Data were collected from vital sign monitors, ventilators and infusion pumps, and automatically recorded by a clinical information system (Copra System GmbH, Sasbachwalden, Germany) introduced to the ICU in 1998. The clinical information system provides staff with complete electronic documentation, order entry (eg, medications) and direct access to laboratory results.

The simplified acute physiology score (SAPS) II 18, therapeutic intervention score-28 (TISS-28) 19 and sequential organ failure assessment (SOFA) scores 20 were calculated daily by the attending physician in charge of the patient. SOFAmax was defined as the maximum SOFA score recorded during the ICU stay. Data recorded prospectively on admission also included age, gender, referring facility, primary and secondary admission diagnoses, and surgical procedures before admission. Sepsis syndromes were defined according to consensus conference definitions 21 and were recorded daily by the attending physician in a special section of the clinical information system. Admission diagnosis was categorised retrospectively on the basis of prospectively recorded codes from the International Classification of Diseases-10 and electronic patient charts. Comorbidities were defined according to the definitions provided in the original SAPS II paper 18. For the purpose of this analysis, the following comorbidities were grouped together to reduce the number of covariates in the final multivariate model: metastatic and non-metastatic cancer; type 1 and type 2 diabetes; and chronic renal failure with or without haemodialysis.

Readmission was defined as admission to the ICU of a patient who had previously been admitted to the ICU during the same hospitalisation period. All admission and discharge dates were available from the clinical information system. Planned admission was defined as an admission after elective surgery, which was scheduled 24 hours before the surgical procedure.

The ICU at the Friedrich Schiller University hospital is a closed surgical ICU operated by the Department of Anesthesiology and Intensive Care Medicine. A consultant intensivist with a special qualification in intensive care medicine is available in-house 24 hours a day. Attending physicians and in-training residents are available throughout the day (on 12-hour shifts). There is no reduction in personnel or in ICU activities during night shifts or at weekends. Rounds are conducted daily by ICU physicians, nursing staff and the operating surgical team. ICU admission and discharge decisions are made by the consultant intensivist on-duty. Due to the absence of step-down or high-dependency units in the institution, patients are discharged from the ICU only when they are haemodynamically stable with an acceptable general condition and adequate organ function.

Data were analysed using SPSS 13.0 for windows (SPSS Inc, Chicago, IL). The Kolmogorov-Smirnov test was used to verify the normality of distribution of continuous variables. Non-parametric tests of comparison were used for variables evaluated as not being normally distributed. Difference testing between groups was performed using a Wilcoxon test, Mann-Whitney U test, chi-squared test and Fisher's exact test as appropriate. A Bonferroni correction was used for multiple comparisons. A Friedmann test was used to compare the evolution of SOFA scores over time.

We performed a multivariate logistic regression analysis, with readmission to the ICU as the dependent factor, of the overall population. Variables included in the logistic regression analysis were age, gender, comorbid diseases, the source of admission, SAPS II and SOFA scores on admission, SOFAmax, the type of surgery undergone, the presence of sepsis syndromes and parameters of organ function on the day of discharge from the ICU. Colinearity between variables was excluded before modelling. Another multivariate logistic regression analysis was performed to identify risk factors for in-hospital mortality in patients who were readmitted to the ICU. To avoid 'over fitting' of the second model due to the low in-hospital mortality event rate, variables were introduced to this model if significantly associated with a higher risk of in-hospital death on a univariate basis at a p < 0.2.

Continuous data are presented as mean ± standard deviation (sd) and categorical data as number and percentage, unless otherwise indicated. All statistics were two-tailed and a p < 0.05 was considered statistically significant.

Of 3169 patients admitted to the ICU during the study period, 173 (5.5%) died in the ICU and 144 (4.5%) were discharged to other hospitals: 2852 patients were discharged to the hospital floor and those patients made up the study group (1828 male (64.1%), mean patient age 62 years). The readmission rate was 13.4% (n = 381): 314 (82.4%) were readmitted once, 39 (10.2%) were readmitted twice and 28 (7.3%) were readmitted more than twice, giving a total of 476 readmission episodes. The first readmission to the ICU occurred within a median of seven days (range = 5 to 14 days) (Figure 1). The characteristics of the study group are presented in Table 1.

Patients who were readmitted to the ICU were older, had a higher incidence of chronic renal failure and sepsis syndromes, were more likely to be unplanned admissions and had higher SAPS II and SOFA scores on initial admission to the ICU compared with patients who were not readmitted. Patients who were readmitted to the ICU underwent more surgical procedures within 24 hours of the initial admission compared with patients who were not readmitted; however, the incidence of major surgical procedures was similar between the two groups. During the weekends, 917 patients (32.2%) were discharged to the hospital ward and 704 patients (24.7%) were discharged to the hospital ward during the night (8 pm to 8 am). There were no differences in the frequencies of weekend (24.4% versus 26.5%; p = 0.375) or nocturnal discharges (32.6% versus 29.1%; p = 0.175) between patients who were not readmitted and those who were readmitted to the ICU.

Of the 476 readmission episodes, 223 (46.8%) were planned and 253 (53.2%) were unplanned postoperative admissions (Table 2). Cardiovascular and respiratory complications were the most common reasons for unplanned readmissions (14.3% and 13%, respectively). On the day of readmission, cardiac surgery, gastrointestinal surgery and neurosurgery were performed in 18.1%, 18.1% and 12.1% of patients, respectively. Unplanned admissions contributed to 30.2% of the initial admissions to the ICU and to about 60% of the second or third readmissions (Table 2).

Gastrointestinal surgery was the most common type of surgery performed within 24 hours of ICU admission in patients who were readmitted to the ICU more than once. Cardiovascular complications necessitating readmission were more frequent during the first readmission, whereas respiratory and gastrointestinal complications were more frequent thereafter. SAPS II scores were higher and TISS-28 scores were lower after second and third readmissions compared with the initial admission.

On initial admission to the ICU, serum bilirubin concentrations, C-reactive protein (CRP) concentrations and platelet counts were similar in all patients, and creatinine concentrations, arterial lactate and leucocyte count were higher in patients who were readmitted to the ICU compared with those who were not (Table 3). The maximum concentrations of serum bilirubin, serum creatinine, leucocyte count, arterial lactate and CRP were higher in patients who were readmitted to the ICU compared with those who were not. Serum creatinine and CRP concentrations within 24 hours of initial discharge from the ICU were higher in patients who were readmitted to the ICU compared with those who were not.

The overall incidence of sepsis syndromes was 9.1% (n = 260). Sepsis syndromes occurred more frequently during the initial admission (14.2% versus 8.3%; p = 0.001) in patients who were readmitted to the ICU. The incidence of sepsis syndromes and mechanical ventilation and the duration of mechanical ventilation were similar during initial and subsequent readmissions. In patients who were readmitted to the ICU, SOFA scores at admission were higher on initial admission to the ICU than on the first readmission; however, the SOFA scores increased steadily over the first few days of the first readmission and remained high during the first two weeks of readmission (Figure 2).

In-hospital mortality was significantly higher in patients readmitted to the ICU (17.1% versus 2.9%; p < 0.001) compared with those that were not. Patients who were readmitted to the ICU more than one week after the initial discharge from the ICU (late readmissions; n = 176) had higher in-hospital mortality rates (22.2% versus 12.7%; p < 0.001) compared with those who were readmitted within 48 hours of initial discharge (early readmission, n = 57). Readmission more than two-times to the ICU was associated with higher ICU mortality (21.4% versus 7.6%; p = 0.004) and in-hospital mortality rates (46.4% versus 17.1%; p < 0.001), and longer ICU length of stay (median = three days (range = one to eight days) versus two day(one to four days); p = 0.02) compared with the first readmission. Hospital mortality was similar for planned and unplanned readmissions (17.6% versus 15.7%; p = 0.667).

Factors associated univariately with a higher risk of ICU readmission included older age, higher SAPS II and SOFA scores on admission, admission from another hospital, unplanned admission, duration of mechanical ventilation, and higher creatinine and CRP concentrations on the day of discharge to the hospital floor (Table 4). In a multivariate analysis, age (odds ratio (OR) = 1.13 per 10 years; 95% confidence intervals (CI) = 1.03 to 1.24; p = 0.025), greater SOFAmax score (OR = 1.04 per point; 95% CI = 1.01 to 1.08; p = 0.04) and higher CRP concentration on the day of discharge to the hospital floor (OR = 1.02; 95% CI = 1.01 to 1.04; p = 0.035) were independently associated with a higher risk of readmission to the ICU.

In patients who were readmitted to the ICU, the presence of cancer, chronic renal failure, gastrointestinal surgery before initial admission and greater SAPS II score were associated univariately with a higher risk of in-hospital mortality (Table 5). In a multivariate analysis with hospital mortality as the dependent variable, SAPS II (OR = 1.02 per point; 95% CI = 1.01 to 1.04; p = 0.045), chronic renal failure (OR = 2.39; 95% CI = 1.01 to 5.2; p = 0.028) and admission after gastrointestinal surgery (OR = 2.6; 95% CI = 1.17 to 5.8; p = 0.02) were independently associated with a higher risk of in-hospital death in these patients.