Published ahead of print on April 17, 2003, doi:10.1164/rccm.200208-785OC
© 2003 American Thoracic Society
Influence of Systemic Inflammatory Response Syndrome and Sepsis on Outcome of Critically Ill Infected PatientsService de Santé Publique, Robert Debré Hospital and Intensive Care Unit, Saint-Louis Hospital, Paris 7 University, Paris, France; Intensive Care Unit, Henri Mondor Hospital, Paris XII University, Créteil, France; General Intensive Care Unit, Department of Anesthesiology and Critical Care Medicine, Hadassah Hebrew University Medical Center, Jerusalem, Israel; Intensive Care Unit, San Raffaele Hospital, Milan, Italy; Medical Surgical Intensive Care Unit, University of Toronto, Toronto, Canada; Intensive Care Unit, Santo Antonio dos Capuchos Hospital, Lisbon, Portugal; Intensive Care Unit, University Hospital of Wales, Cardiff, United Kingdom; Zentrum Anaesthesiologie, Rettungs, und Intensivmedizin, University Hospital, Göttigen, Germany; and Critical Care Center, Parc Taulli University Hospital, Sabadell Hospital, Autonomous University of Barcelona, Barcelona, Spain Correspondence and requests for reprints should be addressed to Corinne Alberti, M.D., Service de Santé Publique, 48 Boulevard Sérurier, 75019 Paris, France. E-mail: corinne.alberti{at}rdb.ap-hop-paris.fr
The clinical significance of the systemic inflammatory response in infected patients remains unclear. We examined risk factors for hospital mortality in 3,608 intensive care unit patients included in the European Sepsis Study. Patients were categorized as having infection without or with (i.e., sepsis) systemic inflammatory response, severe sepsis, and septic shock, on the first day of infection. Hospital mortality varied from 25 to 60% according to sepsis stage, but did not differ between the first two categories (hazard ratio, 0.94; p = 0.55), whereas there was a grading of severity from sepsis to severe sepsis (1.53, p < 10-4) and septic shock (2.64, p < 10-4). Within each stage, mortality was unaffected by the number of inflammatory response criteria. Prognostic factors identified by Cox regression included comorbid conditions, severity of acute illness and acute organ dysfunction, shock, nosocomial infection, and infection caused by aerobic gram-negative bacilli, enterobacteria, Staphylococcus aureus, and infection from a digestive or unknown source. We conclude that whereas the categorization of infection by the presence of organ dysfunction or shock has strong prognostic significance, infection and sepsis have similar outcomes, unaffected by the presence or number of inflammatory response criteria. Refinement of risk stratification of patients presenting with infection and no organ dysfunction is needed.
Key Words: infection • inflammatory response • mortality • risk factors • sepsis Sepsis is a frequent cause of admission to intensive care units (ICUs) and one of the leading causes of death among hospitalized patients (1, 2). Categorization of patients within the subgroups of severity defined by the American College of Chest Physicians/Society of Critical Care Medicine expert panel recommendations (3) has improved the conduct and interpretation of epidemiologic studies and clinical trials in the field of sepsis. Whereas the categorization of infected patients within the subgroups of severe sepsis and septic shock has been well accepted by physicians, the clinical and prognostic significance of the systemic inflammatory response syndrome (SIRS) criteria, whether or not associated with infection (i.e., sepsis), has been questioned since the panel recommendations were issued (4, 5). Consequently, a reevaluation of the clinical significance of the development of SIRS and sepsis in critically ill patients appears warranted. Several studies have examined risk factors for death of patients with sepsis (1, 2, 6–11) and have shown that variables characterizing underlying disease or associated with acute illness were strongly associated with mortality, whereas characteristics of infection itself appeared to have little influence on outcome of patients. However, most of these studies focused on the most severe stages of sepsis (1, 2, 6–10), and the influence of the SIRS criteria on prognosis has been little examined (11). We previously described the epidemiologic characteristics of infection and sepsis in 3,902 unselected patients from a large database of 14,364 patients (the European sepsis database) admitted to 28 ICUs across Europe, Canada, and Israel over a 1-year period (12). To assess the influence of SIRS on the outcome of these critically ill infected patients, we determined their risk factors for hospital mortality after categorizing them in four stages (infection without SIRS, sepsis, severe sepsis, and septic shock); specifically, we examined the influence of SIRS and of the more severe stages of sepsis on outcome of patients, adjusting for comorbid conditions and infection characteristics.
This prospective multicenter cohort study was conducted over a 1-year period in 28 ICUs from Europe, Canada, and Israel (12). For the analyses presented herein, we selected patients having a first episode of infection either on ICU admission or during the ICU stay, and categorized them according to their degree of septic response within the first day of diagnosis of infection. The organization of the study, definitions used, and data collected are described in a previous article (12) (see online supplement for details).
Statistical Analysis Factors associated with hospital mortality were analyzed using Cox proportional hazards model. We analyzed the overall population of infected patients, and the two subgroups of patients presenting with the two least severe stages of infection (i.e., no organ dysfunction associated with infection, with or without SIRS), and with the two most severe stages (severe sepsis or septic shock). To further assess the influence of infection characteristics and host response, we examined separately the subgroup of patients without comorbidities (see online supplement for these complementary analyses). The end point was survival from the first day of infection to hospital discharge. The following variables were considered: demographic and admission characteristics, underlying diseases if any, severity of acute illness and organ failure scores recorded on the first day of infection (15, 16), grading of sepsis, presence of SIRS and number of SIRS criteria (3), and mode of acquisition of infection (community, hospital, and ICU acquired). Cardiac, respiratory, renal, hepatic, hematologic, and neurologic dysfunction (whether acute or preexisting) were defined as an LOD score greater than 0. Infection sites were entered as pulmonary, abdominal, urinary, primary bacteremia, unknown, and multiple sites, with "other sites" as the reference category; other sites included neurologic, ENT (sinusitis, epiglottitis, otitis, tracheitis), thoracic, genital tract, skin and soft tissue, and bone and joint infection. Microorganisms causing infection were entered as presence or absence of: Staphylococcus aureus, other gram-positive cocci, enterobacteria, aerobic gram negative, anaerobes, Candida and fungi, and polymicrobial, with "other microorganisms" as the reference class; the latter category included gram-negative cocci, gram-positive bacilli, mycobacteria, and intracellular organisms. The occurrence of a subsequent ICU-acquired infection was introduced in the Cox model as a time-dependent covariate. The selection of variables for entry into the model was based on the bootstrapping technique (17, 18), which involved resampling the original data (500 samples) with replacement followed by application of a stepwise Cox model to each sample with a backward–forward selection procedure at the 0.05 level. Covariates that were selected in more than 60% of the 500 samples were included in the final set of covariates and were fitted together (18). Variables associated with p-values less than 0.05 were considered statistically significant and were therefore kept in the final model. The absence of a significant increase in the likelihood value on omission of each of the remaining variables selected in 40 to 60% of the 500 samples was checked. The results were expressed as the hazard ratio with their 95% confidence interval (95% CI). All tests were two-sided. Statistical analysis was performed with SAS 8.0 (SAS, Cary, NC) and S-PLUS 2000 (MathSoft, Seattle, WA) software packages.
Study Population During the study period, 14,364 patients were admitted to the 28 ICUs, of whom 3,034 (21.1%; 95% CI, 20.5–21.8%) were infected at ICU admission (Figure 1) . Among patients without infection on admission (n = 11,330), 5,657 stayed more than 24 hours in ICU, of whom 868 (15.3%; 95% CI, 14.4–16.3%) acquired a first episode of infection during the ICU stay. Of the 3,902 patients with a first episode of infection, 294 (8%) patients could not be classified because of missing information and were excluded from the current study. The remaining 3,608 (92%) represented the study population, and were categorized according to the severity of presentation of infection within the first 24 hours of infection (3). There were 1,063 (29%) patients with sepsis (i.e., infection and SIRS), 827 (23%) with severe sepsis, and 1,134 (31%) with septic shock; 584 (16%) patients had infection without SIRS (subsequently referred to as "infection only").
Survival of Infected Patients The median (interquartile range, IQR) age of infected patients was 63 (47–72) years; there were 2,297 (64%) males. Their median Simplified Acute Physiology Score II (SAPS II) was 42 (IQR, 31–54) on ICU admission and was 41 (IQR, 31–53) on the first day of infection (Tables 1 and 2) . Their median (IQR) hospital length of stay was 19 (11–33) days, 20 (10–38) days, 21 (12–36) days, and 15 (5–33) days, respectively, in patients with infection only, sepsis, severe sepsis, and septic shock. The crude overall hospital mortality rate was 40% (95% CI, 38.4–41.6%) (Table 1). Figure 2 shows the cumulative incidences of hospital deaths according to sepsis stages (Figure 2A) and to the number of SIRS criteria met (Figure 2B) in all patients. Figures 3 and 4 show the cumulative incidences of death within each sepsis stage according to the number of SIRS criteria met. Taking "infection only" as the reference class, the hazard ratio of mortality was estimated as 0.94 (95% CI, 0.77–1.15; p = 0.55) in the univariate analysis for patients presenting with sepsis, and 1.53 (95% CI, 1.26–1.86; p < 10-4) and 2.64 (95% CI, 2.21–3.15; p < 10-4), respectively, in patients presenting with severe sepsis and septic shock. With the "none or one SIRS" criterion as the reference class, the hazard ratio of mortality was estimated as 1.54 (1.26–1.89; p < 10-4) in patients with two SIRS criteria, and as 1.70 (1.42–2.03; p < 10-4) and 1.75 (1.45–2.10; p < 10-4), respectively, in patients presenting with three and four SIRS criteria.
Prognostic Factors of Infected Patients Bivariate analyses of variables associated with hospital mortality in all patients reflecting characteristics of patients and of infection are presented in Tables 1 and 2 and in Table 3 , respectively. Multivariable analyses of factors associated with hospital death in the overall population and in the two subgroups with the least severe stages of sepsis (infection only and sepsis) and the two most severe stages (severe sepsis and septic shock) are shown in Table 4 . In the overall population, 15 variables were determined as independent prognostic factors in 60% or more of the 500 bootstrapped samples and 4 additional variables were identified at the last step (male sex, immunosuppression, unknown site of infection, and four SIRS criteria) of the Cox model. It should be noted that the presence of four SIRS criteria was selected at the last step of the model, and appeared as (marginally) protective (hazard ratio, 0.86; 95% CI, 0.77–0.97; p = 0.01), only after adjustment on other prognostic factors, specifically severity of sepsis and severity score.
Among chronic conditions, liver cirrhosis, immunosuppression, and chronic heart failure were identified as associated with a poor prognosis, whereas male sex was a protective factor. Several acute conditions were associated with mortality, including medical admission category, SAPS II score, hematologic, cardiac, and renal dysfunctions, as well as severe sepsis and septic shock. Among characteristics of infection, hospital acquisition of infection (either at ICU admission or during ICU stay), as well as the occurrence of a subsequent ICU-acquired infection, were poor prognostic factors. Infection from an intraabdominal or unknown source, or as a result of aerobic gram-negative bacilli, predicted hospital mortality, whereas infection caused by enterobacteria was associated with a better outcome. In the subgroup of patients having infection with or without SIRS (n = 1,647), fewer prognostic factors were identified, including chronic heart failure, medical admission, severity of illness as reflected by SAPS II and hematologic dysfunction, hospital-acquired infection, infection with aerobic gram-negative and fungi, and a subsequent ICU-acquired infection (Table 4). In this group of mild to moderate severity, the number of SIRS criteria was not associated with mortality. In patients presenting with severe sepsis or shock (n = 1,961), the same variables associated with mortality as in the overall population were selected, except for three (cardiac dysfunction, chronic heart failure, and subsequent ICU-acquired infection, which were retained in only 37, 26, and 11% of the 500 bootstrapped samples, respectively). Septic shock remained an independent prognostic factor in this subgroup. To further characterize the influence of characteristics of infection, we also examined separately the subgroup of the patients without comorbidities (see data presented in Table E1 in the online supplement). In this group, crude hospital mortality was 25 and 35%, respectively, in patients presenting with infection/sepsis and severe sepsis/septic shock. Compared with the overall population, two additional variables were identified as associated with mortality (trauma and infection with S. aureus), whereas hospital- or ICU-acquired infection on admission, as well as subsequent ICU-acquired infection, were no longer identified as independent risk factors of mortality.
Few studies have examined a large group of critically ill infected patients to elucidate the relationships and respective influence on their outcome of the host response to infection, adjusting for underlying diseases, acute severity of illness, and characteristics of infection. In this large multicenter prospective study, we were able to reexamine these questions in an unselected population of more than 3,500 infected patients. A first conclusion from this study is that infection and its associated inflammatory host response in unselected ICU patients still carries a mortality much higher than that recorded in clinical trials of adjuvant therapy of sepsis. In our study, the hospital mortality ranged from about 25% in uncomplicated infection or sepsis to 40% in patients with severe sepsis and 60% for septic shock. Our results thus confirm the prognostic significance of the gradation of severity from sepsis to severe sepsis and shock. However, we found that infection with SIRS (i.e., sepsis) and without SIRS ("infection only") had a similar outcome, and that the presence of SIRS, either considered as a dichotomous variable or by number of criteria, had no prognostic significance in these two subgroups of patients presenting with the least severe stages of host response to infection. The lack of influence of the number of SIRS criteria was also true in all examined subgroups categorized by severity of sepsis. Gradation of the host response to infection into the three stages of sepsis (i.e., SIRS plus infection), severe sepsis, and septic shock has gained wide acceptance since the American College of Chest Physicians/Society of Critical Care Medicine consensus definitions were published (3). However, the definition of SIRS (i.e., at least two criteria among leukocytosis, fever, tachycardia, and tachypnea), has often been criticized because of its lack of specificity and prognostic value (5, 19). Several studies have found that a high proportion of patients fulfilled SIRS criteria during most of their ICU stay, whether infection was documented or not (11, 20–22). Our study confirms that the presence of SIRS, which characterizes sepsis in an infected patient (3), has limited implication in terms of outcome of patients. The crude hospital mortality of patients with infection only and sepsis was similar at 25%. Although the risk of death and cumulative incidence differed significantly among all infected patients, between those having none or one SIRS criterion and those presenting with two or more SIRS criteria (Figure 2B), apparently confirming the findings by Rangel-Frausto and co-workers that the number of SIRS criteria was associated with the risk of death (11), these differences were no longer apparent after stratification for the severity of sepsis stage (Figures 3 and 4). Specifically, in the subgroup of mild to moderate severity, the cumulative incidences of death of infected patients presenting with or without SIRS were almost identical (Figures 3A and 3B). Likewise, although the presence of more than two SIRS criteria was associated with outcome in the univariate analysis of prognostic factors in the overall population, this influence was lost after adjustment for other prognostic variables identified, whether SIRS was considered a dichotomous variable or by number of criteria present, and there was no difference in outcome of patients presenting with none or one, two, three, or four criteria (Table 4). Conversely, the stratification into infection or sepsis, severe sepsis, and septic shock retained its prognostic value, after adjustment on other prognostic factors. Because the distinction between "infection" and "sepsis" (without organ dysfunction) based on two or more SIRS criteria has no significance in terms of prognosis for infected patients, the presence of SIRS is unlikely to fulfill the specific objective of identifying patients at high risk of a poorer outcome than that of patients not fulfilling SIRS criteria. Indeed, in clinical trials to date, the presence of organ dysfunction, qualifying patients for severe sepsis, has been the preferred selection criterion. However, only the characteristics of patients and of infection on the first day of infection were considered in our study, from the viewpoint of overall hospital mortality. The possible transition from one stage of sepsis to another was not examined, and factors associated with worsening or improvement of the sepsis syndromes, and of clinical variables associated with such evolution, should be analyzed to complement our results. It is possible, for example, that the presence of several SIRS criteria may be one means to help identify patients at higher risk of subsequent organ dysfunction, if not mortality, irrespective of other potential risk factors. It should be recalled, however, that the original SIRS criteria and thresholds used were defined empirically. Refining or expanding these criteria appears to be needed, based on objective analysis of factors associated with the occurrence of organ failures and outcome, as suggested by an expert panel (19). With regard to organ failures, the presence of acute hematologic dysfunction (i.e., severe neutropenia or thrombocytopenia) or renal dysfunction was associated with prognosis across all subgroups, in addition to the acute physiology severity score SAPS II. Consequently, these variables should be considered when analyzing results of clinical trials. Shock remained a major and consistent prognostic factor after adjustment on other prognostic variables, both in the overall population and the most severe population, which supports using this clinically simple variable for stratifying septic patients in clinical trials or epidemiologic studies, as already suggested by our group (1). Our results are also consistent with previous reports of prognostic factors in severe sepsis and shock, where the SAPS II score, organ failures or their number, and comorbidities such as chronic liver and cardiovascular failure, and immunosuppression, were found to be associated with a poor outcome (1, 6). In addition to the mode of acquisition of infection, some of the microbiological characteristics of infection (infection caused by aerobic gram-negative bacilli or enterobacteria), and an abdominal or unknown source, also influenced prognosis. As previously suggested in studies of bacteremia (6, 23, 24), infection caused by enterobacteria were protective compared with other etiologies whereas infection caused by aerobic gram-negative bacteria (GNB) were associated with a poor prognosis across all subgroups and after adjustment (Table 4); finally, infection caused by S. aureus was identified as a poor prognostic factor only in patients without comorbidities (see Table E1 in the online supplement). These results highlight the growing threat posed by infection with aerobic GNB and gram-positive cocci, which may reflect a high virulence of aerobic GNB or of S. aureus or possibly a higher rate of resistance to antimicrobial agents and difficult treatment of infection caused by these organisms. Unfortunately, antibiotic therapy or its adequacy was not recorded in our study, and the latter hypothesis cannot be substantiated. Among clinical characteristics of infection, only the intraabdominal site and infection from an unknown source was associated with hospital death, after adjustment for other prognostic factors. Finally, our results emphasize the relatively long duration of hospital stay of infected patients, which ranges from a median of 15 to 21 days for patients according to sepsis stages. The shorter hospital stay associated with septic shock likely reflects the high proportion of early deaths within this subgroup. Given this relatively long duration of hospital stay, it appears necessary that the follow-up of patients included in sepsis studies should be extended beyond 28 days. One strength of this study is the large number of patients included in our database, and the prospective inclusion of an unselected population of all patients with infection or sepsis, which allowed a closer examination of prognostic factors. Most prior epidemiologic studies have focused on one or two of the sepsis stages, usually the most severe ones (1, 2, 7, 10), with one exception (11). The fact that the population was unselected allowed for a clearer identification of the influence of comorbid conditions on outcome of patients, whereas the selection of patients included in clinical trials explains the lower mortality recorded in such trials and limits the generalizability of their findings (25). Our results can help in selecting variables for adjustment of outcome in sepsis studies. We did not wish to present a new "customized model" for sepsis or severe sepsis, as two such models have already been published (26, 27), and there are questions concerning the applicability of such models to independent series of patients. We would rather suggest adjusting for variables of recognized prognostic significance, such as those identified in our study, when analyzing outcome in sepsis studies. One potential limitation of our study is that, as units participating in this study were selected on a voluntary basis, patients may not be representative of all ICU patients with infection or sepsis, and the external validity of our results may be questioned. However, this study is based on a high-quality database from 28 ICUs in academic centers in several countries, and we believe the large number of patients included makes the results of our analyses pertinent and likely applicable to other similar centers, if not all ICUs. Another feature of this study is the use of the bootstrapping method for analyzing risk factors for death, a method allowing reliable selection of the most important variables retained in prognostic models (17, 18). In conclusion, infection still carries a high mortality in unselected intensive care unit patients, varying from 25 to 60%, depending on the presence of organ dysfunction and shock, severity of underlying diseases, and comorbidities, and a few characteristics of infection. Our study can help to select relevant variables of prognostic significance for adjustment in outcome analyses of sepsis studies. Characterization of infected patients by the presence of SIRS criteria has no prognostic implication, and infected patients with or without SIRS and no organ dysfunction have a similar outcome. Within sepsis stages, only the categorization of infected patients by the presence of organ dysfunction or shock has prognostic significance, and the number of SIRS criteria does not influence outcome. More sophisticated indicators of outcome risk in the early stage of infection are needed, especially in the population presenting with infection and no organ dysfunction. Considering the addition of biological markers may be useful for that purpose (4, 19, 28).
European Consortium for Intensive Care Data: H. Burchardi, M.D., Chairman, Göttingen. Paris-Coordinating center: J. R. Le Gall, M.D., Director; E. Lepage, M.D., Ph.D., Biostatistician; C. Alberti, M.D., Biostatistician; Ronan Boulmé, database manager; Viviane Teboul, administrative secretary. Scientific committee: C. Brun-Buisson, M.D., Créteil; P. Loirat, M.D., Suresnes; P. Metnitz, M.D., Wien; B. Misset, M.D., Paris; R. Moreno, M.D., Lisbon; M. Palazzo, M.D., London; M. Smithies, M.D., Cardiff. Participating units: Canada: Victoria Campus, London, C. Martin, M.D., Country Coordinator, L. MacCarthy, Saint Joseph's Hospital, Hamilton, H. Fuller, M.D., S. Spisak, Toronto Hospital, Toronto, J. Granton, M.D., D. Foster, Royal Victoria Hospital, Montreal, S. Magder, M.D., D. Jones; France: Henri Mondor Hospital, Créteil, C. Brun-Buisson, M.D., Country Coordinator, E. Girou, Pharm.D., Saint Joseph Hospital, Paris, J. Carlet, M.D., L. Montuclard, M.D., Cochin Hospital, Paris, J. F. Dhainaut, M.D., S. Bouam, M.D., Saint Louis Hospital, Paris, J. R. Le Gall, M.D., S. Bouam, M.D., Albert Calmette Hospital, Lille, F. Saulnier, M.D., S. Beague, M.D.; Germany: Göttingen Hospital, Göttingen, H. Burchardi, M.D., Country Coordinator, O. Thomas, M.D., Chirurgische Universitätsklinik, Würzburg, H. Reith, M.D., U. Mittelkötter, M.D., Klinikum Kröllwitz, Halle, K. Werdan, M.D., N. Kühn; Israel: Hadassah Hebrew University Medical Center, Jerusalem, C. Sprung, M.D., S. Goodman, M.D.; Italy: Maggiore Hospital, Milano, M. Langer, M.D., Country Coordinator, A. Sicignano, M.D., San Gerardo Tintori Hospital, Monza, R. Fumagalli, M.D., C. Carozzi, Policlinico Umberto Ier, Roma, A. Gasparetto, M.D., M. Antonelli, M.D., San Raffaele Hospital, Milan, D. Guidici, M.D., E. Casaletti, Desio Hospital, Desio, S. Vesconi, M.D., G. Fontana; Portugal: Santo Antonio dos Capuchos Hospital, Lisbon, R. Moreno, M.D., Country Coordinator, E. Pereira, M.D., Unidade de Urgência Medica, S. José Hospital, Lisbon, J. Sá, M.D., Desterro Hospital, Lisbon, E. Silva, M.D., M. J. Serra, M.D.; Spain: Parc Taulli Hospital, Sabadell Barcelona, A. Artigas, M.D., Country Coordinator, A. R. Ochagavia, M.D., Valle Hebron Hospital, Barcelona, J. L. Boveda, M.D., De Getafe Hospital, Madrid, A. Esteban de la Torres, M.D., D. Alia, De Valme Hospital, Sevilla, C. Leon Gil, M.D., M. A. Sanchez, Del Mare Hospital, Barcelona, F. Solsona, M.D., R. Martinez; UK: Charing Cross Hospital, London, M. Palazzo, M.D., Country Coordinator, M. Templeton, University Hospital of Wales, Cardiff, M. Smithies, M.D., T. Jones.
The authors gratefully thank Prof. Lepage for expertise in the study protocol and support, Prof. Sylvie Chevret for helpful reviewing of the manuscript, and all the participating members of the study (see Appendix).
Supported by an educational grant from Roche. This article has an online supplement, which is accessible from this issue's table of contents online at www.atsjournals.org Received in original form August 1, 2002; accepted in final form April 13, 2003
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
TOP
ABSTRACT
METHODS
2 test for categorical data, and Wilcoxon or Kruskal–Wallis tests for continuous data. Cumulative incidences of hospital death from the first day of infection were computed, considering patients discharged alive from hospital as no longer exposed to the risk of death in hospital (
missing data on sepsis stage, n = 104; 
missing data on sepsis stage, n = 294.
1-antitrypsin, and lactoferrin compared with clinical variables. Clin Infect Dis 1999;29:398–407.
