Title: Recognizing Sepsis: Man versus Machines<br/>
Author: Kami Windsor<br/>
<a href='mailto:khu@som.umaryland.edu'>[Click to email author]</a><hr/>
Link: <a href='https://umem.org/educational_pearls/4529/'>https://umem.org/educational_pearls/4529/</a><hr/><p><strong>Background:</strong> Sepsis remains a common entity associated with a relatively high rate of inpatient mortality, with timely recognition and treatment being key to improving patient outcomes. Various screening and warning scores have been created to attempt to identify sepsis and those patients at high risk of mortality earlier, but have limited performance because of suboptimal sensitivity and specificity.</p>
<p><strong>A prospective observational study</strong> compared the performance of a variety of these scores (SIRS, qSOFA, SOFA, MEWS) as well as a machine learning model (MLM) against ED physician gestalt in diagnosing sepsis within the first 15 minutes of ED arrival. </p>
<ul>
<li>2550 patients deemed by EMS or triage nurse as potentially critically-ill
<ul>
<li>Excluded trauma, cardiac arrest, acute MI, stroke activation, patients in labor</li>
</ul>
</li>
<li>Seen by ED attendings (94%) / senior residents (6%) at a single urban academic center
<ul>
<li>Visual analog scale assessment, 0-100% likelihood that patient has sepsis</li>
<li>VAS >50% treated as ED physician gestalt in favor of sepsis</li>
</ul>
</li>
<li>275 patients ultimately with discharge diagnosis of sepsis present on arrival to hospital</li>
<li>Initial VAS outperformed all scores (AUC 0.90; 95% CI 0.88 to 0.92) both at 15 minutes and 1 hour</li>
</ul>
<p>Although not without its limitations, this study highlights the importance and relative accuracy of physician gestalt in recognizing sepsis, with implications for how to develop future screening tools and limit unnecessary exposure to unnecessary fluids and empiric broad spectrum antibiotics.</p>
<p><strong>Bottom Line:</strong> In the era of machine learning models and AI, ED physicians are not obsolete. Even at 15 minutes, without lab results and diagnostics, our assessments lead to appropriate diagnoses and care. In this new normal of prolonged wait times and ED boarding, ED triage and evaluation models that optimize early physician assessment are of the utmost importance.</p>
<fieldset><legend>References</legend><p>Knack SKS, Scott N, Driver BE, Pet al. Early Physician Gestalt Versus Usual Screening Tools for the Prediction of Sepsis in Critically Ill Emergency Patients. Ann Emerg Med. 2024 :S0196-0644(24)00099-4. doi: 10.1016/j.annemergmed.2024.02.009.</p>
</fieldset>