Quality of life before intensive care unit admission is a predictor of survival

The Short-form 36 (SF-36, version 1; ^© 1993 Medical Outcome Trust), a generic, widely used standardized health status questionnaire, was used to measure HRQOL. This measurement contains eight multi-item dimensions: physical functioning, role limitation due to physical problems, bodily pain, general health, vitality, social functioning, role limitation due to emotional problems, and mental health. Answers to the 36 items were transformed, weighed and subsequently scored according to predefined guidelines [6]. Higher scores represent better functioning, with a range from 0 to 100. Furthermore, scores were aggregated to summary measures representing a physical component score (PCS; mainly reflecting physical functioning) and a mental component score (MCS; mainly reflecting social functioning and mental health) [7]. Population scores on PCS and MCS have been standardized on 50 as population mean (SD 10 representing 1) [7]. For the PCS, very high scores indicate no physical limitations, disabilities, or decrements in well being, as well as high energy levels. Very low scores indicate substantial limitations in self-care and in physical, social and role activities, severe bodily pain, or frequent tiredness [7]. For the MCS, very high scores indicate frequent positive effect, absence of psychological distress, and limitations in usual social/role activities caused by emotional problems. Very low scores indicate frequent psychological distress, and substantial social and role disability due to emotional problems [7].

Translation, validation and generating normative data of the Dutch language version of the SF-36 health questionnaire were evaluated in 1998 in community and chronic disease populations [8]. Because most of the patients in our study were unable to complete a questionnaire at the time of admission, proxies had to be used as a surrogate approach. In proxies and patients the same method was used to complete the SF-36. The use of proxies to assess the patients' HRQOL using the SF-36 in the ICU setting was validated in earlier studies conducted by our group [9] and others [10, 11]. HRQOL was measured within 48 hours of ICU admission (estimation of HRQOL up to 4 weeks before admission). All interviews were performed by the same investigator (JH). The average time required to complete the questionnaire was 15 to 20 min. Consideration of multiple items has the advantage of allowing construction of a comprehensive profile of HRQOL, but it may burden the critically ill patient. We used the first question of the SF-36 as a primary approach to estimation of the patient's HRQOL. This is the single-item question pertaining general health status; 'In general, would you say your health is excellent, very good, good, fair, or poor?' [12, 13]. The advantages of such a single-item question are its simplicity and ease of application.

A Pearson's χ² test was used to assess demographic differences between ICU survivors and ICU non survivors. The differences between scores for the single-item question were tested using the χ² test for trend. We examined the relationship between the single-item question on HRQOL before ICU admission and mortality at 6 months after ICU discharge with multivariate logistic regression using the variables known on the first day of ICU admission (APACHE II score), adjusted for age and sex.

To analyze the potential of variables to predict mortality in patient subgroups, we used five statistical models. HRQOL was entered as the response to the single-item question, or as MCS and PCS. In the model A we included the general health item of the SF-36, age and sex. In model B we included both the PCS and MCS from the SF-36, and age and sex. In model C we included APACHE II score, age and sex. In model D we included the general health item of the SF-36, APACHE II score, age and sex. In model E we included both the PCS and MCS from the SF-36, APACHE-II score, age and sex.

To estimate the ability to discriminate between survivors and non-survivors, odds ratios were calculated, receiver operating characteristic analysis was performed and the area under the curve (AUC) was calculated. Classification tables were used to assess the sensitivity for observed deaths being labeled by the models as predicted deaths, specificity for a predicted death being an observed death, and positive and negative predictive values and likelihood ratio. Data were analyzed using SPSS (version 11.5; SPSS Inc., Chicago, IL, USA). All data are expressed as median (interquartile range), unless indicated otherwise.

Using the single-item question on HRQOL as a potential predictor of survival, the AUC for model A (0.719) was comparable to that for the APACHE II score (model C; 0.721) and slightly better than that in model D (AUC = 0.760), in which both factors were combined (Table 2 and Figure 2). Comparable results were obtained when calculating odds ratios (Table 3) and with analysis using MCS and PCS in models B and E. The sensitivity and specificity in model D (sensitivity 0.52 and specificity 0.81) were similar to those in model A (0.45 and 0.80). Similar results were found when using PCS and MCS. In ICU patients (n = 451), sensitivity improved from 0.44 (model C; APACHE II score only) to 0.56 (model E; APACHE II score, and PCS and MCS), respectively. Results for specificity were similar, improving from 0.84 (model C; APACHE II score only) to 0.82 (model E; APACHE II score, and PCS and MCS). Similar results were also found when using the general health item (models A and D; Table 2). The negative and positive predictive values and likelihood ratios are shown in Table 2.