Mortality
The ICU mortality (III), hospital mortality (I–IV), and one-year mortality (III) data were recorded from hospital records and Statistics Finland.
Health-related quality of life
Quality of life was assessed using the generic 15D measure (I) (Sintonen 2001). The 15D questionnaire consists of 15 dimensions: breathing, mental functions, speech (communication), vision, mobility, usual activities, vitality, hearing, eating, elimination, sleeping, distress, discomfort and symptoms, sexual activity, and depression. Respondents were asked to choose the alternative best describing one’s present health status from five alternatives. The responses are divided into 5 levels (1=best, 5=worst). A weighted 15D score is generated on the scale from 1 (no problems on any dimension) to 0 (being dead). The 15D questionnaire is presented in Appendix 1. The 15D questionnaire of HRQoL was mailed for self-administration to the patients admitted to the ICU after December 28, 2002 and known to be alive at 6 months after medical ICU admission. The results from the study population were compared with those from the age- and gender-matched general Finnish population obtained from Health 2000 health examination survey (Aromaa and Koskinen 2004).
Disease severity scores
The SOFA (Vincent et al. 1998) score evaluates status of the following organ systems separately: central nervous system, coagulation, cardiovascular, respiration, hepatic, and renal. The SOFA scores were calculated daily (III), in the first 24 h at ICU admission (I), or daily during the first four days in the ICU (II and IV) to assess the degree of organ dysfunction. The APACHE II score (Knaus et al. 1985) and SAPS II (Le Gall et al. 1993) were calculated in the first 24 h in the ICU to evaluate the severity of illness (I–IV).
63 4.7 Statistical analyses
Kolmogorov-Smirnov test was used to test if a variable is normally distributed. Because most of the main quantitative data was not normally distributed, nonparametric statistics was used.
Nonparametric Mann-Whitney U-test was used for comparisons of continuous variables between two independent samples, e.g. between survivors and nonsurvivors (I–IV).
Chi-square and Fisher’s Exact tests were used for comparisons of categorical variables (I–
IV). The Fisher’s Exact test was used if the number of cases in a sample was lower than five.
Bivariate correlations for continuous, not normally distributed variables were determined with nonparametric Spearman’s correlation, such as those between cell-free plasma DNA or plasma HO-1 concentration and disease severity scores or ED length of stay and the 15D score of HRQoL (I–IV).
Variables significantly associated with plasma DNA or HO-1 concentration were tested by linear regression analysis to determine the independent factors that would best predict the value of the dependent variable (plasma DNA or HO-1). Because plasma DNA and HO-1 values were not normally distributed, log-transformed plasma DNA and HO-1 was used (III–
IV).
Kruskal-Wallis test, a nonparametric test comparing the difference in the median values of two or more samples, was used in Studies III and IV to evaluate whether a difference was present in the plasma DNA or HO-1 values between quartiles of SOFA scores. In Study III, Bonferroni correction was also used because of multiple comparisons.
Kaplan-Meier analysis was applied to evaluate the one-year mortality differences according to the best cut-off value of baseline plasma DNA concentration in Study III.
Multivariate forward logistic regression analysis was used to test the independent effect of the variables on outcome (II–IV).
The discriminative power of plasma DNA and plasma HO-1 regarding mortality was evaluated with ROC analysis, and AUC was calculated with a 95% confidence interval (III–
IV). To evaluate the correct classification rate, sensitivity, specificity, and likelihood ratios, the best predictive cut-off values maximizing the sum of sensitivity and specificity were defined using the GraphROC for Windows (Kairisto and Poola 1995) (III and IV), presented in Table 16.
Table 16. Defining the correct classification rate, sensitivity, specificity, and positive and negative predictive values regarding death.
Condition
Dead Alive
Test result Positive A C A+C
Negative B D B+D
A+B C+D
A=True positive C=False positive B=False negative D=True negative Sensitivity= true positives/(true positives+false negatives)=A/(A+B)
Specificity=true negatives/(true negatives+false positives)=D/(C+D)
Positive prediction value=true positives/(true positives + false positives)=A/(A+C) Negative prediction value=true negatives/(true negatives + false negatives)=D/(D+B) Correct classification rate=(true positives + true negatives)/all=(A+D)/(A+B+C+D)
The standardized mortality ratio (SMR) was calculated as the ratio of actual mortality rate to expected mortality rate (I). The probability of hospital mortality rate was calculated using the SAPS II score (Le Gall et al. 1993).
The analyses were performed using the SPSS 12.0. statistical software (SPSS, Chicago, IL).
The GraphROC for Windows (Kairisto and Poola 1995) was used in ROC analysis. The demographic data are expressed as median and interquartile range (IQR). The 2-tailed level of p<.05 was considered statistically significant in all tests.
In Study IV, Hardy-Weinberg equilibrium was calculated for the HO-1 polymorphisms with the Pearson’s correlation and Fisher’s exact tests. Pairwise LD measures were determined using Haploview 4.0 software. Haplotypes were constructed using the solid spine of LD algorithm and Haploview software (Barrett et al. 2005). In addition to the marker-by-marker analysis, we evaluated the associations also by haplotype-based analysis, which is considered more powerful in detecting associations (Zhang et al. 2002). The Genetic Power Calculator was used for power calculations (Purcell et al. 2003). With a relative risk of 2.5 for the heterozygotes and 3 for minor allele homozygotes regarding hospital mortality, the Study IV sample had 83% power to detect association with the +99G/C polymorphism.
65 5 ETHICAL ASPECTS
The Ethics Committees in each hospital approved the study protocols. A written informed consent was obtained from all the patients who were approached with the HRQoL survey (I) and from the patient or near-relative before enrolling and obtaining the blood tests in Studies II–IV.
The project was partly supported by EVO grants from Helsinki University Hospital, a grant from the Instrumentarium Foundation, and a grant from Duodecim.
The authors of the studies have no potential conflicts of interest to disclose.
6 RESULTS
6.1 Association of emergency department length of stay with mortality and