Time to blood culture positivity as a predictor of clinical outcomes and severity in adults with bacteremic pneumococcal pneumonia

Objectives We aimed to investigate the association between the time to positivity of blood culture (TTP) with clinical outcome and severity of pneumococcal bacteremic pneumonia. Methods Prospective observational study carried out in 278 hospitalized adult CAP patients with positive blood culture for Streptococcus pneumonia (2003–2015). Results A total of 278 cases of bacteremic pneumococcal pneumonia were analyzed, median age 62 (46; 79) years. Fifty-one percent of the cases had PSI IV-V. Twenty-one (8%) died within 30-days after admission. The analysis of the TTP showed that the first quartile of the TTP (9.2h) was the best cut-off for differentiating 2 groups of patients at risk, early (TTP <9.2 h) and late (TTP ≥9.2 h) detection groups (AUC 0.66 [95% CI 0.53 to 0.79]). Early TTP was associated with a statistically significant risk of invasive mechanical ventilation (18% vs. 6%, p = 0.007), longer length of hospital stay (12 days vs. 8 days, p<0.001), higher in-hospital mortality (15% vs. 4%, p = 0.010), and 30-day mortality (15% vs. 5%, p = 0.018). After adjustment for potential confounders, regression analyses revealed early TTP as independently associated with high risk of invasive mechanical ventilation (OR 4.60, 95% CI 1.63 to 13.03), longer length of hospital stay (β 5.20, 95% CI 1.81 to 8.52), higher in-hospital mortality (OR 5.35, 95% CI 1.55 to 18.53), and a trend to higher 30-day mortality (OR 2.47, 95% CI 0.85 to 7.21) to be a contributing factor. Conclusion Our results demonstrate that TTP is an easy to obtain surrogate marker of the severity of pneumococcal pneumonia and a good predictor of its outcome.

Introduction Streptococcus pneumoniae remains the most frequent cause of community-acquired pneumonia (CAP) [1,2]. Bacteremia is documented in 25% of cases [3] and their mortality is 15% to 26% greater than in non-bacteremic patients [4]. The identification of early predictors of worse outcome in patients with bacteremic CAP due to S. pneumoniae is therefore of utmost importance. There is evidence about the association between the high bacterial load and worse clinical outcomes in invasive pneumococcal pneumonia [5,6]. This evidence suggests that determination of pneumococcal load has a clinical utility. Some previous studies suggest that time to positivity (TTP) of blood culture may provide early clues about microorganisms involved and the source of bacteremia [7]. Also, TTP is inversely associated with blood bacterial load (8) and is therefore a reasonable marker of more severe disease and a potential early predictor of mortality.
Two previous studies analyzed the TTP in children and adults with S. pneumoniae bacteremia [8,9]. The experience in children included 175 episodes from different sources (meningitis, pneumonia) and no association was found between TTP and clinical or laboratory parameters except that of the 150 patients seen in the emergency department, those in the first decile were significantly more likely to be admitted to the hospital than were patients in the 10th decile (10 of 15 patients [67%] vs. 5 of 16 patients [31%]; p = 0.02) [9]. In contrast, the experience in adults included 105 episodes and shorter TTP was associated with immunosuppression, severe sepsis and shock, meningitis or ICU admission [8]. However, the number of patients with CAP included in these studies was low and the authors did not evaluate the potential relationship with mortality.
S. pneumoniae is capable of producing a different phenotypic expression, depending on the capsular serotype; it is known that certain serotypes can cause a more invasive disease than others [3]. The higher virulence of these serotypes could be associated with a shorter TTP.
The aim of our study was to evaluate the TTP in prospectively collected episodes of bacteremic CAP due to S. pneumoniae from 2003 to 2015 and to correlate the TTP with the severity of CAP, the S. pneumoniae serotype and the length of hospital stay, in-hospital mortality rate, 30-day mortality rate, ICU admission rate, ICU mortality rate, length of stay in ICU, and need of mechanical ventilation. population of 540,000 inhabitants. We excluded patients who were immunosuppressed, receiving immunosuppressant (those taking >10 mg/day of prednisone or cytotoxic therapy) and all patients known to have human immunodeficiency virus infection.

Data collection and evaluation
At the initial visit, patients underwent a complete clinical history and physical examination. Patients were stratified into risk classes using the validated prediction rule calculated according to the Pneumonia Severity Index (PSI) score [10]. We also calculated the CURB-65 [11] and the sequential organ failure assessment (SOFA) [12] scores at admission. Empirical antibiotic treatment was administered according to the Infectious Disease Society of America/American Thoracic Society (IDSA/ATS) guidelines for management of CAP [13]. All surviving patients were visited at 30-40 days after discharge.

Microbiological evaluation and diagnostic criteria
Regular sampling included sputum specimens, two blood cultures, urine samples for detection of S. pneumoniae (BinaxNOW S. pneumoniae Urinary Antigen Test; Emergo Europe, The Hague, The Netherlands) and Legionella pneumophila serogroup 1 (BinaxNOW L. pneumophila Urinary Antigen Test; Trinity Biotech, Bray, Ireland). Samples from pleural fluid puncture, tracheobronchial aspiration (TBAS) and blind bronchoalveolar lavage (BAL) were obtained according to the judgment of the attending physician.
The etiology of pneumococcal pneumonia was determined in cases with a positive valid sputum culture, positive blood culture; positive pleural fluid and transthoracic needle aspiration cultures; positive urinary antigen for S. pneumoniae; bacterial growth in cultures of TBAS !10 5 CFU/ml, in Protective Brush Sample (PBS) !10 3 CFU/ml, and in BAL !10 4 CFU/ml. For the purpose of this study we only included patients with positive blood culture.
Blood cultures were processed by the BACTEC 9240 system (Becton-Dickinson, MD, USA), and vials were loaded into the machine around the clock. Volumes between 8 to 10 ml of blood samples were inoculated into aerobic and anaerobic vials. The vials used were the resin-containing BACTEC Plus Aerobic/F and BACTEC Plus Anaerobic/F or the non-resincontaining BACTEC Standard 10 Aerobic/F and BACTEC Lytic/10 Anaerobic/F. The incubation period was 5 days before being discarded as negative. The identification of microorganisms isolated from positive cultures was performed according to conventional methods.
Pneumococcal isolates were identified by standard microbiological methods. All strains isolated from normally sterile sites were routinely frozen at -70˚C in skimmed milk until serotype detection was performed. Isolates were serotyped at the Spanish Reference Laboratory for Pneumococci (Majadahonda, Madrid, Spain) by using the Quellung reaction (antisera provided by the Statens Serum Institute Copenhagen, Denmark) and/or dot blot analysis [15].

Definitions
Pneumonia was defined as the presence of a new infiltrate on a chest radiograph together with clinical symptoms that were suggestive of lower respiratory tract infection (e.g., fever, cough, sputum production, pleuritic chest pain).
Prior antibiotic treatment was considered when antibiotics had been taken in the previous month.
Bacteremic pneumococcal pneumonia was defined as the presence of a positive blood culture for S. pneumoniae.
The TTP was determined from the time interval between the start of incubation and the detection of microbial growth in peripheral blood, as documented using an automated monitoring system. When multiple cultures were positive only the shortest TTP was selected for analysis.
Severe CAP was defined according to the IDSA/ATS guidelines criteria [13]. Pulmonary complications were defined by the presence of pleural effusion, empyema, or multilobar infiltrates. Extra-pulmonary complication was defined by the presence of septic shock and acute renal failure.

Clinical outcomes
The primary outcome was in-hospital mortality. Secondary outcomes included length of hospital stay, 30-day mortality, ICU admission, length of stay in ICU, ICU mortality, and need of mechanical ventilation.

Statistical analysis
Data are shown as number and percentage of patients for categorical variables and median (Quartile 1 [Q 1 ]; Quartile 3 [Q 3 ]) for continuous variables with non-normal distribution or mean (standard deviation [SD]) for those with normal distribution. Categorical variables were compared using the X 2 test or the Fisher exact test. Continuous variables were compared using the t-test or the nonparametric Mann-Whitney test. A receiver operating characteristic (ROC) curve was constructed to determine the best cut-point for TTP to predict in-hospital mortality. Youden's index [19] was defined for all points along the ROC curve, and the maximum value of the index was used as a criterion for selecting the optimum cut-off point. Regression analyses [20,21] were used to examine the associations between outcomes (a linear regression analysis for length of hospital stay, two logistic regression analyses for in-hospital mortality and 30-day mortality, and a multinomial logistic regression analysis for non-invasive or invasive mechanical ventilation) and risk factors. In a first step, each risk factor was tested individually. In a second step, all risk factors which showed an association in the univariate model (p<0.10) were added into the multivariate model. Finally, a backward stepwise selection (p in <0.05, p out <0.10) was used to determine factors associated with outcome. The beta coefficient (β) and 95% confidence interval (CI) and the odds ratio (OR) and 95% (CI) were calculated where applicable. The Hosmer-Lemeshow goodness-of-fit test was performed to assess the overall fit of the logistic regression models, the R 2 for the linear regression model, and the Cox and Snell R 2 and the Nagelkerke R 2 for the multinomial logistic regression model. Internal validation of the prediction models was conducted using ordinary nonparametric bootstrapping with 1,000 bootstrap samples and bias-corrected, accelerated 95% CIs [22]. The area under the ROC curve of the multivariate models to predict in-hospital mortality, 30-day mortality, non-invasive and invasive mechanical ventilation were calculated. Simple imputations of random effects were used, if necessary, for variables with missing values. The level of significance was set at 0.05 (2-tailed). All analyses were performed using IBM SPSS Statistics version 22.0 (Armonk, New York, USA).

Participants
During the study period, 4,639 patients were admitted in the Emergency Department with the diagnosis of CAP. Blood cultures were performed on 3,274 (71%) and were positive in 419 (13%). Of these, 301 (72%) were positive for S. pneumoniae, and 23 were excluded from the analysis due to missing TTP, to having human immunodeficiency virus infection and/or who were receiving immunosuppressant. Therefore, 278 cases were finally included in the study.

Time to positivity of S. Pneumoniae
The median TTP of S. pneumoniae in blood culture among the 278 adult patients with pneumococcal CAP was 10.5 (9.2; 11.5) hours (Fig 2). Following Youden's index methodology, we Time to positivity in bacteremic pneumococcal pneumonia selected 9.2 h as the optimal cut-off point for TTP in relation to in-hospital mortality (Fig 3) (53% sensitivity, 77% specificity, 14% positive predictive value, 96% negative predictive value, 2.31 positive likelihood ratio, and 0.61 negative likelihood ratio). Patients were divided into two detection groups: early detection group (TTP <9.

Comparison of characteristics of early and late detection pneumonia
Baseline characteristics comparing cases on early detection group and late detection group appear in Table 1.
We found non-significant differences in the use of previous antibiotics and pneumococcal vaccine between groups. Compared with the late group, cases in the early detection group presented fever at admission less frequently, more days of symptoms, low white blood cell count, low lymphocyte count, higher serum levels of C-reactive protein, and worse oxygenation at admission. Also, the early detection group presented a higher median PSI score, more severe CAP according to the IDSA/ATS definition, and more pulmonary complications.  We found no difference between groups in ICU admission, length of ICU stay or ICU mortality.
We added early detection (TTP <9.2 h) to the CURB-65 and PSI scores to predict mortality, and the effectiveness of the new scores was compared. The addition of early TTP improved

Predictors of length of hospital stay
The simple linear regression analysis revealed several variables significantly associated with length of hospital stay ( Table 3). The variables PCR !15 mg/dl, PSI IV-V, ARDS and early detection (TTP <9.2 h) were those independently associated with length of hospital stay in the multiple analysis. Internal validation of the linear regression model was conducted using bootstrapping with 1,000 samples. The three variables included in the model demonstrated robust results, with small 95% CIs around the original coefficients.

Predictors of in-hospital mortality
The univariate logistic regression analysis revealed several variables significantly associated with in-hospital mortality (Table 4). Among these variables, age !65 years, acute renal failure, septic shock, ARDS, and early detection (TTP <9.2 h) were the variables independently associated with in-hospital mortality in the multivariate analysis.
The area under the ROC curve was 0.91 (95% CI 0.85 to 0.98) for the model predictive of in-hospital mortality. Internal validation of the logistic regression model using bootstrapping with 1,000 samples demonstrated robust results for the five variables included in the model, with small 95% CIs around the original coefficients.

Predictors of 30-day mortality
The univariate logistic regression analysis revealed several variables significantly associated with 30-day mortality ( Table 5). The variables, PSI IV-V, ARDS, acute renal failure, and septic shock were those independently associated with 30-day mortality. There was a trend for early detection (TTP <9.2 h) to be a contributing factor for 30-day mortality.
The area under the ROC curve was 0.88 (95% CI 0.79 to 0.96) for the model predictive of 30-day mortality. Internal validation of the logistic regression model using bootstrapping with 1,000 samples demonstrated robust results for three of four variables included in the model, Time to positivity in bacteremic pneumococcal pneumonia with small 95% CIs around the original coefficients, while the PSI appeared to be less reliable, with a wider 95% CI around the original coefficient.

Predictors of the need for non-invasive or invasive mechanical ventilation
The following risk factors showed significant associations with mechanical ventilation groups in individual multinomial logistic regression and were thus used for the initial multivariate model: alcohol consumption, chronic respiratory disease, lymphocytes, PSI risk class, ARDS, and septic shock and early detection time. Results of the multivariate model are displayed in Table 6. For non-invasive mechanical ventilation, the model shows the OR to be significantly increased if they have prior alcohol consumption, PSI risk class IV-V, and septic shock. The OR for invasive mechanical ventilation, however, was strongly increased with septic shock and early detection (TTP <9.2 h). The area under the ROC curve was 0.75 (95% CI 0.65 to 0.84) for the model predictive of non-invasive mechanical ventilation, and 0.81 (95% CI 0.70 to 0.91) for the model predictive of invasive mechanical ventilation.
Internal validation of the multinomial logistic regression model using bootstrapping with 1,000 samples demonstrated robust results for the four variables included in the model, with small 95% CIs around the original coefficients.

Relationship between time to positivity and pneumococcal serotypes
In order to analyze the association between pneumococcal serotypes and time to positivity, pneumococcal serotypes were divided into three groups according to invasiveness: low (serotypes 3, 6A, 6B, 19A, 19F and 23F), intermediate (  There were no differences in the rate of resistance to penicillin (12% vs. 9%, p = 0.73) nor macrolides (19% vs. 14%, p = 0.28) between groups.

Discussion
A TTP of blood culture shorter than 9.2h in patients with bacteremic pneumococcal pneumonia is independently associated with a more severe disease characterized by a higher C-reactive protein level, a worst oxygenation, and more pulmonary complications which explain the more frequent need of invasive mechanical ventilation in the early detection group. However, we did not find an association between TTP <9.2 h and septic shock, which could be due to the lower number of septic shock patients. In line with this, a TTP <9.2 h was independently associated with the risk of mechanical ventilation, longer length of hospital stay, in-hospital mortality and there was a trend towards higher 30-day mortality. These findings concord with existing data that describe the strong association between the bacterial load and clinical outcomes in invasive pneumococcal diseases [18,23]. These studies were performed with the polymerase chain reaction and showed a higher sensitivity in comparison to blood cultures, although this method cannot measure the viability of the bacteria. Also, our results are in accordance with the results of study by Werno et al. that reported the association of higher pneumococcal load with increased disease severity in adults with CAP [6]. Abbreviations: ARDS = acute respiratory distress syndrome; CI = confidence interval; ICU = intensive care unit; OR = odds ratio; PSI = pneumonia severity index.
Data are shown as estimated ORs (95% CIs) of the explanatory variables in the in-hospital mortality group. The OR is defined as the probability of membership of the group in-hospital mortality divided by the probability of membership of the non-in-hospital mortality group. The P value is based on the null hypothesis that all ORs relating to an explanatory variable equal unity (no effect). a Hosmer-Lemeshow goodness-of-fit test, p = 0.18.
b Predictors from the model can be used to calculate the probability of in-hospital mortality by the following formula: Exp (β)/(1+Exp(β)), where β = -6.42 Time to positivity in bacteremic pneumococcal pneumonia Abbreviations: ARDS = acute respiratory distress syndrome; CI = confidence interval; OR = odds ratio; PSI = pneumonia severity index. Data are shown as estimated ORs (95% CIs) of the explanatory variables in the non-invasive mechanical ventilation and invasive mechanical ventilation groups. The OR is defined as the probability of membership of the groups non-invasive or invasive divided by the probability of membership of the not ventilated group. The P value is based on the null hypothesis that all ORs relating to an explanatory variable equal unity (no effect).
Predictors from the model can be used to calculate the probability of non-invasive mechanical ventilation or invasive mechanical ventilation by the following formulas: Exp(β 1 )/(1+Exp(β 1 )+Exp(β 2 )) and Exp(β 2 )/(1+Exp(β 1 )+Exp(β 2 )), respectively, where β 1 = -4. 13  The concept of early TTP of blood cultures is directly related with a higher bacterial concentration in blood. Interestingly, the first finding of our study is related to the pathogenesis of pneumococcal infection. Patients with blood cultures with rapid growth of pneumococcus could be due to a dysregulation in the host innate immune response, shown by the low number of circulating leukocytes and a high inflammatory response measured by high levels of C-reactive protein [23][24][25][26][27].
These results emphasize that the key to improve the prognosis of patients with bacteremic pneumococcal pneumonia is to improve the host immune response against pneumococcus, and to learn to modulate the inflammatory response of the host. A higher bacterial load could lead to a higher inflammatory response when a β-lactam antibiotic is used [5]; this effect was observed in meningitis and sepsis [28] and was proposed as the key for benefits of corticosteroids, macrolides or fluoroquinoles in adjuvant treatment of pneumonia [29][30][31][32][33].
TTP was better studied in child patients with bacteremic disease by several pathogens such as S. pneumoniae, N. meningitidis and H. influenzae. A higher bacterial load in blood correlates with severity of the disease, also the bacterial load correlates with TTP in catheter related sepsis [34]. In paediatric patients there was no relation among TTP and severity although a positivity correlation was described in adults with pneumococcal invasive disease [8]. A recent small study about the relation of fluorescence rate using blood culture techniques and mortality in patients with invasive pneumococcal pneumonia found a direct association of FR and mortality. However, the authors did not find that TTP correlated with clinical outcomes [35].
Interestingly, no statistically significant differences were observed in the use of previous antibiotics and pneumococcal vaccine in our cohort of patients between groups. Also, there were no differences in resistance to penicillin or macrolides.
We found that early TTP was an independent risk factor for in-hospital mortality; however, only a trend for 30-day mortality was observed. This could be due to an under power analysis.
The prediction models we have presented are the first step in establishing more universal models; to move forward, our prediction models will need to undergo external validations with larger patient cohorts from multiple centers. We were able to apply internal validation techniques to understand how likely these models will be replicable in future studies and at other centers. Bootstrapping techniques were applied and demonstrated that the coefficients obtained from these prediction models were quite robust. PSI was the one factor for 30-day mortality model that the bootstrap results indicated might have limited repeatability in future work. Removal of the PSI from the model did not change which factors were significant predictors of 30-day mortality. However, because of the clinical importance surrounding the PSI, this variable was kept as a factor in this model despite some statistical limitations. In the real world clinical setting where this prediction model could be used, the PSI is an important clinical characteristic that can play a substantial role in decision making.
To the best of our knowledge, with the exception of one study [8] evaluating patients with invasive pneumococcal disease, this is the first study addressing the issue of early TTP specifically in pneumococcal pneumonia in a large adult population.
TTP is an easy to obtain parameter available in all Microbiology laboratories that appears to provide useful prognostic information, and should be reported routinely in order to help clinicians to identify patients at risk of worse outcome that could benefit from more aggressive early management.
Several limitations have to be addressed. First, because the data were collected from a single academic teaching hospital in Spain the results might not be able to be extrapolated to other patients admitted to other types of hospitals in other countries. Second, we could only analyze 278 patients and this sample size may result in a large type II error. Our sample size, however, is large since this is the only study regarding this issue. Our results support the direct relationship of early TTP with severe presentation and worse outcomes in patients with diagnosis of pneumococcal CAP.

Conclusion
In summary, we found evidence that in those patients with pneumococcal pneumonia that had early TTP, they were more severely ill at presentation and had worse outcomes.