
                <td align="center">51 (26%)</td>
                <td align="center">0.05</td>
              </tr>
              <tr>
                <td align="left">Hypertension</td>
                <td align="center">122 (70%)</td>
                <td align="center">138 (71%)</td>
                <td align="center">0.83</td>
              </tr>
              <tr>
                <td align="left">Current smoking</td>
                <td align="center">22 (13%)</td>
                <td align="center">18 (9.3%)</td>
                <td align="center">0.30</td>
              </tr>
              <tr>
                <td align="left">LDL cholesterol (mg/dL)</td>
                <td align="center">113 &#177; 64</td>
                <td align="center">116 &#177; 87</td>
                <td align="center">0.72</td>
              </tr>
              <tr>
                <td align="left">Family history of CAD</td>
                <td align="center">48 (28%)</td>
                <td align="center">42 (22%)</td>
                <td align="center">0.19</td>
              </tr>
              <tr>
                <td align="left">Chronic renal disease</td>
                <td align="center">9 (5.3%)</td>
                <td align="center">7 (3.6%)</td>
                <td align="center">0.45</td>
              </tr>
              <tr>
                <td align="left">Plasma creatinine (mg/dL)</td>
                <td align="center">0.95 (0.80 - 1.20)</td>
                <td align="center">0.80 (0.70 - 1.15)</td>
                <td align="center">0.10</td>
              </tr>
              <tr>
                <td align="left">Current statin therapy</td>
                <td align="center">85 (49%)</td>
                <td align="center">91 (47%)</td>
                <td align="center">0.71</td>
              </tr>
              <tr>
                <td align="left">Current aspirin therapy</td>
                <td align="center">75 (43%)</td>
                <td align="center">76 (39%)</td>
                <td align="center">0.44</td>
              </tr>
              <tr>
                <td align="left" colspan="4">
                  <bold>Chest Pain Characteristics</bold>
                </td>
              </tr>
              <tr>
                <td align="left">Left side location</td>
                <td align="center">137 (79%)</td>
                <td align="center">156 (81%)</td>
                <td align="center">0.70</td>
              </tr>
              <tr>
                <td align="left">Oppressive nature</td>
                <td align="center">97 (57%)</td>
                <td align="center">95 (49%)</td>
                <td align="center">0.14</td>
              </tr>
              <tr>
                <td align="left">Irradiation to neck</td>
                <td align="center">39 (23%)</td>
                <td align="center">51 (26%)</td>
                <td align="center">0.42</td>
              </tr>
              <tr>
                <td align="left">Irradiation to left arm</td>
                <td align="center">57 (33%)</td>
                <td align="center">53 (27%)</td>
                <td align="center">0.24</td>
              </tr>
              <tr>
                <td align="left">Vagal symptoms</td>
                <td align="center">61 (36%)</td>
                <td align="center">78 (40%)</td>
                <td align="center">0.35</td>
              </tr>
              <tr>
                <td align="left">Number of episodes</td>
                <td align="center">1 (1 - 2)</td>
                <td align="center">1 (1 - 3)</td>
                <td align="center">0.81</td>
              </tr>
              <tr>
                <td align="left">Duration (minutes)</td>
                <td align="center">40 (15 - 120)</td>
                <td align="center">40 (10 - 150)</td>
                <td align="center">0.82</td>
              </tr>
              <tr>
                <td align="left">Intensity (1 - 10 scale)</td>
                <td align="center">7.4 &#177; 2.5</td>
                <td align="center">7.1 &#177; 2.6</td>
                <td align="center">0.31</td>
              </tr>
              <tr>
                <td align="left">Relief with nitrate</td>
                <td align="center">84 (50%)</td>
                <td align="center">72 (37%)</td>
                <td align="center">0.02</td>
              </tr>
              <tr>
                <td align="left">Similar to previous infarction</td>
                <td align="center">70 (42%)</td>
                <td align="center">63 (33%)</td>
                <td align="center">0.08</td>
              </tr>
              <tr>
                <td align="left">Worsening with compression</td>
                <td align="center">7 (4.1%)</td>
                <td align="center">26 (13%)</td>
                <td align="center">0.002</td>
              </tr>
              <tr>
                <td align="left">Worsening with position</td>
                <td align="center">24 (14%)</td>
                <td align="center">36 (19%)</td>
                <td align="center">0.23</td>
              </tr>
              <tr>
                <td align="left">Worsening with arm movement</td>
                <td align="center">7 (4.0%)</td>
                <td align="center">16 (8.2%)</td>
                <td align="center">0.097</td>
              </tr>
              <tr>
                <td align="left">Worsening with deep breath</td>
                <td align="center">13 (7.5%)</td>
                <td align="center">36 (19%)</td>
                <td align="center">0.002</td>
              </tr>
              <tr>
                <td align="left" colspan="4">
                  <bold>Laboratory Tests at Admission</bold>
                </td>
              </tr>
              <tr>
                <td align="left">Ischemic changes on ECG</td>
                <td align="center">120 (68%)</td>
                <td align="center">73 (38%)</td>
                <td align="center">&lt; 0.001</td>
              </tr>
              <tr>
                <td align="left">Positive troponin</td>
                <td align="center">116 (66%)</td>
                <td align="center">60 (31%)</td>
                <td align="center">&lt; 0.001</td>
              </tr>
              <tr>
                <td align="left">X-ray and clinical signs of LVF</td>
                <td align="center">26 (15%)</td>
                <td align="center">5 (2.6%)</td>
                <td align="center">&lt; 0.001</td>
              </tr>
              <tr>
                <td align="left">NT-proBNP (pg/mL)</td>
                <td align="center">363 (105 - 1850)</td>
                <td align="center">57 (20 - 235)</td>
                <td align="center">&lt; 0.001</td>
              </tr>
              <tr>
                <td align="left">Plasma glucose (mg/dL)</td>
                <td align="center">120 (97 - 189)</td>
                <td align="center">112 (92 - 145)</td>
                <td align="center">0.22</td>
              </tr>
              <tr>
                <td align="left">C-reactive protein (mg/L)</td>
                <td align="center">7.3 (2.3 - 15)</td>
                <td align="center">5.7 (1.4 - 15)</td>
                <td align="center">0.09</td>
              </tr>
              <tr>
                <td align="left">White cell count</td>
                <td align="center">8.790 &#177; 4.300</td>
                <td align="center">7.701 &#177; 2.865</td>
                <td align="center">0.004</td>
              </tr>
              <tr>
                <td align="left">Hemoglobin (g/dL)</td>
                <td align="center">14.1 &#177; 1.9</td>
                <td align="center">13.7 &#177; 1.7</td>
                <td align="center">0.06</td>
              </tr>
            </tbody>
          </table>
          <table-wrap-foot>
            <fn id="TFN1">
              <p>CAD: coronary artery disease; LVF: left ventricular failure. A family history of CAD implies in the presentation of the disease in a first-degree relative before the age of 55 years (females) or 45 years (males).</p>
            </fn>
          </table-wrap-foot>
        </table-wrap>
        <table-wrap id="t2">
          <label>Table 2</label>
          <caption>
            <title>Intermediates logistic regression models of medical history (Model 1), chest pain characteristics (Model 2) and laboratory tests (Model 3)</title>
          </caption>
          <table frame="hsides" rules="groups">
            <colgroup>
              <col width="50%"/>
              <col width="50%"/>
            </colgroup>
            <thead>
              <tr>
                <th align="left">Variables</th>
                <th align="center">Multivariate significance level</th>
              </tr>
            </thead>
            <tbody>
              <tr>
                <td align="left" colspan="2">
                  <bold>Model 1 (medical history)</bold>
                </td>
              </tr>
              <tr>
                <td align="left">Male gender</td>
                <td align="center">&lt; 0.001</td>
              </tr>
              <tr>
                <td align="left">Age (years)</td>
                <td align="center">&lt; 0.001</td>
              </tr>
              <tr>
                <td align="left">Diabetes</td>
                <td align="center">0.10</td>
              </tr>
              <tr>
                <td align="left">HDL cholesterol</td>
                <td align="center">0.35</td>
              </tr>
              <tr>
                <td align="left">Previous CAD</td>
                <td align="center">0.84</td>
              </tr>
              <tr>
                <td align="left">Plasma creatinine (mg/dL)</td>
                <td align="center">0.95</td>
              </tr>
              <tr>
                <td align="left" colspan="2">
                  <bold>Model 2 (pain characteristics)</bold>
                </td>
              </tr>
              <tr>
                <td align="left">Sensible to manual compression</td>
                <td align="center">0.024</td>
              </tr>
              <tr>
                <td align="left">Sensible to deep breath</td>
                <td align="center">0.037</td>
              </tr>
              <tr>
                <td align="left">Relief with nitrate</td>
                <td align="center">0.045</td>
              </tr>
              <tr>
                <td align="left">Similar to a previous MI</td>
                <td align="center">0.17</td>
              </tr>
              <tr>
                <td align="left">Sensible to arm movement</td>
                <td align="center">0.57</td>
              </tr>
              <tr>
                <td align="left" colspan="2">
                  <bold>Model 3 (laboratory tests)</bold>
                </td>
              </tr>
              <tr>
                <td align="left">Ischemic changes on ECG</td>
                <td align="center">&lt; 0.001</td>
              </tr>
              <tr>
                <td align="left">Positive troponin</td>
                <td align="center">&lt; 0.001</td>
              </tr>
              <tr>
                <td align="left">X-ray or clinical signs of LVF</td>
                <td align="center">0.016</td>
              </tr>
              <tr>
                <td align="left">White cell count</td>
                <td align="center">0.29</td>
              </tr>
              <tr>
                <td align="left">Hemoglobin (g/dL)</td>
                <td align="center">0.67</td>
              </tr>
              <tr>
                <td align="left">NT-proBNP (pg/mL)</td>
                <td align="center">0.81</td>
              </tr>
              <tr>
                <td align="left">C-reactive protein (mg/L)</td>
                <td align="center">0.70</td>
              </tr>
            </tbody>
          </table>
          <table-wrap-foot>
            <fn id="TFN2">
              <p>MI: myocardial infarction; CAD: coronary artery disease; LVF: left ventricular failure.</p>
            </fn>
          </table-wrap-foot>
        </table-wrap>
        <p>Regarding chest pain characteristics, among 14 variables, only five had an association with CAD: relief with nitrates and similarity with previous myocardial infarction. On the other hand, worsening with manual compression, deep breath or arm movement were each more common in patients without CAD (<xref ref-type="table" rid="t1">Table 1</xref>). Of these, relief with nitrates, worsening with manual compression and with deep breath were the three independent predictors in the Intermediate Model 2 (<xref ref-type="table" rid="t2">Table 2</xref>).</p>
        <p>Among the physical examination and laboratory tests, most variables were associated with CAD: ischemic electrocardiogram, positive troponin, and signs of left heart failure were more prevalent in patients with CAD. Also, four numeric variables had higher values in patients with CAD: NT-proBNP, CRP, white cell count, and hemoglobin (<xref ref-type="table" rid="t1">Table 1</xref>). In the Intermediate Model 3, ischemic electrocardiogram, positive troponin, and signs of left heart failure were the independent predictors (<xref ref-type="table" rid="t2">Table 2</xref>).</p>
      </sec>
      <sec>
        <title>Development of a model for CAD prediction</title>
        <p>The eight variables independently associated with CAD in the Intermediate Models 1, 2, and 3 were candidates to the final model, which was built hierarchically in seven steps, defined by clinical reasoning: the first step comprised electrocardiogram and troponin together, followed by the second step that included left ventricular failure. These two first steps represented the severity of the clinical presentation. The third and fourth steps represented intrinsic characteristics of the patients, age, and gender. The fifth, sixth, and seventh steps were related to characteristics of chest pain, which were chosen to be last because of their subjectivity in clinical practice.</p>
        <p>The first step of electrocardiogram and troponin had a -2Log likelihood of 437 (&#967;<sup>2</sup> = 69, p &lt; 0.001), which sequentially improved by the inclusion of left ventricular failure (-2Log likelihood = 427, &#967;<sup>2</sup> = 9.8, p = 0.002), age (-2Log likelihood = 422, &#967;<sup>2</sup> = 4.9, p = 0.02), gender (-2Log likelihood = 401, &#967;<sup>2</sup> = 21, p &lt; 0.001), and relief with nitrates (-2Log likelihood = 394, &#967;<sup>2</sup> = 6.8, p = 0.009). The inclusion of worsening with manual compression (-2 log likelihood = 391, &#967;<sup>2</sup> = 3.2, p = 0.07) and worsening with deep breath (-2 log likelihood = 389, &#967;<sup>2</sup> = 2.3, p = 0.13) did not promote further improvement in the model. Therefore, the first six variables constituted the final model.</p>
        <p>The final model presented good discrimination, with an AUC of 0.80 (95%CI = 0.75 - 0.84) (<xref ref-type="fig" rid="f2">Figure 2A</xref>). The Hosmer-Lemeshow's &#967;<sup>2</sup> of 1.95 indicated that the model was well calibrated (p = 0.98), as shown in the scatter plot of predictive probability <italic>versus</italic> observed prevalence of CAD by deciles (r = 0.99) (<xref ref-type="fig" rid="f2">Figure 2B</xref>). The probability of CAD according to the final model ranged from a minimum of 3% to a maximum of 98%, with patients equally distributed across probabilities. Odds ratio, 95%CIs and regression coefficients, and p values of the final model are depicted in <xref ref-type="table" rid="t3">Table 3</xref>.</p>
        <p>
          <fig id="f2">
            <label>Figure 2 </label>
            <caption>
              <title>Analysis of the model's discrimination and calibration in the derivation sample of 370 patients. Panel A shows significant AUC of the probabilistic model for prediction of obstructive coronary artery disease. Panel B shows a significant correlation between predicted and observed probability of coronary artery disease (CAD). AUC denotes area under the receiver operating characteristic curve.</title>
            </caption>
            <graphic xlink:href="1678-4170-abc-108-04-0304-70037-gf02.jpg"/></fig>
        </p>
        <table-wrap id="t3">
          <label>Table 3</label>
          <caption>
            <title>Final model of logistic regression defining the independent predictors of obstructive coronary artery disease</title>
          </caption>
          <table frame="hsides" rules="groups">
            <colgroup>
              <col width="25%"/>
              <col width="25%"/>
              <col width="25%"/>
              <col width="25%"/>
            </colgroup>
            <thead>
              <tr>
                <th align="left">Variables</th>
                <th align="center">Beta</th>
                <th align="center">Odds Ratio (95%CI)</th>
                <th align="center">p Value</th>
              </tr>
            </thead>
            <tbody>
              <tr>
                <td align="left">Age (each year)</td>
                <td align="center">0.025</td>
                <td align="center">1.03 (1.01 - 1.04)</td>
                <td align="center">0.003</td>
              </tr>
              <tr>
                <td align="left">Relief with nitrates</td>
                <td align="center">0.60</td>
                <td align="center">1.8 (1.1 - 3.0)</td>
                <td align="center">0.016</td>
              </tr>
              <tr>
                <td align="left">Ischemic ECG</td>
                <td align="center">1.10</td>
                <td align="center">3.0 (1.9 - 4.9)</td>
                <td align="center">&lt; 0.001</td>
              </tr>
              <tr>
                <td align="left">Positive troponin</td>
                <td align="center">1.15</td>
                <td align="center">3.2 (1.9 - 5.1)</td>
                <td align="center">&lt; 0.001</td>
              </tr>
              <tr>
                <td align="left">Male gender</td>
                <td align="center">1.16</td>
                <td align="center">3.2 (1.9 - 5.3)</td>
                <td align="center">&lt; 0.001</td>
              </tr>
              <tr>
                <td align="left">Signs of LVF</td>
                <td align="center">1.55</td>
                <td align="center">4.7 (1.6 - 14)</td>
                <td align="center">0.004</td>
              </tr>
              <tr>
                <td align="left">Sensible to deep breath</td>
                <td align="center">----</td>
                <td align="center">----</td>
                <td align="center">0.06</td>
              </tr>
              <tr>
                <td align="left">Sensible to manual compression</td>
                <td align="center">----</td>
                <td align="center">----</td>
                <td align="center">0.18</td>
              </tr>
            </tbody>
          </table>
          <table-wrap-foot>
            <fn id="TFN3">
              <p>LVF: left ventricular failure.</p>
            </fn>
          </table-wrap-foot>
        </table-wrap>
      </sec>
      <sec>
        <title>Incremental value of the full model</title>
        <p>The AUC improved from 0.73 in the first model containing only electrocardiogram and troponin to 0.80 in the full model (95%CI of difference between the areas = 0.03 - 0.10, p = 0.0002). Discrimination progressively improved as variables were added: the AUC was 0.74 in the second model (adding left ventricular failure), 0.76 in the third model (adding age), and 0.79 in the fourth model (adding gender). The integrated discrimination improvement provided by the full model in relation to the first model was 0.09 (p &lt; 0.001), a result of 0.05 of mean increase of probabilities in the group with events plus 0.04 of mean decrease of probabilities in the group free of events.</p>
      </sec>
      <sec>
        <title>Validation by the independent sample</title>
        <p>The validation sample consisted of 100 individuals, 62% males, aged 60 &#177; 13 years, with a 59% prevalence of obstructive CAD. In this group, the AUC was 0.86 (95%CI = 0.79 - 0.93) and Hosmer-Lemeshow's calibration &#967;<sup>2</sup> was 10.1 (p = 0.26) (<xref ref-type="fig" rid="f3">Figure 3A</xref>). As the group was divided into tertiles of model's predicted probability (&lt; 30%, 30 - 60%, &gt; 60%), a progressive increase in disease prevalence was observed (24%, 59%, and 94%, respectively, p for trend &lt; 0.001) (<xref ref-type="fig" rid="f3">Figure 3B</xref>).</p>
        <p>
          <fig id="f3">
            <label>Figure 3</label>
            <caption>
              <title>Analysis of the model's performance in the independent validation sample of 100 patients. Panel A shows a significant AUC of the probabilistic model for prediction of obstructive coronary artery disease (CAD). Panel B indicates a progressive increase in the prevalence of CAD according to tertiles of the model's prediction. AUC denotes area under the receiver operating characteristic curve.</title>
            </caption>
            <graphic xlink:href="1678-4170-abc-108-04-0304-70037-gf03.jpg"/></fig>
        </p>
        <p>Compared with the basic model containing only electrocardiogram and troponin (AUC = 0.78), the increment provided by the full model was +0.07 (95%CI of difference between the areas = 0.004 - 0.14, p = 0.039). The integrated discrimination improvement provided by the full model in relation to the first model was 0.09 (p &lt; 0.0015), a result of 0.02 of mean increase of probabilities in the group with events plus 0.07 of mean decrease of probabilities in the group free of events.</p>
      </sec>
      <sec>
        <title>Sensitivity of the final model to electrocardiogram and troponin</title>
        <p>The entire sample of 470 patients was utilized to test the model's sensitivity to electrocardiogram and troponin. There was no interaction between the model's prediction and presence (or absence) of electrocardiographic/troponin changes (p = 0.48), meaning that the performance of the model was not modified by these variables. The model's AUC of individuals with normal electrocardiogram and troponin (n = 147, 24% with CAD) was 0.74 (95%CI = 0.65 - 0.83), while individuals with either one abnormal (n = 323, 62% of CAD) had an AUC of 0.77 (95%CI = 0.71 - 0.82).</p>
      </sec>
      <sec>
        <title>Prognostic value for 30-day mortality</title>
        <p>In the entire sample of 470 patients, 10 patients (2.1%) died within the first 30 days from initial chest pain, eight during hospitalization and two after discharge. The ability of the model to predict death was shown by an AUC of 0.74 (95%CI = 0.61 - 0.87), similar to the GRACE score prognostic value of 0.72 (95%CI = 0.54 - 0.91, p = 0.83) (<xref ref-type="fig" rid="f4">Figure 4A</xref>). There was no death in the first tertile of this entire sample (CAD probability &lt; 30%), three deaths in the second tertile (30 - 62%), and seven deaths in the third tertile (&gt; 62%, p for trend = 0.006) (<xref ref-type="fig" rid="f4">Figure 4B</xref>).</p>
        <p>
          <fig id="f4">
            <label>Figure 4</label>
            <caption>
              <title>Mortality analysis in the full sample of 470 patients, showing a significant prognostic value of the model, which was originally derived for coronary artery disease (CAD) prediction. Panel A compares the C-index of the model versus GRACE score, indicating similar prediction. Panel B compares the incidence of CAD among tertiles of model's coronary disease prediction. AUC denotes area under the receiver operating characteristic curve.</title>
            </caption>
            <graphic xlink:href="1678-4170-abc-108-04-0304-70037-gf04.jpg"/></fig>
        </p>
      </sec>
      <sec>
        <title>Acute chest pain score</title>
        <p>Points proportional to the regression coefficients were attributed to each positive variable: age (&#946; = 0.025; 0.05 point for each year), relief with nitrates (&#946; = 0.60; 1 point), male gender (&#946; = 1.16; 2 points), ischemic electrocardiogram (&#946; = 1.10; 2 points), positive troponin (&#946; = 1.15; 2 points), and signs of left ventricular failure (&#946; = 1.55; 3 points). The score presented the same AUC as the logistic model. There was a proportional increase in disease prevalence according to score deciles: 11%, 14%, 24%, 37%, 41%, 53%, 59%, 67%, 74%, and 95% (p for linear trend &lt; 0.001) (<xref ref-type="fig" rid="f5">Figure 5</xref>).</p>
        <p>
          <fig id="f5">
            <label>Figure 5</label>
            <caption>
              <title>Prevalence of obstructive coronary artery disease (CAD) according to score's deciles.</title>
            </caption>
            <graphic xlink:href="1678-4170-abc-108-04-0304-70037-gf05.jpg"/></fig>
        </p>
      </sec>
    </sec>
    <sec sec-type="discussion">
      <title>Discussion</title>
      <p>The present study developed and validated a probabilistic model to predict obstructive CAD based on data from the initial presentation of acute chest pain. From a total of 35 candidate variables, a final model of six independent predictors was generated, with good discrimination and calibration for assessing the pretest probability of the disease. Most importantly, the accuracy of the model proved to be superior to the traditional model that uses electrocardiogram and troponin.</p>
      <p>The indication of diagnostic tests should take into account the pretest probability of the disease. However, in the selected setting of coronary care units, virtually all patients with undefined chest pain undergo testing for detecting obstructive CAD, regardless of pretest probability. Since the test will be negative in a significant proportion of patients,<xref ref-type="bibr" rid="B2">2</xref> this approach leads to unnecessarily prolonged hospital stay. Thus, eliminating the need for additional tests in patients with low probability of CAD will improve the efficiency of chest pain protocols. However, validated probabilistic models are not disseminated in this clinical setting, making it hard for the emergency physician to tailor medical decision based on probability. At the most, the probability is evaluated in a binary form, based on whether the electrocardiogram or troponin is altered.</p>
      <p>The use of such a probability model improves accuracy and offers a range of continuous probabilities, approximating medical thinking to the best form of dealing with uncertainty. As William Osler once said, "medicine is the science of uncertainty and the art of probability."</p>
      <p>Our purpose to predict obstructive CAD should not be confused with previous studies that developed neural or logistic models for predicting the clinical diagnosis of myocardial infarction in patients with chest pain.<xref ref-type="bibr" rid="B8">8</xref><sup>-</sup><xref ref-type="bibr" rid="B12">12</xref> Such studies created models from clinical data, symptoms characteristics, and sometimes electrocardiogram, which were tested as predictors of a final diagnosis defined by a systematic analysis of the same variables in addition to markers of myocardial necrosis. Therefore, these mathematical models mainly serve as surrogates of medical thinking or, at the most, predictors of a final impression that will be obtained in a few hours of the initial presentation. In contrast, our model was built to predict the result of imaging tests before they are performed. Since noninvasive or invasive imaging tests aim the diagnosis of obstructive CAD, a model of this kind is clearly useful in efficiently selecting patients for these tests, based on the estimation of the pretest probability of the disease. In addition, the knowledge of a pretest probability permits the calculation of the post-test probability after a noninvasive imaging result is obtained.</p>
      <p>Other scores focus on the risk of adverse events (HEART score,<xref ref-type="bibr" rid="B13">13</xref> TIMI score<xref ref-type="bibr" rid="B14">14</xref> or GRACE score<xref ref-type="bibr" rid="B6">6</xref>). Despite their prognostic value, they are not necessarily good predictors of obstructive CAD<xref ref-type="bibr" rid="B15">15</xref> and physicians are uncomfortable to discharge a patient with acute chest pain with no further testing. Thus, we believe that the calculation of the probability of obstructive CAD would encourage physicians to reduce overuse of imaging studies in patients with low probability, diminishing the phenomena of overdiagnosis and overtreatment. For example, patients with a normal electrocardiogram and negative troponin are known to have a good prognosis. In our study, 50% of these patients had a probability of significant CAD below 20%. Based on favorable prognostic and diagnostic probabilities, these patients could be discharged with no further testing. On the other hand, patients with normal electrocardiogram and troponin may have a significant probability of CAD that can be detected by the model. We should point out that future randomized clinical trials should validate the efficiency and safety of this approach.</p>
      <p>Physicians normally rely on symptoms characteristics (typical or atypical) and traditional risk factors to estimate the chance of CAD in patients with acute chest pain. For example, a diabetic patient with typical chest pain is usually defined as having a high probability of CAD. However, in our study, no risk factors and chest pain characteristic (except for nitrate relief) independently predicted CAD. This is in agreement with previous studies, which indicate that the type of presentation has little influence on the diagnosis in the acute setting. In a comprehensive systematic review, Swap and Nagurney et al.<xref ref-type="bibr" rid="B16">16</xref> showed low likelihood ratios for chest pain characteristics. Seemingly, a recent article by Khan et al.<xref ref-type="bibr" rid="B17">17</xref> demonstrated that most pain characteristics are not associated with coronary disease as the cause of the symptom. Therefore, our data reinforce that the approach to rely on risk factors and symptoms to stratify acute chest pain patients has low accuracy. The utilization of a probabilistic model prevents this type of cognitive error.</p>
      <p>We purposed three easy forms of utilization of the probabilistic model. First, a score based on points attributed to each positive variable, accompanied by a chart relating summed results and probabilities (<xref ref-type="fig" rid="f4">Figure 4</xref>). Considering the low number of variables, five of them of binary nature, the calculation is easily performed. Second, a logistic score within a spreadsheet with the regression formula, containing age as numeric variable and five "yes" or "no" answers. And, most friendly, an application for smartphones. We believe that by offering different forms of calculations, the clinicians will develop a greater interest in using probabilistic models.</p>
      <p>Limitations of the present study should be recognized. The study was performed in a coronary care unit of a specific tertiary hospital, which limits external validity. The population of a chest pain unit is somewhat selected and tends to have a higher prevalence of disease than a general emergency room population. Thus, our model should be further validated for patients with a greater range of clinical presentation. On the other hand, the main purpose of the model is to estimate the pretest probability of hospitalized individuals, which also consist of a large subgroup of real world patients. In this sense, our external validity is not necessarily small; it is just more specific to the tested population.</p>
      <p>We should recognize that our sample size is relatively small in comparison with examples of scores delivered from enormous databanks. We have three arguments in favor of our study of 470 patients: first, its novelty as the first successful attempt to develop such a score, which serves at least as a proof of concept that a multivariate model predicts the pretest probability of the disease. Second, in the absence of a multivariate probabilistic model, physicians use clinical judgment based on probabilistic intuition, which has been proved in different settings to be inferior to multivariate models. Thus, considering the remaining alternative of intuition, it might be a good idea to use such a score, not deterministically, but as a tool to avoid common cognitive biases related to intuition. Third, our sample size was based on <italic>a priori</italic> sample size calculation for the logistic regression and for testing the model with ROC curve. According to this calculation, our number of events was enough to provide the minimum power and precision required. Nevertheless, future reports should improve the precision of our estimates.</p>
      <p>Finally, among patients who underwent noninvasive tests first, only those with positive results had confirmation by angiography. Nevertheless, predicting a negative noninvasive test (as opposed to no disease at all) is sufficient to prevent the patient from staying unnecessarily to undergo the test.</p>
    </sec>
    <sec sec-type="conclusions">
      <title>Conclusion</title>
      <p>The present study developed and validated a novel model to predict obstructive CAD among patients who are admitted with acute chest pain in the coronary care unit. The utilization of such a model should have an impact in preventing overuse of tests, overdiagnosis, and overtreatment while improving the accuracy of pretest assessment of disease probability.</p>
    </sec>
  </body>
  <back>
    <fn-group>
      <fn fn-type="supported-by">
        <p>
          <bold>Sources of Funding</bold>
        </p>
        <p>There were no external funding sources for this study.</p>
      </fn>
      <fn fn-type="other">
        <p>
          <bold>Study Association</bold>
        </p>
        <p>This study is not associated with any thesis or dissertation work.</p>
      </fn>
    </fn-group>
    <ref-list>
      <title>References</title>
      <ref id="B1">
        <label>1</label>
        <mixed-citation>1 Amsterdam EA, Kirk JD, Bluemke DA, Diercks D, Farkouh ME, Garvey JL, et al., on behalf of the American Heart Association Exercise CR, Prevention Committee of the Council on Clinical Cardiology CoCN, Care ICoQo, Research O. Testing of low-risk patients presenting to the emergency department with chest pain: A scientific statement from the american heart association. <italic>Circulation</italic>. 2010;122(17):1756-76.</mixed-citation>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Amsterdam</surname>
              <given-names>EA</given-names>
            </name>
            <name>
              <surname>Kirk</surname>
              <given-names>JD</given-names>
            </name>
            <name>
              <surname>Bluemke</surname>
              <given-names>DA</given-names>
            </name>
            <name>
              <surname>Diercks</surname>
              <given-names>D</given-names>
            </name>
            <name>
              <surname>Farkouh</surname>
              <given-names>ME</given-names>
            </name>
            <name>
              <surname>Garvey</surname>
              <given-names>JL</given-names>
            </name>
            <etal/>
          </person-group>
          <comment>on behalf of the American Heart Association Exercise CR.Prevention Committee of the Council on Clinical Cardiology CoCN.Care ICoQo.Research O</comment>
          <article-title>Testing of low-risk patients presenting to the emergency department with chest pain: A scientific statement from the american heart association</article-title>
          <source>Circulation</source>
          <year>2010</year>
          <volume>122</volume>
          <issue>17</issue>
          <fpage>1756</fpage>
          <lpage>1776</lpage>
        </element-citation>
      </ref>
      <ref id="B2">
        <label>2</label>
        <mixed-citation>2 Hermann LK, Newman DH, Pleasant W, Roianasmtikul D, Lakoff D, Goldberg AS, et al.Yield of routine provocative cardiac testing among patients in an emergency department-based chest pain unit. <italic>JAMA Intern Med</italic>. 2013;173(12):1128-33</mixed-citation>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Hermann</surname>
              <given-names>LK</given-names>
            </name>
            <name>
              <surname>Newman</surname>
              <given-names>DH</given-names>
            </name>
            <name>
              <surname>Pleasant</surname>
              <given-names>W</given-names>
            </name>
            <name>
              <surname>Roianasmtikul</surname>
              <given-names>D</given-names>
            </name>
            <name>
              <surname>Lakoff</surname>
              <given-names>D</given-names>
            </name>
            <name>
              <surname>Goldberg</surname>
              <given-names>AS</given-names>
            </name>
            <etal/>
          </person-group>
          <article-title>Yield of routine provocative cardiac testing among patients in an emergency department-based chest pain unit</article-title>
          <source>JAMA Intern Med</source>
          <year>2013</year>
          <volume>173</volume>
          <issue>12</issue>
          <fpage>1128</fpage>
          <lpage>1133</lpage>
        </element-citation>
      </ref>
      <ref id="B3">
        <label>3</label>
        <mixed-citation>3 Redberg RF. Coronary ct angiography for acute chest pain. N Engl J Med. 2012;367(4):375-6.</mixed-citation>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Redberg</surname>
              <given-names>RF</given-names>
            </name>
          </person-group>
          <article-title>Coronary ct angiography for acute chest pain</article-title>
          <source>N Engl J Med</source>
          <year>2012</year>
          <volume>367</volume>
          <issue>4</issue>
          <fpage>375</fpage>
          <lpage>376</lpage>
        </element-citation>
      </ref>
      <ref id="B4">
        <label>4</label>
        <mixed-citation>4 Kachalia A, Mello MM. Defensive medicine-legally necessary but ethically wrong?: Inpatient stress testing for chest pain in low-risk patients. <italic>JAMA Intern Med</italic>. 2013;173(4):1056-7.</mixed-citation>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Kachalia</surname>
              <given-names>A</given-names>
            </name>
            <name>
              <surname>Mello</surname>
              <given-names>MM</given-names>
            </name>
          </person-group>
          <article-title>Defensive medicine-legally necessary but ethically wrong?: Inpatient stress testing for chest pain in low-risk patients</article-title>
          <source>JAMA Intern Med</source>
          <year>2013</year>
          <volume>173</volume>
          <issue>4</issue>
          <fpage>1056</fpage>
          <lpage>1057</lpage>
        </element-citation>
      </ref>
      <ref id="B5">
        <label>5</label>
        <mixed-citation>5 Pencina MJ, D' Agostino RB, D' Agostino RB, Vasan RS. Evaluating the added predictive ability of a new marker: From area under the roc curve to reclassification and beyond. <italic>Stat Med</italic>. 2008;27(2):157-72.</mixed-citation>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Pencina</surname>
              <given-names>MJ</given-names>
            </name>
            <name>
              <surname>D' Agostino</surname>
              <given-names>RB</given-names>
            </name>
            <name>
              <surname>D' Agostino</surname>
              <given-names>RB</given-names>
            </name>
            <name>
              <surname>Vasan</surname>
              <given-names>RS</given-names>
            </name>
          </person-group>
          <article-title>Evaluating the added predictive ability of a new marker: From area under the roc curve to reclassification and beyond</article-title>
          <source>Stat Med</source>
          <year>2008</year>
          <volume>27</volume>
          <issue>2</issue>
          <fpage>157</fpage>
          <lpage>172</lpage>
        </element-citation>
      </ref>
      <ref id="B6">
        <label>6</label>
        <mixed-citation>6 Granger CB, Goldberg RJ, Dabbous O, Pieper KS, Eagle KA, Cannon CP, et al. Predictors of hospital mortality in the global registry of acute coronary events. <italic>Arch Intern Med</italic>. 2003;163(19):2345-53.</mixed-citation>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Granger</surname>
              <given-names>CB</given-names>
            </name>
            <name>
              <surname>Goldberg</surname>
              <given-names>RJ</given-names>
            </name>
            <name>
              <surname>Dabbous</surname>
              <given-names>O</given-names>
            </name>
            <name>
              <surname>Pieper</surname>
              <given-names>KS</given-names>
            </name>
            <name>
              <surname>Eagle</surname>
              <given-names>KA</given-names>
            </name>
            <name>
              <surname>Cannon</surname>
              <given-names>CP</given-names>
            </name>
            <etal/>
          </person-group>
          <article-title>Predictors of hospital mortality in the global registry of acute coronary events</article-title>
          <source>Arch Intern Med</source>
          <year>2003</year>
          <volume>163</volume>
          <issue>19</issue>
          <fpage>2345</fpage>
          <lpage>2353</lpage>
        </element-citation>
      </ref>
      <ref id="B7">
        <label>7</label>
        <mixed-citation>7 Peduzzi P, Concato J, Kemper E, Holford TR, Feinstein AR. A simulation study of the number of events per variable in logistic regression analysis. <italic>J Clin Epidemiol</italic>.1996;49(12):1373-9.</mixed-citation>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Peduzzi</surname>
              <given-names>P</given-names>
            </name>
            <name>
              <surname>Concato</surname>
              <given-names>J</given-names>
            </name>
            <name>
              <surname>Kemper</surname>
              <given-names>E</given-names>
            </name>
            <name>
              <surname>Holford</surname>
              <given-names>TR</given-names>
            </name>
            <name>
              <surname>Feinstein</surname>
              <given-names>AR</given-names>
            </name>
          </person-group>
          <article-title>A simulation study of the number of events per variable in logistic regression analysis</article-title>
          <source>J Clin Epidemiol</source>
          <year>1996</year>
          <volume>49</volume>
          <issue>12</issue>
          <fpage>1373</fpage>
          <lpage>1379</lpage>
        </element-citation>
      </ref>
      <ref id="B8">
        <label>8</label>
        <mixed-citation>8 Wang SJ, Ohno-Machado L, Fraser HS, Kennedy RL. Using patient-reportable clinical history factors to predict myocardial infarction. <italic>Comput Biol Med</italic>. 2001;31(1):1-13.</mixed-citation>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Wang</surname>
              <given-names>SJ</given-names>
            </name>
            <name>
              <surname>Ohno-Machado</surname>
              <given-names>L</given-names>
            </name>
            <name>
              <surname>Fraser</surname>
              <given-names>HS</given-names>
            </name>
            <name>
              <surname>Kennedy</surname>
              <given-names>RL</given-names>
            </name>
          </person-group>
          <article-title>Using patient-reportable clinical history factors to predict myocardial infarction</article-title>
          <source>Comput Biol Med</source>
          <year>2001</year>
          <volume>31</volume>
          <issue>1</issue>
          <fpage>1</fpage>
          <lpage>13</lpage>
        </element-citation>
      </ref>
      <ref id="B9">
        <label>9</label>
        <mixed-citation>9 Selker HP, Griffith JL, Patil S, Long WJ, D'Agostino RB. A comparison of performance of mathematical predictive methods for medical diagnosis: Identifying acute cardiac ischemia among emergency department patients. <italic>J Investig Med</italic>. 1995;43(5):468-76</mixed-citation>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Selker</surname>
              <given-names>HP</given-names>
            </name>
            <name>
              <surname>Griffith</surname>
              <given-names>JL</given-names>
            </name>
            <name>
              <surname>Patil</surname>
              <given-names>S</given-names>
            </name>
            <name>
              <surname>Long</surname>
              <given-names>WJ</given-names>
            </name>
            <name>
              <surname>D'Agostino</surname>
              <given-names>RB</given-names>
            </name>
          </person-group>
          <article-title>A comparison of performance of mathematical predictive methods for medical diagnosis: Identifying acute cardiac ischemia among emergency department patients</article-title>
          <source>J Investig Med</source>
          <year>1995</year>
          <volume>43</volume>
          <issue>5</issue>
          <fpage>468</fpage>
          <lpage>476</lpage>
        </element-citation>
      </ref>
      <ref id="B10">
        <label>10</label>
        <mixed-citation>10 Pozen MW, D'Agostino RB, Selker HP, Sytkowski PA, Hood WB Jr. A predictive instrument to improve coronary-care-unit admission practices in acute ischemic heart disease. A prospective multicenter clinical trial. N Engl J Med. 1984;310(20):1273-8.</mixed-citation>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Pozen</surname>
              <given-names>MW</given-names>
            </name>
            <name>
              <surname>D'Agostino</surname>
              <given-names>RB</given-names>
            </name>
            <name>
              <surname>Selker</surname>
              <given-names>HP</given-names>
            </name>
            <name>
              <surname>Sytkowski</surname>
              <given-names>PA</given-names>
            </name>
            <name>
              <surname>Hood</surname>
              <given-names>WB</given-names>
              <suffix>Jr</suffix>
            </name>
          </person-group>
          <article-title>A predictive instrument to improve coronary-care-unit admission practices in acute ischemic heart disease. A prospective multicenter clinical trial</article-title>
          <source>N Engl J Med</source>
          <year>1984</year>
          <volume>310</volume>
          <issue>20</issue>
          <fpage>1273</fpage>
          <lpage>1278</lpage>
        </element-citation>
      </ref>
      <ref id="B11">
        <label>11</label>
        <mixed-citation>11 Harrison RF, Kennedy RL. Artificial neural network models for prediction of acute coronary syndromes using clinical data from the time of presentation. <italic>Ann Emerg Med</italic>. 2005;46(5):431-9.</mixed-citation>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Harrison</surname>
              <given-names>RF</given-names>
            </name>
            <name>
              <surname>Kennedy</surname>
              <given-names>RL</given-names>
            </name>
          </person-group>
          <article-title>Artificial neural network models for prediction of acute coronary syndromes using clinical data from the time of presentation</article-title>
          <source>Ann Emerg Med</source>
          <year>2005</year>
          <volume>46</volume>
          <issue>5</issue>
          <fpage>431</fpage>
          <lpage>439</lpage>
        </element-citation>
      </ref>
      <ref id="B12">
        <label>12</label>
        <mixed-citation>12 Goodacre S, Locker T, Morris F, Campbell S. How useful are clinical features in the diagnosis of acute, undifferentiated chest pain? <italic>Acad Emerg Med</italic>.. 2002;9(3):203-8.</mixed-citation>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Goodacre</surname>
              <given-names>S</given-names>
            </name>
            <name>
              <surname>Locker</surname>
              <given-names>T</given-names>
            </name>
            <name>
              <surname>Morris</surname>
              <given-names>F</given-names>
            </name>
            <name>
              <surname>Campbell</surname>
              <given-names>S</given-names>
            </name>
          </person-group>
          <article-title>How useful are clinical features in the diagnosis of acute, undifferentiated chest pain?</article-title>
          <source>Acad Emerg Med</source>
          <year>2002</year>
          <volume>9</volume>
          <issue>3</issue>
          <fpage>203</fpage>
          <lpage>208</lpage>
        </element-citation>
      </ref>
      <ref id="B13">
        <label>13</label>
        <mixed-citation>13 Backus BE, Six AJ, Kelder JC, Bosschaert MAR, Mast EG, Mosterd A, et al. A prospective validation of the heart score for chest pain patients at the emergency department. <italic>Int J Cardiol</italic>.2013;168(3):2153-8.</mixed-citation>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Backus</surname>
              <given-names>BE</given-names>
            </name>
            <name>
              <surname>Six</surname>
              <given-names>AJ</given-names>
            </name>
            <name>
              <surname>Kelder</surname>
              <given-names>JC</given-names>
            </name>
            <name>
              <surname>Bosschaert</surname>
              <given-names>MAR</given-names>
            </name>
            <name>
              <surname>Mast</surname>
              <given-names>EG</given-names>
            </name>
            <name>
              <surname>Mosterd</surname>
              <given-names>A</given-names>
            </name>
            <etal/>
          </person-group>
          <article-title>A prospective validation of the heart score for chest pain patients at the emergency department</article-title>
          <source>Int J Cardiol</source>
          <year>2013</year>
          <volume>168</volume>
          <issue>3</issue>
          <fpage>2153</fpage>
          <lpage>2158</lpage>
        </element-citation>
      </ref>
      <ref id="B14">
        <label>14</label>
        <mixed-citation>14 Body R, Carley S, McDowell G, Ferguson J, Mackway-Jones K. Can a modified thrombolysis in myocardial infarction risk score outperform the original for risk stratifying emergency department patients with chest pain? <italic>Emerg Med J</italic>. 2009;26(2):95-9.</mixed-citation>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Body</surname>
              <given-names>R</given-names>
            </name>
            <name>
              <surname>Carley</surname>
              <given-names>S</given-names>
            </name>
            <name>
              <surname>McDowell</surname>
              <given-names>G</given-names>
            </name>
            <name>
              <surname>Ferguson</surname>
              <given-names>J</given-names>
            </name>
            <name>
              <surname>Mackway-Jones</surname>
              <given-names>K</given-names>
            </name>
          </person-group>
          <article-title>Can a modified thrombolysis in myocardial infarction risk score outperform the original for risk stratifying emergency department patients with chest pain?</article-title>
          <source>Emerg Med J</source>
          <year>2009</year>
          <volume>26</volume>
          <issue>2</issue>
          <fpage>95</fpage>
          <lpage>99</lpage>
        </element-citation>
      </ref>
      <ref id="B15">
        <label>15</label>
        <mixed-citation>15 Barbosa CE, Viana M ,Brito M, Sabino G, Garcia M, Maraux M,et al. Accuracy of the grace and timi scores in predicting the angiographic severity of acute coronary syndrome.Arq Bras Cardiol.2012;9(3):818-24.</mixed-citation>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Barbosa</surname>
              <given-names>CE</given-names>
            </name>
            <name>
              <surname>Viana</surname>
              <given-names>M</given-names>
            </name>
            <name>
              <surname>Brito</surname>
              <given-names>M</given-names>
            </name>
            <name>
              <surname>Sabino</surname>
              <given-names>G</given-names>
            </name>
            <name>
              <surname>Garcia</surname>
              <given-names>M</given-names>
            </name>
            <name>
              <surname>Maraux</surname>
              <given-names>M</given-names>
            </name>
            <etal/>
          </person-group>
          <article-title>Accuracy of the grace and timi scores in predicting the angiographic severity of acute coronary syndrome</article-title>
          <source>Arq Bras Cardiol</source>
          <year>2012</year>
          <volume>9</volume>
          <issue>3</issue>
          <fpage>818</fpage>
          <lpage>824</lpage>
        </element-citation>
      </ref>
      <ref id="B16">
        <label>16</label>
        <mixed-citation>16 Swap CJ, Nagurney JT. Value and limitations of chest pain history in the evaluation of patients with suspected acute coronary syndromes. <italic>JAMA</italic>. 2005;294(20):2623-9.</mixed-citation>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Swap</surname>
              <given-names>CJ</given-names>
            </name>
            <name>
              <surname>Nagurney</surname>
              <given-names>JT</given-names>
            </name>
          </person-group>
          <article-title>Value and limitations of chest pain history in the evaluation of patients with suspected acute coronary syndromes</article-title>
          <source>JAMA</source>
          <year>2005</year>
          <volume>294</volume>
          <issue>20</issue>
          <fpage>2623</fpage>
          <lpage>2629</lpage>
        </element-citation>
      </ref>
      <ref id="B17">
        <label>17</label>
        <mixed-citation>17 Khan NA, Daskalopoulou SS, Karp I, Eisenberg MJ, Pelletier R, Tsadek MA, et al. Sex differences in acute coronary syndrome symptom presentation in young patients. <italic>JAMA Intern Med</italic>. 2013;173(20):1863-71.</mixed-citation>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Khan</surname>
              <given-names>NA</given-names>
            </name>
            <name>
              <surname>Daskalopoulou</surname>
              <given-names>SS</given-names>
            </name>
            <name>
              <surname>Karp</surname>
              <given-names>I</given-names>
            </name>
            <name>
              <surname>Eisenberg</surname>
              <given-names>MJ</given-names>
            </name>
            <name>
              <surname>Pelletier</surname>
              <given-names>R</given-names>
            </name>
            <name>
              <surname>Tsadek</surname>
              <given-names>MA</given-names>
            </name>
            <etal/>
          </person-group>
          <article-title>Sex differences in acute coronary syndrome symptom presentation in young patients</article-title>
          <source>JAMA Intern Med</source>
          <year>2013</year>
          <volume>173</volume>
          <issue>20</issue>
          <fpage>1863</fpage>
          <lpage>1871</lpage>
        </element-citation>
      </ref>
    </ref-list>
  </back>
  <sub-article article-type="translation" id="s1" xml:lang="pt">
    <front-stub>
      <article-categories>
        <subj-group subj-group-type="heading">
          <subject>Artigos Originais</subject>
          <subj-group>
            <subject>Coronariopatia Aguda</subject>
          </subj-group>
        </subj-group>
      </article-categories>
      <title-group>
        <article-title>Um Modelo Multivariado para Predi&#231;&#227;o de Doen&#231;a Coronariana Obstrutiva em Pacientes com Dor Tor&#225;cica Aguda: Desenvolvimento e Valida&#231;&#227;o</article-title>
      </title-group>
      <contrib-group>
        <contrib contrib-type="author">
          <name>
            <surname>Correia</surname>
            <given-names>Luis Cl&#225;udio Lemos</given-names>
          </name>
          <xref ref-type="aff" rid="aff3">1</xref>
          <xref ref-type="aff" rid="aff4">2</xref>
          <xref ref-type="corresp" rid="c2"/>
        </contrib>
        <contrib contrib-type="author">
          <name>
            <surname>Cerqueira</surname>
            <given-names>Maur&#237;cio</given-names>
          </name>
          <xref ref-type="aff" rid="aff3">1</xref>
        </contrib>
        <contrib contrib-type="author">
          <name>
            <surname>Carvalhal</surname>
            <given-names>Manuela</given-names>
          </name>
          <xref ref-type="aff" rid="aff3">1</xref>
        </contrib>
        <contrib contrib-type="author">
          <name>
            <surname>Ferreira</surname>
            <given-names>Felipe</given-names>
          </name>
          <xref ref-type="aff" rid="aff3">1</xref>
        </contrib>
        <contrib contrib-type="author">
          <name>
            <surname>Garcia</surname>
            <given-names>Guilherme</given-names>
          </name>
          <xref ref-type="aff" rid="aff4">2</xref>
        </contrib>
        <contrib contrib-type="author">
          <name>
            <surname>Silva</surname>
            <given-names>Andr&#233; Barcelos da</given-names>
          </name>
          <xref ref-type="aff" rid="aff3">1</xref>
        </contrib>
        <contrib contrib-type="author">
          <name>
            <surname>S&#225;</surname>
            <given-names>Nicole de</given-names>
          </name>
          <xref ref-type="aff" rid="aff3">1</xref>
        </contrib>
        <contrib contrib-type="author">
          <name>
            <surname>Lopes</surname>
            <given-names>Fernanda</given-names>
          </name>
          <xref ref-type="aff" rid="aff3">1</xref>
        </contrib>
        <contrib contrib-type="author">
          <name>
            <surname>Barcelos</surname>
            <given-names>Ana Clara</given-names>
          </name>
          <xref ref-type="aff" rid="aff3">1</xref>
        </contrib>
        <contrib contrib-type="author">
          <name>
            <surname>Noya-Rabelo</surname>
            <given-names>M&#225;rcia</given-names>
          </name>
          <xref ref-type="aff" rid="aff3">1</xref>
          <xref ref-type="aff" rid="aff4">2</xref>
        </contrib>
      </contrib-group>
      <aff id="aff3">
        <label>1</label>
        <institution content-type="original">Escola Bahiana de Medicina e Sa&#250;de P&#250;blica; Salvador, BA - Brasil</institution>
      </aff>
      <aff id="aff4">
        <label>2</label>
        <institution content-type="original">Hospital S&#227;o Rafael, Salvador, BA - Brasil</institution>
      </aff>
      <author-notes>
        <corresp id="c2"><bold>Correspond&#234;ncia: Lu&#237;s Cl&#225;udio Lemos Correia</bold>, Av. Princesa Leopoldina, 19/402. CEP 40150-080, Gra&#231;a, Salvador, BA - Brasil. E-mail: <email>lccorreia@cardiol.br</email>; <email>lccorreia@terra.com.br</email>
				</corresp>
        <fn fn-type="con">
          <p>
            <bold>Contribui&#231;&#227;o dos autores</bold>
          </p>
          <p>Concep&#231;&#227;o e desenho da pesquisa: Correia LCL, Cerqueira M, Carvalhal M, Ferreira F, Garcia G, Silva AB, S&#225; N, Lopes F, Barcelos AC, Noya-Rabelo M; Obten&#231;&#227;o de dados: Cerqueira M, Carvalhal M, Ferreira F, Silva AB, S&#225; N, Lopes F, Barcelos AC, Noya-Rabelo M; An&#225;lise e interpreta&#231;&#227;o dos dados: Correia LCL, Cerqueira M, Garcia G, Silva AB, S&#225; N, Lopes F, Barcelos AC, Noya-Rabelo M; An&#225;lise estat&#237;stica: Correia LCL, Cerqueira M, Ferreira F, Garcia G, Noya-Rabelo M; Reda&#231;&#227;o do manuscrito: Correia LCL, Cerqueira M, Carvalhal M, Ferreira F, Silva AB, S&#225; N, Lopes F, Barcelos AC, Noya-Rabelo M; Revis&#227;o cr&#237;tica do manuscrito quanto ao conte&#250;do intelectual importante: Correia LCL, Cerqueira M, Carvalhal M, Garcia G, Silva AB, S&#225; N, Lopes F, Barcelos AC, Noya-Rabelo M.</p>
        </fn>
        <fn fn-type="conflict">
          <p>
            <bold>Potencial conflito de interesses</bold>
          </p>
          <p>Declaro n&#227;o haver conflito de interesses pertinentes.</p>
        </fn>
      </author-notes>
      <abstract>
        <title>Resumo</title>
        <sec>
          <title>Fundamento:</title>
          <p>Atualmente, n&#227;o existe um modelo multivariado validado para predizer a probabilidade de doen&#231;a coronariana obstrutiva em pacientes com dor tor&#225;cica aguda.</p>
        </sec>
        <sec>
          <title>Objetivo:</title>
          <p>Desenvolver e validar um modelo multivariado para predizer doen&#231;a arterial coronariana (DAC) com base em vari&#225;veis avaliadas &#224; admiss&#227;o na unidade coronariana (UC) devido a dor tor&#225;cica aguda.</p>
        </sec>
        <sec>
          <title>M&#233;todos:</title>
          <p>Foram estudados um total de 470 pacientes, 370 utilizados como amostra de deriva&#231;&#227;o e os subsequentes 100 pacientes como amostra de valida&#231;&#227;o. Como padr&#227;o de refer&#234;ncia, a angiografia foi necess&#225;ria para descartar DAC (estenose &#8805; 70%), enquanto a angiografia ou um teste n&#227;o invasivo negativo foi utilizado para confirmar a doen&#231;a. Foram testadas como preditoras 13 vari&#225;veis basais relacionadas &#224; hist&#243;ria m&#233;dica, 14 caracter&#237;sticas de desconforto tor&#225;cico e oito vari&#225;veis relacionadas ao exame f&#237;sico ou testes laboratoriais.</p>
        </sec>
        <sec>
          <title>Resultados:</title>
          <p>A preval&#234;ncia de DAC foi de 48%. Por regress&#227;o log&#237;stica, seis vari&#225;veis permaneceram como preditoras independentes de DAC: idade, g&#234;nero masculino, al&#237;vio com nitrato, sinais de insufici&#234;ncia card&#237;aca, e eletrocardiograma e troponina positivos. A &#225;rea sob a curva (<italic>area under the curve</italic>, AUC) deste modelo final foi de 0,80 (intervalo de confian&#231;a de 95% [IC95%] = 0,75 - 0,84) na amostra de deriva&#231;&#227;o e 0,86 (IC95% = 0,79 - 0,93) na amostra de valida&#231;&#227;o. O teste de Hosmer-Lemeshow indicou uma boa calibra&#231;&#227;o em ambas as amostras (p = 0,98 e p = 0,23, respectivamente). Em compara&#231;&#227;o com o modelo b&#225;sico contendo eletrocardiograma e troponina, o modelo completo ofereceu um incremento na AUC de 0,07 tanto na amostra de deriva&#231;&#227;o (p = 0,0002) quanto na de valida&#231;&#227;o (p = 0,039). A melhoria na discrimina&#231;&#227;o integrada foi de 0,09 nas amostras de deriva&#231;&#227;o (p &lt; 0,001) e valida&#231;&#227;o (p &lt; 0,0015).</p>
        </sec>
        <sec>
          <title>Conclus&#227;o:</title>
          <p>Um modelo multivariado foi derivado e validado como uma ferramenta acurada para estimar a probabilidade pr&#233;-teste de DAC em pacientes com dor tor&#225;cica aguda. </p>
        </sec>
      </abstract>
      <kwd-group xml:lang="pt">
        <title>Palavras-chave:</title>
        <kwd>Doen&#231;a Arterial Coronariana</kwd>
        <kwd>M&#233;todos</kwd>
        <kwd>Dor no Peito</kwd>
        <kwd>Modelos Estat&#237;sticos</kwd>
        <kwd>Angiografia Coron&#225;ria</kwd>
        <kwd>Troponina</kwd>
        <kwd>Eletrocardiografia</kwd>
      </kwd-group>
    </front-stub>
    <body>
      <sec sec-type="intro">
        <title>Introdu&#231;&#227;o</title>
        <p>A dor tor&#225;cica aguda &#233; uma das causas mais comuns de visitas a departamentos de emerg&#234;ncia. Uma vez que esta dor pode representar uma manifesta&#231;&#227;o cl&#237;nica de isquemia card&#237;aca, a alta do paciente &#233; normalmente condicionada a um teste negativo para doen&#231;a arterial coronariana (CAD) obstrutiva.<xref ref-type="bibr" rid="B1">1</xref> No entanto, a efici&#234;ncia desta estrat&#233;gia defensiva &#233; contestada por um baixo rendimento dos testes card&#237;acos, uma vez que apenas uma parte dos pacientes acaba apresentando DAC obstrutiva e uma parte ainda menor precisar&#225; de revasculariza&#231;&#227;o.<xref ref-type="bibr" rid="B2">2</xref> Al&#233;m disso, a realiza&#231;&#227;o de testes de rotina n&#227;o &#233; suportada por evid&#234;ncia de efeito ben&#233;fico<xref ref-type="bibr" rid="B3">3</xref> e pode ter consequ&#234;ncias n&#227;o intencionais: excesso de diagn&#243;stico e de tratamento de doen&#231;a coronariana n&#227;o relacionada aos sintomas, hospitaliza&#231;&#227;o prolongada, procedimentos invasivos desnecess&#225;rios por resultados falso-positivos de testes e aumento de despesas m&#233;dicas.<xref ref-type="bibr" rid="B4">4</xref>
				</p>
        <p>Com isto, uma abordagem mais racional seria indicar testes adicionais com base em probabilidade pr&#233;-teste. Tradicionalmente, esta avalia&#231;&#227;o pr&#233;-teste &#233; restrita ao eletrocardiograma e marcadores de necrose. No entanto, o uso de um modelo multivariado tem o potencial de melhorar a acur&#225;cia e fornecer uma faixa mais cont&#237;nua de probabilidades. A fim de desenvolver e validar um modelo multivariado para predizer DAC com base em vari&#225;veis avaliadas no momento da admiss&#227;o na unidade coronariana, 370 pacientes consecutivos foram estudados. Trinta e cinco vari&#225;veis foram testadas como candidatas preditoras de DAC obstrutiva a fim de gerar um modelo final que foi subsequentemente validado em uma amostra de 100 pacientes.</p>
      </sec>
      <sec sec-type="methods">
        <title>M&#233;todos</title>
        <sec>
          <title>Sele&#231;&#227;o da amostra</title>
          <p>Durante um per&#237;odo de 30 meses consecutivos, todos os pacientes admitidos na unidade coronariana do nosso hospital foram inclu&#237;dos no estudo. As admiss&#245;es ocorreram sempre que o julgamento m&#233;dico reconheceu qualquer chance de uma etiologia coronariana, independentemente do eletrocardiograma ou troponina. O &#250;nico crit&#233;rio de exclus&#227;o foi a recusa do paciente em participar. Como definido <italic>a priori,</italic> os primeiros 370 pacientes foram utilizados como amostra de deriva&#231;&#227;o e os 100 pacientes seguintes como amostra de valida&#231;&#227;o. O estudo foi aprovado por um comit&#234; de revis&#227;o institucional e todos os sujeitos deram consentimento informado para participar do estudo.</p>
        </sec>
        <sec>
          <title>Preditores de DAC obstrutiva</title>
          <p>&#192; admiss&#227;o no momento basal, tr&#234;s conjuntos de vari&#225;veis foram registrados como candidatos &#224; predi&#231;&#227;o de DAC obstrutiva. O primeiro foi composto por 13 vari&#225;veis relacionadas &#224; hist&#243;ria m&#233;dica, como idade, g&#234;nero, hist&#243;ria pr&#233;via de DAC, fatores de risco para DAC e comorbidades; o segundo incluiu 14 caracter&#237;sticas de desconforto tor&#225;cico e o terceiro foi composto por oito vari&#225;veis relacionadas ao exame f&#237;sico ou a testes b&#225;sicos de admiss&#227;o, incluindo sinais f&#237;sicos e radiol&#243;gicos de insufici&#234;ncia card&#237;aca esquerda, altera&#231;&#245;es isqu&#234;micas ao eletrocardiograma (invers&#227;o da onda T &#8805; 1 mm ou desvio din&#226;mico de ST &#8805; 0,5 mm), troponina positiva (&gt; 99&#730; percentil da popula&#231;&#227;o geral; Ortho-Clinical Diagnostics, Rochester, NY, EUA), fragmento N-terminal do pept&#237;deo natriur&#233;tico tipo B (NT-proBNP, ensaio fluorescente ligado &#224; enzima, Biom&#233;rieux, Fran&#231;a), prote&#237;na C reativa ultrassens&#237;vel (PCR; nefelometria, Dade-Behring, EUA), contagem de c&#233;lulas brancas, glicose plasm&#225;tica e hemoglobina. Os exames laboratoriais foram realizados em plasma coletado &#224; apresenta&#231;&#227;o na sala de emerg&#234;ncia. A hist&#243;ria m&#233;dica e as caracter&#237;sticas da dor tor&#225;cica foram registradas por tr&#234;s investigadores (M.C., A.M.C. e R.B.) treinados para entrevistar os pacientes de forma sistem&#225;tica, a fim de diminuir a ocorr&#234;ncia de vi&#233;s e melhorar a reprodutibilidade. Sinais radiol&#243;gicos de fal&#234;ncia ventricular e eletrocardiograma foram interpretados pelo mesmo investigador s&#234;nior (L.C.).</p>
        </sec>
        <sec>
          <title>Defini&#231;&#227;o de desfechos</title>
          <p>O desfecho prim&#225;rio a ser previsto pelo modelo foi o diagn&#243;stico de DAC obstrutiva, definido por testes subsequentes realizados durante a interna&#231;&#227;o. Os dados dos desfechos foram coletados por tr&#234;s investigadores (M.C., A.M.C. e R.B.) e julgados por um quarto investigador (L.C.). Para a avalia&#231;&#227;o diagn&#243;stica, os pacientes foram submetidos a angiografia coronariana invasiva ou teste provocativo n&#227;o invasivo (resson&#226;ncia magn&#233;tica nuclear de perfus&#227;o e tomografia computadorizada por emiss&#227;o de f&#243;ton &#250;nico ou ecocardiografia de estresse com dobutamina), a crit&#233;rio do cardiologista assistente. No caso de um teste n&#227;o invasivo positivo, os pacientes realizavam angiografia para confirma&#231;&#227;o. Com base neste algoritmo diagn&#243;stico, a DAC obstrutiva foi definida como uma estenose &#8805; 70% &#224; angiografia. Um teste n&#227;o invasivo normal (tamanho do defeito isqu&#234;mico &lt; 5% do mioc&#225;rdio ventricular esquerdo) indicou a aus&#234;ncia de DAC obstrutiva e nenhum teste adicional foi necess&#225;rio. Independentemente de testes coronarianos, os pacientes foram classificados como apresentando ou n&#227;o DAC obstrutiva se um dos seguintes diagn&#243;sticos dominantes era confirmado por imagem: pericardite, embolia pulmonar, dissec&#231;&#227;o a&#243;rtica ou pneumonia. Secundariamente, o modelo foi testado para a predi&#231;&#227;o de morte dentro de 30 dias da admiss&#227;o.</p>
        </sec>
        <sec>
          <title>An&#225;lise estat&#237;stica</title>
          <p>A an&#225;lise estat&#237;stica est&#225; representada na <xref ref-type="fig" rid="f6">Figura 1</xref>. A amostra inicial de 370 pacientes consecutivos foi utilizada para deriva&#231;&#227;o do modelo. Em primeiro lugar, associa&#231;&#245;es univariadas entre DAC obstrutiva e caracter&#237;sticas basais foram testadas pelo teste <italic>t</italic> de Student n&#227;o pareado para vari&#225;veis num&#233;ricas e teste do qui-quadrado de Pearson para vari&#225;veis categ&#243;ricas. As vari&#225;veis num&#233;ricas sem distribui&#231;&#227;o normal foram expressas por mediana e intervalo interquartil e comparadas pelo teste n&#227;o param&#233;trico de Mann-Whitney. Em segundo lugar, vari&#225;veis com um valor de p &lt; 0,10 na an&#225;lise univariada foram inclu&#237;das na an&#225;lise de regress&#227;o log&#237;stica multivariada para a predi&#231;&#227;o de DAC obstrutiva.</p>
          <p>
            <fig id="f6">
              <label>Figura 1</label>
              <caption>
                <title>Fluxograma da an&#225;lise estat&#237;stica. DAC: doen&#231;a arterial coronariana.</title>
              </caption>
              <graphic xlink:href="1678-4170-abc-108-04-0304-70037-gf01-pt.jpg"/></fig>
          </p>
          <p>Modelos multivariados foram desenvolvidos pelo m&#233;todo <italic>stepwise</italic>, for&#231;ando todas as vari&#225;veis selecionadas na regress&#227;o e eliminando as menos significativas a cada etapa, de acordo com o teste estat&#237;stico Wald. Inicialmente, tr&#234;s modelos intermedi&#225;rios foram constru&#237;dos de acordo com o tipo de vari&#225;vel preditora (hist&#243;ria m&#233;dica, caracter&#237;sticas da dor tor&#225;cica ou exame f&#237;sico/testes laboratoriais). Preditores independentes (p &lt; 0,05) em cada modelo intermedi&#225;rio foram inclu&#237;dos como covari&#225;veis no modelo final. Este modelo final foi constru&#237;do hierarquicamente, com a ordem de imputa&#231;&#227;o das vari&#225;veis definida por racioc&#237;nio cl&#237;nico. A melhoria do modelo a cada etapa foi descrita pela diminui&#231;&#227;o do -2Log da probabilidade.</p>
          <p>A discrimina&#231;&#227;o foi avaliada pela &#225;rea sob a curva (<italic>area under the curve,</italic> AUC) caracter&#237;stica de opera&#231;&#227;o do receptor (<italic>receiver operating characteristic</italic>, ROC), enquanto a calibra&#231;&#227;o foi avaliada pelo teste de Hosmer-Lemeshow e pela correla&#231;&#227;o entre a preval&#234;ncia preditiva e observada da doen&#231;a de acordo com decis de predi&#231;&#227;o. O valor incremental do modelo completo em rela&#231;&#227;o ao modelo mais b&#225;sico foi avaliado atrav&#233;s de uma compara&#231;&#227;o das duas AUC pelo teste de DeLong. Al&#233;m disso, a <italic>melhoria da discrimina&#231;&#227;o integrada</italic> pelo modelo completo foi descrita de acordo com o m&#233;todo de Pencina.<xref ref-type="bibr" rid="B5">5</xref>
					</p>
          <p>Posteriormente, 100 pacientes consecutivos serviram como amostra de valida&#231;&#227;o. Nessa amostra, a discrimina&#231;&#227;o da DAC foi testada pela AUC. Uma vez que a an&#225;lise da calibra&#231;&#227;o por decis n&#227;o seria apropriada em uma amostra de 100 pacientes, a preval&#234;ncia observada de DAC foi comparada entre tercis de predi&#231;&#227;o da DAC. O valor incremental do modelo completo em rela&#231;&#227;o ao modelo mais b&#225;sico foi avaliado atrav&#233;s de uma compara&#231;&#227;o das duas AUC pelo teste de DeLong. A <italic>melhoria da discrimina&#231;&#227;o integrada</italic> pelo modelo completo foi tamb&#233;m descrita nesta amostra.</p>
          <p>Em uma an&#225;lise de sensibilidade, a amostra total de 470 pacientes foi utilizada para testar se o desempenho do modelo mudou de acordo com a presen&#231;a ou aus&#234;ncia de altera&#231;&#245;es eletrocardiogr&#225;ficas ou de troponina. Para esta an&#225;lise, um termo de intera&#231;&#227;o foi testado por regress&#227;o log&#237;stica. A amostra total foi tamb&#233;m utilizada para testar o valor progn&#243;stico do modelo. A AUC para predi&#231;&#227;o de mortalidade em 30 dias foi descrita e comparada com o escore GRACE<xref ref-type="bibr" rid="B6">6</xref> como representante de um modelo criado especificamente para inten&#231;&#227;o progn&#243;stica. O teste de DeLong foi utilizado para comparar as AUCs.</p>
          <p>A signific&#226;ncia estat&#237;stica foi definida como alfa &lt; 0,05. Para as vari&#225;veis num&#233;ricas com distribui&#231;&#227;o normal, foi utilizada m&#233;dia e desvio padr&#227;o, enquanto uma distribui&#231;&#227;o n&#227;o normal implicou no uso de mediana e intervalo interquartil. SPSS, vers&#227;o 21.0, foi o programa utilizado para a an&#225;lise estat&#237;stica.</p>
        </sec>
        <sec>
          <title>Escore de dor tor&#225;cica aguda</title>
          <p>A fim de gerar um escore de predi&#231;&#227;o de DAC, foram atribu&#237;dos pontos a cada vari&#225;vel positiva, proporcionais aos seus coeficientes de regress&#227;o no modelo final. A preval&#234;ncia de DAC obstrutiva foi descrita de acordo com decis de escores. Alternativamente, a f&#243;rmula de regress&#227;o final foi utilizada para criar uma calculadora log&#237;stica, oferecida como uma planilha do Excel (arquivo eletr&#244;nico) ou aplicativo para smartphones (Estar&#225; dispon&#237;vel num futuro pr&#243;ximo).</p>
        </sec>
        <sec>
          <title>Determina&#231;&#227;o do tamanho da amostra</title>
          <p>Como descrito acima, duas amostras consecutivas de pacientes foram selecionadas: um conjunto de deriva&#231;&#227;o e outro de valida&#231;&#227;o. Para o conjunto de deriva&#231;&#227;o, o tamanho da amostra foi planejado para permitir a inclus&#227;o de pelo menos 10 covari&#225;veis no modelo de regress&#227;o log&#237;stica. O c&#225;lculo foi baseado nos seguintes pressupostos: preval&#234;ncia de DAC obstrutiva de 30% e necessidade de 10 eventos para cada covari&#225;vel no modelo de regress&#227;o log&#237;stica.<xref ref-type="bibr" rid="B7">7</xref> Portanto, um m&#237;nimo de 300 pacientes seria necess&#225;rio e, como medida de seguran&#231;a, n&#243;s planejamos incluir um total de 370 indiv&#237;duos. A amostra de valida&#231;&#227;o foi definida para testar a acur&#225;cia discriminat&#243;ria pela an&#225;lise da curva ROC. Com base no pressuposto de uma AUC de 0,70, para oferecer um poder de 90% para rejeitar a hip&#243;tese nula de AUC igual a 0,50, com um alfa de 5%, um m&#237;nimo de 85 pacientes seria necess&#225;rio. Com isto, planejamos incluir 100 pacientes no conjunto de valida&#231;&#227;o.</p>
        </sec>
      </sec>
      <sec sec-type="results">
        <title>Resultados</title>
        <sec>
          <title>Popula&#231;&#227;o da amostra para o modelo de deriva&#231;&#227;o</title>
          <p>No total, 370 pacientes foram estudados, com m&#233;dia de idade de 60 &#177; 16 anos, 57% homens, 33% com hist&#243;ria pr&#233;via de doen&#231;a coronariana. A mediana do tempo decorrido entre o in&#237;cio dos sintomas e a primeira avalia&#231;&#227;o cl&#237;nica no hospital foi de 4 horas (intervalo interquartil = 1,8 - 13 horas). &#192; apresenta&#231;&#227;o, 52% dos pacientes apresentavam altera&#231;&#245;es isqu&#234;micas ao eletrocardiograma e 48% apresentavam troponina positiva. Investiga&#231;&#245;es adicionais, de acordo com o protocolo do estudo, identificaram DAC obstrutiva em 176 pacientes (preval&#234;ncia de 48%). Todos os casos tiveram confirma&#231;&#227;o diagn&#243;stica por angiografia coronariana invasiva. Em rela&#231;&#227;o aos 194 pacientes sem DAC, 74 foram classificados por angiografia negativa, 105 por um teste n&#227;o invasivo negativo e 15 tiveram outro diagn&#243;stico dominante (quatro com embolia pulmonar, dois com dissec&#231;&#227;o da aorta, sete com pericardite e dois com pneumonia).</p>
        </sec>
        <sec>
          <title>Preditores de DAC obstrutiva</title>
          <p>Entre as 13 vari&#225;veis relacionadas &#224; hist&#243;ria m&#233;dica, apenas quatro estiveram associadas &#224; DAC obstrutiva: idade mais avan&#231;ada, maior preval&#234;ncia do g&#234;nero masculino, hist&#243;ria pr&#233;via de DAC e uma tend&#234;ncia maior ao diabetes (<xref ref-type="table" rid="t4">Tabela 1</xref>). Quando estas quatro vari&#225;veis foram inclu&#237;das no modelo de regress&#227;o log&#237;stica, idade e sexo masculino permaneceram estatisticamente significativos (Modelo Intermedi&#225;rio 1) (<xref ref-type="table" rid="t5">Tabela 2</xref>).</p>
          <table-wrap id="t4">
            <label>Tabela 1</label>
            <caption>
              <title>Compara&#231;&#227;o da hist&#243;ria m&#233;dica, caracter&#237;sticas da dor tor&#225;cica e testes laboratoriais entre pacientes com e sem doen&#231;a arterial coronariana obstrutiva</title>
            </caption>
            <table frame="hsides" rules="groups">
              <colgroup>
                <col width="25%"/>
                <col width="25%"/>
                <col width="25%"/>
                <col width="25%"/>
              </colgroup>
              <thead>
                <tr style="border-bottom-width:thin;border-bottom-style:solid">
                  <th align="center" colspan="4">Doen&#231;a Coronariana Obstrutiva</th>
                </tr>
                <tr>
                  <th align="center">
									</th>
                  <th align="center">Sim (n = 176)</th>
                  <th align="center">N&#227;o (n = 194)</th>
                  <th align="center">Valor de p</th>
                </tr>
              </thead>
              <tbody>
                <tr>
                  <td align="left" colspan="4">
                    <bold>Hist&#243;rico M&#233;dico</bold>
                  </td>
                </tr>
                <tr>
                  <td align="left">Idade (anos)</td>
                  <td align="center">63 &#177; 14</td>
                  <td align="center">57 &#177; 16</td>
                  <td align="center">&lt; 0,001</td>
                </tr>
                <tr>
                  <td align="left">G&#234;nero masculino</td>
                  <td align="center">121 (69%)</td>
                  <td align="center">90 (46%)</td>
                  <td align="center">&lt; 0,001</td>
                </tr>
                <tr>
                  <td align="left">&#205;ndice de massa corporal (kg/m2)</td>
                  <td align="center">28 &#177; 4,8</td>
                  <td align="center">28 &#177; 5,9</td>
                  <td align="center">0,61</td>
                </tr>
                <tr>
                  <td align="left">Hist&#243;ria de DAC</td>
                  <td align="center">68 (39%)</td>
                  <td align="center">55 (28%)</td>
                  <td align="center">0,03</td>
                </tr>
                <tr>
                  <td align="left">Diabetes</td>
                  <td align="center">62 (36%)</td>
                  <td align="center">51 (26%)</td>
                  <td align="center">0,05</td>
                </tr>
                <tr>
                  <td align="left">Hipertens&#227;o</td>
                  <td align="center">122 (70%)</td>
                  <td align="center">138 (71%)</td>
                  <td align="center">0,83</td>
                </tr>
                <tr>
                  <td align="left">Tabagismo atual</td>
                  <td align="center">22 (13%)</td>
                  <td align="center">18 (9,3%)</td>
                  <td align="center">0,30</td>
                </tr>
                <tr>
                  <td align="left">LDL-colesterol (mg/dL)</td>
                  <td align="center">113 &#177; 64</td>
                  <td align="center">116 &#177; 87</td>
                  <td align="center">0,72</td>
                </tr>
                <tr>
                  <td align="left">Hist&#243;ria familiar de DAC</td>
                  <td align="center">48 (28%)</td>
                  <td align="center">42 (22%)</td>
                  <td align="center">0,19</td>
                </tr>
                <tr>
                  <td align="left">Doen&#231;a renal cr&#244;nica</td>
                  <td align="center">9 (5,3%)</td>
                  <td align="center">7 (3,6%)</td>
                  <td align="center">0,45</td>
                </tr>
                <tr>
                  <td align="left">Creatinina plasm&#225;tica (mg/dL)</td>
                  <td align="center">0,95 (0,80 - 1,20)</td>
                  <td align="center">0,80 (0,70 - 1,15)</td>
                  <td align="center">0,10</td>
                </tr>
                <tr>
                  <td align="left">Tratamento atual com estatina</td>
                  <td align="center">85 (49%)</td>
                  <td align="center">91 (47%)</td>
                  <td align="center">0,71</td>
                </tr>
                <tr>
                  <td align="left">Terapia atual com AAS</td>
                  <td align="center">75 (43%)</td>
                  <td align="center">76 (39%)</td>
                  <td align="center">0,44</td>
                </tr>
                <tr>
                  <td align="left" colspan="4">
                    <bold>Caracter&#237;sticas da Dor Tor&#225;cica</bold>
                  </td>
                </tr>
                <tr>
                  <td align="left">Localiza&#231;&#227;o no lado esquerdo</td>
                  <td align="center">137 (79%)</td>
                  <td align="center">156 (81%)</td>
                  <td align="center">0,70</td>
                </tr>
                <tr>
                  <td align="left">Natureza opressiva</td>
                  <td align="center">97 (57%)</td>
                  <td align="center">95 (49%)</td>
                  <td align="center">0,14</td>
                </tr>
                <tr>
                  <td align="left">Irradia&#231;&#227;o para o pesco&#231;o</td>
                  <td align="center">39 (23%)</td>
                  <td align="center">51 (26%)</td>
                  <td align="center">0,42</td>
                </tr>
                <tr>
                  <td align="left">Irradia&#231;&#227;o para o bra&#231;o esquerdo</td>
                  <td align="center">57 (33%)</td>
                  <td align="center">53 (27%)</td>
                  <td align="center">0,24</td>
                </tr>
                <tr>
                  <td align="left">Sintomas vagais</td>
                  <td align="center">61 (36%)</td>
                  <td align="center">78 (40%)</td>
                  <td align="center">0,35</td>
                </tr>
                <tr>
                  <td align="left">N&#250;mero de epis&#243;dios</td>
                  <td align="center">1 (1 - 2)</td>
                  <td align="center">1 (1 - 3)</td>
                  <td align="center">0,81</td>
                </tr>
                <tr>
                  <td align="left">Dura&#231;&#227;o (minutos)</td>
                  <td align="center">40 (15 - 120)</td>
                  <td align="center">40 (10 - 150)</td>
                  <td align="center">0,82</td>
                </tr>
                <tr>
                  <td align="left">Intensidade (escala de 1 - 10)</td>
                  <td align="center">7,4 &#177; 2,5</td>
                  <td align="center">7,1 &#177; 2,6</td>
                  <td align="center">0,31</td>
                </tr>
                <tr>
                  <td align="left">Al&#237;vio com nitrato</td>
                  <td align="center">84 (50%)</td>
                  <td align="center">72 (37%)</td>
                  <td align="center">0,02</td>
                </tr>
                <tr>
                  <td align="left">Semelhante a infarto anterior</td>
                  <td align="center">70 (42%)</td>
                  <td align="center">63 (33%)</td>
                  <td align="center">0,08</td>
                </tr>
                <tr>
                  <td align="left">Piora com compress&#227;o</td>
                  <td align="center">7 (4,1%)</td>
                  <td align="center">26 (13%)</td>
                  <td align="center">0,002</td>
                </tr>
                <tr>
                  <td align="left">Piora com a posi&#231;&#227;o</td>
                  <td align="center">24 (14%)</td>
                  <td align="center">36 (19%)</td>
                  <td align="center">0,23</td>
                </tr>
                <tr>
                  <td align="left">Piora com movimento do bra&#231;o</td>
                  <td align="center">7 (4,0%)</td>
                  <td align="center">16 (8,2%)</td>
                  <td align="center">0,097</td>
                </tr>
                <tr>
                  <td align="left">Piora com inspira&#231;&#227;o profunda</td>
                  <td align="center">13 (7,5%)</td>
                  <td align="center">36 (19%)</td>
                  <td align="center">0,002</td>
                </tr>
                <tr>
                  <td align="left" colspan="4">
                    <bold>Testes Laboratoriais &#224; Admiss&#227;o</bold>
                  </td>
                </tr>
                <tr>
                  <td align="left">Altera&#231;&#245;es isqu&#234;micas ao ECG</td>
                  <td align="center">120 (68%)</td>
                  <td align="center">73 (38%)</td>
                  <td align="center">&lt; 0,001</td>
                </tr>
                <tr>
                  <td align="left">Troponina positiva</td>
                  <td align="center">116 (66%)</td>
                  <td align="center">60 (31%)</td>
                  <td align="center">&lt; 0,001</td>
                </tr>
                <tr>
                  <td align="left">Sinais cl&#237;nicos e radiogr&#225;ficos de IVE</td>
                  <td align="center">26 (15%)</td>
                  <td align="center">5 (2,6%)</td>
                  <td align="center">&lt; 0,001</td>
                </tr>
                <tr>
                  <td align="left">NT-proBNP (pg/mL)</td>
                  <td align="center">363 (105 - 1850)</td>
                  <td align="center">57 (20 - 235)</td>
                  <td align="center">&lt; 0,001</td>
                </tr>
                <tr>
                  <td align="left">Glicose plasm&#225;tica (mg/dL)</td>
                  <td align="center">120 (97 - 189)</td>
                  <td align="center">112 (92 - 145)</td>
                  <td align="center">0,22</td>
                </tr>
                <tr>
                  <td align="left">Prote&#237;na C reativa (mg/L)</td>
                  <td align="center">7,3 (2,3 - 15)</td>
                  <td align="center">5,7 (1,4 - 15)</td>
                  <td align="center">0,09</td>
                </tr>
                <tr>
                  <td align="left">Contagem de c&#233;lulas brancas</td>
                  <td align="center">8,790 &#177; 4,300</td>
                  <td align="center">7,701 &#177; 2,865</td>
                  <td align="center">0,004</td>
                </tr>
                <tr>
                  <td align="left">Hemoglobina (g/dL)</td>
                  <td align="center">14,1 &#177; 1,9</td>
                  <td align="center">13,7 &#177; 1,7</td>
                  <td align="center">0,06</td>
                </tr>
              </tbody>
            </table>
            <table-wrap-foot>
              <fn id="TFN4">
                <p>DAC: doen&#231;a arterial coronariana; IVE: insufici&#234;ncia ventricular esquerda. ECG: eletrocardiograma. Hist&#243;ria familiar de DAC implica na ocorr&#234;ncia da doen&#231;a em um parente de primeiro grau antes da idade de 55 anos (se do sexo feminino) ou 45 anos (se do sexo masculino)</p>
              </fn>
            </table-wrap-foot>
          </table-wrap>
          <table-wrap id="t5">
            <label>Tabela 2</label>
            <caption>
              <title>Modelos intermedi&#225;rios de regress&#227;o log&#237;stica da hist&#243;ria m&#233;dica (Modelo 1), caracter&#237;sticas da dor tor&#225;cica (Modelo 2) e testes laboratoriais (Modelo 3)</title>
            </caption>
            <table frame="hsides" rules="groups">
              <colgroup>
                <col width="50%"/>
                <col width="50%"/>
              </colgroup>
              <thead>
                <tr>
                  <th align="left">Vari&#225;veis</th>
                  <th align="center">N&#237;vel de signific&#226;ncia multivariada</th>
                </tr>
              </thead>
              <tbody>
                <tr>
                  <td align="left" colspan="2">
                    <bold>Modelo 1 (hist&#243;ria m&#233;dica)</bold>
                  </td>
                </tr>
                <tr>
                  <td align="left">G&#234;nero masculino</td>
                  <td align="center">&lt; 0,001</td>
                </tr>
                <tr>
                  <td align="left">Idade (anos)</td>
                  <td align="center">&lt; 0,001</td>
                </tr>
                <tr>
                  <td align="left">Diabetes</td>
                  <td align="center">0,10</td>
                </tr>
                <tr>
                  <td align="left">HDL-colesterol</td>
                  <td align="center">0,35</td>
                </tr>
                <tr>
                  <td align="left">DAC pr&#233;via</td>
                  <td align="center">0,84</td>
                </tr>
                <tr>
                  <td align="left">Creatinina plasm&#225;tica (mg/dL)</td>
                  <td align="center">0,95</td>
                </tr>
                <tr>
                  <td align="left" colspan="2">
                    <bold>Modelo 2 (caracter&#237;sticas da dor)</bold>
                  </td>
                </tr>
                <tr>
                  <td align="left">Sens&#237;vel &#224; compress&#227;o manual</td>
                  <td align="center">0,024</td>
                </tr>
                <tr>
                  <td align="left">Sens&#237;vel &#224; respira&#231;&#227;o profunda</td>
                  <td align="center">0,037</td>
                </tr>
                <tr>
                  <td align="left">Al&#237;vio com nitrato</td>
                  <td align="center">0,045</td>
                </tr>
                <tr>
                  <td align="left">Semelhante a IAM anterior</td>
                  <td align="center">0,17</td>
                </tr>
                <tr>
                  <td align="left">Sens&#237;vel &#224; movimenta&#231;&#227;o do bra&#231;o</td>
                  <td align="center">0,57</td>
                </tr>
                <tr>
                  <td align="left" colspan="2">
                    <bold>Modelo 3 (exames laboratoriais)</bold>
                  </td>
                </tr>
                <tr>
                  <td align="left">Altera&#231;&#245;es isqu&#234;micas ao ECG</td>
                  <td align="center">&lt; 0,001</td>
                </tr>
                <tr>
                  <td align="left">Troponina positiva</td>
                  <td align="center">&lt; 0,001</td>
                </tr>
                <tr>
                  <td align="left">Sinais cl&#237;nicos ou radiogr&#225;ficos de IVE</td>
                  <td align="center">0,016</td>
                </tr>
                <tr>
                  <td align="left">Contagem de c&#233;lulas brancas</td>
                  <td align="center">0,29</td>
                </tr>
                <tr>
                  <td align="left">Hemoglobina (g/dL)</td>
                  <td align="center">0,67</td>
                </tr>
                <tr>
                  <td align="left">NT-proBNP (pg/mL)</td>
                  <td align="center">0,81</td>
                </tr>
                <tr>
                  <td align="left">Prote&#237;na C reativa (mg/L)</td>
                  <td align="center">0,70</td>
                </tr>
              </tbody>
            </table>
            <table-wrap-foot>
              <fn id="TFN5">
                <p>IAM: infarto agudo do mioc&#225;rio; CAD: doen&#231;a arterial coronariana; IVE: insufici&#234;ncia ventricular esquerda; ECG: eletrocardiograma.</p>
              </fn>
            </table-wrap-foot>
          </table-wrap>
          <p>Quanto &#224;s caracter&#237;sticas de dor tor&#225;cica, apenas cinco entre 14 vari&#225;veis tiveram associa&#231;&#227;o com DAC: al&#237;vio com nitratos e similaridade com infarto agudo do mioc&#225;rdio pr&#233;vio. Por outro lado, uma piora com compress&#227;o manual, inspira&#231;&#227;o profunda ou movimento do bra&#231;o foi, cada qual, mais comum em pacientes sem DAC (<xref ref-type="table" rid="t4">Tabela 1</xref>). Destes, al&#237;vio com nitratos e piora com a compress&#227;o manual e com a inspira&#231;&#227;o profunda foram os tr&#234;s preditores independentes no Modelo Intermedi&#225;rio 2 (<xref ref-type="table" rid="t5">Tabela 2</xref>).</p>
          <p>Entre o exame f&#237;sico e os testes laboratoriais, a maioria das vari&#225;veis estiveram associadas com DAC: eletrocardiograma isqu&#234;mico, troponina positiva e sinais de insufici&#234;ncia card&#237;aca esquerda foram mais prevalentes em pacientes com DAC. Al&#233;m disso, quatro vari&#225;veis num&#233;ricas apresentaram valores mais elevados em pacientes com DAC: NT-proBNP, PCR, contagem de c&#233;lulas brancas e hemoglobina (<xref ref-type="table" rid="t4">Tabela 1</xref>). No Modelo Intermedi&#225;rio 3, os preditores independentes foram eletrocardiograma isqu&#234;mico, troponina positiva e sinais de insufici&#234;ncia card&#237;aca esquerda (<xref ref-type="table" rid="t5">Tabela 2</xref>).</p>
        </sec>
        <sec>
          <title>Desenvolvimento de um modelo para predi&#231;&#227;o de DAC</title>
          <p>As oito vari&#225;veis independentemente associadas com DAC nos Modelos Intermedi&#225;rios 1, 2 e 3 foram candidatas ao modelo final, que foi constru&#237;do hierarquicamente em sete etapas definidas por racioc&#237;nio cl&#237;nico: a primeira etapa incluiu juntamente eletrocardiograma e troponina, seguida pela segunda etapa que incluiu insufici&#234;ncia ventricular esquerda. Estas duas primeiras etapas representaram a gravidade da apresenta&#231;&#227;o cl&#237;nica. As etapas terceira e quarta representaram as caracter&#237;sticas intr&#237;nsecas dos pacientes, idade e g&#234;nero. A quinta, sexta e s&#233;tima etapa estiveram relacionadas com as caracter&#237;sticas da dor tor&#225;cica, que foram escolhidas para serem &#250;ltimas devido &#224; sua subjetividade na pr&#225;tica cl&#237;nica.</p>
          <p>A primeiro etapa do eletrocardiograma e da troponina apresentou um -2Log de probabilidade de 437 (&#967;<sup>2</sup> = 69, p &lt; 0,001), que na sequ&#234;ncia melhorou com a inclus&#227;o de insufici&#234;ncia ventricular esquerda (-2Log de probabilidade = 427, &#967;<sup>2</sup> = 9,8, p = 0,002), idade (-2Log de probabilidade = 422, &#967;<sup>2</sup> = 4,9, p = 0,02), g&#234;nero (-2Log de probabilidade = 401, &#967;<sup>2</sup> = 21, p &lt; 0,001) e al&#237;vio com nitratos (-2Log de probabilidade = 394, &#967;<sup>2</sup> = 6,8, p = 0,009). A inclus&#227;o de piora com compress&#227;o manual (-2Log de probabilidade = 391, &#967;<sup>2</sup> = 3,2, p = 0,07) e piora com inspira&#231;&#227;o profunda (-2Log de probabilidade = 389, &#967;<sup>2</sup> = 2,3, p = 0,13) n&#227;o promoveu melhora adicional ao modelo. Por conseguinte, as primeiras seis vari&#225;veis constitu&#237;ram o modelo final.</p>
          <p>O modelo final apresentou boa discrimina&#231;&#227;o, com uma AUC de 0,80 (IC95% = 0,75 - 0,84) (<xref ref-type="fig" rid="f7">Figura 2A</xref>). O &#967;<sup>2</sup>de Hosmer-Lemeshow de 1,95 indicou que o modelo estava bem calibrado (p = 0,98), como mostra o gr&#225;fico de dispers&#227;o da probabilidade preditiva <italic>versus</italic> a preval&#234;ncia observada de DAC por decis (r = 0,99) (<xref ref-type="fig" rid="f7">Figura 2B</xref>). A probabilidade de DAC de acordo com o modelo final variou de um m&#237;nimo de 3% a um m&#225;ximo de 98%, com pacientes igualmente distribu&#237;dos ao longo das probabilidades. Raz&#245;es de chance, IC95% e coeficientes de regress&#227;o, al&#233;m dos valores de p do modelo final est&#227;o representados na <xref ref-type="table" rid="t6">Tabela 3</xref>.</p>
          <p>
            <fig id="f7">
              <label>Figura 2</label>
              <caption>
                <title>An&#225;lise do modelo discriminat&#243;rio e calibra&#231;&#227;o na amostra de deriva&#231;&#227;o de 370 pacientes. O Painel A mostra uma AUC significativa do modelo probabil&#237;stico na predi&#231;&#227;o de doen&#231;a arterial coronariana (DAC) obstrutiva. O Painel B mostra uma correla&#231;&#227;o significativa entre a probabilidade preditora e observada de DAC. AUC denota &#225;rea sob a curva caracter&#237;stica de opera&#231;&#227;o do receptor (receiver operating characteristic).</title>
              </caption>
              <graphic xlink:href="1678-4170-abc-108-04-0304-70037-gf02-pt.jpg"/></fig>
          </p>
          <table-wrap id="t6">
            <label>Tabela 3</label>
            <caption>
              <title>Modelo final de regress&#227;o log&#237;stica definindo os preditores independentes de doen&#231;a arterial coronariana obstrutiva</title>
            </caption>
            <table frame="hsides" rules="groups">
              <colgroup>
                <col width="25%"/>
                <col width="25%"/>
                <col width="25%"/>
                <col width="25%"/>
              </colgroup>
              <thead>
                <tr>
                  <th align="left">Vari&#225;veis</th>
                  <th align="center">Beta</th>
                  <th align="center">Raz&#227;o de chances (IC95%)</th>
                  <th align="center">Valor de p</th>
                </tr>
              </thead>
              <tbody>
                <tr>
                  <td align="left">Idade (cada ano)</td>
                  <td align="center">0,025</td>
                  <td align="center">1,03 (1,01 - 1,04)</td>
                  <td align="center">0,003</td>
                </tr>
                <tr>
                  <td align="left">Al&#237;vio com nitratos</td>
                  <td align="center">0,60</td>
                  <td align="center">1,8 (1,1 - 3,0)</td>
                  <td align="center">0,016</td>
                </tr>
                <tr>
                  <td align="left">ECG isqu&#234;mico</td>
                  <td align="center">1,10</td>
                  <td align="center">3,0 (1,9 - 4,9)</td>
                  <td align="center">&lt; 0,001</td>
                </tr>
                <tr>
                  <td align="left">Troponina positiva</td>
                  <td align="center">1,15</td>
                  <td align="center">3,2 (1,9 - 5,1)</td>
                  <td align="center">&lt; 0,001</td>
                </tr>
                <tr>
                  <td align="left">G&#234;nero masculino</td>
                  <td align="center">1,16</td>
                  <td align="center">3,2 (1,9 - 5,3)</td>
                  <td align="center">&lt; 0,001</td>
                </tr>
                <tr>
                  <td align="left">Sinais de IVE</td>
                  <td align="center">1,55</td>
                  <td align="center">4,7 (1,6 - 14)</td>
                  <td align="center">0,004</td>
                </tr>
                <tr>
                  <td align="left">Sens&#237;vel &#224; respira&#231;&#227;o profunda</td>
                  <td align="center">----</td>
                  <td align="center">----</td>
                  <td align="center">0,06</td>
                </tr>
                <tr>
                  <td align="left">Sens&#237;vel &#224; compress&#227;o manual</td>
                  <td align="center">----</td>
                  <td align="center">----</td>
                  <td align="center">0,18</td>
                </tr>
              </tbody>
            </table>
            <table-wrap-foot>
              <fn id="TFN6">
                <p>IVE: insufici&#234;ncia ventricular esquerda; ECG: eletrocardiograma.</p>
              </fn>
            </table-wrap-foot>
          </table-wrap>
        </sec>
        <sec>
          <title>Valor incremental do modelo completo</title>
          <p>A AUC melhorou de 0,73 no primeiro modelo contendo apenas eletrocardiograma e troponina para 0,80 no modelo completo (IC95% de diferen&#231;a entre as &#225;reas = 0,03 - 0,10, p = 0,0002). A discrimina&#231;&#227;o melhorou progressivamente &#224; medida que vari&#225;veis foram adicionadas: a AUC foi de 0,74 no segundo modelo (com adi&#231;&#227;o de insufici&#234;ncia ventricular esquerda), 0,76 no terceiro modelo (com adi&#231;&#227;o de idade) e 0,79 no quarto modelo (com adi&#231;&#227;o de g&#234;nero). A melhoria discriminat&#243;ria integrada fornecida pelo modelo completo em rela&#231;&#227;o ao primeiro modelo foi de 0,09 (p &lt; 0,001), um resultado de 0,05 de aumento m&#233;dio de probabilidades no grupo com eventos mais 0,04 de redu&#231;&#227;o m&#233;dia de probabilidades no grupo livre de eventos.</p>
        </sec>
        <sec>
          <title>Valida&#231;&#227;o por amostra independente</title>
          <p>A amostra de valida&#231;&#227;o consistiu de 100 indiv&#237;duos, 62% dos quais homens, com m&#233;dia de idade de 60 &#177; 13 anos e uma preval&#234;ncia de 59% de DAC obstrutiva. Neste grupo, a AUC foi de 0,86 (IC95% = 0,79 - 0,93) e o &#967;<sup>2</sup> da calibra&#231;&#227;o de Hosmer-Lemeshow foi de 10,1 (p = 0,26) (<xref ref-type="fig" rid="f8">Figura 3A</xref>). Como o grupo foi dividido em tercis de probabilidade preditiva do modelo (&lt; 30%, 30 - 60%, &gt; 60%), um aumento progressivo na preval&#234;ncia da doen&#231;a foi observado (24%, 59% e 94%, respectivamente, p para tend&#234;ncia &lt; 0,001) (<xref ref-type="fig" rid="f8">Figura 3B</xref>).</p>
          <p>
            <fig id="f8">
              <label>Figura 3</label>
              <caption>
                <title>An&#225;lise do desempenho do modelo na amostra de valida&#231;&#227;o independente de 100 pacientes. O Painel A mostra uma AUC significativa do modelo probabil&#237;stico na predi&#231;&#227;o de doen&#231;a arterial coronariana (DAC) obstrutiva. O Painel B indica um aumento progressivo na preval&#234;ncia de DAC de acordo com tercis de predi&#231;&#227;o do modelo. AUC denota &#225;rea sob a curva caracter&#237;stica de opera&#231;&#227;o do receptor (receiver operating characteristic)</title>
              </caption>
              <graphic xlink:href="1678-4170-abc-108-04-0304-70037-gf03-pt.jpg"/></fig>
          </p>
          <p>Em compara&#231;&#227;o ao modelo b&#225;sico contendo apenas eletrocardiograma e troponina (AUC = 0,78), o incremento oferecido pelo modelo completo foi de +0,07 (IC95% de diferen&#231;a entre as &#225;reas = 0,004 - 0,14, p = 0,039). A melhoria discriminat&#243;ria integrada fornecida pelo modelo completo em rela&#231;&#227;o ao primeiro modelo foi de 0,09 (p &lt; 0,0015), um resultado de 0,02 de aumento m&#233;dio de probabilidades no grupo com eventos mais 0,07 de redu&#231;&#227;o m&#233;dia de probabilidades no grupo livre de eventos. </p>
        </sec>
        <sec>
          <title>Sensibilidade do modelo final ao eletrocardiograma e &#224; troponina</title>
          <p>A amostra total de 470 pacientes foi utilizada para testar a sensibilidade do modelo ao eletrocardiograma e &#224; troponina. N&#227;o houve intera&#231;&#227;o entre o modelo de predi&#231;&#227;o e a presen&#231;a (ou aus&#234;ncia) de altera&#231;&#245;es eletrocardiogr&#225;ficas/troponina (p = 0,48), significando que o desempenho do modelo n&#227;o foi modificado por essas vari&#225;veis. A AUC do modelo de indiv&#237;duos com eletrocardiograma e troponina normais (n = 147, 24% com DAC) foi de 0,74 (IC95% = 0,65 - 0,83), enquanto os indiv&#237;duos com anormalidade em um destes testes (n = 323, 62% de DAC) apresentaram uma AUC de 0,77 (IC95% = 0,71 - 0,82).</p>
        </sec>
        <sec>
          <title>Valor progn&#243;stico para mortalidade em 30 dias</title>
          <p>Em toda a amostra de 470 pacientes, 10 pacientes (2,1%) morreram nos primeiros 30 dias a partir da dor tor&#225;cica inicial, oito durante a interna&#231;&#227;o e dois ap&#243;s a alta hospitalar. A habilidade do modelo em predizer &#243;bito foi mostrada por uma AUC de 0,74 (IC95% = 0,61 - 0,87), semelhante ao valor progn&#243;stico do escore GRACE de 0,72 (IC95% = 0,54 - 0,91, p = 0,83) (<xref ref-type="fig" rid="f9">Figura 4A</xref>). N&#227;o houve &#243;bito no primeiro tercil da amostra total (probabilidade de DAC &lt; 30%), enquanto que tr&#234;s &#243;bitos ocorreram no segundo tercil (30 - 62%) e sete &#243;bitos no terceiro tercil (&gt; 62%, p para tend&#234;ncia = 0,006) (<xref ref-type="fig" rid="f9">Figura 4B</xref>).</p>
          <p>
            <fig id="f9">
              <label>Figura 4</label>
              <caption>
                <title>An&#225;lise da mortalidade na amostra total de 470 pacientes mostrando valor progn&#243;stico significativo do modelo, que foi originalmente derivado para predi&#231;&#227;o de doen&#231;a arterial coronariana(DAC). O Painel A compara o &#237;ndice C do modelo versus o escore GRACE, indicando predi&#231;&#227;o semelhante. O Painel B compara a incid&#234;ncia de DAC entre tercis de modelo de predi&#231;&#227;o de doen&#231;a coronariana. AUC denota &#225;rea sob a curva caracter&#237;stica de opera&#231;&#227;o do receptor (receiver operating characteristic).</title>
              </caption>
              <graphic xlink:href="1678-4170-abc-108-04-0304-70037-gf04-pt.jpg"/></fig>
          </p>
        </sec>
        <sec>
          <title>Escore de dor tor&#225;cica aguda</title>
          <p>Pontos proporcionais aos coeficientes de regress&#227;o foram atribu&#237;dos a cada vari&#225;vel positiva: idade (&#946; = 0,025; 0,05 ponto para cada ano), al&#237;vio com nitratos (&#946; = 0,60; 1 ponto), g&#234;nero masculino (&#946; = 1,16; 2 pontos), eletrocardiograma isqu&#234;mico (&#946; = 1.10; 2 pontos), troponina positiva (&#946; = 1.15; 2 pontos) e sinais de insufici&#234;ncia ventricular esquerda (&#946; = 1,55; 3 pontos). O escore apresentou a mesma AUC do que o modelo log&#237;stico. Houve um aumento proporcional na preval&#234;ncia da doen&#231;a de acordo com decis de escore: 11%, 14%, 24%, 37%, 41%, 53%, 59%, 67%, 74% e 95% (p para tend&#234;ncia linear &lt; 0,001) (<xref ref-type="fig" rid="f10">Figura 5</xref>).</p>
          <p>
            <fig id="f10">
              <label>Figura 5</label>
              <caption>
                <title>Preval&#234;ncia de doen&#231;a arterial coronariana (DAC) obstrutiva de acordo com decis de escore.</title>
              </caption>
              <graphic xlink:href="1678-4170-abc-108-04-0304-70037-gf05-pt.jpg"/></fig>
          </p>
        </sec>
      </sec>
      <sec sec-type="discussion">
        <title>Discuss&#227;o</title>
        <p>O presente estudo desenvolveu e validou um modelo probabil&#237;stico para predi&#231;&#227;o de DAC obstrutiva, baseado em dados de apresenta&#231;&#227;o inicial de dor tor&#225;cica aguda. De um total de 35 vari&#225;veis candidatas, um modelo final de seis preditores independentes foi gerado, com boa discrimina&#231;&#227;o e calibra&#231;&#227;o para avaliar a probabilidade pr&#233;-teste da doen&#231;a. Mais importante ainda, a acur&#225;cia do modelo provou ser superior ao modelo tradicional que utiliza eletrocardiograma e troponina.</p>
        <p>A indica&#231;&#227;o de testes diagn&#243;sticos deve levar em conta a probabilidade pr&#233;-teste da doen&#231;a. No entanto, no cen&#225;rio selecionado de unidades coronarianas, virtualmente todos os pacientes com dor tor&#225;cica indefinida s&#227;o submetidos a testes para detec&#231;&#227;o de DAC obstrutiva, independentemente da probabilidade pr&#233;-teste. Uma vez que o teste ser&#225; negativo em uma propor&#231;&#227;o significativa de pacientes,<xref ref-type="bibr" rid="B2">2</xref> esta abordagem conduz a uma estadia hospitalar prolongada desnecess&#225;ria. Assim, a elimina&#231;&#227;o da necessidade de testes adicionais em pacientes com baixa probabilidade de DAC ir&#225; melhorar a efici&#234;ncia dos protocolos de dor tor&#225;cica. No entanto, modelos probabil&#237;sticos validados n&#227;o s&#227;o disseminados neste cen&#225;rio cl&#237;nico, tornando dif&#237;cil para o m&#233;dico de emerg&#234;ncia adaptar a decis&#227;o m&#233;dica baseada em probabilidade. A probabilidade &#233;, no m&#225;ximo, avaliada de uma forma bin&#225;ria, baseada em altera&#231;&#227;o do eletrocardiograma ou troponina.</p>
        <p>A utiliza&#231;&#227;o de tal modelo de probabilidade melhora a acur&#225;cia e oferece uma gama cont&#237;nua de probabilidades, aproximando o pensamento m&#233;dico &#224; melhor forma de lidar com a incerteza. Como disse uma vez William Osler, "a medicina &#233; a ci&#234;ncia da incerteza e a arte da probabilidade".</p>
        <p>O nosso objetivo em predizer DAC obstrutiva n&#227;o deve ser confundido com estudos anteriores que desenvolveram modelos neurais ou log&#237;sticos para predizer o diagn&#243;stico cl&#237;nico do infarto do mioc&#225;rdio em pacientes com dor tor&#225;cica.<xref ref-type="bibr" rid="B8">8</xref><sup>-</sup><xref ref-type="bibr" rid="B12">12</xref> Esses estudos criaram modelos a partir de dados cl&#237;nicos, caracter&#237;sticas de sintomas e, por vezes, eletrocardiograma, que foram testados como preditores de um diagn&#243;stico final definido por uma an&#225;lise sistem&#225;tica das mesmas vari&#225;veis, al&#233;m de marcadores de necrose mioc&#225;rdica. Por conseguinte, estes modelos matem&#225;ticos servem principalmente como substitutos ao pensamento m&#233;dico ou, no m&#225;ximo, preditores de uma impress&#227;o final que ser&#225; obtida poucas horas ap&#243;s apresenta&#231;&#227;o inicial. Por outro lado, nosso modelo foi constru&#237;do para predizer o resultado de exames de imagem antes de serem realizados. Uma vez que exames de imagem n&#227;o invasivos ou invasivos objetivam o diagn&#243;stico de DAC obstrutiva, um modelo deste tipo &#233; claramente &#250;til para a sele&#231;&#227;o eficiente de pacientes para esses testes, com base na estimativa da probabilidade pr&#233;-teste da doen&#231;a. Al&#233;m disso, o conhecimento de uma probabilidade pr&#233;-teste permite o c&#225;lculo da probabilidade p&#243;s-teste ap&#243;s a obten&#231;&#227;o de um resultado de imagem n&#227;o invasivo.</p>
        <p>Outros escores est&#227;o focados no risco de eventos adversos (escore HEART,<xref ref-type="bibr" rid="B13">13</xref> escore TIMI<xref ref-type="bibr" rid="B14">14</xref> ou escore GRACE<xref ref-type="bibr" rid="B6">6</xref>). Apesar dos seus valores progn&#243;sticos, eles n&#227;o s&#227;o necessariamente bons preditores de DAC obstrutiva<xref ref-type="bibr" rid="B15">15</xref> e os m&#233;dicos se sentem desconfort&#225;veis em dar alta a um paciente com dor tor&#225;cica aguda sem testes adicionais. Assim, acreditamos que o c&#225;lculo da probabilidade de DAC obstrutiva incentivaria os m&#233;dicos a reduzirem a utiliza&#231;&#227;o excessiva de estudos de imagem em pacientes com baixa probabilidade, diminuindo os fen&#244;menos de excesso de diagn&#243;stico e excesso de tratamento. Por exemplo, sabe-se que pacientes com eletrocardiograma normal e troponina negativa t&#234;m um bom progn&#243;stico. Em nosso estudo, 50% desses pacientes apresentaram uma probabilidade de DAC significativa abaixo de 20%. Com base em probabilidades de diagn&#243;stico e progn&#243;stico favor&#225;veis, estes pacientes poderiam receber alta sem testes adicionais. Por outro lado, pacientes com eletrocardiograma e troponina normais podem ter uma probabilidade significativa de DAC que pode ser detectada pelo modelo. Devemos salientar que ensaios cl&#237;nicos randomizados futuros devem validar a efici&#234;ncia e seguran&#231;a desta abordagem.</p>
        <p>Os m&#233;dicos normalmente se baseiam nas caracter&#237;sticas dos sintomas (t&#237;picas ou at&#237;picas) e em fatores de risco tradicionais para estimar a probabilidade de DAC em pacientes com dor tor&#225;cica aguda. Por exemplo, um paciente diab&#233;tico com dor tor&#225;cica t&#237;pica &#233; geralmente definido como tendo uma alta probabilidade de DAC. No entanto, em nosso estudo, nenhum fator de risco ou caracter&#237;stica da dor tor&#225;cica (exceto para o al&#237;vio de nitrato) emergiu como preditor independente de DAC. Isto est&#225; de acordo com estudos anteriores que indicam que o tipo de apresenta&#231;&#227;o tem pouca influ&#234;ncia sobre o diagn&#243;stico no cen&#225;rio agudo. Em uma revis&#227;o sistem&#225;tica abrangente, Swap e Nagurney<xref ref-type="bibr" rid="B16">16</xref> mostraram baixas raz&#245;es de verossimilhan&#231;a para caracter&#237;sticas da dor tor&#225;cica. Aparentemente, um artigo recente de Khan et al.<xref ref-type="bibr" rid="B17">17</xref> demonstraram que a maioria das caracter&#237;sticas da dor n&#227;o est&#225; associada com doen&#231;a coronariana como a causa do sintoma.<xref ref-type="bibr" rid="B17">17</xref> Portanto, nossos dados refor&#231;am que a abordagem de confiar em fatores de risco e sintomas para estratificar pacientes com dor tor&#225;cica aguda &#233; de baixa acur&#225;cia. A utiliza&#231;&#227;o de um modelo probabil&#237;stico impede este tipo de erro cognitivo.</p>
        <p>N&#243;s propusemos tr&#234;s formas simples de utiliza&#231;&#227;o do modelo probabil&#237;stico. Primeiro, um escore baseado em pontos atribu&#237;dos a cada vari&#225;vel positiva, acompanhado de um gr&#225;fico relacionando resultados e probabilidades (<xref ref-type="fig" rid="f9">Figura 4</xref>). Tendo em conta o baixo n&#250;mero de vari&#225;veis, cinco delas de natureza bin&#225;ria, o c&#225;lculo &#233; facilmente executado. Segundo, um escore log&#237;stico dentro de uma planilha com a f&#243;rmula de regress&#227;o, contendo a idade como vari&#225;vel num&#233;rica e cinco respostas "sim" ou "n&#227;o". E, mais amig&#225;vel, um aplicativo para smartphones. Acreditamos que com a oferta de diferentes formas de c&#225;lculos, os m&#233;dicos ir&#227;o desenvolver um interesse maior na utiliza&#231;&#227;o de modelos probabil&#237;sticos.</p>
        <p>Limita&#231;&#245;es no presente estudo devem ser reconhecidas. O estudo foi realizado em uma unidade coronariana de um hospital terci&#225;rio espec&#237;fico, o que limita a sua valida&#231;&#227;o externa. A popula&#231;&#227;o de uma unidade de dor tor&#225;cica &#233; um tanto selecionada e tende a ter uma maior preval&#234;ncia da doen&#231;a do que uma popula&#231;&#227;o em uma sala de emerg&#234;ncia. Portanto, nosso modelo deve ser adicionalmente validado para pacientes com uma maior variedade de apresenta&#231;&#227;o cl&#237;nica. Por outro lado, o principal objetivo do modelo &#233; estimar a probabilidade pr&#233;-teste de indiv&#237;duos hospitalizados, que tamb&#233;m consiste de um grande subgrupo de pacientes do mundo real. Neste sentido, nossa valida&#231;&#227;o externa n&#227;o &#233; necessariamente pequena, &#233; apenas mais espec&#237;fica para a popula&#231;&#227;o testada.</p>
        <p>N&#243;s devemos reconhecer que o tamanho de nossa amostra &#233; relativamente pequeno em compara&#231;&#227;o com exemplos de escores fornecidos a partir de enormes bancos de dados. Temos tr&#234;s argumentos a favor do nosso estudo de 470 pacientes: em primeiro lugar, a sua novidade como a primeira tentativa bem-sucedida de desenvolver um escore como este, que serve pelo menos como uma prova de conceito de um modelo preditivo multivariado de probabilidade pr&#233;-teste da doen&#231;a. Segundo, na aus&#234;ncia de um modelo probabil&#237;stico multivariado, os m&#233;dicos utilizam o julgamento cl&#237;nico baseado em intui&#231;&#227;o probabil&#237;stica, que tem sido demonstrado em diferentes cen&#225;rios como sendo inferior a modelos multivariados. Assim, levando em conta a alternativa restante da intui&#231;&#227;o, talvez seja uma boa ideia a utiliza&#231;&#227;o deste escore, n&#227;o de uma forma determin&#237;stica, mas como uma ferramenta para evitar o vi&#233;s cognitivo comum relacionado &#224; intui&#231;&#227;o. Terceiro, o tamanho da nossa amostra foi baseado <italic>a priori</italic> no c&#225;lculo do tamanho da amostra para a regress&#227;o log&#237;stica e para testar o modelo com a curva ROC. De acordo com este c&#225;lculo, o nosso n&#250;mero de eventos foi suficiente para fornecer um poder m&#237;nimo e a precis&#227;o exigida. No entanto, futuros relatos devem melhorar a precis&#227;o de nossas estimativas.</p>
        <p>Por &#250;ltimo, entre os pacientes que foram inicialmente submetidos a testes n&#227;o invasivos, apenas aqueles com resultados positivos tiveram confirma&#231;&#227;o pela angiografia. Mesmo assim, a predi&#231;&#227;o de um teste n&#227;o invasivo negativo (em oposi&#231;&#227;o a nenhuma doen&#231;a) &#233; suficiente para evitar que o paciente permane&#231;a desnecessariamente para ser submetido ao teste.</p>
      </sec>
      <sec sec-type="conclusions">
        <title>Conclus&#227;o</title>
        <p>O presente estudo desenvolveu e validou um novo modelo para predizer DAC obstrutiva em pacientes admitidos &#224; unidade coronariana com dor tor&#225;cica aguda. A utiliza&#231;&#227;o deste modelo dever&#225; ter um impacto na preven&#231;&#227;o da utiliza&#231;&#227;o excessiva de testes e do excesso de diagn&#243;stico e tratamento, ao mesmo tempo que dever&#225; melhorar a acur&#225;cia da avalia&#231;&#227;o pr&#233;-teste da probabilidade da doen&#231;a.</p>
      </sec>
    </body>
    <back>
      <fn-group>
        <fn fn-type="supported-by">
          <p>
            <bold>Fontes de financiamento</bold>
          </p>
          <p>O presente estudo n&#227;o teve fontes de financiamento externas.</p>
        </fn>
        <fn fn-type="other">
          <p>
            <bold>Vincula&#231;&#227;o acad&#234;mica</bold>
          </p>
          <p>N&#227;o h&#225; vincula&#231;&#227;o deste estudo a programas de p&#243;s-gradua&#231;&#227;o.</p>
        </fn>
      </fn-group>
    </back>
  </sub-article>
</article>
