By studying simple and routinely used clinical and biological markers of infection in patients with confirmed nosocomial infection compared to non-infected ICU patients we determined that three parameters, temperature > 38.6°C, PCT > 1.86 ng/mL, and CRP > 88 mg/L, could perform well in discriminating infected from non-infected patients (AUCs of 0.88, 0.84, and 0.88 respectively). The complementary sensitivity/specificity profiles of each marker (76%/94% for fever, 68%/91% for PCT, and 92%/70% for CRP) allowed the construction of a composite score (score = 0.068 × D0 PCT + 0.005 × D0 CRP + 0.7 × temperature) more discriminating and highly specific than each single component (AUC of 0.90 and sensitivity of 97%).
Given the absence of any gold standard of infection we used the same methodology as Povoa et al.  and distinguished between patients with a confirmed diagnosis of noscomial infection, with all the limitations inherent to diagnosis of infection in the ICU, and patients discharged from the ICU without being treated by antimicrobials and therefore very unlikely to be infected.
Although one of the most frequently measured parameters in the ICU setting and non invasive, body temperature remains a poor indicator of infection . Temperature can be influenced by a number of non-infectious factors, such as non-infectious causes of fever and antipyretic therapy . Nevertheless, in the present study, temperature appeared to perform reasonably well for identifying nosocomial infection, with an AUC of 0.88. The best temperature cut-off value was 38.6°C, resulting in a sensitivity-specificity of 76%/94%.
Our data support the view of some authors that leucocyte count has little value in discriminating patients with nosocomial infection . In our study, AUC of WBC was 0.62, indicating that WBC was close to the line of non discrimination.
PCT is secreted as part of the systemic inflammatory response to infection and serum values are greatly based on the type and severity of infection. Interpretation of the literature is further complicated by frequent discrepancies or variations in the choice of the cut-off value of PCT, etiologies of infection, severity of infection, and study populations [11, 12]. Serum values of PCT vary greatly based on the type and severity of infection. The highest PCT concentrations have been reported in patients with septic shock, and patients with severe sepsis had significantly higher PCT levels than patients with sepsis or SIRS .
In the present study, we used an ultrasensitive assay for PCT, capable of measuring low levels to identify even “subclinical” inflammatory states before the development of clinically evident sepsis. The best cut-off value of PCT for identifying ICU-acquired infection was 1.86 ng/mL, with high specificity (91%) but low sensitivity (68%).
Few studies have analysed the behaviour of PCT in nosocomial ICU-acquired infection. In cardiac surgery patients, PCT measurement was found a reliable marker for diagnosis of infection: with a cut-off of 1 ng/mL allowing a sensitivity of 85% and a specificity of 95% . In another study, PCT was useful in the diagnosis of VAP with a cuff-off value of 3.9 ng/mL allowing a specificity of 100% at the cost of a sensitivity of 41% .
In a recent study assessing PCT monitoring in the early diagnosis of nosocomial infection, PCT at D0 was the best predictor of proven infection. A cut-off value of 0.44 ng/mL provided sensitivity and specificity of 65.2% and 83%, respectively for discriminating patients with proven nosocomial infection from clinically suspected but non-proven nosocomial infection .
Several reports suggested that PCT should replace CRP as a marker of infection in the ICU setting [17, 18]. In certain situations, especially to differentiate bacteremic from non-bacteremic infections, PCT was reported to be superior to CRP . A cut-off value of 0.4 ng/mL was associated with a negative predictive value of 98%. However, well-designed studies have shown that PCT is neither a better nor an earlier diagnostic marker of infection than CRP [20–22].
In a previous meta-analysis performed by Tang et al., PCT could not reliably differentiate sepsis from other non-infectious causes of systemic inflammatory response syndrome in critically ill adult patients . On the other hand, in another review, PCT concentration was found to be better than CRP for diagnosis of bacterial infection . However, this review included studies across a wide range of age group, clinical setting and spectrum diseases: 46% were paediatric patients and 57% did not have SIRS. Additionally, in some clinical situations of infectious origin commonly found in ICU, PCT can be normal or even undetectable early course of infections , localised infections , or subacute endocarditis [19, 26, 27].
CRP is an acute-phase protein, member of the pentraxin family of proteins, whose hepatic synthesis is triggered by cytokine release due to any cause of inflammation, infectious or not. In the present study, CRP D0 had the highest AUC (0.88) and a cut-off of 88 mg/L had a sensitivity of 92% and a specificity of 70% to identify patients with ICU-acquired infection. Povoa et al. studied the role of CRP to detect infections in critically ill patients . In that study, the combination of CRP > 87 mg/L and body temperature > 38.2°C was associated with infection diagnosis with a specificity of 100%. A study using the methodology we based our study upon, found that daily CRP monitoring could be used as a marker of infection prediction . Patients presenting maximum daily CRP variation > 4.1 mg/dL and a CRP level > 87 mg/L had an 88% risk of infection.
As all preceding studies we have confirmed that no single parameter or biomarker can reliably assist the clinician in diagnosing infection in the ICU. This is most probably due to: the lack of any possible gold standard for the diagnosis of infection and the extreme heterogeneity of infection in the ICU as to causal pathogen, underlying diseases, host-response, therapy, evolution and outcome. Facing these problems, our subsequent combination of diagnostic makers appears a useful approach to improve the accuracy in diagnosing nosocomial infection in ICU patients. Our results showed that combining PCT, CRP and temperature D0 was discriminant (highest AUC = 0.90) and highly specific (specificity of 97%). We found in our study a complementarity of CRP and PCT, with low specificity/high sensitivity for PCT and high sensitivity/low specificity for CRP. Such a combined clinical and multibiomarker approach for prediction has been proposed in another highly heterogenous complex disease in the ICU: adult respiratory distress syndrome (ARDS). Ware et al., studying the prediction of mortality in the ARDS cohort from the NHLBI studies, found that a combination of biomarkers and clinical predictors was superior to clinical predictors or biomarkers alone .
Some limitations of the present investigation should be noted. Our findings are based on a single centre study, one should be cautious in or extrapolating these data. Our findings cannot be generalised to specific diseases (pancreatitis, burns) or settings (cardiovascular surgical patients, neonatal/paediatric patients). Due to our study design, our findings might be only applicable to patients with late-onset nosocomial infection. Likewise, our focus on late-onset nosocomial infection led to a small number (17%) of documented infections which could also be a limitation.
It must also be acknowledged that CRP at D0 had a negative predictive value higher than our composite score (97% vs. 94%), showing that low CRP, under the threshold of 88 mg/L could assist clinicians in eliminating the diagnosis of nosocomial infection. However, the composite score had by far the highest positive predictive value (87% vs. 47%), suggesting that it could best be used in encouraging clinicians to initiate antimicrobial therapy when faced with a suspected diagnosis of nosocomial infection and a CRP over 88 mg/L.