We designed an interventional randomized, controlled, open label, parallel-group, trial in two French EDs. The two recruiting centers were a university hospital and a community hospital. POCUS was added to the usual diagnosis workup in the experimental group. POCUS, performed in B mode only with a curvilinear probe, assessed the major spots and search for main anomalies. This study followed the CONSORT reporting guidelines [12].

Since there is no international consensus-based guideline, the exploration protocol was collectively designed after a literature review. It was focused on aorta (aneurysm, aortic dissection), gallbladder (lithiasis, cholecystitis), kidneys (hydronephrosis), bladder, small bowel loops, appendix and ovaries (Table 1). POCUS was performed using Mindray TE7 or Philips CX50 with a curvilinear probe (3.5–5 MHz) or a linear probe (7–10 MHz) for the appendix exploration. It was performed by a trained EP. Theses EP previously attend a refresher course [12], and were not necessarily the EP in charge of the patient. The investigators have previously completed a validated training program. It could be a certified one-year faculty-based training program or a short training session (two days). The study protocol was published [11].

We included patients strictly over 17 years old presenting to the ED with acute abdominal pain, when an EP trained in POCUS was present. They were identified from the referral system. An informed consent was obtained before randomization. It was a convenience sample since an EP trained in POCUS was not always available. Exclusion criteria were a documented end-of-life, immediate need of life-support therapy, pregnant or breast-feeding women and patient under guardianship.

The final diagnosis was established a posteriori by an adjudication committee composed of three independent experts in EM, radiology and abdominal surgery. The committee was blind regarding the group (POCUS vs control). They had access to all data from patient files including advanced imaging results excepted the preliminary diagnosis made by the emergency physician in charge of the patient and POCUS results. The preliminary diagnosis was made by the treating EP before any radiologist-performed imaging study (including ultrasound, CT-scan, MRI). The preliminary diagnosis was based on clinical examination and results of laboratory tests in both arms, associated with POCUS in the intervention arm. The treating EP were not blind of the POCUS results in the intervention arm.

The primary endpoint was the proportion of exact preliminary diagnosis. The preliminary diagnosis made by the EP was considered exact/correct when it was similar to the final diagnosis made by adjudication committee. Diagnosis were chosen in a predefined list. The diagnostic criteria were not specified since the adjudication committee was composed of experts. Non-specific abdominal pain was defined as an acute abdominal pain of under 7 days’ duration, and with no diagnosis after examination and baseline investigations. Secondary endpoints were the time between admission at the ED and diagnosis, ED length of stay, diagnostic accuracy for non-specific abdominal pain, prescription of biological and radiological exams during the ED length of stay and hospitalization rate. A post-hoc analysis on diagnostic performance was performed on a sub-group of patients with diagnosis accessible to US.

Patients were randomized 1:1 to POCUS or control group by a computed-based program in random block sizes and stratified by centre. Randomization list were generated using SAS software.

Based on previous studies [7,8,9,10], a correct diagnostic rate of 57% was expected in the control group and 74% in the experimental group. With an alpha value of 0.05 and a power level of 80%, 244 patients were required. A 5% attrition rate (patients randomized but presenting an exclusion criteria) was expected, thus 256 randomized patients were needed.

The primary endpoint was compared between the two groups using a mixed model taking into account the recruiting centre. The delays were compared by a mixed linear generalized model adjusted on the recruiting centres. Sensitivity, specificity, positive and negative predictive values were estimated with their 95% confidence intervals. The rates of readmission and hospitalization were compared using a logistic generalized mixed model adjusted on the centre. In the experimental group (with POCUS), duration and self-assessed difficulty of POCUS were described by means and standard deviations.