点击显示 收起
Clinical Epidemiology Unit, Hpital Robert Debree
Department of Medical Biostatistics, Hpital Saint-Louis
Medical Intensive Care Unit, Hpital Saint-Louis, Assistance Publique—Hpitaux de Paris, Universitee Paris VII, Paris
Medical Intensive Care Unit, Hpital Henri-Mondor, Assistance Publique-Hpitaux de Paris, Universitee Paris XII, Creeteil, France
Istituto di Anestesiologia e Rianimazione, Universita Cattolica del Sacro Cuore, Policlinico A. Gemelli, Rome, Italy
General Intensive Care Unit, Department of Anesthesiology and Critical Care Medicine, Hadassah Hebrew University Medical Center, Jerusalem, Isral
Critical Care-Trauma Centre, London Health Sciences Centre, London, Ontario, Canada
Intensive Care Unit, Santo Antonio dos Capuchos Hospital, Lisboa, Portugal
Intensive Care Unit, Parc Tauli Hospital, Red Gina, Spain
Intensive Care Unit, Charing Cross Hospital, London, United Kingdom
Universitatsklinik und Poliklinik fur Innere Medizin III, Klinikum Krollwitz der Martin-Luther-Universitat Halle-Wittenberg, Halle, Germany
ABSTRACT
Rationale: The systemic inflammatory response syndrome has low specificity to identify infected patients at risk of worsening to severe sepsis or shock. Objective: To examine the incidence of and risk factors for worsening sepsis in infected patients. Methods: A 1-year inception cohort study in 28 intensive care units of patients (n = 1,531) having a first episode of infection on admission or during the stay. Measurements and main results: The cumulative incidence of progression to severe sepsis or shock was 20% and 24% at Days 10 and 30, respectively. Variables independently associated (hazard ratio [HR]) with worsening sepsis included: temperature higher than 38.2°C (1.6), heart rate greater than 120/minute (1.3), systolic blood pressure higher than 110 mm Hg (1.5), platelets higher than 150 x 109/L (1.5), serum sodium higher than 145 mmol/L (1.5), bilirubin higher than 30 eol/L (1.3), mechanical ventilation (1.5), and five variables characterizing infection (pneumonia [HR 1.5], peritonitis [1.5], primary bacteremia [1.8], and infection with gram-positive cocci [1.3] or aerobic gram-negative bacilli [1.4]). The 12 weighted variables were included in a score (Risk of Infection to Severe Sepsis and Shock Score, range 0eC49), summarized in four classes of "low" (score 0eC8) and "moderate" (8.5eC16) risk (9% and 17% probability of worsening, respectively), and of "high" (16.5eC24) and "very high" (score > 24) risk (31% and 55% probability, respectively). Conclusions: One of four patients presenting with infection/sepsis worsen to severe sepsis or shock. A score estimating this risk, using objectively defined criteria for systemic inflammatory response syndrome, could be used by physicians to stratify patients for clinical management and to test new interventions.
Key Words: infection, intensive care units, multivariable models, risk prediction, sepsis, septic shock
Sepsis is a common and complex entity, with marked heterogeneity of patients affected and wide variations in outcome (1, 2). For more than 10 years, the severity of sepsis has been graded, according to the American College of Chest Physicians/Society of Critical Care Medicine (ACCP/SCCM) classification (3), in three groups of increasing severity (i.e., sepsis, the combination of infection and a systemic inflammatory response, or SIRS), severe sepsis, and septic shock, which are viewed as a continuum of risk (4). Although somewhat subjective, this classification has been useful in epidemiologic studies or clinical trials, even if criteria defining groups have been slightly modified across studies (5, 6).
One of the main purposes of identifying SIRS and sepsis—the lower severity group—among the different stages of the systemic response to infection is to help identify patients at risk of progression to a more severe stage, for early therapeutic intervention, and possible enrolment into clinical trials of new therapies (3). However, the original criteria used to define SIRS were broad, limited in number, and subjectively determined. The definition has therefore been much criticized for its poor specificity and lack of prognostic significance (7eC9), and refined and more comprehensive definitions are awaited (7, 10).
In this study, we used the European Sepsis Database (9, 11) to estimate the probability of progression from the early stage of sepsis to a more severe one during the intensive care unit (ICU) stay, considering mortality without progression and ICU discharge as competing risks and examine the influence of comorbidities, physiologic disturbances, and infection characteristics on the risk of progression from infection and sepsis to severe sepsis or septic shock. We thus objectively selected variables and their respective cutoffs for inclusion into a revised definition of the SIRS, and developed a score of risk for worsening sepsis (the Risk of Infection to Severe Sepsis and Shock Score, or RISSC). The use of such a score could help stratify patients in epidemiologic studies and clinical trials of early sepsis. Some of these results have been presented previously in abstract form (12).
METHODS
Patients
This prospective cohort study was conducted over a 1-year period in 28 ICUs of eight countries from Europe, Canada, and Israel (11). All adult patients (age 18 years) consecutively admitted in the participating ICUs during a 1-year period (May 1, 1997, to April 30, 1998) were enrolled. For the analyses presented herein, we selected patients having a first infectious episode occurring either on ICU admission or during ICU stay. Infected patients were categorized by the investigators according to one of the three sepsis stages on the first day of diagnosis of infection (3). In a previous study, we found that patients presenting with a first episode of infection at ICU admission or during ICU stay had a similar outcome (9). Similarly, outcome did not differ between patients having infection with (i.e., sepsis group) or without (i.e., infection only) the usual criteria defining SIRS, in the absence of severe sepsis or shock. Therefore, in the present study, all patients with a first episode of infection independently of its time of occurrence were analyzed together. In addition, patients having infection with or without SIRS and no severe sepsis were grouped under the category of "infection/sepsis."
Data Collection and Definitions
The methods used in this study have been described elsewhere (9, 11). Briefly, data were collected by a trained investigator in each participating unit, using standardized forms and a database software (Microsoft FoxPro 5.0; Microsoft, Redmond, WA), according to a detailed operating manual specifying all definitions used. The following patient data were collected: demographics, ICU admission characteristics and underlying diseases, including major preexisting comorbidities and immunodepression (detailed in the online supplement). Clinical and physiologic data at time of infection were taken as the worst value recorded during the first 24 hours of diagnosis. All patients were screened for infection at ICU admission and daily during the ICU stay. Infection was defined on the basis of clinical history and symptoms, physical examination, laboratory findings, and administration of antiinfective therapy. Infection was characterized according to its mode of acquisition (community-, hospital-, or ICU-acquired) (13), its source, documentation (microbiologic or clinical only), etiologic microorganism(s), and grading of sepsis severity (3) at presentation of infection. SIRS was defined, independent of the source of infection, as the presence of at least two of four criteria (temperature, heart rate, respiratory rate, and leukocytosis) defined by the ACCP/SCCM conference (3); in patients receiving mechanical ventilation, this variable was substituted for a high respiratory rate. Organ dysfunctions characterizing severe sepsis, when not directly related to the primary cause of infection and otherwise unexplained, were defined as a logistic organ dysfunction score greater than 0 (14). The sepsis stage was assessed on the first day of diagnosis of infection, then daily until ICU discharge or Day 30, whichever occurred first, and mortality at ICU and hospital discharge was recorded.
Statistical Analysis
Results are presented as percentages for categorical variables and median (quartiles) or mean (± SD) for continuous variables. Statistical analyses were performed using SAS 8.2 (SAS Inc., Cary, NC) and S-plus 2,000 (MathSoft Inc., Seattle, WA) software packages.
Incidence Estimates of Sepsis Progression
The time to occurrence of the most severe sepsis stage was estimated using cumulative incidences up to ICU Day 30, considering death without progression as a competing risk with regard to the progression of sepsis (15, 16).
Risk Factors for Progression of Sepsis and Development of the RISSC
Variables associated with progression of infection/sepsis to severe sepsis or shock were analyzed using the regression model for the subdistributions of competing risks developed by Fine and Gray (S-plus library, available at http://biowww.dfci.harvard.edu/~gray) (17) (see online supplement for additional details). The following variables were studied: demographic and admission characteristics, underlying diseases if any, clinical and biological variables recorded on the first day of infection, and characteristics of infection (site and microbiology). To identify cutpoints defining severity levels for each continuous covariate, a nonparametric spline estimation was performed using the generalized additive model procedure (18). The cutpoints were used to define dummy variables for individual levels of each variable. Variables with p value less than 0.3 in bivariate analysis were first entered in the multivariable model, and stepwise backward-forward selection was used. Results were expressed as hazard ratio with 95% confidence intervals. All tests were two-sided. Bootstrapping of a large number of subsamples was used to validate the weight of each variable retained in the final model (19eC22). The score of risk for progression from infection to severe sepsis was developed from the regression coefficients associated with each variable. Finally, the score was divided into four risk classes for stratification in clinical use. The internal validity of the score was also tested on several subgroups of patients. In addition, because excluding patients who died without worsening sepsis may appear questionable, we rerun the score analysis after including this subgroup of patients within the group that worsened (see online supplement).
RESULTS
Study Population
Of the 14,364 patients admitted to the 28 ICUs during the study period, 459 (3.2%) were excluded because of missing information or incomplete follow-up, and 3,443 patients with a first episode of infection occurring on admission (n = 2,688) or during ICU stay (n = 755) remained in the cohort analyzed (Figure 1). There were 1,531 (44.5%) patients presenting with infection/sepsis, 795 (23.1%) with severe sepsis, and 1,117 (32.4%) with septic shock. Their median (quartiles) age was 63 (43eC73) years and their Simplified Acute Physiology Score II on ICU admission was 42 (32eC55). The overall crude ICU and hospital mortality of the 3,443 patients with infection was 34% (n = 1,178) and 41% (n = 1,402), respectively. According to the sepsis stage at presentation of infection, the hospital mortality rate was 26% in the infection/sepsis group, 42% in the severe sepsis group, and 61% in the septic shock group.
From Infection to Severe Sepsis or Septic Shock: Incidence and Mortality
The clinical characteristics of the 1,531 patients initially presenting with infection/sepsis are shown in Table E1 (see online supplement), which also compares patients worsening to those having died without worsening from sepsis to severe sepsis. A total of 167 patients (11%) developed severe sepsis and 201 (13%) developed septic shock before ICU discharge or Day 30, whereas 1,163 did not worsen within the 30-day follow-up period (Figure 2). The cumulative probability of severe sepsis or shock, accounting for the competitive risks of death or survival to ICU discharge before 30 days, is shown in Figure 3. At Day 28, the probability estimate of progression (24.2%) approximately equates the sum of crude incidences of severe sepsis or shock. Hospital mortality ranged from 17 to 26% for patients presenting with infection/sepsis and not worsening, or returning to the former stage after experiencing a more severe one, whereas those patients evolving to severe sepsis or shock and not returning to a former stage had a mortality ranging from 47 to 97% (Figure 2).
From Infection to Severe Sepsis or Septic Shock: Risk Factors
In bivariate analyses (Table 1), we found that liver cirrhosis was the only comorbidity favoring the progression from infection to severe sepsis. In addition to all the original SIRS criteria, other physiologic variables (with corresponding threshold defined by spline analysis) were found associated with progression—systolic arterial blood pressure (< 110 mm Hg), platelet count (< 150 x 109/L), serum sodium (> 145 mmol/L), bilirubin (> 30 eol/L), and blood urea (> 15 mmol/L). For variables included in the original SIRS criteria, the cutpoints were defined as follows: a temperature greater than 38.2°C, a heart rate higher than 120/minute, and blood leukocyte count lower than 4.0 x 109/L. Among infection characteristics, pneumonia, peritonitis, primary bacteremia, and infection caused by gram-positive cocci or aerobic gram-negative bacilli were also identified as risk factors for worsening sepsis.
Twelve variables were retained in the final multivariable model (Table 2), including six physiologic variables (temperature, heart rate, systolic blood pressure, platelet count, serum sodium, and bilirubin), as well as mechanical ventilation (used in place of respiratory rate in patients on a ventilator), three infection sites (pneumonia, peritonitis, and primary bacteremia) and two categories of microorganisms (gram-positive cocci and aerobic gram-negative bacilli).
The RISSC for Progression From Infection to Severe Sepsis
The score derived from the regression coefficient of each variable retained in the final model ranges from 0 to 49, with highest weights attributed to hyperthermia, primary bacteremia, and mechanical ventilation. In the bootstrap analyses, the median weight attributed to each variable was close to that of the original model (Table 2).
In our cohort of 1,531 patients with infection/sepsis, the score ranged from 0 to 38, with a mean (± SD) value of 15 (± 7). Figure 4 shows the prevalence of the eight four-point classes of the score in this population, obtained after merging the subgroups with a score greater than 28, and the corresponding proportion of patients worsening to severe sepsis. The cumulative incidences of progression to severe sepsis in these eight subgroups are shown in Figure E1. To simplify the score for clinical use, this information was summarized in four risk strata (Figure 5): the "low" (score 0eC8) and "moderate" (score 8.5eC16) risk groups, for which the cumulative estimated risk of progression to severe sepsis was 9% and 17%, and the "high" (score 16.5eC24) and "very high" risk (score > 24) groups, associated with a risk of progression to severe sepsis of 31% and 55%, respectively (Table 3). Corresponding curves for various subgroups of patients are shown in the online supplement.
DISCUSSION
A major purpose of introducing the definition for SIRS was to help early identification of infected patients at risk of worsening (3). In this study, we analyzed the rate and risk for progression of patients with infection or sepsis to the more severe stages of severe sepsis or shock, and reexamined the predictive value of the SIRS criteria in this regard. We thus refined and complemented these criteria to select a set of variables associated with the risk of sepsis progression, which were then used to develop a score (the RISSC). The score includes 12 objectively defined variables characterizing infection and associated physiologic disturbances. This score should help physicians caring for critically ill patients to stratify infected patients according to their probability of worsening at initial presentation and identify high-risk patients potentially justifying more aggressive and early intervention as well as for the design of clinical trials of new therapeutic approaches in sepsis.
Only Rangel-Frausto and colleagues (4, 23) previously analyzed the rate of progression from infection and sepsis to severe sepsis and septic shock. In this study, we provide incidence estimates of these progressions in patients from a large database of European ICUs (9, 11) using appropriate statistical modeling, accounting for death without progression and ICU discharge as competing events (15, 17).
Incidence of Sepsis Progression
Approximately one of four patients presenting with infection/sepsis worsened to a more severe stage during the 30-day follow-up period (Figure 2). Worsening occurred mostly in the first 10 days after inclusion: the cumulative incidence of severe sepsis was estimated at approximately 20% and 24% at 10 and 28 days after onset of infection/sepsis, respectively (Figure 3). These estimates are much lower than those reported by Rangel-Frausto and colleagues (23), who found a 64% probability at Day 14 and 72% overall of developing severe sepsis after the occurrence of sepsis (4). This may be due to differences in design and population studied, but mostly to analytic methods (16, 17), as discussed in the online supplement. In addition, both initial and secondary sepsis (i.e., return to sepsis after severe sepsis and septic shock) were considered by Rangel-Frausto and colleagues (23), whereas we modeled only the first transition from sepsis to severe sepsis or shock.
The Grading and Progression of Sepsis and its Impact on Mortality
Our study confirmed the hierarchical continuum of severity across the three stages of sepsis, severe sepsis, and shock. Patients presenting with infection/sepsis had an overall mortality rate of 26% (Figure 2), comparable to rates recorded in similar patients by others (4, 24). This is markedly lower than the 61% and 42% overall hospital mortality recorded in patients presenting with septic shock or severe sepsis, respectively, but is consistent with the recognized prognostic importance of shock and organ failures (4, 25eC32). Nevertheless, the mortality rate of patients with infection/sepsis varied between 17% and 97%, depending on whether sepsis was the most severe stage recorded during the follow-up period or patients experienced a more severe stage. Interestingly, patients returning to a milder stage after experiencing severe sepsis or shock had a mortality rate only marginally higher (23%) than that of patients not worsening (17%, Figure 2). Our results therefore confirm the grading of mortality according to the hierarchical sepsis stages for both improvement and worsening of sepsis.
Development of the Risk Score
Although the 1992 ACCP/SCCM conference definitions (3) have been useful in epidemiologic investigations and clinical trials, the criteria for SIRS are broad and insufficiently specific, and revision of these criteria is being actively sought (7, 8, 10, 33eC35). A recent expert panel suggested complementing the SIRS criteria with several other variables in the sepsis definition (10, 33), including predisposing factors and comorbidities, characteristics of infection, physiologic variables characterizing the host response, and manifestations of organ dysfunction. However, their relative weight remains uncertain. The panel also suggested including biological markers of the inflammatory response (e.g., cytokines, procalcitonin) to improve the specificity of the definition, and, ultimately, genetic predisposing factors to identify high-risk patients (33). It was also hoped that a more rigorous approach to disease stratification could be used through development of models rather than consensus (34).
Based on objective selection of variables and definition of cut-points, we developed a new set of variables and a score (RISSC) to estimate the risk of worsening sepsis in critically ill patients with infection. Twelve variables were retained in the final multivariable model (Table 2), including six physiologic variables. In addition to three of the four variables (temperature, heart rate, respiratory rate or mechanical ventilation) included in the original SIRS definition, three others were included (platelet count, serum sodium, bilirubin). The spline analyses allowed to define the predictive cutpoints for temperature (> 38.2°C), heart rate (> 120/minute), and leukocyte count (< 4.0 x 109/L) more accurately than in the original SIRS definition. Interestingly, the cutoff for heart rate we found associated with sepsis progression was higher (120/minute) than that empirically selected by the ACCP/SCCM conference (90/minute). Conversely, a relatively high cutoff was determined for systolic blood pressure (110 mm Hg), suggesting that patients having a marginally low blood pressure in the context of infection might need close monitoring. Of note, leukocytosis was not retained in the final model. Mechanical ventilation at onset of sepsis had a strong weight, and tachypnea did not remain in the model after adjusting for the former. Major comorbidities had little influence on sepsis progression (Table 1). Only liver cirrhosis was associated with progression of sepsis, but this variable was not retained in the final model. We also examined whether having one or more comorbidity influenced the risk of sepsis progression, but this was not the case (data shown in the online supplement). Comorbidities therefore appear more strongly associated with the risk of (late) mortality in sepsis (25, 36, 37) probably though interaction between preexisting organ failures and sepsis-related organ dysfunction characterizing severe sepsis, rather than with the risk of early progression from infection to severe sepsis (see online supplement for complementary analyses of comorbidities and expanded discussion).
The RISSC also includes five variables related to infection characteristics. Among these, two can be easily obtained from the initial evaluation of the patient (pneumonia and peritonitis), whereas the three others require simple microbiological documentation (bacteremia, gram-positive cocci, and aerobic gram-negative infection). A simplified score, omitting these variables, can, be developed, however. Using this approach, we found that the number of points attributed to the other variables did not change (data shown in the online supplement, Table E3).
For routine clinical risk stratification of patients, we first split the score in eight four-point classes, subsequently simplified into four subclasses of risk (Table 3). This classification provided a risk estimate for progression varying from 9% and 17% in the "low" and "moderate" risk classes, to 31% and 55% in the two higher risk classes. Depending on the context and objectives, physicians/investigators may wish to select or exclude patients at low risk or at higher risk.
The strengths of our study include its prospective design and conduct on a relatively large number of unselected patients with infection or sepsis from 28 ICUs in academic centers in several countries. These features favor the applicability of our findings to routine clinical practice. However, our study has potential limitations. We used bootstrapping analysis rather than a split-sample approach to confirm the weight attributed to each variable and assess the internal validity of our model, because the latter approach results in less accurate estimates of the internal validity compared with bootstrap (38). In the split-sample approach, the performance of the model is underestimated because only one part (usually one-third) is used to construct the model, and the other for validation; thus, the estimate can be unstable. In contrast, with regular bootstrapping, the model is constructed and validated in datasets with 100% of the patients (20, 21, 38). In addition, we tested the validity of the score in various subgroups of patients (see online supplement). Nevertheless, it would be useful to validate our score on an external database of critically ill infected patients. The ability of the score to fulfill its objectives also needs to be confirmed in clinical practice.
Although we used objective methods to select variables and define their cutoff, not all variables that might be considered for inclusion into a risk score of sepsis progression were considered in our analysis (33, 34). For example, we had no information on acute-phase proteins (e.g., procalcitonin, C-reactive protein) or on cytokines levels in our population. There is also increasing evidence that the patients' genetic predisposition is an important contributor to the presentation and outcome of severe infections (39), which might account in part for the heterogeneity of septic patients. Incorporating such variables into a sepsis risk score is of potential importance and might be feasible in a not-too-distant future (33, 34).
To summarize, we found that approximately one-fourth of our patients with infection or sepsis evolve to a more severe stage. A larger number of variables than previously incorporated in the SIRS definition, including physiologic measures and characteristics of infection, were objectively identified as associated with the risk of progression from infection to severe sepsis or shock. These variables can be incorporated into a score to stratify patients according to the estimated probability of worsening sepsis, at time of diagnosis of infection.
Acknowledgments
Participants to the European Sepsis Study are listed in the Appendix (see online supplement).
Funded in part by unrestricted grants from Roche laboratories and Glaxo-Smith-Kline.
This article has an online supplement, which is accessible from this issue's table of contents at www.atsjournals.org
REFERENCES
Knaus WA, Sun X, Nystrom PO, Wagner DP. Evaluation of definitions for sepsis. Chest 1992;101:1656eC1662.
Angus DC, Linde-Zwirble WT, Lidicker J, Clermont G, Carcillo J, Pinsky MR. Epidemiology of severe sepsis in the United States: analysis of incidence, outcome, and associated costs of care. Crit Care Med 2001;29:1303eC1310.
Bone RC, Balk RA, Cerra FB, Dellinger EP, Fein AM, Knaus WA, Schein RM, Sibbald WJ, the ACCP/SCCM Consensus Conference Committee. Definitions for sepsis and organ failure and guidelines for the use of innovative therapies in sepsis. Chest 1992;101:1656eC1662.
Rangel-Frausto MS, Pittet D, Costigan MD, Hwang T, Davis CS, Wenzel RP. The natural history of the systemic inflammatory response syndrome. JAMA 1995;273:117eC123.
Sands KE, Bates DW, Lanken PN, Graman PS, Hibberd PL, Kahn KL, Parsonnet J, Panzer R, Orav EJ, Snydman DR, et al. Epidemiology of sepsis syndrome in 8 academic medical centers. JAMA 1997;277:234eC240.
Bernard GR, Vincent JL, Laterre PF, LaRosa SP, Dhainaut J-F, Lopez-Rodriguez A, Steingrub JS, Garber GE, Helterbrand JD, Ely EW, et al. Efficacy and safety of recombinant human activated protein C for severe sepsis. N Engl J Med 2001;344:699eC709.
Abraham E, Matthay MA, Dinarello CA, Vincent JL, Cohen J, Opal SM, Glauser M, Parsons P, Fisher CJ, Repine JE. Consensus conference definitions for sepsis, septic shock, acute lung injury, and acute respiratory distress syndrome: time for a reevaluation. Crit Care Med 2000;28:232eC235.
Pittet D, Rangel-Frausto S, Li N, Tarara D, Costigan M, Rempe L, Jebson P, Wenzel RP. Systemic inflammatory response syndrome, sepsis, severe sepsis and septic shock: incidence, morbidities and outcomes in surgical ICU patients. Intensive Care Med 1995;21:302eC309.
Alberti C, Brun-Buisson C, Goodman SV, Guidici D, Granton J, Moreno R, Smithies M, Thomas O, Artigas A, Le Gall JR. Influence of systemic inflammatory response syndrome and sepsis on outcome of critically ill infected patients. Am J Respir Crit Care Med 2003;168:77eC84.
Vincent JL. Sepsis definitions. Lancet 2002;2:135.
Alberti C, Brun-Buisson C, Burchardi H, Martin C, Goodman S, Artigas A, Sicignano A, Palazzo M, Moreno R, Boulmee R, et al. The European Sepsis Group. Epidemiology of sepsis and infection in ICU patients from an international multicenter cohort study. Intensive Care Med 2002;28:108eC121. [Published erratum appears in Intensive Care Med 2002;28:525eC526.].
Alberti C, Brun-Buisson C, Chevret S, Le Gall JR, for the European Sepsis Study Group. Risk of progression of infection or sepsis to severe sepsis or shock: the RISSC score. Intensive Care Med 2003;29:S98.
Garner JS, Jarvis WR, Emori TG, Horan TC, Hughes JM. CDC definitions for nosocomial infections, 1988. Am J Infect Control 1988;16:128eC140.
Le Gall JR, Klar J, Lemeshow S, Saulnier F, Alberti C, Artigas A, Teres D, The ICU Scoring Group. The logistic organ dysfunction system: a new way to assess organ dysfunction in the intensive care unit. JAMA 1996;276:802eC810.
Pepe MS, Mori M. Kaplan-Meier, marginal or conditional probability curves in summarizing competing risks failure time data Stat Med 1993;12:737eC751.
Alberti C, Meetivier F, Landais PJ, Thervet E, Legendre C, Chevret S. Improving estimates of event incidence over time in populations exposed to other events: application to three large databases. J Clin Epidemiol 2003;56:536eC545.
Fine JP, Gray RJ. A proportional hazards model for the subdistribution of a competing risk. J Am Stat Assoc 1999;94:496eC509.
Hastie T, Tibshirami R. Generalized additive models for medical research. Stat Methods Med Res 1995;4:187eC196.
Altman DG, Andersen PK. Bootstrap investigation of the stability of a Cox regression model. Stat Med 1989;8:771eC783.
Altman DG, Royston P. What do we mean by validating a prognostic model Stat Med 2000;19:453eC473.
Efron B, Tibshirani RJ. An introduction to the bootstrap. New York: Chapman & Hall; 1993.
Harrell FE Jr, Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 1996;15:361eC387.
Rangel-Frausto MS, Pittet D, Hwang T, Woolson RF, Wenzel RP. The dynamics of disease progression in sepsis: Markov modeling describing the natural history and the likely impact of effective antisepsis agents. Clin Infect Dis 1998;27:185eC190.
Brun-Buisson C. The epidemiology of the systemic inflammatory response. Intensive Care Med 2000;26:S64eCS74.
Brun-Buisson C, Doyon F, Carlet J, Dellamonica P, Gouin F, Lepoutre A, Mercier JC, Offenstadt G, Reegnier B. Incidence, risk factors, and outcome of severe sepsis and septic shock in adults. A multicenter prospective study in intensive care units. JAMA 1995;274:968eC974.
Brun-Buisson C, Doyon F, Carlet J. Bacteremia and severe sepsis in adults: a multicenter prospective survey in ICUs and wards of 24 hospitals. Am J Respir Crit Care Med 1996;154:617eC624.
Kieft H, Hoepelman AIM, Zhou W, Rozenberg-Arski M, Struyvenberg A, Verhoef J. The sepsis syndrome in a Dutch University Hospital: clinical observations. Arch Intern Med 1993;153:2241eC2247.
Pittet D, Thieevent B, Wenzel RP, Li N, Auckenthaler R, Suter PM, Thievent B. Bedside prediction of mortality from bacteremic sepsis: a dynamic analysis of ICU patients. Am J Respir Crit Care Med 1996;153:684eC693.
Ponce de Leon-Rosales SP, Molinar-Ramos F, Dominguez-Cherit G, Rangel-Frausto MS, Vazquez-Ramos VG. Prevalence of infections in intensive care units in Mexico: a multicenter study. Crit Care Med 2000;28:1316eC1321.
Salvo I, de Cian W, Musico M, Langer M, Piadena R, Wolfler A, Montant C, Magni E, the Sepsis Study Group. The Italian SEPSIS study: preliminary results on the incidence and evolution of SIRS, sepsis, severe sepsis and septic shock. Intensive Care Med 1995;21:S244eCS249.
Vallees J, Leon C, Alvarez-Lerma F. The Spanish Collaborative Group for Infections in Intensive Care Units of SEMIUC. Nosocomial bacteremia in critically ill patients: a multicenter study evaluating epidemiology and prognosis. Clin Infect Dis 1997;24:387eC395.
Annane D, Aegerter P, Jars-Guincestre MC, Guidet B. Current epidemiology of septic shock: the CUB-Rea network. Am J Respir Crit Care Med 2003;168:165eC172.
Levy MM, Fink MP, Marshall JC, Abraham E, Angus D, Cook D, Cohen J, Opal SM, Vincent JL, Ramsay G. 2001 SCCM/ESICM/ACCP/ATS/SIS International Sepsis Definitions Conference. Crit Care Med 2003;31:1250eC1256.
Marshall JC, Vincent JL, Fink MP, Cook DJ, Rubenfeld G, Foster D, Fisher CJ Jr, Faist E, Reinhart K. Measures, markers, and mediators: toward a staging system for clinical sepsis: a report of the Fifth Toronto Sepsis Roundtable, Toronto, Ontario, Canada, October 25eC26, 2000. Crit Care Med 2003;31:1560eC1567.
Sibbald WJ, Vincent JL. Round table conference on clinical trials for the treatment of sepsis. Crit Care Med 1995;23:394eC399.
Brun-Buisson C, Meshaka P, Pinton P, Vallet B. EPISEPSIS: a reappraisal of the epidemiology and outcome of severe sepsis in French intensive care units. Intensive Care Med 2004;30:580eC588.
Pittet D, Thieevent B, Wenzel RP, Li N, Gurman G, Suter PM. Importance of pre-existing comorbidities for prognosis of septicemia in critically ill patients. Intensive Care Med 1993;19:265eC272.
Steyerberg EW, Harrell FE Jr, Borsboom GJ, Eijkemans MJ, Vergouwe Y, Habbema JD. Internal validation of predictive models: efficiency of some procedures for logistic regression analysis. J Clin Epidemiol 2001;54:774eC781.
Schaaf BM, Boehmke F, Esnaashari H, Seitzer U, Kothe H, Maass M, Zabel P, Dalhoff K. Pneumococcal septic shock is associated with the interleukin-10eC1082 gene promoter polymorphism. Am J Respir Crit Care Med 2003;168:476eC480.