Why POCUS Quality Assurance Matters, and Why It Will Matter More

It is common for POCUS programmes to lack a structured quality assurance system. This was manageable while scan volumes were small and every number on the screen was produced by a human. Both of those things are now changing. A national survey across five specialties at the US Department of Veterans Affairs, found that most specialty groups had no POCUS image archiving, no specialty-specific credentialing policy, and no quality assurance process (Resop et al., The Ultrasound Journal 2025).

What makes POCUS different from every other imaging modality

Formal imaging separates two skills between two people. A technician acquires the images; a radiologist interprets them. Each is trained, assessed and credentialed for one half of the job. POCUS enables a single clinician to both acquire and interpret images at the bedside, in real time. As Joshua Guttman, an emergency physician at Emory University who writes on POCUS governance, puts it, this “dual requirement underscores the need for a tailored credentialing process that assures both skills are present” (Peachtree POCUS). This has two implications:

1. Competence is application-specific. Guttman’s example is precise: a cardiologist may be highly skilled at cardiac POCUS and have no training in lung POCUS. The two look like the same modality and share almost none of the same acquisition or interpretive skill. This differs from other credentialed procedures, where technique is broadly uniform across specialties and proof of supervised competency covers it. A trauma surgeon credentialed for eFAST is not thereby competent at cardiac or lung scanning.
2. The riskiest group is not the untrained one. Clinicians with no POCUS training know not to scan unsupervised. Guttman identifies the concerning group as those with informal or incomplete training, “who may develop a false sense of confidence due to limited experience”. A training roster will not catch them, because they have had training. Quality review catches them, because their images and interpretations show it.

What QA is actually for

Clinicians new to POCUS often worry about malpractice exposure, on the assumption that their images and readings will be judged against a sonographer’s and a radiologist’s. Guttman examined the published case law and found the anxiety largely unsupported. Across three analyses of emergency department malpractice claims, covering 659 cases from 1987 to 2007, 120 from 2008 to 2012 and 276 from 2012 to 2021, the recurring allegation was failure to perform an ultrasound rather than failure to interpret one. A 2020 review of 131 malpractice cases across internal medicine, critical care, paediatrics and family practice found none involving POCUS at all.

His conclusion is unambiguous: “I found virtually no malpractice cases in the U.S. stemming from poor POCUS performance. Instead, the few POCUS-related cases should encourage more liberal use of POCUS in practice” (MedPage Today).

The Canadian closed-case series often cited in this context supports this conclusion. Of 15 POCUS-related cases in the Canadian Medical Protective Association repository, only five were civil suits, and seven concerned failure to perform a POCUS exam when it was indicated (Prager et al., The Ultrasound Journal 2024).

So the honest case for quality assurance is not defensive. QA exists “to make sure we’re delivering safe, effective care that improves patient outcomes”, and in POCUS specifically “it ensures images are good enough to make decisions, interpretations are accurate, and limitations are recognized and documented” (Peachtree POCUS).

Two further reasons matter to anyone running a programme. Credentialing frameworks count quality-reviewed examinations rather than examinations performed, so a department without QA cannot credential its clinicians against any recognised threshold. And Focused and Ongoing Professional Practice Evaluation, the mechanisms hospitals already use for procedural oversight, apply to POCUS and give a POCUS lead a concrete basis for requesting protected time and staffing.

Why QA in POCUS will only matter more

Clinicians are increasingly acting on numbers generated by AI. Automated measurement is now routine in POCUS: automated ejection fraction, automated B-line counts, automated bladder volume, automated gestational age from blind sweeps. None of these tools are cleared for autonomous diagnosis and each keeps a clinician as the decision-maker, but the measurement itself is no longer a human act, which quietly undoes the dual-skill assumption credentialing is built on.

Regulatory clearance does not guarantee quality. Across 1,357 AI devices cleared through December 2025, 2.5% were linked to registered prospective trials and 0.2% evaluated patient-centred outcomes (Abulibdeh et al., PLOS Digit Health 2026). Among 950 AI-enabled devices authorised through November 2024, roughly 43% of recall events occurred within a year of authorisation, most commonly for diagnostic or measurement errors (Dai et al., JAMA Health Forum 2025).

Performance also moves after deployment. A model trained on one manufacturer’s images may not perform equally well on another’s (Vega et al., npj Digit Med 2025), and a 2026 intensive care review describes degradation over time as an inevitability given changes such as the introduction of new machines, recommending dual validation by AI and clinician as the standard approach (Wu, Arntfield & Millington, J Intensive Care Med 2026).

Regulators are moving from encouraging monitoring to requiring it. Article 26 of the EU AI Act places duties on deployers of high-risk AI, including assigning human oversight to competent individuals, monitoring operation and retaining logs for at least six months, applying to AI in medical devices from August 2028. The Royal College of Radiologists published post-deployment monitoring guidance in March 2026 asking departments to audit AI outputs against a reference standard and track AI–human disagreement rates. Ultrasound’s own societies have not caught up: AIUM’s practice-topic page on artificial intelligence states that there are currently no related practice parameters.

The dependency underneath it all: a scan that was never saved cannot be reviewed

That single constraint explains why many programs lack QA processes and a scalable credentialing approach. This same gap will shortly prevent departments from answering questions about AI performance that they are going to be asked.