Zworth Reading

Accuracy of Point-of-Care Ultrasound Versus Consultative Echocardiography to Identify Right Ventricular Dysfunction in Emergency Department Patients With Pulmonary EmbolismLOE 4 (retrospective cohort)

Annals of Emergency Medicine

In ED patients with confirmed PE, how does EP-performed POCUS compare to cardiologist-interpreted echo for detecting RV dysfunction?

POCUS was 76.8% sensitive and 85.9% specific for RV dysfunction compared to consultative echo (kappa 0.63). Accuracy improved for moderate-severe RV dysfunction. Interestingly, agreement declined through residency training , with senior residents performing worse than juniors.

Single centre retrospective design with inherent selection bias. Patients needed both POCUS and formal echo, which likely enriches for sicker patients or diagnostic uncertainty. The reference standard (cardiologist echo) is imperfect. The declining accuracy with training is counterintuitive and unexplained. Possibly reflects overconfidence or different case mix (or more junior residents being supervised by staff). Sample size adequate (194 patients, 97 operators). The 77% sensitivity means roughly 1 in 4 patients with RV dysfunction would be missed by POCUS.

Bottom line: This study has several limitations, but is the best evidence we have on this question to date. It found that POCUS has moderate accuracy for RV dysfunction in PE useful for ruling in moderate-severe dysfunction but not reliable enough to rule out RV strain when making risk stratification decisions. Personally, I find POCUS most useful in the undifferentiated patient with shock or dyspnea. POCUS won't necessarily rule out PE, but if I do see signs of significant RV strain that may take me down the PE diagnostic pathway. Would be curious to here thoughts from some of my more POCUS-minded readers on this paper.

Age-Adjusted D-Dimer Cutoff Levels to Rule Out Deep Vein ThrombosisKnowledge translation — structured appraisal of primary literature

EMOttawa  ·  LOE 3 (prospective diagnostic accuracy study) — the underlying ADJUST-DVT study

EMOttawa reviews the ADJUST-DVT study validating age-adjusted D-dimer thresholds (age × 10 μg/L for patients >50) for DVT exclusion. The post rates methodology 4/5 and usefulness 5/5.

Bottom line: The ADJUST-DVT study was well done, and suggests that an age-adjusted D-Dimer approach is safe for ruling out DVT.

Mechanical CPR Device Use and Cardiac Arrest Survival in EMS AgenciesLOE 3 (prospective registry with interrupted time series analysis)

Circulation  ·  Resuscitation / EMS

In EMS agencies treating OHCA, does introduction of mechanical CPR devices improve survival compared to the pre-device era?

Among 49 EMS agencies that introduced mechanical CPR devices (31,914 OHCAs), there was no improvement in favorable neurological survival (8.9% before vs 8.3% after) or survival to discharge (11.0% vs 10.0%). The interrupted time series showed no change in intercept or slope after device introduction.

Large registry study (CARES) with appropriate approach for this question. Limitations: retrospective and observational, cannot account for unmeasured confounders, device use protocols varied across agencies, and registry data quality is variable. Within those caveats, the signal is consistent with prior RCT data.

Bottom line: This study adds to the existing body of evidence suggesting no clear survival benefit from mechanical CPR devices at the EMS system level. Given the retrospective design, interpret with appropriate caution, but the consistency with the RCT literature (LINC, PARAMEDIC) is notable. These devices may still have niche roles (prolonged transport, hypothermic arrest, cath lab, sustained quality CPR with limited responders), but the data continues to argue against prioritizing them over high-quality manual CPR.

---

Large Language Models Provide Accurate but Potentially Unsafe Answers to Multimodal Critical Care Medicine Board Review QuestionsLOE 4 (observational study using validated item bank)

Critical Care Medicine  ·  Critical Care / Medical Education / AI

How accurate is ChatGPT-4o on multimodal critical care board questions, and how often are its answers potentially harmful?

ChatGPT-4o answered 74.9% of 183 board-style questions correctly - slightly higher than pooled clinician responses at 71.1% (p=0.03). Strengths: question comprehension (87.4%), pulmonary (91.7%), surgery/trauma (87.5%), neurology (81.8%). Weaknesses: image interpretation (61.7%), reasoning (68.3%), critical care ultrasound (51.1%). Critically, 33.3% of responses were judged to have potential for clinical harm, often driven by incorrect image interpretation and downstream treatment recommendations.

Important category of study and this one appears well-done. The results aren't surprising. LLMs do some things very well (text comprehension, recall of common clinical patterns) and struggle in others (image interpretation, reasoning through ambiguity). The 33% harm-potential rate is the most important number here. And if LLMs underperform on board questions (where there is a structured stem, complete information, and a single correct answer) they will almost certainly perform worse in the real ED, where presentations are messy, information is incomplete, and there is rarely a single right answer.

Bottom line: LLMs are useful but require clinical caution. The question you ask matters more than the tool itself. 'Are there diagnoses or management options I haven't considered?' is a far better use than 'What is the diagnosis?' or 'What is the next step?' Treat LLM output as a prompt to think, not a clinical recommendation. The technology will improve, but the principle of who owns the decision will not change.

Comparison of Intraosseous Line Placement Location and Rates of Return of Spontaneous Circulation and Survival to Discharge Among Patients with Out-of-Hospital Cardiac ArrestLOE 4 (retrospective observational)

Resuscitation

Retrospective single-county EMS analysis comparing humeral vs tibial IO placement in OHCA. Humeral IO was associated with substantially higher rates of ROSC (adjusted OR ~2.5) and survival to discharge. The adjusted model included age, sex, initial rhythm, and compression fraction.

Retrospective design with significant confounding by indication. Providers who choose humeral sites are likely systematically different from those who choose tibial (more experienced, different protocols, patient factors driving site choice. Single county limits generalizability and the survival analysis is underpowered (n=49 survivors). All caveats noted, the effect size is also very large for an intervention that affects drug delivery by seconds to minutes. That said, the pharmacokinetic rationale is plausible (humeral IO drains centrally faster than tibial), and the signal is worth taking seriously as hypothesis-generating.

Bottom line: Interesting and biologically plausible signal, but retrospective design and confounding mean this cannot establish that humeral IO improves outcomes. The clinical tension is real: in arrest, time is everything and you should place IO at the site you can access fastest. But if humeral really does deliver drugs more effectively, that's an argument for building proficiency with humeral placement so it becomes the fastest site for more providers. We need a properly designed prospective study with protocolized site selection to know.