C-reactive protein, an early marker of community-acquired sepsis resolution: a multi-center prospective observational study

The SACiUCI study (Sepsis Adquirida na Comunidade e internada em Unidade de Cuidados Intensivos) was a prospective, multiple-center, observational study, designed to evaluate the epidemiology of CAS. Details of study design, definitions, data collection and management are provided elsewhere [29].

Briefly, all patients who were ≥ 18 years old and newly admitted with CAS to the participating ICU (see Additional file 1 for a list of the participating ICU) from December 2004 to November 2005 were consecutively enrolled. Almost two thirds of the patients with sepsis were admitted from the emergency room of the hospital of the participating ICU, and the remaining came from emergency departments of other hospitals. Patients were followed-up until death or hospital discharge. Only the first ICU admission was included. Since this observational study did not require any deviation from routine medical practice, Institutional Review Boards approved the study design and waived the need of informed consent.

Infection was defined as a pathologic process caused by the invasion of normally sterile tissue or fluid or body cavity by a pathogenic or potentially pathogenic microorganism [1] and/or clinically suspected infection, plus the prescription of antimicrobial therapy. Community-acquired infection was defined as the onset of infection prior to hospital admission or not present at admission but becomes evident in the first 48 hrs [30]. Presence of sepsis was defined according to the American College of Chest Physicians/Society of Critical Care Medicine Consensus Conference criteria [31].

The presence of health-care associated infection (HCAI) was defined at hospital admission according to the presence of the following criteria: home infusion therapy (including antibiotics) or home wound care; chronic dialysis or chemotherapy within 30 days; hospitalization for 2 days or more in the preceding 90 days; residence in a nursing home or extended care facility [32, 33].

Data were collected prospectively either using pre-printed case report forms, a specific base software or on-line through the study web page.

Data collection included demographic data and comorbid diseases. Clinical and laboratory data at the time of hospital admission was recorded. The Simplified Acute Physiology Score (SAPS) II [34] was calculated. Microbiological and clinical infectious data were reported. C-reactive protein, body temperature, white blood cell count (WBC) and the Sequential Organ Failure Assessment (SOFA) score [35] were evaluated during the first five days of ICU stay. Blood samples were obtained from an arterial line at ICU admission and subsequently every morning.

For purposes of the time dependent analysis Day 1 (D1) was defined as the day of ICU admission. The following days were successively named as D2, D3 up to D5.

The withdrawal of the inflammatory stimulus results in a sharp decrease of the CRP serum concentration in a way similar to a first-order elimination kinetics [36, 37]. Consequently, relative CRP variations are more informative that absolute changes. As a result we assessed a new variable, CRP-ratio, which was calculated in relation to D1 CRP concentration. Subsequently, we classified patients according to three patterns of CRP-ratio response to antibiotic therapy: fast reponse, if D5 CRP value was less than 40% of D1 value; slow response, if D5 CRP was between 40% and 80% of D1; and no response, if D5 CRP was higher than 80% of D1 [10, 11, 19]. The CRP-ratio patterns were a modification of others previously published [19, 20, 27].

The evolution of CRP, CRP-ratio, temperature and WBC during the five days of CAS course was analyzed comparing survivors and non-survivors.

Data entry was performed by a single investigator in each participating center and consistency was assessed with a rechecking procedure of a 10% random sample of patients (see Acknowledgements). Data were screened in detail (see Acknowledgements) for missing information, implausible and outlying values.

Continuous variables were expressed as means and standard deviations (SD) or median and interquartile range (IQR) if the distribution was clearly assymetric. Comparisons between groups were performed with two-tailed unpaired Student's t-test, one-way ANOVA, Mann-Whitney U or Kruskal-Wallis H tests for continuous variables according to data distribution. Fisher's exact test and Chi-square test was used to carry out comparisons between categorical variables as appropriate.

For the statistical analysis of the patient's status at hospital discharge as function of a longitudinal covariate (five measurements of CPR) we used a two-step approach. First, we modeled the CRP measurements as a function of time (five days of CRP measurement). We used a linear mixed model that allowed us to predict an intercept and a slope for each patient. This step is conceptually similar to fitting a linear regression, Mean_CRP = α + β*day, for each patient and obtaining an individual intercept (α) and slope (β) per patient. The intercept describes the initial CRP value and the slope describes the CPR rate of change per day for a specific patient. We assumed a normal distribution of the random effects. Once the model was fitted, the estimates for the intercepts and slopes were obtained as the best linear unbiased predictors, BLUPs [38]. Figure 1 shows the observed values of CRP and predicted values by the model for a randomly selected group of patients.

For the second step, we started by computing the relative rate of CRP change, CRP-ratio, per day for each patient which is obtained by dividing the slopes by the intercepts. This quantity gives the percentage of change per day in CRP, relative to the first measurement (intercept). Then, we used the individual CRP-ratios and intercepts, together with SAPS II and sepsis severity (uncomplicated sepsis, severe sepsis and septic shock), as covariates in a logistic regression for hospital mortality. This model allowed us to compute the odds ratios (OR) associated with the CRP profile (expressed as an intercept and a rate of change per day, the CRP-ratio) adjusted for patient's severity (SAPS II and sepsis severity). Because the covariates "intercept" and "CRP-ratio" are estimates and not observed values, the standard errors provided by the software are not correctly computed. Thus, we obtained estimated of the correct standard errors for the regression coefficients through bootstraping. A similar approach was used for the five days of measurements of temperature and WBC. However, WBC was skewed and we used a transformation by taking the natural log of WBC (log-WBC) to comply with the normal assumption.

Additionally, we analysed the association between hospital mortality and CRP by considering the patterns of CRP-ratio response. Only patients with available measurements at D1 and D5 were considered for this analysis. We computed the OR of hospital mortality for the CRP-ratio patterns using a logistic regression and adjusting for SAPS II and sepsis severity.

For all logistic models, we checked the Hosmer and Lemeshow goodness-of-fit and computed the receiver operating characteristic (ROC) curve to evaluate model discrimination. Data were analyzed using PASW v.18.0 for MAC (SPSS, Chicago, IL, USA) and R (Development Core Team. R: A Language and Environment for Statistical Computing. Vienna, Austria: 2005). All statistics were two-tailed and significance level was set at 0.05.

A total of 891 patients with CAS were included (Table 1); 209 patients were classified as having HCAI. The lungs were the most common site of infection (61%), followed by the abdomen (18%) and urinary tract (7%). Cultures were positive in 40% of the patients, with Streptococcus pneumoniae (21%), Escherichia coli (18%) and methicillin sensitive Staphylococcus aureus (12%) the three most common isolated microorganisms. On ICU admission, 9% of CAS patients presented with uncomplicated sepsis, 40% severe sepsis and 51% septic shock. The overall ICU and hospital mortality rate of CAS was 30% (N = 265) and 38% (N = 338), respectively.

At D1 (Table 1), CRP of survivors and non-survivors was not statistically different, 19.8 ± 12.5 mg/dL vs. 20.7 ± 12.8 mg/dL (P = 0.367). When we compared CRP (Figure 2) of survivors and non-survivors at the different time points, we found that CRP of non-survivors was significantly higher since D3 onwards (P < 0.001, for D3, D4 and D5).

After adjusting for SAPS II and the severity of sepsis, the initial value of CRP remained not significantly associated with hospital mortality (OR_initial = 1.01, confidence interval (CI)_95% = (0.99, 1.02), P = 0.297). On the contrary, the course of CRP measured as the relative changes during the first five ICU days, was significantly associated with hospital mortality (OR_CRP-ratio= 1.03, CI_95% = (1.02, 1.04), P < 0.001). For example, a patient with an average decrease of CRP concentration of 10% per day has 32% less chances of dying when compared to a patient with the same SAPS II score and the same severity of sepsis but with no CRP decrease. The ability of the model to predict hospital outcome assessed by the area under the ROC curve was 0.77 (CI_95% = (0.73; 0.80)).

To assess the impact of the variable HCAI, we added this variable to our model to study its interaction with the initial value of CRP and CRP-ratio. The interactions were not significant (P = 0.579 and P = 0.182, respectively), indicating that there is no evidence in our patient population of a different behaviour of CRP for this specific subgroup of septic patients.

The same analysis, restricted to the subgroup of septic shock patients, presented similar results (data not shown).

At D1, temperature and WBC of survivors and non-survivors were not statistically different (Table 1), 37.5 ± 1.3°C vs. 37.5 ± 1.1°C (P = 0.799) and median (IQR) 12.4 (8.5, 18.7) × 10³/mm³ vs. 12.4 (6.9, 18.5) × 10³/mm³ (P = 0.496), respectively. Time dependent analysis of temperature from D1 to D5 showed no association between temperature evolution and survival (P = 0.360). Temperature decreased likewise in survivors and non-survivors. Similarly, no differences were found on the WBC course along the first five days of the ICU stay (P = 0.594).

Community-acquired sepsis patients were divided according to three patterns of CRP-ratio response to antibiotic therapy. Two hundred died or were discharged before D5. One hundred and thirty five patients had no CRP recorded at D1 or D5. Five hundred, fifty-six patients had complete data for the analysis. One hundred, ninety-nine (36%) patients were classified as fast response, 149 (27%) as slow response and 208 (37%) as no response.

The ICU mortality rate was significantly different according to the patterns of CRP-ratio response: fast response 14%, slow response 20% and no response 32% (P < 0.001). After adjusting for SAPS II and severity of sepsis, the odds of death for slow or no response patients were 1.6 (CI_95% (0.9, 2.9), P = 0.119) and 3.0 (CI_95% (1.8, 5.1), P < 0.001), respectively, when compared with the fast response ones. By D3, median CRP-ratio (5^th and 95^th percentiles) was 0.81 (0.40, 1.30), 0.95 (0.62, 1.48) and 1.22 (0.70, 6.64) in patients with fast response, slow response, no response patterns, respectively (P < 0.001).

The hospital mortality was also significantly associated with the pattern of response (P = 0.001). Fast responders have lower mortality (23%) when compared with slow and no responders (30% and 41%, respectively). After adjusting for SAPS II and severity of sepsis, the pattern of CRP-ratio response remained significantly associated with mortality. No responders had a significantly increase on the odds of mortality (OR = 2.5, CI_95% = (1.6, 4.0), P < 0.001) when compared with fast responders. Slow responders showed a non-significant increase on the odds of mortality in comparison with the fast responders (OR = 1.5, CI_95% = (0.9, 2.5), P = 0.124).

The same analysis, restricted to the subgroup of septic shock patients, presented similar results (data not shown).

Temperature decreased significantly faster for the fast and slow response patterns groups when compared with the no response pattern group (P = 0.004 and P < 0.001, respectively) (Figure 3). However, the D1 and D5 temperature of patients with the fast and slow response CRP-ratio patterns was very similar (D1: 37.4 ± 1.1°C and 37.5 ± 1.0°C and D5: 37.3 ± 0.8°C and 37.5 ± 0.9°C for fast and slow response, respectively), while the temperature of the patients with no response CRP-ratio pattern increased slightly from 37.6 ± 1.1°C at D1 to 37.8 ± 0.9°C at D5.

On the other side, no significant time trend was observed regarding WBC for the fast response pattern (P = 0.137) and no significant differences were found for the slow and no response pattern groups (P = 0.680 and P = 0.619, respectively) in comparison to the fast response group (Figure 3).

The time dependent analysis of the SOFA score of the three patterns of CRP-ratio response from D1 to D5 showed that these patterns of evolution were significantly different (fast vs. slow responders, P = 0.003 and fast vs. no responders, P < 0.001) (Figure 4). The SOFA scores of patients with the fast and slow CRP-ratio response patterns decreased from 7.0 ± 3.7 and 6.9 ± 3.4 at D1 to 5.1 ± 3.6 and 5.6 ± 3.9 at D5. Opposed to this were the SOFA scores of the patients with no response, in which the CRP-ratio pattern remained almost unchanged, from 6.8 ± 3.3 at D1 to 6.8 ± 3.9 at D5.