Acute diarrhea in children is still one of the leading causes of morbidity and mortality, particularly in developing countries1; however, diarrheal diseases are significant clinical problems in industrialized countries.2-8 In Italy viruses (particularly rotaviruses) play the major role in the etiology of acute diarrhea,9-11 whereas the most commonly isolated bacteria are minor Salmonella spp., followed by Campylobacter.11 In infants <3 months old, or in immunodepressed patients, untreated minor Salmonella infections may be associated with metastatic localization to bone, lung, meninges and ear12; these patients are usually treated with antibiotics. Unfortunately clinical data or laboratory tests (including search for occult blood or leukocytes in stools13 are often inaccurate in differentiating bacterial from viral gastroenteritis (GE); only culture allows a reliable diagnosis. The diagnostic accuracy of C-reactive protein (CRP), erythrocyte sedimentation rate (ESR) and white blood cell (WBC) count and their usefulness in the clinical differentiation of viral from bacterial GE in children have been investigated in a recent paper by Borgnolo et al.14 We have tried to confirm the data obtained by Borgnolo et al.14 in a group of 248 consecutive children with acute GE admitted to our pediatric department. Methods. From January, 1996, to June, 1998, 260 children were admitted to our pediatric department for acute GE. A rectal swab culture was performed in 247 children for bacterial culture of Salmonella and Shigella spp., with standard laboratory techniques. Rotavirus and adenovirus were searched for using the rapid latex tests Rotalatex and Adenolatex (Bouty, Milan, Italy). At the time of admission we also determined CRP concentration (standard nephelometric method), ESR (standard Westergren method) and WBC count (using a standard automated cell counter). Maximal body temperature, age, days of hospitalization, days with diarrhea, presence of vomiting and need for parenteral rehydration were also recorded. Statistical analysis was performed with standard computer software (Microsoft Excel and SPSS 6.0 for Windows). Receiver operating characteristic curve analysis was performed with GraphRoc for Windows 2.0. Results. Of 247 children admitted with acute GE, 29 (11.7%) had a confirmed bacterial infection (all minor Salmonella spp.), 121 (48.9%) had a confirmed viral infection (90 rotavirus, 8 adenovirus, 23 rotavirus and adenovirus combined), and 97 (39.3%) had no demonstrable pathogen. For analysis of the positive and negative predictive values, we pooled the data of the patients with viral and undetermined GE, given the impossibility of distinguishing clinically between the two conditions. The receiver operating characteristic (ROC) curve of CRP, ESR and WBC are shown in Figure 1; the area under the ROC curve was 0.8674 for CRP, 0.7575 for ESR and 0.4963 for WBC count (a test with no power of discrimination has an area under the curve of 0.5000). The differences between the three ROC curves are significant (CRP vs. ESR, P = 0.03; CRP vs. WBC, P < 0.0001; ESR vs. WBC, P < 0.001).Fig. 1: ROC curve for CRP, ESR and WBC.CRP was equal or higher than 2 mg/dl in 24 of 29 (82.7%) children with Salmonella infection, in 21 of 121 (17.3%) children with a viral infection and in 22 of 97 (22.7%) children with no identifiable pathogen. With a value of 2.0 or greater, the sensitivity of CRP was 79.3%, with a specificity of 80.3%; the positive predictive value (i.e. the percentage of children with a CRP ≥2.0 who are expected to have a Salmonella GE) is 34.8%, whereas the negative predictive value (i.e. the percentage of children with a CRP <2.0 who are not expected to have Salmonella GE) was 96.7%. A summary of the efficiency data with thresholds of 1.2, 2.0 and 3.5 mg/dl is reported in Table 1.TABLE 1: Efficiency data of CRP at various threshold levels Efficiency data for WBC count (at a value ≥12 000/ml) and ESR (at a value of ≥20 mm/h) were always inferior to CRP, with positive predictive values of 27 and 12% and negative predictive values of 94 and 92%, respectively; sensitivity was 63 and 83%, with a specificity of 77 and 25% for WBC count and ESR. ESR was ≥20 mm/h in 21 of 30 (70%) children with bacterial GE, in 35 of 120 (29.2%) of those with a viral infection and in 32 of 102 (31.4%) of children with GE of undetermined etiology. A WBC count higher than 12 000/ml was observed in 5 of 30 (16.7%) bacterial infections, in 17 of 118 (14.4%) viral infections and in 24 of 97 (24.7%) undetermined etiology patients. Mean values ± SD of CRP, ESR, WBC, maximal temperature, age, days of hospitalization and days with diarrhea were compared in the three groups of patients. Multivariate ANOVA showed a highly significant difference among the three groups when CRP, ESR and WBC were considered (P < 0.001); univariate F tests for CRP and ESR also showed highly significant differences among the three groups (P < 0.001), but not for WBC (P = NS); age was also significantly different (P = 0.027), being lower children in with viral GE. Maximal recorded body temperature, total hospital stay and days with diarrhea were not significantly different among the three groups. Vomiting was present in 57% of children with Salmonella GE, vs. 77% in viral GE and 78% in undetermined GE (chi-square, 6.11; P = 0.047); parenteral rehydration was required in 87% of bacterial GE, 80% of viral GE and 69% of undetermined GE (chi square, 5.79; P = 0.055). An attempt at identifying a useful linear discriminant function by stepwise analysis of the three variables (CRP, ESR and WBC) yielded a function with a very low predictive value (<50% of correctly predicted cases). Discussion. In children with acute GE Borgnolo et al.14 showed increased CRP values in patients with bacterial infections, suggesting that a value >1.2 mg/dl should be considered sufficient evidence to obtain a culture and start empiric antibiotic treatment. These authors found an area under the ROC curve that was similar, although slightly better, than ours (0.9011 vs. 0.8674), and their sensitivity and specificity were similar (0.77 and 0.89 in their data vs. 0.93 and 0.70 in our data at the 1.2-mg/dl threshold, respectively). However, our interpretation of the data is not the same. We observed in our patients a positive predictive value of only 0.29 at the 1.2 mg/dl (and 0.35 at the 2.0) threshold, compared with values of 0.91 and 0.97 in the study by Borgnolo.14 This means clinically that fewer than one in three of our children with a CRP value ≥1.2 actually had Salmonella infection. This contrasting result is due to the very different a priori probability of a bacterial infection in our group (0.12 vs. 0.47 in the patients of Borgnolo et al.14). Furthermore in their statistical analysis they compared bacterial vs. viral GE alone, excluding children with GE of unknown cause, thus giving an even higher (0.60) a priori probability of bacterial infection. This prior probability, although similar to previously reported data in Italy (0.60),11 is much higher than that reported in another study in a hospital setting (0.05).15 Because like Borgnolo et al.14 we did not screen for uncommon bacteria, including Aeromonas hydrophila, Plesiomonas shigelloides, Bacillus cereus, Campylobacter spp. and invasive Escherichia coli subgroups, it is possible that the two PCR outliers in the undetermined cause group actually represent infections by these bacteria. As in many other conditions in medicine, the positive and negative predictive values of a test are substantially influenced by the a priori probability of the condition being tested, such that even high sensitivity and specificity coupled with a good area under the ROC curve are not sufficient to yield a reliable discrimination in practice. In conclusion we suggest that WBC count and ESR values are of no practical utility in the discrimination of bacterial GE in children; furthermore our results do not support the use of CRP as a single screening test to distinguish between Salmonella and nonbacterial GE, at least in conditions of low prevalence of bacterial infections. However, the very good negative predictive value of a low (<2.0) CRP value may allow clinicians to reliably exclude Salmonella as a cause of gastroenteritis. Gian Franco Meloni, M.D. Paolo A. Tomasi, M.D., Ph.D. Paolo Spanu. M.D. Sergio Piga, M.D. Andrea Porcu, M.D. Clinica Pediatrica e Neonatologica; Università degli Studi di Sassari; Sassari (GFM, PS, SP, AP) Servizio di Pronto Soccorso ed Accettazione; Ospedale "S. Giovanni di Dio"; Olbia (PAT) Italy