Corresponding author and reprint requests: U. Hohenthal, Department of Medicine, Turku University Hospital, Kiinamyllynkatu 4-8, 20520 Turku, Finland E-mail:firstname.lastname@example.org
Previous studies on the usefulness of C-reactive protein (CRP) in patients with community-acquired pneumonia (CAP) have yielded somewhat inconsistent results. Our aim was to assess the value of CRP in estimating the severity and complications of CAP. CRP levels during the first 5 days of hospitalization were measured in 384 adult patients with CAP, and the data were evaluated using comprehensive statistical analyses. Significantly higher CRP levels on admission were detected in Pneumonia Severity Index (PSI) classes III–V than in classes I and II (p <0.001). An increment of 50 mg/L CRP on admission was associated with a 1.22-fold odds for a patient to be in PSI classes III–V as compared with classes I and II (OR 1.22, 95% CI 1.11–1.34; p <0.001). CRP levels were significantly higher in bacteraemic pneumonia than in non-bacteraemic pneumonia (p <0.001). An increment of 50 mg/L CRP was associated with a 1.67-fold odds for a patient to be bacteraemic (OR 1.67, 95% CI 1.46–1.92; p <0.001). CRP levels >100 mg/L on day 4 after the admission were significantly associated with complications (p <0.01). There was a trend for an association between the level of CRP on admission and the time to reach clinical stability (p <0.01). In conclusion, CRP may be valuable for revealing the development of complications in CAP. It may also be useful to assess the disease severity, thus being complementary to the assessment of the PSI. In our patients, high CRP levels were associated with a failure to reach clinical stability.
Many studies have shown the clinical utility of C-reactive protein (CRP) concentration as an acute-phase reactant in patients with various infections, including septicaemia, meningitis and infective endocarditis [1–8]. Previous studies have also shown that the CRP level may contribute to establishing the diagnosis of community-acquired pneumonia (CAP) [9–13]. In addition, the use of CRP as a tool in the aetiological diagnosis of CAP has been investigated in a number of studies, but the results have been discordant [11,14–18]. Similarly, the results of the studies evaluating the use of CRP as a prognostic factor have been somewhat inconsistent [14,17–23], although the CRP levels have been found to be higher in bacteraemic pneumonia than in non-bacteraemic pneumonia [14,15,17].
In the present study, we sought to evaluate whether the level of CRP on admission could be used to estimate the severity and potential aetiology of CAP. We also assessed whether the CRP levels during the first days of hospitalization were associated with the clinical stabilization of the patient or the development of complications in CAP.
Patients and Methods
The patients included in this study are the same as those included in a prospective clinical study of the aetiology of CAP in 384 consecutive hospitalized patients treated at the Department of Infectious Diseases, Turku University Hospital, Turku, Finland, between December 1999 and 2004 . The diagnostic criteria for CAP included a new infiltrate on the chest radiograph in a patient with either fever or clinical signs/symptoms of lower respiratory tract infection, or both . Exclusion criteria included immunosuppression of the patient, an emerging alternative diagnosis during the follow-up, pneumonia caused by tuberculosis or aspiration, and hospitalization within the previous 10 days.
The microbiological evaluation included blood culture, sputum culture, urine antigen tests for Streptococcus pneumoniae and Legionella pneumophila, detection of a respiratory virus antigen in a nasopharyngeal sample, PCR tests for Mycoplasma pneumoniae, Chlamydophila pneumoniae, L. pneumophila, influenza A virus and influenza B virus from a throat swab sample, and detection of antibodies against M. pneumoniae, C. pneumoniae and L. pneumophila in serum samples .
The severity of CAP was assessed using the Pneumonia Severity Index (PSI) . The clinical stability of the patient was defined on the basis of an approach described by Halm et al. . Death, transfer to the intensive-care unit (ICU) and empyema were defined as complications.
CRP levels were determined by an immunoturbidometric method (Tina-quant®) on a Hitachi 917 automated biochemistry analyser (Roche Diagnostics GmbH, Mannheim, Germany). The CRP levels were studied on admission, on a daily basis until the rising trend of CRP level turned into a declining trend, and subsequently as clinically indicated two to five times a week during the hospitalization period. In addition, CRP level was studied at the time of clinical stability. The CRP level on admission was defined as CRP1, and the CRP value at the time of clinical stability as CRP2. Patients gave written informed consent, and the study was approved by the Ethics Committee of Turku University Central Hospital.
The associations of CRP levels with baseline characteristics other than age were statistically tested using a two-sample t-test. Associations between age and CRP levels were studied using Pearson’s correlation coefficient. The two-sample t-test was also applied with other two group comparisons of the CRP mean levels. One-way ANOVA with Tukey’s adjustment in post hoc comparisons was used when comparing more than two groups for CRP mean levels. In multivariate analysis, when covariate-adjusted group comparisons were performed, the analysis was carried out using linear model techniques. Predictive associations of CRP levels and dichotomous clinical classifications were analysed using logistic regression models and receiver operating characteristic (ROC) analysis.
Finally, the predictive associations of CRP levels with the time of stabilization were examined using Cox’s proportional hazard regression analysis. The quantifications of the analyses were performed by giving the mean values or differences in the mean values in the case of t-test and ANOVA, or by giving the ORs, specificity, sensitivity and area under the ROC curve in the case of logistic analyses, or by giving hazard ratios. Statistical confidence of these quantifications was reported by giving the 95% CIs among the estimates. A significance level of 0.05 was used as the limit for significance. Statistical computing was performed with the SAS System for Windows, Release 9.1.3/2004.
Of the 384 patients included, 201 were male and 183 female. The mean age of the patients was 49.8 years (standard deviation (SD) 19.2 years). Various baseline characteristics of the patients are presented in Table 1. The distributions of the patients according to PSI classes and aetiological agents are presented in tables 2 and 3. S. pneumoniae was the only aetiological agent in 95 (24.7%) patients, M. pneumoniae in 36 (9.4%) patients, C. pneumoniae in 24 (6.3%) patients, another bacterial agent in seven (1.8%) patients, and a respiratory virus in 27 (7.0%) patients. Two aetiological agents were identified in 20 (5.2%) patients, and the aetiology remained unidentified in 175 (45.6%) patients. Blood culture was positive in 55 patients (52 S. pneumoniae, two Staphylococcus aureus, one Streptococcus pyogenes). Of all patients, 35 (9.1%) were transferred to the ICU and 13 (3.3%) died during the hospitalization.
Table 1. Mean levels of C-reactive protein on admission (CRP1) according to baseline patient characteristics, bacteraemia, admission to intensive-care unit ICU), and death
COPD, chronic obstructive pulmonary disease; SD, standard deviation.
(n =201) (n =183)
199 (125) 204 (112)
Yes (n =167) No (n =217)
209 (127) 195 (111)
Yes (n =17) No (n =367)
230 (110) 200 (119)
Yes (n =35) No (n =349)
185 (128) 203 (118)
Yes (n =70) No (n =314)
180 (126) 206 (116)
Yes (n =15) No (n =369)
317 (115) 197 (116)
Yes (n =28) No (n =356)
224 (140) 200 (117)
Yes (n =54) No (n =330)
199 (98) 202 (122)
Yes (n =125) No (n =259)
238 (128) 184 (110)
Yes (n =110) No (n =274)
178 (91) 211 (127)
Yes (n =55) No (n =329)
328 (128) 181 (103)
Yes (n =52) No (n =55)
337 (126) 238 (135)
Admission to ICU
Yes (n =35) No (n =349)
270 (163) 194 (111)
Yes (n =13) No (n =371)
217 (131) 201 (118)
Table 2. Mean levels of C-reactive protein on admission (CRP1) according to Pneumonia Severity Index (PSI) classes
PSI I n =124
PSI II n =113
PSI III n =59
PSI IV n =73
PSI V n =15
SD, standard deviation.
PSI I and II combined
PSI III–V combined
Table 3. Comparison of the mean levels of C-reactive protein on admission (CRP1) with respect to the aetiology of community-acquired pneumonia in 384 patients. Below diagonal, 95% CIs for differences between means; above diagonal, p-values. Overall ANOVA p-value <0.001 for CRP1 level
The mean CRP1 level was 201 mg/L (SD 119 mg/L) (range 1–650 mg/L, n =384). There were significant differences in the mean CRP1 levels between the patients with alcoholism, smoking or preceding antibiotic treatment as baseline characteristics and those without these characteristics (Table 1). The association between CRP1 level and age did not reach statistical significance (r =0.10, p 0.051).
The mean CRP1 level was significantly higher among the patients who were transferred to the ICU than among the patients who were not (p <0.001) (Table 1). In prediction of ICU admission, the area under the ROC curve for CRP1 was 0.639 (95% CI 0.520–0.58). Correspondingly, the area under the ROC curve for PSI was 0.885 (95% CI 0.847–0.923).
No significant difference was observed in the mean CRP1 levels between the patients who died and the patients who survived (p 0.998) (Table 1).
In univariate analysis, significant (p <0.001) differences were observed in CRP1 levels between the patients belonging to different PSI classes (Table 2). The differences in CRP1 levels were also significant (p <0.001) when PSI groups I and II combined were compared with PSI groups III–V combined.
In multivariate analysis, the differences between the PSI groups remained significant after adjustment for age, sex, cardiovascular disease, bacteraemic pneumonia and aetiological agents (p 0.004). In pairwise comparison of the groups, significant differences in CRP1 levels were observed only between PSI class V and either PSI class I (p 0.002) or PSI class II (p 0.022). The difference in CRP1 levels was also significant (p 0.023) when PSI groups I and II combined were compared with PSI groups III–V combined.
In the logistic regression analysis, an increment of 50 mg/L of the CRP1 level was associated with a 1.22-fold odds for a patient to be in PSI classes III–V as compared with being in classes I and II (OR 1.22, 95% CI 1.11–1.34; p <0.001).
The mean CRP1 levels were significantly higher in the patients with bacteraemic pneumonia than in those with non-bacteraemic pneumonia (p <0.001). In addition, the difference was significant between the patients with bacteraemic and non-bacteraemic pneumococcal pneumonia (p <0.001) (Table 1).
An increment of 50 mg/L of the CRP1 level was associated with a 1.67-fold odds for a patient to be bacteraemic (OR 1.67, 95% CI 1.46–1.92; p <0.001).
ROC analysis was used to find a cut-off point of CRP1 level that could predict bacteraemia with a sensitivity of at least 75% combined with the best possible specificity. The concentration of 230 mg/L was identified as such a cut-off point. The CRP1 levels were ≥230 mg/L for 43 of the 55 patients with bacteraemic pneumonia (sensitivity 78%) and <230 mg/L for 243 of the 324 patients with non-bacteraemic pneumonia (specificity 75%) (area under the ROC curve 0.812). In PSI classes I and II, the sensitivity and specificity of this cut-off point value were 72% and 80%, respectively (area under the ROC curve 0.783).
Significant (p <0.001) differences in mean CRP1 level were observed between the patients belonging to various aetiological groups (Table 3). The CRP1 level was significantly higher in patients with pneumococcal pneumonia than in those with any other aetiology, except for a bacterial agent other than M. pneumoniae or C. pneumoniae.
CRP2 and clinical stabilization
The mean CRP2 level was 80 mg/L (SD 52 mg/L) (range 1–321 mg/L, n =336). The mean duration between the admission and the day when the patient was stabilized was 4.6 days (SD 3.2 days) (range 1–27 mg/L).
Significant differences in CRP2 level were observed between the different PSI classes (p 0.022) and the aetiology of CAP (p 0.029). In pairwise comparisons of the aetiological agents, the difference was significant only between the patients with pneumococcal pneumonia and those with Mycoplasma pneumonia. CRP2 level was not significantly associated with any of the underlying conditions studied (all p-values ≥0.103), or with the age of the patient (r =0.05, p 0.398).
There was a trend for an association between CRP1 level and the time to reach clinical stability. In the Cox regression analysis, it was found that for an increment of 50 mg/L of the CRP1 level, the risk for the patient to remain unstabilized increased by 6% (hazard ratio 1.06, 95% CI 1.02–1.11; p 0.005).
CRP during the follow-up
The CRP levels during the first 5 days of hospitalization regarding the different PSI classes, aetiology of CAP, bacteraemic and non-bacteraemic disease and the development of complications are given in Fig. 1.
The median time to reach clinical stability was 4 days. Of all 384 patients, 23 were discharged, four died, and three were transferred to another hospital in <4 days after admission. On day 4 after admission, the CRP level was examined in 272 (76.8%) of the remaining 354 patients. The risk of being admitted to the ICU or the need to change the initial antimicrobial treatment was significantly higher in the patients with CRP levels >100 mg/L on day 4 after admission, and a significantly smaller number of these patients than of those whose CRP levels were ≤100 mg/L had been stabilized at that time-point (Table 4). The CRP level was >100 mg/L in all four patients who developed empyema.
Table 4. Comparison between patients with community-acquired pneumonia who had C-reactive protein (CRP) levels ≤100 mg/L and >100 mg/L on day 4 after admission
CRP ≤100 mg/L n =172
CRP >100 mg/L n =100
CRP missing n =82
Data are number (%) of patients, unless otherwise indicated.
ICU, intensive-care unit.
aNone of the patients was admitted to the ICU later than day 4.
bTwo patients were admitted to the ICU later than day 4.
cTwo patients died on day 4 and two patients later.
dAll patients died later than day 4.
Mean duration of hospitalization,days (SD) (range)
7.2 (4.0) (4–36)
13.1 (7.8) (6–42)
8.2 (4.4) (4–29)
Patients clinically stabilized
Patients admitted to the ICU
Change of antimicrobial treatment
In this study, we analysed the utility of CRP in the evaluation of patients with CAP, focusing particularly on disease severity. In this respect, the positive correlation between high CRP levels and severe disease was one of the main findings here. This was manifested by the significantly higher CRP1 levels in patients who were transferred to the ICU than in those who were not, as well as by the significant association between the CRP1 levels and different PSI classes. Moreover, the CRP1 levels in PSI classes I and II combined differed significantly from those in PSI classes III–V combined. These are clinically the two most important groups to be differentiated, as the mortality risk and the ICU admission rates are known to be highest in high PSI classes [25,27–29].
According to several studies, mortality and admission rates to the ICU are higher in patients with bacteraemic pneumonia than in those with non-bacteraemic pneumonia [29–31]. Consequently, it is vitally important that patients with bacteraemia are rapidly recognized, as they should be admitted to hospital. In the present study, bacteraemic patients had significantly higher CRP1 levels than non-bacteraemic patients. A cut-off point of 230 mg/L CRP1 predicted bacteraemia with a sensitivity of 78% and a specificity of 75%. The use of this cut-off point may be especially useful in patients belonging to PSI classes I and II, for whom outpatient treatment is commonly considered to be appropriate [25,32]. High CRP levels on admission should arouse a suspicion of bacteraemia even in this group of patients. In the present work, the CRP1 level was >230 mg/L in 13 (72%) of the 18 bacteraemic patients in PSI classes I and II.
Procalcitonin measurements were not used here. In previous studies, procalcitonin has not been shown to be superior to CRP as a diagnostic marker in community-aquired infections but has seemed to be a better marker of bacteraemia [4,33].
In addition, CRP proved valuable as a follow-up test: the CRP levels fell rapidly in accordance with the clinical recovery of the patient. This is in line with the few previous studies, in which the usefulness of CRP level as a follow-up test was evaluated [16,20,22,34,35]. An important finding in our patients was that the CRP level of >100 mg/L on day 4 after admission was suggestive of a treatment failure or development of complications. It is of note that only 19% of the patients with CRP levels >100 mg/L on day 4 had been stabilized, and 21% of them needed treatment in the ICU.
As far as we know, the present study is the first to evaluate the CRP level at the time of clinical stabilization, and shows a significant association between CRP2 level and PSI classes. The significant association observed between CRP1 level and the length of the time to reach clinical stability further substantiates the connection between high CRP levels and severe disease.
Previous studies evaluating the behaviour of CRP with respect to the aetiology of CAP have yielded somewhat inconsistent results. A few studies have found no difference in the CRP levels between different aetiologies [16,17], whereas others have found significantly higher CRP levels only in bacteraemic pneumococcal pneumonia [14,15]. In one study, the CRP levels were significantly higher in L. pneumophila pneumonia than in any other group . In another study, significantly higher CRP levels were detected in pneumococcal pneumonia or L. pneumophila pneumonia than in other aetiological groups . Here, CRP1 levels in pneumococcal pneumonia differed significantly from those in all other aetiologies except for a bacterial agent other than M. pneumoniae and C. pneumoniae. Although significant differences in the CRP1 levels between the different causative agents were observed, the range within each group was wide, with the lowest values <80 mg/L and the highest values >200 mg/L. The high variation and large overlap in the values of the patients with different microbial pathogens indicate that the CRP concentration is not reliable for guiding decisions regarding the aetiology of CAP in a clinical setting.
In conclusion, our results show that the CRP test may be valuable as a tool with which to reveal the development of complications during the treatment of CAP. The positive correlation observed between high CRP levels and serious illness suggests that this biomarker may also be useful for assessing disease severity in patients with CAP. Towards this end, the determination of CRP is complementary to, but cannot substitute for, assessment of the PSI. In our patients, high CRP levels were associated with a failure to reach clinical stability. The results presented here show that the CRP test is not reliable for assessing the aetiology of CAP in an individual patient.
This study was supported by the Maud Kuistila Foundation and the Turku University Central Hospital Research Fund. The authors state that there are no dual or conflicting of interests regarding this article.