A retrospective study was conducted in order to examine the predictive power of three different scoring systems for BSI in a hematological unit. The literature regarding predictive power of IPS in hematology/oncology patients is sparse. Most of the information about the epidemiology of BSIs and outcome is extrapolated from studies of hospitalized patients in general hospitals or from studies of patients in University hospitals  or, more specifically, from studies of neutropenic cancer patients . In our study, the incidence density of BSIs (21.99 per 1,000 patient-days at risk) was higher than that reported by previous studies (11.2 per 1000 patient-days at risk) . Furthermore in the current study the crude mortality (5.9%) was lower than that reported elsewhere . One of the major problems associated with comparison studies, is the difficulty in interpretation caused by variability in study designs and efficacy or outcome measures . Furthermore, many patients with neutropenia have complex medical profiles, leading to increased physician concern and a tendency to modify regimens during the treatment course of febrile episodes. Such modifications generally are made in a setting of inadequate information or lack of definitive diagnosis. It has been estimated that empiric regimens may be modified in 40% to 60% of cases .
Furthermore, we found that the most common isolates were gram-negative (44.4%) and gram positive (44.4%) microorganisms. Our findings are very similar to those reported in other studies . It is noteworthy that a considerable shift in the spectrum of pathogens isolated from blood cultures of febrile neutropenic cancer patients in the USA has emerged since 1995, with Gram-positive cocci increasing from 62% to 76% of isolates , Gram-negative bacilli declining from 21.5% to 14% and fungi declining from 15 to 8%. Based on consecutive clinical trials conducted by the European Organization on the Research and treatment of cancer (EORTC) between 1973 and 2000, a shift toward the predominance of Cram-positive bacteria has also been reported however, during the most recent time period (1998 - 2000), Gram- positive and Gram-negative isolates have been involved in the same magnitude .
A number of epidemiological factors may have contributed to this changing pattern. Firstly, this period corresponds to the uptake of widespread empiric antibiotic therapy for febrile neutropenia especially fluroquinolones and third/fourth generation cephalosporines), with resulting selection pressure for gram-positive bacteria . Secondly, the use of more intensive treatments with more severe oral mucositis may provide a frequent portal of entry for gram-positive organisms. Thirdly, increased utility of medium to long-term venous access devises for patients with malignancy may increase gram-positive blood stream infections (particularly staphylococcal infections . Finally, local infection control practices may impact upon the number of infections and spectrum of causative organisms. Further investigation of the data of patients with BSIs showed that most patients had received third/fourth generation cephalosporines (53%) and quinolones (35.3%), and they had central venous (11%) and Hickman catheters. Our findings highlight that the ongoing co-operation between haematologists, oncologists and infectious disease specialists is important to detect trends in epidemiology, which can be used to design empirical antibiotic regimens and guide infection control policies .
The mean IPS score was higher in patients with a BSI. This is a common finding in the two other researches supporting the discriminative adequacy of the tool [13, 28]. The two previous studies have confirmed the predictive power of the IPS for the onset of a Healthcare Associated Infection (HCAI). The current study extends the use of IPS as a predictor of the onset of a BSI. Additionally, the IPS separated infected from non-infected patients already on the first day of evaluation . This finding enables the therapeutic team to early detect the infected patients, as well as to avoid the administration of unnecessary antibiotics to the patients that are unlikely considered to be infected. Further studies with a bigger sample would be useful to give more evidence to this argument.
Between the three different prognostic scoring systems, IPS had the best sensitivity in predicting BSI. It is worth noting that IPS as a risk factor for BSIs in cancer patients has never being studied before. A previous study has used IPS to predict the need for mechanical ventilation and the duration of mechanical ventilation  and for the prediction of HCAI in the ICU patients . In a previous study  the cutoff point for the prediction of a HAI was 14 with a positive predictive value of 53.6% and a negative predictive value of 89.5%. In our study the cutoff point for the prediction of a BSI was 10 with a positive predictive value of 37% and a negative predictive value of 92.5%. The patients were considered as unlikely to be infected with a BSI if they had a score < 10 points. An IPS score >10 represents an argument against a BSI onset in oncology-hematology patients and should be considered as a complementary tool for the early detection of a BSI. In the present study the best cut-off value of IPS for the prediction of a BSI was 10 as opposed to 14 that has been previously reported from Martini et al.  and Peres-Bota . One possible explanation is that the present research has been conducted in a sample of hematology/oncology patients who had a different nosological profile compared to the ICU patients. Furthermore Martini et al.  have used a priori the cut-off value 14 without testing its sensitivity and specificity as opposed to our study that has used ROC analysis to find the best cut-off value. A further research is needed to compare IPS scores between several categories of patients. A multi-centered study would be useful to test the reliability of the tool in patients with a similar pathology.
The AUC value of IPS is associated with a good test as it is above 0.70 with means that the probability that the patients will be correctly classified as an infected person is 72%. It may be considered that a bigger sample of bloodstream infected person would help us to make safer conclusions. Furthermore it could be argued that the tool is more sensitive in predicting the onset of a HAI compared with the onset of a BSI. In conclusion the calculation of the IPS score is simple and the variables used can be obtained routinely, without additional costs. Additionally it is easy to be used from the clinicians.
To reduce the risk of BSIs and associated mortality in the hematology population, a number of strategies may be implemented including the routine use of IPS. Early identification of a changing spectrum and antimicrobial sensitivity of isolates should always be monitored. Frequently, a multi-disciplinary approach is recommended, to facilitate certain interventions such as typing of isolates, ward cleaning and structural modification of clinical areas. Guidelines for empirical antibiotic therapy for febrile neutropenia must be based upon the risk of infection in specific hematology patients and the local epidemiology and susceptibility of infecting isolates.