[Clinic predictive model and validation of anaerobic bacteremias (including polymicrobial bacteremias)]

Anaerobic bacteremias are uncommon. There is no agreement on their clinical predictability and the usefulness of anaerobic blood cultures. The objective of this study was to develop and validate a model for the prediction of anaerobic bacteremias.

The developing model was created with 984 bacteremias (45 anaerobic bacteremias) during 1985-1986 and 1996-1997. The validation model was made with 320 bacteremias during 2005-2006.

Independent multivariate predictors of true anaerobic bacteremia were used to develop a model stratifying patients with scores of 0 to 13 points(p), which were: unknown focus OR 3.46 (CI: 1.13-10.54) 3 p; abdominal and skin focus OR 14.85 (CI: 6.37-34.62) 6p; hypotension OR 1.99 (CI: 0.98-4.04) 2p; absence of vascular manipulations OR 2.62 (CI: 1.04-6.60) 2p and age over 60 years OR 3.21 (CI: 1.19-8.67) 3p. In the derivation sets group with more than 7p the model had Sensitivity: 77.8%, Specificity: 78.3%, PPV:14.7%, and a NPV of 98.6%. The area under curve was ROC=0.84 (SE=0.011), 95% CI: 0.82-0.86 with an anaerobic bacteremia prevalence of 4.6%. The validation set was studied analysing 320 bacteremias. Of these, 83.6% (95% CI: 71.19%-92.23%) of anaerobic bacteremias had more than 7 points, and 72.7% had more than 9 points. There was 26.4% (95% CI: 21.2%-32.15%) aerobic and facultative anaerobic bacteremias with more than 7 points, and only the 11.7% with 9 or more points. The area under the curve was, ROC=0.82 (SE=0.02), 95% CI:0.78-0.86, and estimated prevalence, 2%.

Abdominal and skin focus OR 14,85; unknown focus OR 3,46; hypotension OR 1,99; absence of vascular manipulations OR 2,62 and age over 60 years enable us to make a predictive clinical model of probability of anaerobic bacteremia with a high sensitivity and specificity. The model particularly has a significant predictive negative value due to the low prevalence of anaerobic bacteremia.