<!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="editorial" dtd-version="1.1" specific-use="sps-1.9" xml:lang="pt">
  <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 specific-use="scielo-v3" pub-id-type="publisher-id">WJDWpS78653m9YqZPvdmwkG</article-id>
      <article-id specific-use="scielo-v2" pub-id-type="publisher-id">S0066-782X2025000500101</article-id>
      <article-id pub-id-type="other">00101</article-id>
      <article-id pub-id-type="doi">10.36660/abc.20250300</article-id>
      <article-id pub-id-type="publisher-id">abc.20250300</article-id>
      <article-categories>
        <subj-group subj-group-type="heading">
          <subject>Editorial</subject>
        </subj-group>
      </article-categories>
      <title-group>
        <article-title>Voc&#234; Vai Avis&#225;-los Quando Estiverem Errados? Vi&#233;s de Automa&#231;&#227;o e Intelig&#234;ncia Artificial na Medicina</article-title>
      </title-group>
      <contrib-group>
        <contrib contrib-type="author">
          <contrib-id contrib-id-type="orcid">0000-0002-3711-1754</contrib-id>
          <name>
            <surname>Bittencourt</surname>
            <given-names>Marcio Sommer</given-names>
          </name>
          <xref ref-type="aff" rid="aff1">
            <sup>1</sup>
          </xref>
          <xref ref-type="corresp" rid="c1"/>
        </contrib>
        <aff id="aff1">
          <label>1</label>
          <institution content-type="orgname">University of Pittsburgh</institution>
          <institution content-type="orgdiv1">Department of Medicine</institution>
          <institution content-type="orgdiv2">Division of Cardiology</institution>
          <addr-line>
            <named-content content-type="city">Pittsburgh</named-content>
            <named-content content-type="state">PA</named-content>
          </addr-line>
          <country country="US">EUA</country>
          <institution content-type="original">Division of Cardiology, Department of Medicine, University of Pittsburgh, Pittsburgh, PA &#8211; EUA</institution>
        </aff>
      </contrib-group>
      <author-notes>
        <corresp id="c1"><label>Correspond&#234;ncia:</label><bold>Marcio Sommer Bittencourt &#8226;</bold> UPMC Presbyterian Hospital |200 Lothrop Street | Suite A-429 |Pittsburgh, PA <postal-code>15213</postal-code> E-mail: <email>msbittencourt@mail.harvard.edu</email> </corresp>
      </author-notes>
      <pub-date publication-format="electronic" date-type="pub">
        <day>14</day>
        <month>07</month>
        <year>2025</year>
      </pub-date>
      <pub-date date-type="collection" publication-format="electronic">
        <year>2025</year>
      </pub-date>
      <volume>122</volume>
      <issue>05</issue>
      <elocation-id>e20250300</elocation-id>
      <history>
        <date date-type="received">
          <day>23</day>
          <month>04</month>
          <year>2025</year>
        </date>
        <date date-type="rev-recd">
          <day>02</day>
          <month>05</month>
          <year>2025</year>
        </date>
        <date date-type="accepted">
          <day>02</day>
          <month>05</month>
          <year>2025</year>
        </date>
      </history>
      <permissions>
        <license license-type="open-access" xlink:href="https://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</license-p>
        </license>
      </permissions>
      <kwd-group xml:lang="pt">
        <title>Palavras-chave</title>
        <kwd>Intelig&#234;ncia Artificial</kwd>
        <kwd>Cardiologia</kwd>
        <kwd>Vi&#233;s</kwd>
      </kwd-group>
      <counts>
        <fig-count count="0"/>
        <table-count count="0"/>
        <equation-count count="0"/>
        <ref-count count="5"/>
      </counts>
    </article-meta>
  </front>
  <body>
    <p>O uso de ferramentas de suporte &#224; decis&#227;o e de automa&#231;&#227;o na pr&#225;tica m&#233;dica n&#227;o &#233; algo recente. As primeiras ferramentas utilizavam modelos de regress&#227;o simples para reduzir a heterogeneidade e oferecer um cuidado mais estruturado. Na cardiologia, exemplos iniciais de ferramentas de suporte &#224; decis&#227;o inclu&#237;ram o escore de risco de Framingham para estratifica&#231;&#227;o de risco na preven&#231;&#227;o prim&#225;ria e os ajustes automatizados de doses para heparina ou varfarina. Desde ent&#227;o, os processos de automa&#231;&#227;o tornaram-se mais sofisticados e atualmente est&#227;o amplamente disseminados em diversos aspectos rotineiros da pr&#225;tica m&#233;dica. Na cardiologia, incluem interpreta&#231;&#245;es preliminares automatizadas de tra&#231;ados de eletrocardiograma (ECG), bem como medi&#231;&#245;es e an&#225;lises de imagens automatizadas em ecocardiografia, medicina nuclear e resson&#226;ncia magn&#233;tica card&#237;aca. Outros sistemas automatizados de suporte &#224; decis&#227;o cl&#237;nica auxiliam na redu&#231;&#227;o de erros de prescri&#231;&#227;o, alertando os profissionais de sa&#250;de sobre intera&#231;&#245;es medicamentosas ou dosagens incorretas, por exemplo. Com os avan&#231;os recentes da intelig&#234;ncia artificial (IA), as expectativas quanto ao futuro dessas ferramentas cresceram de forma expressiva. No entanto, as poss&#237;veis implica&#231;&#245;es negativas t&#234;m recebido consideravelmente menos aten&#231;&#227;o.</p>
    <p>Os sistemas de suporte &#224; decis&#227;o podem aprimorar a tomada de decis&#227;o m&#233;dica e os desfechos dos pacientes.<sup><xref ref-type="bibr" rid="B1">1</xref></sup> No entanto, esses sistemas n&#227;o s&#227;o infal&#237;veis e podem gerar resultados incorretos. A maioria dos estudos que avaliam seu desempenho baseia-se em m&#233;tricas m&#233;dicas convencionais, como sensibilidade, especificidade, acur&#225;cia e &#225;rea sob a curva <italic>receiver operating characteristic</italic>. Contudo, &#233; ainda mais importante avaliar o impacto dessas ferramentas no mundo real, especialmente considerando a forma como os profissionais interagem com elas. Essa quest&#227;o torna-se particularmente relevante &#224; luz do Regulamento de Intelig&#234;ncia Artificial da Uni&#227;o Europeia, que exige supervis&#227;o humana para sistemas de IA considerados de alto risco. O tema adquire ainda mais relev&#226;ncia com o uso disseminado de <italic>chatbots</italic> de IA generativa, uma vez que a atual gera&#231;&#227;o de ferramentas complexas de IA tem contribu&#237;do para uma tend&#234;ncia crescente de depend&#234;ncia excessiva de processos de automa&#231;&#227;o imperfeitos.</p>
    <p>Essa depend&#234;ncia excessiva pode ser compreendida a partir de dois conceitos relacionados, por&#233;m distintos: vi&#233;s de automa&#231;&#227;o e complac&#234;ncia com a automa&#231;&#227;o. O vi&#233;s de automa&#231;&#227;o refere-se &#224; tend&#234;ncia humana de confiar nos resultados fornecidos por ferramentas automatizadas em detrimento de informa&#231;&#245;es n&#227;o automatizadas, como o julgamento pessoal ou a contribui&#231;&#227;o de outros indiv&#237;duos. A complac&#234;ncia com a automa&#231;&#227;o, por sua vez, &#233; uma manifesta&#231;&#227;o diferente do mesmo fen&#244;meno. Ela descreve uma sensa&#231;&#227;o de confian&#231;a ou satisfa&#231;&#227;o excessiva em rela&#231;&#227;o aos resultados de sistemas automatizados, o que leva &#224; redu&#231;&#227;o da vigil&#226;ncia e &#224; aceita&#231;&#227;o acr&#237;tica de seus resultados.<sup><xref ref-type="bibr" rid="B2">2</xref></sup></p>
    <p>Um exemplo simples na cardiologia envolve o uso de softwares de pr&#233;-leitura de ECGs. Devido &#224; depend&#234;ncia excessiva dos processos automatizados, o cl&#237;nico pode sentir-se compelido a incluir no laudo conclus&#245;es que n&#227;o consegue identificar com seguran&#231;a nos tra&#231;ados, pressupondo que o software seja mais preciso do que sua pr&#243;pria interpreta&#231;&#227;o. Alternativamente, o profissional pode tornar-se t&#227;o injustificadamente dependente do sistema que deixa de avaliar adequadamente os segmentos ST em casos de supra de ST &#8212; a menos que essas altera&#231;&#245;es sejam sinalizadas pelo pr&#243;prio sistema. Esse comportamento reflete uma redu&#231;&#227;o da vigil&#226;ncia decorrente da confian&#231;a excessiva na ferramenta de suporte &#224; decis&#227;o, associada a uma sensa&#231;&#227;o inadequada de seguran&#231;a na interpreta&#231;&#227;o automatizada. As consequ&#234;ncias desses erros podem ser significativas. Um estudo de 2004 sobre ferramentas de interpreta&#231;&#227;o automatizada de ECG constatou que mais de dois ter&#231;os dos tra&#231;ados diagnosticados incorretamente como fibrila&#231;&#227;o atrial n&#227;o foram corrigidos pelo m&#233;dico que interpretava o exame. Isso resultou no uso inadequado de antiarr&#237;tmicos e anticoagula&#231;&#227;o em um ter&#231;o dos casos com diagn&#243;stico automatizado incorreto.<sup><xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B4">4</xref></sup></p>
    <p>Embora seja necess&#225;rio reconhecer que as ferramentas de automa&#231;&#227;o evolu&#237;ram desde 2004, sistemas com maior acur&#225;cia podem, paradoxalmente, aumentar o vi&#233;s de automa&#231;&#227;o, uma vez que sua suposta confiabilidade pode levar a uma depend&#234;ncia ainda maior. Al&#233;m disso, uma extensa literatura indica que o risco de vi&#233;s de automa&#231;&#227;o aumenta &#224; medida que a tarefa se torna mais dif&#237;cil e complexa.<sup><xref ref-type="bibr" rid="B5">5</xref></sup> Com a cont&#237;nua evolu&#231;&#227;o da IA, as ferramentas de suporte est&#227;o se tornando mais complexas e sendo aplicadas a tarefas cada vez mais desafiadoras. Portanto, &#233; fundamental que os profissionais de sa&#250;de compreendam as potenciais consequ&#234;ncias do vi&#233;s de automa&#231;&#227;o e da complac&#234;ncia com a automa&#231;&#227;o, bem como as estrat&#233;gias dispon&#237;veis para mitigar esses riscos.</p>
    <p>Os principais fatores que contribuem para o vi&#233;s de automa&#231;&#227;o incluem a suposi&#231;&#227;o de que m&#225;quinas e sistemas automatizados s&#227;o infal&#237;veis. Al&#233;m disso, a maioria dos sistemas automatizados apresenta maior facilidade de uso em compara&#231;&#227;o com m&#233;todos tradicionais (p.ex., &#233; mais simples relatar a frequ&#234;ncia card&#237;aca calculada pelo software em um ECG do que calcul&#225;-la manualmente), e os seres humanos tendem naturalmente a preferir a efici&#234;ncia. Por fim, temos a tend&#234;ncia de confiar mais em sistemas que n&#227;o compreendemos completamente &#8212; uma quest&#227;o especialmente relevante para a atual gera&#231;&#227;o de ferramentas de automa&#231;&#227;o baseadas em IA, cujos par&#226;metros subjacentes frequentemente s&#227;o desconhecidos pelo usu&#225;rio.</p>
    <p>Para mitigar as consequ&#234;ncias do vi&#233;s de automa&#231;&#227;o, s&#227;o necess&#225;rias interven&#231;&#245;es multifacetadas. Essas interven&#231;&#245;es devem abranger o usu&#225;rio da ferramenta de automa&#231;&#227;o, o design da pr&#243;pria ferramenta e os processos organizacionais que influenciam seu uso. Do ponto de vista do usu&#225;rio, estrat&#233;gias abrangentes de treinamento e educa&#231;&#227;o s&#227;o fundamentais para aumentar a conscientiza&#231;&#227;o sobre as limita&#231;&#245;es e potenciais erros dos sistemas automatizados. Isso inclui tanto a educa&#231;&#227;o geral sobre ferramentas de automa&#231;&#227;o baseadas em IA quanto treinamentos espec&#237;ficos de cada ferramenta, a fim de ajudar os usu&#225;rios a reconhecer quando uma determinada ferramenta tende a apresentar bom &#8212; ou mau &#8212; desempenho. Esse treinamento direcionado pode fomentar um pensamento mais cr&#237;tico, permitindo que os usu&#225;rios avaliem e questionem melhor os resultados automatizados, reduzindo, assim, o risco de complac&#234;ncia com a automa&#231;&#227;o.</p>
    <p>Sob a perspectiva do design, &#233; essencial garantir que as ferramentas sejam devidamente validadas, considerando o componente de intera&#231;&#227;o humana. Mais do que relatar apenas m&#233;tricas de desempenho, como acur&#225;cia, essas ferramentas devem tamb&#233;m ser avaliadas com base em seu impacto no cuidado cl&#237;nico ap&#243;s a implementa&#231;&#227;o. As consequ&#234;ncias dos erros podem ser mais relevantes do que sua frequ&#234;ncia &#8212; por exemplo, deixar de identificar um infarto agudo do mioc&#225;rdio com supradesnivelamento do segmento ST &#233; muito mais cr&#237;tico do que n&#227;o detectar extrass&#237;stoles ventriculares. Compreender com que frequ&#234;ncia os profissionais cometem erros por a&#231;&#227;o ou omiss&#227;o ao utilizar essas ferramentas tamb&#233;m &#233; fundamental. Al&#233;m disso, os desenvolvedores devem trabalhar em estreita colabora&#231;&#227;o com os usu&#225;rios finais para obter feedback sobre o desempenho em situa&#231;&#245;es reais e identificar oportunidades de melhoria. Fechar esse ciclo de feedback entre desenvolvedores e usu&#225;rios &#233; essencial para aprimorar tanto a qualidade dos processos automatizados quanto sua implementa&#231;&#227;o bem-sucedida.</p>
    <p>Por fim, as organiza&#231;&#245;es de sa&#250;de devem estabelecer processos que garantam a implementa&#231;&#227;o das ferramentas de automa&#231;&#227;o em contextos adequados. Por exemplo, uma ferramenta de interpreta&#231;&#227;o de ECG pode ser mais &#250;til em um departamento de emerg&#234;ncia remoto, sem a presen&#231;a de um cardiologista, do que em um grande centro de cardiologia. As institui&#231;&#245;es precisam considerar o contexto em que cada ferramenta ser&#225; utilizada, bem como os protocolos, fluxos de trabalho, treinamentos e certifica&#231;&#245;es necess&#225;rios, a fim de implementar essas ferramentas de forma eficaz, minimizando o risco de vi&#233;s de automa&#231;&#227;o.</p>
  </body>
  <back>
    <ref-list>
      <title>Refer&#234;ncias</title>
      <ref id="B1">
        <label>1</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Garg</surname>
              <given-names>AX</given-names>
            </name>
            <name>
              <surname>Adhikari</surname>
              <given-names>NK</given-names>
            </name>
            <name>
              <surname>McDonald</surname>
              <given-names>H</given-names>
            </name>
            <name>
              <surname>Rosas-Arellano</surname>
              <given-names>MP</given-names>
            </name>
            <name>
              <surname>Devereaux</surname>
              <given-names>PJ</given-names>
            </name>
            <name>
              <surname>Beyene</surname>
              <given-names>J</given-names>
            </name>
            <etal/>
          </person-group>
          <article-title>Effects of Computerized Clinical Decision Support Systems on Practitioner Performance and Patient Outcomes: A Systematic Review</article-title>
          <source>JAMA</source>
          <year>2005</year>
          <volume>293</volume>
          <issue>10</issue>
          <fpage>1223</fpage>
          <lpage>1238</lpage>
          <pub-id pub-id-type="doi">10.1001/jama.293.10.1223</pub-id>
        </element-citation>
        <mixed-citation>1 Garg AX, Adhikari NK, McDonald H, Rosas-Arellano MP, Devereaux PJ, Beyene J, et al. Effects of Computerized Clinical Decision Support Systems on Practitioner Performance and Patient Outcomes: A Systematic Review. JAMA. 2005;293(10):1223-38. doi: 10.1001/jama.293.10.1223.</mixed-citation>
      </ref>
      <ref id="B2">
        <label>2</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Parasuraman</surname>
              <given-names>R</given-names>
            </name>
            <name>
              <surname>Manzey</surname>
              <given-names>DH</given-names>
            </name>
          </person-group>
          <article-title>Complacency and Bias in Human Use of Automation: An Attentional Integration</article-title>
          <source>Hum Factors</source>
          <year>2010</year>
          <volume>52</volume>
          <issue>3</issue>
          <fpage>381</fpage>
          <lpage>410</lpage>
          <pub-id pub-id-type="doi">10.1177/0018720810376055</pub-id>
        </element-citation>
        <mixed-citation>2 Parasuraman R, Manzey DH. Complacency and Bias in Human Use of Automation: An Attentional Integration. Hum Factors. 2010;52(3):381-410. doi: 10.1177/0018720810376055.</mixed-citation>
      </ref>
      <ref id="B3">
        <label>3</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Bailey</surname>
              <given-names>NR</given-names>
            </name>
            <name>
              <surname>Scerbo</surname>
              <given-names>MW</given-names>
            </name>
          </person-group>
          <article-title>Automation-Induced Complacency for Monitoring Highly Reliable Systems: The Role of Task Complexity, System Experience, and Operator Trust</article-title>
          <source>Theor Issues Ergon Sci</source>
          <year>2007</year>
          <volume>8</volume>
          <issue>4</issue>
          <fpage>321</fpage>
          <lpage>348</lpage>
          <pub-id pub-id-type="doi">10.1080/14639220500535301</pub-id>
        </element-citation>
        <mixed-citation>3 Bailey NR, Scerbo MW. Automation-Induced Complacency for Monitoring Highly Reliable Systems: The Role of Task Complexity, System Experience, and Operator Trust. Theor Issues Ergon Sci. 2007;8(4):321-48. doi: 10.1080/14639220500535301.</mixed-citation>
      </ref>
      <ref id="B4">
        <label>4</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Singh</surname>
              <given-names>IL</given-names>
            </name>
            <name>
              <surname>Sharma</surname>
              <given-names>HO</given-names>
            </name>
            <name>
              <surname>Parasuraman</surname>
              <given-names>R</given-names>
            </name>
          </person-group>
          <article-title>Effects of Training and Automation Reliability on Monitoring Performance in a Flight Simulation TaskProc</article-title>
          <source>Hum. Factors Ergon Soc Annu Meet</source>
          <year>2000</year>
          <volume>44</volume>
          <issue>13</issue>
          <fpage>53</fpage>
          <lpage>56</lpage>
          <pub-id pub-id-type="doi">10.1177/154193120004401314</pub-id>
        </element-citation>
        <mixed-citation>4 Singh IL, Sharma HO, Parasuraman R. Effects of Training and Automation Reliability on Monitoring Performance in a Flight Simulation TaskProc. Hum. Factors Ergon Soc Annu Meet. 2000;44(13):53-6. doi: 10.1177/154193120004401314.</mixed-citation>
      </ref>
      <ref id="B5">
        <label>5</label>
        <element-citation publication-type="journal">
          <person-group person-group-type="author">
            <name>
              <surname>Goddard</surname>
              <given-names>K</given-names>
            </name>
            <name>
              <surname>Roudsari</surname>
              <given-names>A</given-names>
            </name>
            <name>
              <surname>Wyatt</surname>
              <given-names>JC</given-names>
            </name>
          </person-group>
          <article-title>Automation Bias: A Systematic Review of Frequency, Effect Mediators, and Mitigators</article-title>
          <source>J Am Med Inform Assoc</source>
          <year>2012</year>
          <volume>19</volume>
          <issue>1</issue>
          <fpage>121</fpage>
          <lpage>127</lpage>
          <pub-id pub-id-type="doi">10.1136/amiajnl-2011-000089</pub-id>
        </element-citation>
        <mixed-citation>5 Goddard K, Roudsari A, Wyatt JC. Automation Bias: A Systematic Review of Frequency, Effect Mediators, and Mitigators. J Am Med Inform Assoc. 2012;19(1):121-7. doi: 10.1136/amiajnl-2011-000089.</mixed-citation>
      </ref>
    </ref-list>
  </back>
  <sub-article article-type="translation" id="S1" xml:lang="en">
    <front-stub>
      <article-id pub-id-type="doi">10.36660/abc.20250300i</article-id>
      <article-categories>
        <subj-group subj-group-type="heading">
          <subject>Editorial</subject>
        </subj-group>
      </article-categories>
      <title-group>
        <article-title>Will You Tell Them When They Are Wrong? Automation Bias and Artificial Intelligence in Medicine</article-title>
      </title-group>
      <contrib-group>
        <contrib contrib-type="author">
          <contrib-id contrib-id-type="orcid">0000-0002-3711-1754</contrib-id>
          <name>
            <surname>Bittencourt</surname>
            <given-names>Marcio Sommer</given-names>
          </name>
          <xref ref-type="aff" rid="aff2">
            <sup>1</sup>
          </xref>
          <xref ref-type="corresp" rid="c2"/>
        </contrib>
        <aff id="aff2">
          <label>1</label>
          <addr-line>
            <named-content content-type="city">Pittsburgh</named-content>
            <named-content content-type="state">PA</named-content>
          </addr-line>
          <country country="US">USA</country>
          <institution content-type="original">Division of Cardiology, Department of Medicine, University of Pittsburgh, Pittsburgh, PA &#8211; USA</institution>
        </aff>
      </contrib-group>
      <author-notes>
        <corresp id="c2"><label>Mailing Address:</label><bold>Marcio Sommer Bittencourt &#8226;</bold> UPMC Presbyterian Hospital |200 Lothrop Street | Suite A-429 |Pittsburgh, PA <postal-code>15213</postal-code> E-mail: <email>msbittencourt@mail.harvard.edu</email> </corresp>
      </author-notes>
      <kwd-group xml:lang="en">
        <title>Keywords</title>
        <kwd>Artificial Intelligence</kwd>
        <kwd>Cardiology</kwd>
        <kwd>Bias</kwd>
      </kwd-group>
    </front-stub>
    <body>
      <p>The use of decision support tools and automation in medical practice is not new. Early tools employed simple regression models to reduce heterogeneity and offer more structured care. In cardiology, initial examples of decision support tools included the Framingham risk scores for primary prevention risk stratification and automated sliding scale dose adjustments for heparin or warfarin. Since then, automation processes have become more sophisticated and are now widespread in many routine aspects of medical practice. In cardiology, these include automated preliminary interpretations of electrocardiography (ECG) tracings as well as automated measurements and image analysis in echocardiography, nuclear medicine, and cardiac magnetic resonance imaging. Other automated clinical decision support systems help reduce prescribing errors by alerting clinicians to drug-drug interactions or incorrect dosages, for example. With recent advances in artificial intelligence (AI), expectations for the future of such tools have grown dramatically. Yet, the potential negative implications of these technologies have received considerably less attention.</p>
      <p>Clinical decision support systems can improve medical decision-making and patient outcomes.<sup><xref ref-type="bibr" rid="B1">1</xref></sup> However, these systems are not flawless and may produce incorrect outputs. Most studies assessing their performance rely on conventional medical metrics such as sensitivity, specificity, accuracy, and area under the receiver operating characteristic curve. Yet, it is far more important to evaluate the real-world impact of these tools, particularly in light of how clinicians interact with them. This is especially relevant given that the European Union Artificial Intelligence Act mandates human oversight for high-risk AI systems. The issue becomes even more critical with the widespread use of generative AI chatbots, as the current generation of complex AI tools has led to a growing trend of overreliance on imperfect automation.</p>
      <p>This overreliance can be understood through two related but distinct concepts: automation bias and automation complacency. Automation bias refers to the human tendency to rely on outputs from automated tools rather than on non-automated cues, such as personal judgment or input from other individuals. Automation complacency, on the other hand, is a different expression of the same phenomenon. It describes a sense of undue confidence or satisfaction with the output of automated systems, which leads to reduced vigilance and the uncritical acceptance of their results.<sup><xref ref-type="bibr" rid="B2">2</xref></sup></p>
      <p>A simple example in cardiology involves the use of pre-reading software for ECGs. Due to overreliance on automated processes, a clinician may feel compelled to include in the report conclusions they cannot confidently identify in the tracings, assuming the software is more accurate than their own interpretation. Alternatively, the clinician may become so unjustifiably dependent on the software that they fail to assess ST segments properly in cases of ST elevation &#8212; unless those abnormalities are flagged by the system. This reflects reduced vigilance resulting from overreliance on the decision support tool and a misplaced sense of confidence in the automated interpretation. The consequences of such errors can be significant. A 2004 study on automated ECG interpretation tools found that more than two-thirds of tracings incorrectly diagnosed as atrial fibrillation were not corrected by the interpreting physician. This led to inappropriate use of antiarrhythmic drugs and anticoagulation in one-third of the cases with incorrect automated diagnoses.<sup><xref ref-type="bibr" rid="B3">3</xref>,<xref ref-type="bibr" rid="B4">4</xref></sup></p>
      <p>While it is fair to acknowledge that automation tools have improved since 2004, higher-accuracy systems may increase automation bias since their perceived reliability can lead to greater overreliance. Furthermore, a substantial body of literature indicates that the risk of automation bias rises with increasing task difficulty and complexity.<sup><xref ref-type="bibr" rid="B5">5</xref></sup> As AI continues to evolve, supporting tools are becoming more complex and are being applied to increasingly challenging tasks. Therefore, clinicians should understand the potential consequences of automation bias and automation complacency as well as the strategies available to mitigate these risks.</p>
      <p>The primary drivers of automation bias include the human assumption that machines and automated systems are infallible. In addition, most automated systems are easier to use than traditional methods (e.g., it is simpler to report the software-calculated heart rate in an ECG than to calculate it manually), and humans naturally prefer efficiency. Finally, we tend to rely more heavily on systems we do not fully understand &#8212; an issue particularly relevant to the current generation of AI-based automation tools, where the underlying parameters are often unknown to the user.</p>
      <p>To mitigate the consequences of automation bias, multifaceted interventions are required. These interventions must address the user of the automation tool, the technical design of the tool itself, and the organizational processes that influence its use. From the user's perspective, comprehensive training and educational strategies are essential to raise awareness of the limitations and potential errors of automated systems. This includes both general education about AI-based automation tools and tool-specific training to help users recognize when a particular tool is likely to perform well &#8212; or poorly. Such targeted training can foster more critical thinking, enabling users to better assess and question automated outputs, thereby reducing the risk of automation complacency.</p>
      <p>From a design perspective, it is essential to ensure that tools are properly validated, considering the human interaction component. Rather than reporting only performance metrics such as accuracy, these tools must also be evaluated based on their impact on clinical care once implemented. The consequences of errors may be more important than their frequency &#8212; for example, missing a case of ST-elevation myocardial infarction is far more critical than overlooking premature ventricular contractions. Understanding how often clinicians commit errors of commission or omission when using these tools is also key. Additionally, developers must work closely with end users to gather feedback on real-world performance and identify opportunities for improvement. Closing this feedback loop between developers and users is vital to enhancing both the quality of automated processes and their successful implementation.</p>
      <p>Finally, health care organizations must establish processes that ensure automation tools are implemented in appropriate settings. For example, an ECG interpretation tool may be more valuable in a remote emergency department without an on-site cardiologist than in a large cardiology center. Institutions must consider the context in which each tool will be used, along with the necessary protocols, workflows, training, and certification, in order to implement these tools effectively while minimizing the risk of automation bias.</p>
    </body>
  </sub-article>
</article>
