Lindholm, Maj-Lis PhD, RN*; Träff, Stefan MD†; Granath, Fredrik PhD‡; Greenwald, Scott D. PhD§; Ekbom, Anders MD, PhD‡; Lennmarken, Claes MD, PhD†; Sandin, Rolf H. MD, PhD*
Section Editor(s): Feldman, Jeffrey M.
From the *Department of Physiology and Pharmacology, Section for Anesthesia and Intensive Care, Karolinska Institutet, Stockholm, Sweden; †Department of Anesthesia and Intensive Care, University Hospital, Linköping, Sweden; ‡Department of Medicine, Clinical Epidemiology Unit, Karolinska Institutet, Stockholm, Sweden; and §Aspect Medical Systems, Norwood, Massachusetts.
Accepted for publication August 2, 2008.
Supported by FORSS (The Research Council in South-East Sweden) financed study personnel who collected raw data. BIS monitors were loaned and sensors provided by Aspect Medical Systems. Aspect Medical Systems also financed downloading of BIS trends by their local representative in Sweden, and contributed to financing the dedicated study personnel. Scott D. Greenwald, PhD, of Aspect Medical Systems, assisted with statistics and computing in collaboration with Fredrik Granath, PhD, Except Scott D. Greenwald, the other authors have not been employed or supported by any commercial sponsor.
Address correspondence and reprint requests to Maj-Lis Lindholm, PhD, RN, Department of Anesthesia and Intensive Care, Länssjukhuset, S-39185 Kalmar, Sweden. Address e-mail to email@example.com.
BACKGROUND: A correlation between deep anesthesia (defined as time with Bispectral Index (BIS) <45; TBIS <45) and death within 1 yr after surgery has previously been reported. In order to confirm or refute these findings, we evaluated TBIS <45 as an independent risk factor for death within 1 and 2 yr after surgery and also the impact of malignancy, the predominant cause of death in the previous report.
METHODS: Mortality within 2 yr after surgery, causes of death and the occurrence of malignant disease at the time of surgery were identified in a cohort of 4087 BIS-monitored patients. Statistically significant univariate predictors of mortality were identified. In order to allow for comparison with previous data, the following multivariate analysis was first done without, and thereafter with, preexisting malignancy status, the predominant cause of death.
RESULTS: One-hundred-seventy-four (4.3%) patients died within 1 yr and another 92 during the second year (totaling 6.5% in 2 yr). TBIS <45 was a significant predictor of 1- and 2-yr mortality when preexisting malignant disease was not among the co-variates (hazard ratio [HR] 1.13 [1.01–1.27] and 1.18 [1.08–1.29], respectively). Further exploration confined the significant relation between postoperative mortality and TBIS <45 to patients with preexisting malignant diagnoses associated with extensive surgery and less favorable prognosis. The most powerful predictors of 2-yr mortality in the model, including preexisting malignancy, were ASA physical score class IV (HR 19.3 [7.31–51.1]), age >80 yr (HR 2.93 [1.79–4.79]), and preexisting malignancy associated with less favorable prognosis (HR 9.30 [6.60–13.1]).When the initial multivariate regression was repeated using preexisting malignancy status among the co-variates in the model, the previously significant relation between 1, and 2-yr mortality and TBIS <45 did not reach statistical significance.
CONCLUSION: Using a similar set of co-variates as in previous work, we confirmed the statistical relation between 1-yr mortality and TBIS <45, and we extended this observation to 2-yr mortality. However, this relation is sensitive to the selection of co-variates in the statistical model, and a randomized study is required to demonstrate that there really is a causal impact from and TBIS <45 on postoperative mortality and, if it does, the effect is probably very weak in comparison with co-morbidity as assessed by ASA physical score, the preexisting malignancy status at surgery and age.
In a previous study in which 5.5% of the 1064 patients died within 1 yr after surgery, the most significant predictor of mortality was co-morbidity with a relative risk (hazard ratio; HR) of 16.1 However, two factors that are possible to modify during anesthesia, intraoperative hypotension and cumulative deep hypnotic level (i.e., the time in each case that the Bispectral Index [BIS] was <45 [TBIS <45]), were also identified as significant independent predictors of mortality. The relative risk for TBIS <45 was 1.244 per hour, but it was not determined if the relation between deep anesthesia and increased mortality was causal or coincidental. The most prevalent cause of death was malignancy (52%) and modulation of the immune system from deep anesthesia was among the proposed mechanisms. The aim of this investigation was to confirm or refute these unexpected findings and to assess the impact of preexisting malignancy, the dominant cause of death in the population studied by Monk et al.1
After IRB approval and oral informed consent, 5056 consecutive patients, prospectively enrolled in a study to evaluate the impact of BIS monitoring on the incidence of awareness, were studied.2 The patients underwent noncardiac surgery requiring endotracheal intubation and/or muscle relaxants and were given inhaled or IV general anesthesia at the discretion of the anesthesiologist. BIS monitoring (Aspect 2000™, BIS revision 3.4, Aspect Medical Systems, Norwood, MA) and traditional monitoring, including end-tidal anesthetic gas concentration, were used. The BIS sensor was placed on the forehead according to the manufacturer’s recommendation. Thirty second smoothing was used. The anesthesiologists were instructed to keep the BIS value between 40 and 60. BIS data trends, consisting of sequential 1-min BIS averages, were recorded automatically. Patients were not included if they were <16-yr-of-age, unable to communicate or impossible to monitor because of the surgical field. For 246 patients who had surgery more than once (up to 6; a total of 960 procedures) within the study period, only data from the last procedure were used. This reduced the remaining cohort by 704 cases, leaving 1 single anesthesia for 4352 unique patients to be assessed. Official data from The Swedish Board for Health and Welfare were used to identify patients who died within 2 yr after surgery.
The analysis was performed in three steps. As a first step, we compared the nonsurvivors to the survivors regarding gender, ASA physical status, the type of surgery, age, Body Mass Index (BMI), smoker status, case duration and intraoperative BIS values. Statistically significant univariate predictors of mortality were identified. To mimic the analysis in the previous study by Monk et al.,1 malignancy as a covariate was not entered at this stage. In the multivariate model evaluating the significant univariate variables, data were incomplete for 265 patients, therefore, this analysis was performed in 4087 subjects. The main reason for missing data was absence of information on preoperative BMI in patients requiring emergency surgery (n = 239), 9 patients were lost to follow-up due to emigration, and in another 17 cases various data had mistakenly not been recorded.
In a second step, we examined the relationship between TBIS <45 and mortality within 2 yr with regard to malignant disease. Data on malignancy were obtained from The National Registry of Cancer in Sweden (data in this registry are subjected to several control points for accuracy). Malignancies considered as requiring more extensive surgery (defined as an estimated duration of surgery >90 min) and, being associated with shorter life expectancy, were confined to group M2 (ICD-7 numbers 150–157, 162, 175, 180, and 181; defined in Table 1). The remaining malignant diagnoses were allocated to group M1. Group M0 consisted of patients with no preexisting malignant disease at the time of surgery or within 1 mo thereafter. Occurrence of malignant disease was confirmed by histopathological diagnosis. Finally, the initial analysis was repeated and this time preexisting malignancy status was included among the covariates.
The statistical analysis of the risk of dying within 2 yr after surgery was analyzed by means of Cox’s proportional hazard models. The model applied included the identified risk factors: age, gender, BMI, ASA, type of surgery, (categorized as presented in Table 2), and TBIS <45 as a continuous variable. In a further model, we also adjusted and stratified for cancer status before surgery. Results are presented as HR together with 95% confidence intervals. All analyses were performed using the SAS software (SAS Institute, Cary, NC).
Inhaled anesthesia was given in 95% of the cases, and the remaining 5% had IV anesthesia. Statistically significant univariate predictors of mortality, mortality rates, and types of surgery are given in Table 2. Significant risk factors associated with mortality within 2 yr are shown in Table 3. By far, the strongest predictor of death was preoperative morbidity as assessed by ASA physical status. TBIS <45 was among the six independent predictors of mortality when malignancy status was not among the covariates (Table 3). The HR for 1 yr mortality was 1.13 for every hour with BIS <45 and the corresponding HR for 2 yr was 1.18. Notably, case duration was not among the significant covariates.
Malignancy was the cause of death in 190 of the 266 patients (71%) who died within 2 yr after surgery (Table 4). One-hundred-ninety-nine patients had a diagnosis of malignant disease before the operation. Preexisting malignances at the time of surgery are found in Table 1. The average BIS value in the whole cohort of 4087 patients was 37. Adjusted for preexisting malignancy of all types, TBIS <45 was not significantly related to 2-yr mortality (P < 0.09). The categorization of malignancy (M0, M1, and M2) used in this study was a significant predictor of 2-yr mortality. Relative to group M0, the HR for 2 yr mortality in group M1 was 3.4 (CI 2.4–5.0; P < 0.0001) and 9.3 (CI 6.6–13.1; P < 0.0001) in group M2. In an analysis stratified according to preexisting malignancy (Table 5) it is shown that the significant association between TBIS <45 and mortality was confined to group M2 and that this association was more pronounced and reached significance for the second year of follow-up. Group M2 comprised 366 individuals (9.0%) accounting for 55% of the mortality cases (n = 146) in this study.
Using Cox’s Proportional Hazards Modeling, we identified factors that predicted 1-yr and 2-yr postoperative mortality. We found ASA physical status to be the most powerful predictor of mortality within 2 yr postoperatively. This is similar to the previous study by Monk et al.1 in which co-morbidity as assessed by the Charlson index3 was the dominant predictor of 1-yr mortality. Among the significant factors related to mortality were the duration of deep anesthesia (i.e., time with BIS <45). We found that patients who died within 1 yr after surgery spent 24% longer time with BIS <45 compared to survivors, corresponding to a HR of 1.13 for every hour with BIS <45. For 2-yr mortality, we found a corresponding HR of 1.18. In these analyses, preexisting malignancy at the time of surgery was not specifically accounted for in order to allow for comparison with previous data.1 When the preexisting malignancy status was among the covariates, a statistical relation between TBIS <45 and mortality was confined to 2-yr mortality in patients with malignant disease associated with poorer life expectancy and major surgery. This group of patients constituted 9.0% of the total study population, had a disseminated disease at the time of surgery in 74% of cases and accounted for 55% of the 2-yr mortality. No significant relationship between TBIS <45 and 1, or 2-yr postoperative mortality was found in patients without malignant disease or with malignancies considered associated with less poor prognosis.
When preexisting malignancy status was not specifically among the covariates, we found a similar statistical relation between TBIS <45 and 1-yr postoperative mortality as previously reported by Monk et al. (HR = 1.244/h).1 Monk et al. suggested that their unexpected finding of a relation between TBIS <45 and 1 yr mortality could have been due to effects on the immune system caused by anesthetic management, i.e., deep anesthetic level.1 Indeed, anesthetic, opioids, surgery, blood transfusions, temperature changes, pain and psychological stress have all been implicated as contributing to immunosuppression in the perioperative period.4 In accordance with the study by Monk et al., we found malignancy (75%; 52% in1) and cardiovascular disease (17%; 17% in1) to be the predominant causes of mortality among patients who died during the first year after surgery. We also found that 75% of our patients who died within 2 yr had a preexisting malignant disease at the time of surgery. We hypothesized that, even if malignancy is among the 19 medical conditions comprising the Charlson co-morbidity index which was used by Monk et al. to assess concomitant disease, this index, as well as ASA physical score used in our initial analysis, may not fully control for survival among cancer patients after surgery.3 Our following analysis confirmed that preexisting malignant disease was a significant predictor of 2-yr mortality, especially for malignancies associated with more extensive surgery and reduced life expectancy (HR 9.30; CI 6.60–13.1) and also for patients with malignancies considered less ominous (HR 3.42; CI 2.35–4.96) when compared to patients without malignant disease. The importance of considering preexisting malignancy status is further demonstrated by the fact that, when the initial Cox regression was repeated with preexisting malignancy status among the covariates, the relation between 2-yr mortality and TBIS <45 was no longer significant.
Thus, not unexpectedly, our data indicate that the preexisting malignancy status is of utmost importance for postoperative survival and that the relation between TBIS <45 and mortality need not be causal. Still, it could be argued that our data indicate that spending a long amount of time at low BIS values may impair survival in some cancer patients. However, it should be noted that the patients in this study received relatively deep anesthesia. The average BIS during maintenance in the whole cohort was 37, which is well below the threshold of 45 used in the risk-estimating calculations. Even if case duration was not an independent risk factor, the significant relation between TBIS <45 and 2-yr mortality we found in a small fraction of patients may reflect extensive, lengthy surgery in patients with limited life expectancy due to preexisting malignant disease.
Two-hundred-forty-six patients had surgery more than once during the study period. These patients were given anesthesia on 960 occasions. Forty-two of the patients who died had surgery more than once. It is not known if a potentially causal relation between deep anesthesia and postoperative mortality would be related to cumulative time with deep anesthesia. In this study, we chose to use only the last anesthetic within the study period. In addition, we cannot exclude the possibility of multiple surgeries unknown to us between the time when inclusion of patients ended and completion of the 2-yr follow-up.
Patients with preexisting malignant disease requiring more extensive surgery and being associated with less favorable life expectancy were allocated to a separate category in our analysis. A validated prognostic tool, the tumor, node and metastatic system (TNM),5 could not be used since the major part of the studied cohort was included before TNM data became available in Sweden in 2002. In the absence of a validated estimate of life expectancy due to malignant disease, we used an arbitrary definition. Even if another definition may have altered the result, we think that the types of malignant disease allocated to groups M1 and M2, respectively, are generally regarded to have different life expectancies. This view is supported by the fact that malignancy status according to our definition was a powerful independent predictor of 2-yr mortality. Finally, since the patients, on average, received rather deep anesthesia, potential positive effects on survival from light anesthesia may have been obscured.
To conclude, by using a similar set of covariates as in previous work, we confirm the statistical relation between 1-yr mortality and TBIS <45 and extend this observation to 2-yr mortality. However, this relation is sensitive to the selection of covariates in the statistical model. A randomized study is required to demonstrate that a causal impact from TBIS <45 on postoperative mortality really exists. If it does, the effect is probably very weak in comparison with co-morbidity as assessed by ASA physical score, the preexisting malignancy status at surgery, and age.