Published ahead of print on August 16, 2007, doi:10.1164/rccm.200702-307OC
© 2007 American Thoracic Society doi: 10.1164/rccm.200702-307OC
Free and Total Cortisol Levels as Predictors of Severity and Outcome in Community-acquired Pneumonia1 Department of Endocrinology, William Harvey Research Institute, Barts and The London, Queen Mary's School of Medicine and Dentistry, London, United Kingdom; 2 Department of Pneumology, 3 Department of Internal Medicine, and 4 Department of Endocrinology, University Hospital Basel, Basel, Switzerland; and 5 NETRIA, St. Bartholomew's Hospital, London, United Kingdom Correspondence and requests for reprints should be addressed to Mirjam Christ-Crain, M.D., University Hospital, Petersgraben 4, CH-4031 Basel, Switzerland. E-mail: m.christ-crain{at}unibas.ch
Rationale: High cortisol levels are of prognostic value in sepsis. The predictive value of cortisol in pneumonia is unknown. Routinely available assays measure serum total cortisol (TC) and not free cortisol (FC). Whether FC concentrations better reflect outcome is uncertain. Objectives: To investigate the predictive value of TC and FC in community-acquired pneumonia (CAP). Methods: Preplanned subanalysis of a prospective intervention study in 278 patients presenting to the emergency department with CAP. Measurements and Main Results: TC, FC, procalcitonin, C-reactive protein, leukocytes, clinical variables, and the pneumonia severity index (PSI) were measured. The major outcome measures were PSI and survival. TC and FC, but not C-reactive protein or leukocytes, increased with increasing severity of CAP according to the PSI (P < 0.001). TC and FC levels on presentation in patients who died during follow-up were significantly higher as compared with levels in survivors. In a receiver operating characteristic analysis to predict survival, the area under the receiver operating characteristic curve (AUC) was 0.76 (95% confidence interval, 0.70–0.81) for TC and 0.69 (0.63–0.74) for FC. This was similar to the AUC of the PSI (0.76 [0.70–0.81]), and better as compared with C-reactive protein, procalcitonin, or leukocytes. In univariate analysis, only TC, FC, and the PSI were predictors of death. In multivariate analysis, the predictive potential of TC equaled the prognostic power of PSI points. Conclusions: Cortisol levels are predictors of severity and outcome in CAP to a similar extent to the PSI, and are better than routinely measured laboratory parameters. In CAP, the prognostic accuracy of FC is not superior to TC. Clinical trial registered with www.controlled-trials.com (ISRCTN04176397).
Key Words: pneumonia • cortisol • prognosis
High cortisol levels reflect a higher degree of stress (1). The effects of cortisol are directed toward the acute provision of energy, protection against excessive inflammation, and improvement in hemodynamic status (2, 3). Therefore, appropriate activation of the hypothalamic–pituitary–adrenal axis during illness is essential for survival, and parallels the degree of stress. Several studies have shown a positive correlation between serum total cortisol and severity of critical illness as well as risk of death (4–10). The activation of the hypothalamic–pituitary–adrenal axis in critically ill patients is characterized by an increase in free cortisol levels and a possibly smaller increase in total cortisol levels (11). The free cortisol, rather than the protein-bound fraction, is responsible for the physiologic function (11). Accordingly, serum free cortisol concentrations, as compared with serum total cortisol concentrations, might offer a better reflection of the degree of stress and the subsequent activation of the hypothalamo–pituitary–adrenal axis in critically ill patients (11). However, whether free cortisol concentrations are better correlated with outcome such as death or treatment failure remains an open question. The major cause and precursor of sepsis and critical illness is respiratory tract infections. Community-acquired pneumonia (CAP) is the major infection-related cause of death in developed countries (12, 13). In the assessment and management of CAP, predictors of outcome are crucial to estimate disease severity, and to guide therapeutic options such as the need for hospital or intensive care presentation, the intensity of work-up, the choice and route of antimicrobial agents, and the suitability for discharge (14, 15). The value of cortisol concentrations to predict outcome in CAP is currently unknown. This is the first study to evaluate the prognostic value of cortisol levels, measured on presentation, in the prediction of severity of disease and outcome in a well-defined cohort of patients with CAP (16), and to compare the prognostic value of free and total cortisol concentrations (17).
Setting and Study Population Data from 278 patients of an initial cohort of 302 patients presenting to the emergency department with CAP were analyzed. The primary objective of the study was to evaluate antibiotic duration by procalcitonin guidance as compared with standard recommended guidelines (16). Baseline data were similar in both cohorts (Table 1). A predefined secondary endpoint was the assessment of prognostic factors in CAP.
Briefly, consecutive patients with CAP admitted from November 2003 through February 2005 to the University Hospital Basel (Basel, Switzerland), a 950-bed tertiary care hospital, were included. Patients had to be more than 18 years of age with suspected CAP as the principal diagnosis on presentation. Exclusions included patients with cystic fibrosis or active pulmonary tuberculosis, hospital-acquired pneumonia, or severely immunocompromised patients. Patients were examined on presentation to the emergency department by a resident supervised by a board-certified specialist in internal medicine. Baseline assessment included clinical data and vital signs, comorbid conditions, and routine blood tests. CAP was defined by the presence of one or several of the following recently acquired respiratory signs or symptoms: cough, sputum production, dyspnea, core body temperature exceeding 38.0°C, auscultatory findings of abnormal breath sounds and rales, leukocyte count greater than 10 x 109 or less than 4 x 109 cells/L and an infiltrate on chest radiograph (13). The pneumonia severity index (PSI), a validated severity scoring system for patients with pneumonia, was calculated (18). Chest radiographs were screened by the physician in charge and reviewed by a senior radiologist, unaware of clinical or laboratory findings. The study was approved by the local ethics committee for human studies and written informed consent was obtained from all patients.
Outcome
Measurement of Total and Free Cortisol Levels Free cortisol was measured in house by a newly developed assay (19). The fractions of free and bound cortisol from the serum samples were first separated by equilibrium dialysis, followed by measurement of the concentration of the free fraction from the dialysate, using an enzyme immunoassay (NETRIA, St. Bartholomew's Hospital). Dialysis tubing was cut into sacs 7 mm in length with a knot tied at one end. Cartridges were assembled by inserting the sacs into LP-3 tubes. Eight hundred microliters of buffer solution (1% hydrolyzed gelatin in phosphate-buffered saline [0.05 M phosphate, 0.15 M sodium chloride]) was pipetted into the space between the inner wall of the LP-3 tube and the outer aspect of the dialysis sac. Undiluted sample (200 µl) was then pipetted directly into the dialysis sac. Excess tubing was folded over the mouth of the LP-3 tube and a stopper was inserted to complete the cartridge. All cartridges were placed on an elevated rotator in a 37°C heated chamber and left overnight. Dialysate was then collected and the enzyme immunoassay was performed. The calculated sensitivity of the enzyme immunoassay (96-well microtiter plates triple-coated with gammaglobulin, donkey anti-sheep immunoglobulin, and anti-cortisol antiserum in a 1:30,000 dilution) was 0.37 ± 0.49 nmol/L (mean ± SD). The interassay coefficient of variance was 3.5% at 2.1 nmol/L, 6.8% at 24 nmol/L, and 10.0% at 52 nmol/L. The intraassay coefficient of variance was assessed by precision profile as follows: 10% at 1 nmol/L, 6% at 10 nmol/L, 5% at 20 nmol/L, and 7% at 100 nmol/L. Assay time was about 90 minutes.
Measurement of Other Laboratory Parameters
Statistical Analysis Univariate and multivariate analysis to predict the binary end point, death, was performed by including the PSI score, C-reactive protein, leukocyte count, procalcitonin, and total and free cortisol levels. To assess the influence of these variables on death or clinical failure, we produced Kaplan-Meier survival curves: comparison between the groups was done by log-rank test. Levels that were nondetectable were assigned a value equal to the lower limit of detection for the assay. All testing was two-tailed and P values less than 0.05 were considered to indicate statistical significance.
Patients Detailed baseline characteristics of the study population are summarized in Table 1. The mean age of the 278 patients was 69 ± 17 years; 71 (25.5%) were smokers and 56 (20.1%) had been pretreated with antibiotics. Overall, 87% of patients had relevant comorbidities. No patient was taking etomidate, barbiturates, or muscle relaxants: 21 patients (7.6%) were taking sedative drugs, which may interact centrally with the pituitary–adrenal axis (20). Total and free cortisol levels in patients with and without these drugs were not significantly different. The mean PSI of all patients was 99.1 ± 35.5 points: 22 patients (7.9%) were in PSI class I, 37 (13.3%) were in PSI class II, 52 (18.7%) were in PSI class III, 120 (43%) were in PSI class IV, and 47 (17%) were in PSI class V. No patient on presentation was undergoing intravenous steroid treatment; 27 patients were receiving oral corticosteroid treatment on presentation (corresponding to a median dose of 5 mg of prednisone daily; range, 2.5–50 mg daily). Therefore, all analyses were performed by including and excluding these patients, respectively.
Total and Free Cortisol Levels and Severity of CAP
Four patients (1.4%) were hypotensive, with a systolic blood pressure less than 90 mm Hg at presentation to the emergency department. Total and free cortisol levels in these patients were 1,304.5 (834–1,805) and 160.1 (85.3–279.8) nmol/L, respectively. Because we report data on presentation to the emergency department, the consensus criteria to define septic shock, which include adequate fluid resuscitation over several hours (21, 22), do not apply and no patient was formally classified as having septic shock.
There was a significant correlation between total and free cortisol levels (r = +0.71; P < 0.001). The correlation was still significant when taking into account only patients with hypoalbuminemia (i.e., albumin
Total and Free Cortisol Levels as Prognostic Marker for Outcome For the subsequent analyses, only steroid-naive patients were considered. On presentation, total and free cortisol levels in patients who died during follow-up were significantly higher as compared with levels in survivors (1,141 [941–3,355] and 88.4 [51.9–124.1] vs. 697 [429–1,081] and 52.4 [35.0–88.9] nmol/L; P < 0.001 and 0.001, respectively) (Figure 2). Albumin levels tended to be lower in nonsurvivors (27.7 ± 6.8 g/L) compared with survivors (30.8 ± 6.2 g/L; P = 0.07). The respective values were not significant for procalcitonin (0.6 [0.4–2.2] vs. 0.5 [0.2–2.4] µg/L; P = 0.08), for C-reactive protein (145 [101–204] vs. 132 [66–212] mg/L; P = 0.64), or for total leukocyte count (13.5 [11.4–16.5] x 109 vs. 12.1 [9.0–15.6] x 109 cells/L; P = 0.18).
To evaluate the potential to predict death from CAP, a receiver operating characteristic (ROC) analysis was performed in which sensitivity was calculated with those patients who died before follow-up (n = 28) and specificity was assessed with those patients who survived until follow-up (n = 223). The area under the ROC curve (AUC) and likelihood ratios for various cut-off points are summarized in Tables 2 and 3. The prognostic accuracy of total, more than free, cortisol to predict death was superior to that of C-reactive protein, leukocyte count, and procalcitonin, and was comparable to that of the PSI score (Figure 3). With an optimal calculated total cortisol threshold of 960 nmol/L, sensitivity and specificity were 75.0 and 71.7% with positive and negative likelihood ratios of 2.65 and 0.35, respectively. For free cortisol with an optimal threshold of 80.3 nmol/L, sensitivity was 64.3% and specificity was 71.6%, with positive and negative likelihood ratios of 2.27 and 0.50, respectively. Results for the various parameters in predicting treatment failure were similar (data not shown).
To estimate the additive value of total cortisol in combination with the PSI score we calculated a logistic regression model combining the PSI score and the total cortisol concentration. With an AUC of 0.79 (95% confidence interval, 0.71–0.87) the combined model tended to improve the prognostic accuracy compared with the PSI alone (P = 0.22). When we performed a comparison of survival among patients with total and free cortisol levels below and above these calculated optimal cutoff values by Kaplan-Meier survival curves, patients with total or free cortisol levels above this cutoff had significantly lower survival rates compared with patients with levels below the cutoff (P < 0.0001 and P < 0.001, respectively; Figure 4).
When entering total and free cortisol, procalcitonin, C-reactive protein, leukocytes, and PSI in a univariate regression analysis, only increasing total cortisol (P < 0.001) and free cortisol levels (P = 0.004) and a rise in PSI (P < 0.001) were predictors of death (Table 4). In a multivariate analysis, the predictive power of total cortisol equaled the prognostic power of the PSI (Table 4). The results for the different parameters to predict treatment failure were similar (data not shown).
In addition, we also calculated whether particularly low (as well as high) cortisol levels, as defined by Cooper and coworkers (23) and Annane and coworkers (24), were associated with poor outcome. Overall, 54 patients (19%) had total cortisol levels below 414 nmol/L, whereas 30 patients (11%) had total cortisol levels below 275 nmol/L. Of the 30 patients with total cortisol levels below 275 nmol/L, 2 died (6.7%), whereas of the 54 patients with a total cortisol level below 414 nmol/L, 3 patients died (5.6%); overall mortality was 11.2%. This difference in mortality between patients with low cortisol levels and patients overall was not statistically significant.
Our study has three main findings. First, cortisol concentrations measured on presentation predict the severity and outcome of CAP. Second, the prognostic accuracy of cortisol levels is as high as the PSI and higher as compared with commonly measured laboratory parameters. Third, the prognostic accuracy of free cortisol levels is not superior to that of total cortisol in this paradigm. The release of cortisol in acute illness is essential for cardiovascular, metabolic, and immunologic homeostasis. The rationale for a presumed prognostic value of cortisol is due mainly to its positive correlation with the stress level associated with illness. A higher level of proinflammatory cytokines in patients with worse outcome, that is, nonsurvivors, could additionally lead to a more pronounced increase in cortisol as the strongest known natural inhibitor of inflammation (25). However, levels of C-reactive protein, as one example of an inflammatory mediator, were similar in survivors and nonsurvivors. Indeed, in critically ill patients a strong association of cortisol with outcome has previously been reported (4–10). Conversely, low unstimulated cortisol levels in septic patients can also predict absolute or relative adrenal insufficiency (24, 26). On the basis of our study, serum cortisol concentrations are also a prognostic parameter in mild to severe CAP. Interestingly, in this study low cortisol levels were not associated with poor outcome in our study cohort. Indeed, the mortality among patients with relatively low levels was about half of that in the cohort as a whole, although possibly due to the small numbers involved this did not attain statistical significance. It is possible that in our series the absence of patients taking any form of adrenolytic drug excluded the possibility that there was any form of iatrogenic, and hence deleterious, hypoadrenalism. Alternatively, there may be two factors in operation, one directly associating serum cortisol with mortality, and one operating at low cortisol levels to increase mortality, such that at low cortisol levels these cancel each other out. Whatever the explanation, the overriding factor associated with poor outcome was increased cortisol, however measured. Indeed, there is a debate concerning the definition of relative adrenal insufficiency, its treatment, and the identification of patients at risk (1). In a small study, low-dose hydrocortisone infusion appeared to be beneficial (27); however, these preliminary data need to confirmed in a larger study before it can be recommended. Our findings of an association of high cortisol levels with worse outcome question the rationale of hydrocortisone treatment in patients with CAP. However, our study population was different from the population included in the study by Confalonieri and coworkers (27). Whereas they included mainly mechanically ventilated patients in an intensive care unit, we investigated patients presenting to an emergency department with all classes of PSI. Alternatively, it may also be speculated that despite the marked increase in circulating cortisol in severe CAP, this concentration is not high enough to counterbalance the inflammatory response. In general, it is assumed that the biologically active fraction of cortisol to which tissues are exposed is free cortisol. Specifically, in critically ill patients there is a fall in binding proteins such that total cortisol levels no longer accurately reflect the free fraction (11, 28). Our newly developed assay for free cortisol used in this study has been validated (19), and has certain advantages over other, commercially available assays. In a standardized model of acute major surgical stress, circulating free cortisol levels showed a more pronounced increase with major stress as compared with total cortisol levels, followed by a more pronounced decrease after resolution of major stress. Although total cortisol levels show a similar increase with major stress as compared with adrenocorticotropic hormone stimulation, the increase in free cortisol was more pronounced than with the surgical stress (19). Nevertheless, although such an assay might have been predicted to better correlate with clinical outcomes compared with total cortisol, free cortisol in our study was no better predictor than total cortisol. However, although albumin levels were slightly lower in nonsurvivors compared with survivors of CAP, the binding protein levels did not correlate strongly with pneumonia severity as assessed with the PSI. This was in contrast to our previous study including patients with major surgical stress. Therefore, in less critically ill patients such as our patients with CAP, fluctuations of cortisol-binding proteins (such as albumin and cortisol-binding globulin) may become less relevant, diminishing any potential advantages of an assay measuring the free hormone. Only 5% of the patients in our study had serum albumin levels below 2 g/L and in only 19% was it below 2.5 g/L. Taken together, these findings suggest that the prediction of outcome is not primarily related to the free fraction of cortisol in CAP. A key decision for a clinician is whether to admit a patient with CAP (29). This decision is complex and depends on many variables, including estimates of the severity of illness. It often relies on the clinician's judgment; however, the interpretation of clinical signs and symptoms lacks standardization and validation and is prone to interobserver variability (30, 31). In addition, physicians continue to be conservative and commonly overestimate the risk of death for patients with CAP (32). Clinical signs such as body temperature or routinely used laboratory parameters (i.e., C-reactive protein or leukocyte count) are of only limited value to predict disease severity in CAP (33). Thus, clinical judgment alone can be misleading in estimating disease severity, thereby leading to under- as well as overestimation of the severity of CAP. For this reason, prognostic scoring rules have been developed to predict severity of CAP and outcome, with the PSI being a well-validated prognostic classification score (18, 34–37). Limitations of the PSI include a potential overemphasis on age and the fact that, for clinical ease, the PSI dichotomizes continuous values such as heart rate or oxygen saturation into normal and abnormal values. The intraobserver variation of the PSI is reported to be about 10%, with most patients misclassified into the high-risk classes IV and V (38). In addition, the PSI is best validated for assessing patients with a low mortality risk who may be suitable for home management rather than for those with severe CAP at the time of hospital presentation (37). The major limitation for the routine use of the PSI, however, is its laborious calculation. In a study validating the predictive potential of various indices in 731 patients with CAP the PSI score could be calculated in only 70% of all patients (39) restricting its widespread adoption. The American Thoracic Society guidelines do not offer any algorithm for the clinical assessment of disease severity (13, 40). In this context, there is a need for readily measurable parameters predicting the severity level and outcome of CAP. According to our data, the simple measurement of total cortisol provides equivalent prognostic information as the complex 20-variable severity index. This is remarkable. It therefore represents an additional and easy-to-determine prognostic tool. As a new indicator for those patients with a worse prognosis, it allows such patients to be targeted for more intensive therapy. As the combination of cortisol measurement and the PSI score tended to have an even higher prognostic accuracy, it is advisable to use several clinical and laboratory parameters, which may mirror different physiologic aspects, in the complex task of prognostic assessment and treatment decisions. C-reactive protein has been reported to be a useful marker for predicting disease severity in patients with pneumonia (41). In contrast, in our study C-reactive protein could not differentiate between different severities of CAP, as defined by the PSI, and was not a significant independent predictor of outcome. It should be recognized that C-reactive protein is a rather nonspecific marker of acute-phase inflammation, and is thus subject to the influence of many other factors. IL-6, a key stimulator of hepatic C-reactive protein release, has also been suggested as a useful marker in the determination of the severity of CAP (42). Measurement of plasma cytokines such as IL-6 is, however, a difficult and cumbersome process, partly because of the short plasma half-life and the presence of blocking factors (43). Most recently, D-dimers have also been proposed as prognostic parameter in CAP (44), but these were not measured in our study. Finally, we and others have proposed that procalcitonin is a novel and useful marker of disease severity in community-acquired pneumonia (16, 45). However, the present data suggest, rather, that procalcitonin is useful as a diagnostic and not primarily as a prognostic tool able to guide decisions on antibiotic therapy (16, 46). The following limitations of our study should be mentioned. First, this study had not been prospectively designed to use cortisol concentrations as a primary endpoint, and only single presentation cortisol levels were measured (16). Second, serial cortisol levels show large variations during the day (47), although during infective illness the circadian pattern of cortisol is usually lost (2). In our study, serum cortisol levels were measured at the time of presentation, that is, at different times during the day, which could limit the predictive accuracy of a single cortisol value on presentation to the hospital. A standardized cortisol value determined at the same time of day in all patients would most probably have had an even higher prognostic accuracy. It is also possible that serial measurements over the course of the disease may add prognostic information, and show differences in the values of free and total cortisol (48). Third, we did not assess adrenal function based on the response to injection of synthetic adrenocorticotropin, as used in other studies. However, because these patients did not have septic shock, it seems unlikely that they experienced relative adrenal insufficiency. Indeed, as noted above, there was actually a trend for decreased mortality among patients with relatively low serum cortisol levels. In conclusion, our findings indicate that cortisol levels are significant independent predictors of outcome in mild to severe CAP; the prognostic accuracy of total and free cortisol is as high as that of the PSI and better as compared with routinely measured laboratory parameters. Free cortisol levels are not superior to total cortisol levels.
The authors thank the staff of the clinics of emergency medicine, internal medicine, and endocrinology and the department of clinical chemistry, notably Fausta Chiaverio, Martina-Barbara Bingisser, Maya Kunz, Ursula Schild, Vreni Wyss, David Miedinger, Jörg Leuppi, and Charly Nusbaumer, for support during this study.
Supported by the Swiss Foundation of Medical and Biological Stipends (SSMBS, PASMA-114617/1) (to M.C.-C.). Originally Published in Press as DOI: 10.1164/rccm.200702-307OC on August 16, 2007 Conflict of Interest Statement: None of the authors has a financial relationship with a commercial entity that has an interest in the subject of this manuscript. Received in original form February 23, 2007; accepted in final form August 9, 2007
This article has been cited by other articles:
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
TOP
ABSTRACT
AT A GLANCE COMMENTARY
2.5 g/L [11]) (r = +0.68; P < 0.001). Total and free cortisol levels correlated with other markers of infection, that is, procalcitonin (r = +0.30 and +0.37; P < 0.001 for both), C-reactive protein (r = +0.20 and +0.27; P < 0.001 for both), and total leukocyte count (r = +0.14 and r = +0.17; P < 0.05 and P < 0.01, respectively). 
