Causes of death and determinants of outcome in critically ill patients

On admission to the ICU, pre-ICU data were documented in a standardised study protocol by the intensivist in charge. Pre-ICU data included the following: demographic variables (age and gender), admission diagnosis, type of admission (emergency or elective), referring unit (emergency department, operating theatre, recovery room, ward, or other ICU), type of disease (surgical or non-surgical), anatomical region of surgical intervention (cardiac, abdominal, traumatological, thoraco-abdominal, extremity, thoracic, neuro-, or spinal surgery), preoperative American Society of Anesthesiologic classification [12], specific data on cardiac surgery patients (preoperative ejection fraction, time on cardiopulmonary bypass, aortic cross-clamp time, and reperfusion time), history of pre-existent chronic diseases (chronic obstructive pulmonary disease, coronary heart disease, myocardial infarction within the preceding six months, myocardial infarction longer than six months before ICU admission, stable angina pectoris, unstable angina pectoris, congestive heart failure, treated chronic arterial hypertension, untreated chronic arterial hypertension, chronic renal insufficiency, chronic renal insufficiency requiring haemodialysis, liver cirrhosis, Child-Pugh classification of liver cirrhosis [13], insulin-dependent diabetes mellitus, non-insulin-dependent diabetes mellitus, healed tumour disease, malignant tumour disease, gastroduodenal ulcer disease, cerebrovascular insufficiency, status post-transient ischemic attack, prolonged ischemic neurological deficit or apoplectic insult, tetra- or paraplegia, other neurological pathology, psychiatric disease, immunosuppression, chemotherapy or radiation therapy during the preceding six months, chronic therapy with corticosteroids, and obesity), and the number of pre-existent chronic diseases. Presence or absence of pre-existent chronic diseases was documented in a binary fashion (0 = absent, 1 = present).

Any new complication or additional diagnosis that arose during the ICU stay was documented on a daily basis by one of three senior intensivists. Data included need for re-operation, massive transfusion, continuous veno-venous haemofiltration, or extracorporeal membrane oxygenation, as well as new-onset arrhythmias, SIRS (systemic inflammatory response syndrome), infection, sepsis, septic shock, acute respiratory distress syndrome, partial or global respiratory insufficiency, acute delirium, or critical illness polyneuropathy.

After discharge of the patient, data documentation was completed by one of the senior intensivists. Data documented at patient discharge included the Therapeutic Intervention Severity Score [14] and Simplified Acute Physiology Score (SAPS) II [15], which were both calculated from the worst physiological and laboratory parameters during the first 24 hours after ICU admission; highest multiple organ dysfunction syndrome (MODS) score (Appendix) during the ICU stay; worst PaO₂/FiO₂ ratio; creatinine, aspartate, alanine aminotransferase, and bilirubin serum concentrations during the ICU stay; duration of ICU stay in days; patient mortality; and type of unit the patient was transferred to (ward, cardiac surgical intermediate care unit, surgical intermediate care unit, other ICU, or other hospital). For all patients who died in the ICU, the cause of death was documented.

In January 2005, the demographic data of the study population were transferred to the Institute of Management and Quality Control of the university hospital, which documented the following data from all study patients: number of admissions to the ICU, hospital mortality, institution the patient was discharged to from hospital (home, other hospital, or rehabilitation unit), and causes of in-hospital death of critically ill patients after discharge from the ICU. At the same time, mortality data (death rate and cause of death according to the International Statistical Classification of Diseases and Related Health Problems [ICD-10] code [16]) were delivered by the 'Austrian Statistical Institution' as well as the 'Tumour Register' of South Tyrol. Using these data, mortality within one year after ICU admission and days of survival after ICU discharge were calculated for each study patient.

Before entry into the computer database, each case report was reviewed by a member of the study committee (senior intensivist or coworker). At the end of the electronic documentation of all study patients, plausibility tests were performed for each study variable to detect and correct typing mistakes that occurred during data entry or processing.

All codes and definitions of study variables were established before the beginning of the study and were uniformly documented as standard operating procedures for study data documentation. Study-related definitions are summarised in Table 1[16‐21]. Cause of death was defined as the primary pathology (irrespective of its duration) leading to death of the patient or to the decision to withhold or withdraw intensive care therapy. Thus, cause of death did not necessarily have to be identical to admission diagnosis. To reduce inter-investigator variability to a minimum, all study-relevant decisions on cause of death, occurrence of new complications in the ICU, as well as any decision to withhold or withdraw intensive care treatment were made exclusively by one of three senior intensivists heading the ICU and in charge of the study.

The primary study endpoint was to evaluate the causes of death of critically ill patients in the ICU, in the hospital after discharge from the ICU, and within one year after admission to the ICU. The secondary study endpoint was to define risk factors for death in the ICU, in the hospital after discharge from the ICU, and within one year after admission to the ICU.

Descriptive statistical methods were used to analyse demographic and clinical data of the study population, as well as causes of death. Three separate logistic regression analyses applying forward conditioning variables only were used to examine the association between study variables and ICU mortality (first analysis, denominator: death in the ICU), in-hospital mortality (second analysis, denominator: death in the hospital after discharge from the ICU), and mortality within one year after admission to the ICU (third analysis, denominator: death after hospital discharge and within one year after ICU admission). Variable selection for the three models was separately based on univariate comparisons. In each analysis, variables that were statistically significant at α = 0.05 in univariate comparisons were introduced into a multivariate model; covariates significant at <0.05 were retained in the model. To evaluate associations between single-organ functions and outcome variables, MODS score was not directly entered into the statistical model but was divided into its seven components (lungs, kidney, cardiovascular system, liver, coagulation, gastrointestinal tract, and central nervous system). According to the score (Appendix), these components were again subdivided into unaffected organ function (0 points), organ dysfunction (1 point), and organ failure (2 points). In both models, the MODS score was tested as contrasts of failure versus unaffected organ function, as well as organ dysfunction versus unaffected organ function. Tests for differences between study subgroups were performed using the unpaired Student t, χ² , or Mann-Whitney U-rank sum tests, as appropriate. Kaplan-Meier curves together with the log-rank sum test were used to illustrate cumulative survival rates for patients with and without central nervous system failure or cardiovascular failure in the ICU. A standard statistical program was used for all analyses of this study (SPSS 12.0 for Windows; SPSS Inc., Chicago, IL, USA). Data are given as mean values ± standard deviation unless stated otherwise.

During the observation period, a total of 4,055 critically ill patients were admitted to the ICU, of whom 3,700 were included in the statistical analysis (Figure 1). Tables 2 to 4 give characteristics of the study population.

ICU mortality was 9.5% (353/3,700) for the study population for which full data sets were available one year after ICU admission and 8.7% (353/4,055) for all patients treated in the ICU during the given period. ICU mortality in patients admitted to the ICU because of infection, sepsis, or septic shock was 10.1%, 17.5%, and 53.3%, respectively.

ICU survivors had a significantly shorter ICU stay than did non-survivors (7.6 ± 9.5 versus 11.7 ± 11.5 days, p < 0.001). No study patient died within the first day of ICU therapy. Twelve percent of non-survivors died within the first 3 days in the ICU, and 52.7% died within the first week after ICU admission. In end-of-life-decisions, treatment was withdrawn in 54.7% (193/353) of patients who died in the ICU. Table 5 summarises the causes of death of critically ill patients in the ICU. Acute, refractory MODS was the most frequent cause of death. Figure 2 presents the relationship between the number of failing organs and mortality at ICU discharge, hospital discharge, and one year after ICU admission. When the MODS score reached 14 points (failure of all seven evaluated organ systems) (n = 6), ICU mortality was 100%.

Independent risk factors for death in the ICU are shown in Table 6. Central nervous system failure and cardiovascular failure were the two most important risk factors for death in the ICU. Figure 3 displays Kaplan-Meier curves with the log-rank sum test for ICU patients with and without central nervous system failure (a) or cardiovascular failure (b). Patients with central nervous system or cardiovascular failure had a significantly higher ICU mortality rate than did patients without central nervous system failure (7.7% versus 54.2%, p < 0.001) or cardiovascular failure (1.4% versus 40.5%, p < 0.001). When compared with patients who were admitted from the operating theatre, emergency department, normal ward, or other ICUs, patients who were admitted to the ICU from the recovery room were older (61.9 ± 19 versus 59 ± 19 years, p = .002), had more pre-existent diseases (2.8 ± 1.9 versus 2.4 ± 1.6, p < 0.001), a higher American Society of Anesthesiologists classification (3.4 ± 0.9 versus 3.2 ± 0.9, p < 0.001), and a higher SAPS II (39.7 ± 17.9 versus 35.6 ± 14.8, p < 0.001).

Table 5 summarises the most frequent causes of death of critically ill patients who died in the hospital after discharge from the ICU. Malignant tumour disease was the most frequent cause of in-hospital death of critically ill patients after discharge from the ICU. Table 6 presents independent risk factors for death of critically ill patients in the hospital.

After discharge from the hospital, mortality within one year after admission to the ICU was 8.9% (286/3,203). Overall mortality of critically ill patients within one year after admission to the ICU was 21.2% (783/3,700). One-year mortality of patients admitted to the ICU because of infection, sepsis, or septic shock was 33.6%, 42.3%, and 66.9%, respectively.

In patients who died within one year after admission to the ICU, the mean time after discharge from the hospital to death was 133 ± 108 days. Table 5 displays the most frequent causes of death of critically ill patients who died within one year after ICU admission. Tumour disease was the most frequent cause of death. Fifty-five percent of the non-survivors died from the same condition that prompted admission to the ICU.

Table 6 shows independent risk factors for death of critically ill patients within one year after admission to the ICU. The number of ICU admissions was the most important risk factor for death. Patients who were treated more than once in the ICU were significantly more likely to die within the first year after ICU admission than patients who required only one admission to the ICU (30.7% versus 21%, p = 0.026).