This is a prospective cohort study performed in four different ICUs with a total of 34 beds; i.e., Anesthesia and Reanimation ICU (12 beds), Internal Diseases ICU (9 beds), Neurology ICU (7 beds), and Neurosurgery ICU (6 beds) in the Medical Faculty of Gazi University between October 2010 and October 2011. Approval was obtained from the Ethical Committee on 14 October 2010 (numbered with B-10-0-IEG-0-15-00-01/67444). The patients aged over 18 years, staying in an ICU for at least ≥48 hours, and diagnosed as BSI acquired from an ICU based on the criteria issued by Center of Disease Control and Prevention (CDC) [14] were included in the study. The patients treated for BSI for less than five days were not included in the study. The concepts, items, and measures used in the study are summarised in [Table/Fig-1] [15].

The patients were followed from admission to the ICUs until the end of treatment for the first BSI attack or until the 14^th day of treatment when the duration of treatment was extended for various reasons. APACHE II score [11], SOFA score [12], and diagnoses of sepsis, severe sepsis, and septic shock were recorded on admission to the ICUs, retrospectively 48 hours before development of BSI, retrospectively on the day when BSI developed, and on day 3, day 7, and the last day of treatment. Diagnoses of sepsis, severe sepsis, and septic shock were made according to definitions determined at “2001 International Sepsis Conference” [16].

Data were analysed with SPSS.15.0. Kolmogorov Smirnov test was used to determine whether collected data were normally distributed. Normally distributed continuous scores were analysed with t-test and data non-normally distributed were analysed with Mann-Whitney U test. Data about categorical variables were analysed with Fisher’s-exact test or contingency coefficient. The variables found to be significant in univariate analyses were included into the logistic regression analysis.

Characteristics of the blood stream infection sample: Success and failure groups; The study included 70 patients. Treatment was successful in 20 patients (28.6%) (success group), but it failed in 50 patients (71.4%) (failure group). Of 20 patients treated successfully, 11 (55.0%) had a cure for BSI and 9 (45.0%) had clinical improvement.

The univariate tests results (t-test or Fisher-exact tests) for differences between the “successful” and “failed” treatment groups can be seen in [Table/Fig-2]. In addition to the t-tests for means and Fisher-exact tests for 2×2 tables listed in [Table/Fig-2], the contingency coefficient was calculated for the causative agents or sources and the primary foci of the infection (infection in urinary, respiratory, cardiovascular, gastrointestinal or in intra-abdominal system, or infection related to prosthesis or implant), The characteristics of the BSI observed in the sample are listed in [Table/Fig-3]. There were statistically significant differences in neither in causative agents/sources (p=0.816) nor in the primary foci of the infection (p=0.725) between success and failure groups. Median test (Mann-Whitney-U test) results for APACHE II and SOFA score are presented in [Table/Fig-4,5].

As [Table/Fig-2] shows, no statistically significant differences between success and failure groups were found in any of the following variables: age (range: 19-92), gender (48.6% male), number of comorbidities (range: 0-6), presence of immunosuppression (25.7%), Charlson comorbidity index (range: 0-12), nor the time of catheter removal among catheter-related BSI cases (range: 0-4). Statistically significant differences between failure and success groups were found in days elapsed between the appearance of BSI and initiation of appropriate treatment (range: 0-13).

APACHE II and SOFA scores: In the failure group, APACHE II score was significantly higher 48 hours before the BSI development, on the day of BSI development, on day 3, and day 7, and the last day of treatment [Table/Fig-4]. SOFA score was significantly higher on day 3, on day 7, and the last day of treatment [Table/Fig-5].

Clinical changes in bacteraemia: On the day of BSI initiation, 10 patients (20%) had severe sepsis, and seven patients (14%) had a septic shock in the failure group, and three patients (15%) had severe sepsis, and one patient (5%) had a septic shock in the success group. On the third, the seventh, and the last days of treatment, the number of patients with severe sepsis were 13 (26%), 11 (28%), and 5 (10%), respectively, and the number of patients with septic shock were 9 (18%), 6 (15%), and 32 (64%), respectively, in the failure group. In the success group, only one patient (5%) had a septic shock on the third day of treatment, and the other patients had neither severe sepsis nor septic shock on the third, the seventh, or the last days of treatment.

Microbiological response: After 48-hours of appropriate treatment, the control blood culture was taken from 11 (55%) patients in the successful treatment group, and all of them (100%) had a microbiological response. In the failed treatment group, a microbiological response was present in 17 (40%) of 42 (84%) patients who were taken a control blood culture from. Of the 25 (60%) patients without a microbiological response, 5 (20%) had breakthrough bacteraemia, 14 (56%) had a superinfection, and 6 (24%) had both breakthrough bacteraemia and superinfection. In the remaining 17 patients from whom blood culture could not be taken (24%), the microbiological response was obscure.

Predictors of the treatment failure: Predictors of the treatment failure in the failure group are presented in [Table/Fig-6]. Presence of another infection focus (79%), and superinfection development (40%) were the most frequently observed conditions [Table/Fig-6].

Effects of initial treatment approach on treatment outcome: Of 70 cases of BSI, 40 (57%) were given empirical treatment, and 30 (43%) were given treatment directed towards the causative agent [Table/Fig-7]. All inappropriate treatments initiated in the failure group (16%) were related to the fact that the causative agents were resistant.

Effects of causative agents on the selection of initial treatment: Significantly higher rates of BSI with S.aureus were found with empirical treatment (p=0.027), and significantly higher rates of BSI with Candida spp. were found with treatment targeting the causative agent (p=0.016).

Multivariate analysis: Data were analysed with logistic regression analysis. Age, gender, and the variables found to be significant [Table/Fig-2] in addition to SOFA and APACHE II scores were included in the analysis. The results of the logistic regression analysis are listed in [Table/Fig-8].

According to the results, the rate of treatment failure was 1.15 times higher in the patients with higher APACHE II scores on the third day of treatment and 1.53 times higher in the patients more prolonged time elapsing from the onset of BSI to initiation of appropriate treatment.

In this study, the patients were evaluated in two aspects. First, the success and treatment groups were compared regarding factors likely to influence treatment outcomes. Results showed that higher APACHE II scores on the second day of treatment increased treatment failure by 1.15 times and that the longer time elapsing from the onset of BSI to the initiation of appropriate treatment increased treatment failure by 1.53 times. Second, the analyses showed that the presence of a concurrent infection focus and super infection were the most frequent predictors of treatment failure.

Several studies have investigated when the disease severity should be measured to reveal the real contribution of BSI to mortality in ICUs. Studies have shown that presence of severe disease on admission to the ICU [2,17-20]; on the day when BSI developed [9,13,21,22]; and 24-hours before BSI development [23,24] increase mortality. Based on their meta-analysis of 51 studies, McGregor JC et al., concluded that the optimal time to measure the severity of illness is immediately before the actual onset of BSI while they also recommended that most appropriate time should be 48-hours before the index blood sample is obtained for culture [25]. Another meta-analysis recommended that data should be obtained on admission to the ICU and 24 hours before BSI development [26]. The present authors recorded APACHE II and SOFA scores on the admission to the ICUs, 48-hours before the onset of BSI, and on the 2^nd day of BSI development. Also, APACHE II and SOFA scores were calculated on the 3^rd, the 7^th, and the last days of treatment, which has not been done in earlier studies. According to the results, APACHE II and SOFA scores on admission to the ICUs did not significantly differ between the success and the failure groups, which is in line with the literature [5,27]. However, the failure group had higher APACHE II scores 48 hours before the onset of BSI and on the day when BSI developed compared to the success group. This suggests that the patients failing to respond to treatment had a more severe disease 48 hours before BSI development. In this group, APACHE II and SOFA scores on the 3^rd, the 7^th, and the last days of treatment were also higher. The results show that the APACHE II scores can be more indicative than the SOFA scores on the 3^rd day of treatment.

The selection of appropriate empirical treatment at the onset of BSI reduces mortality [18,28-30]. Consistent with the literature, the rate of initiation of appropriate empirical treatment was significantly higher in the success group in this study [18,28-30]. Not only the effects of selection of appropriate treatment but also the time of initiation of the treatment on mortality has been studied. Garrouste-Orgeas M et al., found that late initiation of the treatment increased the mortality by 2.6 times [5]. The critical time elapsing from collection of blood culture specimens to initiation of appropriate treatment was reported to be 52 hours by Lodise TP et al., and 72 hours by Anderson DJ et al., [31,32]. In the present study, longer duration increased treatment failure by 1.53 times. The current study showed that the day when BSI developed, and the following day were critical to institute appropriate treatment.

The type of causative agent was found to be important in the early initiation of appropriate treatment. The BSI cases due to S.aureus were subjected to empirical treatment more frequently while cases with Candida were subjected more frequently to the treatment targeting causative agents. This suggests that infections due to S.aureus had a more noticeable clinical onset and, thus, they were more frequently treated with empirical antibiotics whereas those due to Candida species had a less overt clinical course and, consequently, were treated with antibiotics targeting causative agents. Moreover, it has been emphasised that the presence of a resistant causative agent causes selection of an inappropriate empirical treatment leading to late initiation of an appropriate treatment [33].

In the current study, the most common three predictors of treatment failure were the presence of a concurrent infection focus, superinfection, and the inability to bring the infection focus under control due to conditions other than antibiotic therapy, and breakthrough bacteraemia. Presence of a concurrent infection focus and superinfection increased the likelihood of treatment failure. Development of infection in more than one focus is not uncommon in ICU patients. In a study by Meduri GU et al., on causes of fever and pulmonary infiltration in ICU patients, 82% of the cases had one or more accompanying infection foci, the most common concurrent infections being pneumonia and catheter infection [34]. Temiz E et al., reported that 41.6% of the cases with a catheter-related urinary tract infection had an accompanying disease [35]. In the present study, 76% of the patients in the failure group had a concurrent infection focus, and BSI was most frequently accompanied by respiratory system infection followed by urinary tract infection. It should always be kept in mind that patients with fever and central venous catheters can have CR-BSI.

Firstly, the study period was limited to 12 months and four different ICUs in one hospital. Due to these constraints, the sample included 70 patients. With more extended period for data collection and inclusion of more hospitals into the study, the sample size could have been increased. Large sample size would have provided more variation in the cases and allowed more detailed analyses. On the other hand, a longer time period and a larger number of hospitals could have increased the bias caused by hospital-related factors like differences in workload and clinical practices. Similarly, long observation periods lasting several years can lead to unwanted variance in the data, since clinical practices, treatment procedures, and other hospital-related factors tend to change as time passes. Secondly, the findings may not be directly generalisable to different ICUs and hospitals worldwide or even in Turkey. In future studies, a larger sample of hospitals could be included, which would allow us to investigate variation among different hospitals and ICUs.

The most important independent variable which had an influence on treatment failure in cases of BSI is the late institution of appropriate treatment. High APACHE II scores detected on the 3^rd day of treatment can be a warning about treatment failure. Besides, one of the predictors of treatment failure was the presence of other infections or superinfection. Therefore, careful follow-up of ICU patients and enhanced compliance with precautions of infection control are essential for the detection of multiple infection foci and superinfection.