Introduction The diagnosis of Spontaneous Bacterial Peritonitis (SBP) is based (via consensus) on a neutrophil count of 250/mm³. Traditionally this has been ascertained manually via light microscopy and counting chambers. Current AASLD and EASL guidelines cite recent promising validation studies of modern automated cell counters showing high diagnostic sensitivity and specificity when compared to the “gold standard” of light microscopy.

A retrospective local analysis of the correlation between synchronous automated and manual ascitic leucocyte counts in cirrhotic patients was undertaken in our NHS Trust.

Method All automated cell counts performed during 2014 were included in the analysis. 162 ascitic samples were identified and concurrent manual cells counts were sought. The automated counter used was the Beckman Coulter UniCel DxH 800™ cellular analysis system and the microscopy was performed using the NanoEnTek™ disposable haemocytometer. An automated leucocyte count of ≥ 0.3 × 10^–9and a manual neutrophil count of ≥250 × 10^–6 were deemed positive. Corresponding bacteriology results and serum albumin-ascites gradients (SAAG) were analysed.

Robust linear regression, Bland-Altman and ROC analyses were undertaken in addition to calculation of sensitivity, specificity, and positive/negative predictive value of automated counts.

Results 56 diagnostic paracenteses on cirrhotic patients with both automated and manual cell counts were identified. Robust linear regression of the automated count on the manual count gave an R²of 0.4 with a positive bias (non-zero intercept). Bland-Altman analysis confirmed this bias and identified spontaneous outliers. ROC analysis gave an optimal operating point of sensitivity 0.78 and specificity 0.8 at an automated count threshold of 0.35 × 10⁹/L.

The prevalence of SBP was 7.14% with 4 cases of SBP by manual neutrophil count. 19 results were deemed false positive by the automated counter: none was falsely negative: 33 were truly negative. Sensitivity of the automated counter was 100% (95% CI: 40.23–100%) and specificity was 63.46% (95% CI: 48.96% > 76.37%). Positive predictive value was 17.39% (95% CI: 5.06–38.8%) and negative predictive value was 100% (95% CI: 89.32–100%). No organisms were seen on Gram stain. Three cases of bacterascites with neutrophil counts of <250/mm³all grew coagulase negative Staphylococci. No cases of culture-positive SBP were identified in this series.

Conclusion In contrast to previous published validation data assessing the diagnostic accuracy of automated cell counters, local experience of automated leucocyte counts in the rapid assessment of SBP has shown an unacceptably low specificity and positive predictive value. This would lead to significant overtreatment if this assay replaced the traditional manual method in its current form.

Disclosure of interest None Declared.