A total of 238 patients with clinical diagnoses of dengue were screened, and 178 entered the study. Follow-up was completed by telephone in 139 (78%) participants and in 39 (22%) only by reviewing clinical records (Fig. 2). There were 44.4% men, with a median age of 16.2 (range 2–85) years, a median of 5 (0–10) days of fever onset, and none with pulse pressure < 20 mmHg at enrollment. While none of the participants were clinically classified as severe dengue, more than half (n = 98, 55%) were clinically classified as dengue with warning signs. Those with warning signs were more often female (61.2%), 30 years old or younger (87.8%), with fever onset of one week or more (40.8%), and with thrombocytopenia (78.5%). A dengue IgM rapid test was performed in almost all (155/178, 87%) participants, with positive results in 64.5% (100/155) of them. A dengue IgG rapid test was reported less frequently than IgM (54% of participants), with positive results in 67.7% (65/96). The most frequent hemogram anomaly was thrombocytopenia (73.6%), followed by leukopenia (51.7%), lymphopenia (16.8%), and monocytosis (13.4%). The highest degree of hemoconcentration was 59.2% in an IgM-positive subject aged 32 years old, with a minimum hematocrit of 18.2% and a maximum of 29.0% (Table 1). Other clinical tests were seldom requested, including the following: aspartate aminotransferase/alanine aminotransferase (n = 45), blood creatinine (n = 17), blood urea nitrogen (n = 9), malaria thick smear (n = 4), blood sodium (n = 2), and chest X-ray (n = 1).

Flowchart of participant selection and follow-up

A total of 2121 images were obtained by the study physician and quality scored by the radiologist. Of these, 1053 (49.6%) corresponding to 161 patients were of suitable quality (score 3 or more) to be interpreted for evidence of plasma leakage by the expert radiologist (Fig. 3). Overall quality was independent of the patient’s age, sex, and dengue classification. Except in the right lung apex where the quality was higher for younger patients, in the right lung base where quality was higher for men, and the gallbladder wall where quality was higher for women (see Additional File 2 for results of image quality by patient characteristics). The median fasting time before POCUS was 3 h (IQR: 1–5).

Quality score of point-of-care ultrasound images by anatomical site. ND not done

Plasma leakage was observed in 85.1% (95% CI: 78.6–90.2%) of participants and was more frequent in the abdomen, manifesting as ascites (hepatorenal or splenorenal or pelvic) and/or gallbladder wall thickening (Table 1) (Fig. 4). The appearance of the gallbladder was reported as normal in 84/121 (69.4%) subjects, followed by striated and thickened in 11 (9.1%), contracted in 10 (8.2%), with a “honeycomb” pattern and thickened in 7 (5.8%), with pericholecystic fluid and striated in 4 (3.3%), “honeycomb” and pericholecystic fluid in 3 (2.5%) and thickened in 2 (1.7%). Splenorenal ascites, pelvic ascites, gallbladder wall thickening, and pericholecystic fluid were more frequent in subjects classified as dengue with warning signs, but only the latter was statistically significant (P = 0.02) (Table 1).

Point-of-care ultrasound findings suggesting plasma leakage. a RPLAPS right posterolateral alveolar or pleural syndromes-point demonstrating pleural effusion (arrow) and lung atelectasis (star); b LPLAPS left posterolateral alveolar or pleural syndromes-point demonstrating lung atelectasis (star), pleural effusion (left arrow), and splenorenal ascites (right arrow); c hepatorenal ascites (top arrow), right pleural effusion (bottom arrow), and perirenal fluid (star); d pelvis demonstrating abdominal ascites (top arrow) and liquid in rectovesical pouch (bottom arrow); e gallbladder with calipers measuring gallbladder wall (star) thickness and demonstration of pericholecystic fluid (arrow); f pericardial effusion (arrow)

The study physician detected plasma leakage in 76 (47.2%) subjects, which corresponds to a Kappa value of 0.25 (95% CI: 0.15–0.35) considered “fair”. The disagreement was particularly observed in the ascites, where the study physician detected a smaller proportion of the splenorenal (4/25 vs 16/25, Kappa = 0.06) and hepatorenal (5/39 vs 34/39, Kappa = 0.04) ascites than the radiologist. There was improved concordance for right pleural effusion (5/45 vs 4/45, Kappa = 0.18), pericardial effusion (3/48 vs 3/48, Kappa = 0.28), pelvic ascites (23/146 vs 32/146, Kappa = 0.48) and even better for the B-lines, left pleural effusion and thickened gallbladder wall, where the study physician and radiologist detected a similar proportion (2/96 vs 3/96, Kappa = 0.79), (2/23 vs 1/23, Kappa = 0.64), and (7/77 vs 8/77, Kappa = 0.62), respectively.

A decreased frequency of plasma leakage was observed in the age group 30 to 59 when compared to those ≤ 18 years old. Patients ≤ 18 years old showed more pelvic ascites (27/87, 31%) than older patients (5/60, 8.3%) (P = 0.001). In contrast, patients > 18 years old were more likely to show B-lines (4/68, 5.8%) than younger patients (0/94) (P = 0.01) There was no association between POCUS evidence of plasma leakage and sex, dengue clinical classification, leukopenia, lymphopenia, or monocytosis. Plasma leakage was more frequent in those seeking care with 7 to 10 days of fever onset who had hemoconcentration and thrombocytopenia. Of these, thrombocytopenia and age had a statistically significant association in the multivariate model (Table 2). The final multivariate model demonstrated a good fit (P = 0.16) with McFadden’s R2 of 0.20.

In two subjects, the treating physician concluded that the diagnosis was not dengue; one was diagnosed with malaria due to Plasmodium falciparum and had no POCUS evidence of plasma leakage, while the other was diagnosed with hepatitis A and had evidence of hepatorenal ascites. Two other study participants were diagnosed as having coincident dengue and another disease; one was diagnosed with malaria due to P. vivax with the presence of pelvic ascites, and another with pneumonia with no signs of plasma leakage. All participants were alive at day 14 after enrollment, 66 (37.1%) were treated as outpatients, 100 (56.2%) were hospitalized, and 12 (7.7%) required referral to a higher level of care due to the severity of their condition. POCUS evidence of plasma leakage was associated with hospital admission or referral (cOR = 7.5, 95% CI: 2.7–20.4, P < 0.0001), as was thrombocytopenia (6.1, 95% CI: 3–12.6, P < 0.0001), a unit increase in pulse pressure (1.04, 95% CI: 1–1.07, P = 0.02) and degree of hemoconcentration (1.07, 95% CI: 1.01–1.13, P = 0.01) (Table 3). All these variables, except hemoconcentration, were associated with hospital admission or referral to a higher level of care in the multivariate model, while older age was associated with outpatient care (Table 3). The final model demonstrated a good fit (P = 0.7) with McFadden’s R2 of 0.27. The predicted probability of being hospitalized if there was POCUS evidence of plasma leakage on admission was 0.70 (95% CI: 0.63–0.76) and 0.32 (95% CI: 0.13–0.51) without POCUS evidence of plasma leakage. The total number of POCUS anatomical sites positive for plasma leakage was associated with hospital admission or referral to a higher level of care (Table 3) but did not enter the final model because it was correlated with POCUS evidence of plasma leakage. A single positive POCUS result (regardless of the anatomical site) showed 93.3% sensitivity and 33.3% specificity for hospitalization/referral. With each additional positive site, sensitivity decreased, and specificity increased. Of the POCUS findings, hepatorenal ascites (cOR = 3.8, 95% CI: 0.5–29.3, P = 0.2) showed an increased risk of hospital admission or referral to a higher level of care, but it did not reach statistical significance (Table 3).