Volatile Versus Total Intravenous Anesthesia for Cancer Prognosis in Patients Having Digestive Cancer Surgery

A Nationwide Retrospective Cohort Study

Kanako Makito, M.D., M.P.H.; Hiroki Matsui, M.P.H.; Kiyohide Fushimi, M.D., Ph.D.; Hideo Yasunaga, M.D., Ph.D.


Anesthesiology. 2020;133(4):764-773. 

In This Article

Materials and Methods

Data Source

Patient data were extracted from the Japanese Diagnosis Procedure Combination database, the details of which have been previously described.[11] Briefly, the database includes administrative claims data and the following detailed patient data: age; sex; body mass index (BMI); diagnoses and comorbidities at admission; complications after admission recorded with text data in the Japanese language and encoded by International Classification of Diseases and Related Health Problems, 10th Revision (ICD-10) codes; medical procedures encoded by Japanese original codes; tumor node metastasis classification of malignant tumors and cancer stage; medications; activities of daily living at admission (converted to Barthel Index); and discharge status. According to a previous validation study of the database, the recorded diagnoses of several common diseases (including malignant tumors, cardiac diseases, stroke, and renal diseases) have moderate sensitivity and high specificity, whereas the recorded procedures and drugs have high sensitivity and specificity.[11] The database includes administrative data on 7 million inpatients per year, accounting for approximately 50% of all acute care inpatients in Japan. More than 1,000 hospitals participate in the database voluntarily, and approximately 300 hospitals also provide outpatient data.

The requirement for informed consent was waived because the study was based on a secondary analysis of anonymous administrative data. This study was approved by the Institutional Review Board at The University of Tokyo (Institutional Review Board number 3501).

Patient Selection

From the database, we obtained the records of patients who had elective esophagectomy, gastrectomy, hepatectomy, cholecystectomy, pancreatectomy, colectomy, or rectal cancer surgery from July 1, 2010 to March 31, 2018 at 218 hospitals that provided outpatient data. The inclusion criteria were an age of greater than or equal to 18 yr at the time of the first surgery with volatile anesthesia or total intravenous anesthesia. We excluded patients who had anesthesia multiple times during the study period, those who were diagnosed with a benign tumor or a malignant potential tumor, those who had spinal anesthesia, and those in whom nitrous oxide was used without volatile anesthesia. A malignant potential tumor is a tumor that is reported to be associated with a risk of malignancy, including intraductal papillary mucinous neoplasm, mucinous cystic neoplasm, and Crohn's disease.


The exposure variable was having volatile anesthesia or total intravenous anesthesia. The patients were divided into two groups: (1) those who had volatile anesthesia using desflurane, sevoflurane, or isoflurane with/without nitrous oxide and (2) those who had propofol-based total intravenous anesthesia.

Confounding Variables and Outcomes

We extracted information on baseline characteristics including age at the time of the first elective surgery, sex, BMI, length of stay, smoking status (current/past smoker or nonsmoker), admission date of cancer recurrence, date of death, comorbidities at admission, complications after admission, type of surgery (esophagectomy, gastrectomy, hepatectomy, cholecystectomy, pancreatectomy, colectomy, or rectal cancer surgery), year of surgery, cancer stage, use of epidural anesthesia, use of morphine, use of oxycodone, preoperative chemotherapy, preoperative radiotherapy, postoperative chemotherapy, postoperative radiotherapy, preoperative renal replacement therapy, intraoperative blood transfusion, type of hospital (academic or nonacademic), Barthel Index at admission, and hospital volume. The patients were categorized into four age groups (younger than 59, 60 to 69, 70 to 79, and 80 yr or older) because more than 65% of the patients were aged 60 to 79 yr. The BMI was divided into five categories based on the World Health organization classifications of underweight (less than 18.5 kg/m2), normal weight (18.5 to 24.9 kg/m2), overweight (25.0 to 29.9 kg/m2), and obese (30.0 kg/m2 or more). The tumor, node, metastasis cancer stages were determined by postoperative pathology and divided into 0 or I, II, III, and IV.

The Barthel Index is frequently used to measure performance in activities of daily living, with scores ranging from 0 to 100 points (higher scores indicate less disability). This index includes 10 items of mobility and self-care functions. We divided the Barthel Index into two groups (0 to 95 and 100) because more than 90% of the patients had a Barthel Index of 100.

For comorbidities at admission, each ICD-10 code for a comorbidity was converted into a Charlson Comorbidity Index score, which is widely used as a validated measure to predict in-hospital morbidity and mortality for each patient.[12] All patients were diagnosed with cancer; therefore, the lowest Charlson Comorbidity Index score was 2.

Hospital volume was defined as the average number of surgeries performed at each hospital annually and was divided into three groups containing almost equal numbers of patients.

Perioperative complications were defined as the occurrence of the following diseases during the first perioperative period: cerebral infarction or hemorrhage (ICD-10 codes I60 to I64), acute coronary events (I21, I22, and I252), heart failure (I50), pulmonary embolism (I26), acute and subacute hepatic failure (K720), acute renal failure (N17), sepsis (A40 and A41), wound infection (T793 and T814), pneumonia (J12 to J18 and J69), and urinary tract infection (N390, T835, and N30). We also searched for anastomotic leakage using Japanese text.

The primary outcomes were recurrence-free survival and overall survival.

Statistical Analysis

No statistical power calculation was conducted before the study because the sample size in our study was based on secondary use of administrative claims data and a fixed available sample.

The patient characteristics and type of surgery in each group are described using number and proportion for categorical variables and mean with SD for continuous variables. Standardized differences were used to compare the distribution of baseline covariates between treatment groups in observational studies. Small differences in the absolute standardized differences (less than 0.10) suggest balanced baseline characteristics between patients in the volatile anesthesia and total intravenous anesthesia groups.[13]

Kaplan–Meier survival analysis was used to compare overall survival and recurrence-free survival between the two groups. Cox proportional hazard regression models were used to compare the relationship between total intravenous anesthesia and overall survival or recurrence-free survival with adjustment for the following baseline variables: age, sex, BMI, smoking status, Charlson Comorbidity Index score, cancer stage, preoperative adjuvant therapy, postoperative adjuvant therapy, preoperative renal replacement therapy, preoperative or intraoperative blood transfusion, preoperative use of morphine or oxycodone, type of hospital (academic or nonacademic), hospital volume, Barthel Index at admission, and at least one postoperative complication. We used the Schoenfeld residuals test and complementary log plots to assess the proportional hazards assumption. The proportional hazards assumption was not violated in any of our analyses. Some data regarding the BMI, cancer stage, and Barthel Index at admission were missing. We used a complete case analysis for these missing values. Follow-up was censored on March 31, 2018 or the last outpatient record.

Observational studies have the potential for residual confounders due to measured and unmeasured baseline characteristics, which can lead to incorrect associations between the type of anesthesia and outcomes. One strategy to address this limitation is the use of an instrumental variable analysis designed to adjust for unmeasured confounding between two groups, allowing the achievement of a pseudo-randomized controlled trial.[14] An instrumental variable analysis requires the following assumptions: (1) the instrumental variable is independent of the unmeasured confounding; (2) the instrumental variable is strongly associated with the treatment; and (3) the instrumental variable is associated with the outcome only indirectly through its effect on the treatment. The type of anesthesia performed mainly depends on the physician's preference; therefore, we used the proportion of total intravenous anesthesia use at each hospital as an instrumental variable. The proportion of total intravenous anesthesia use was defined as the number of patients who had total intravenous anesthesia divided by the number of all patients in each hospital. We conducted two-stage residual inclusions for the instrumental variable analyses to compare recurrence-free survival or overall survival between the volatile anesthesia and total intravenous anesthesia groups. We fit a first stage logistic model that predicts treatment assignment (volatile anesthesia and total intravenous anesthesia) with the instrumental variable and the aforementioned variables to estimate the probability of having total intravenous anesthesia. Next, the second stage model was fitted by regressing these outcomes on the performance of total intravenous anesthesia in a Cox regression model, along with the residuals from the first-stage model and the other variables.

We confirmed that the proportion of total intravenous anesthesia use at each hospital was not a weak instrument using a partial F test, with an F statistic of more than 10.[15]

Two additional approaches were performed as sensitivity analyses. First, we used Cox regression analyses after propensity score matching. We estimated the propensity score using a logistic regression model for the receipt of total intravenous anesthesia, incorporating the baseline characteristics of the aforementioned variables without postoperative adjuvant therapy and at least one postoperative complication. We set a caliper at 0.2 SD of the estimated logit of the propensity score and performed one-to-one propensity score matching of patients between the types of anesthesia using the nearest neighbor method without replacement. We estimated the balance in the propensity score–matched cohort using standardized differences. Second, we performed an instrumental variable analysis using the proportion of total intravenous anesthesia at each of 47 prefectures as another instrumental variable.

Subgroup analyses were performed for the type of cancer surgery. We evaluated the association between total intravenous anesthesia and the primary outcomes using Cox regression analyses. A two-tailed P value of less than 0.05 was considered statistically significant. All analyses were performed using Stata/MP 16.0 (StataCorp, USA).

A Priori Versus Post Hoc Analyses

As a priori analyses, we planned to perform Cox regression analyses and instrumental variable analyses using the proportion of total intravenous anesthesia use at each hospital as an instrumental variable in all patients to evaluate the association between the type of anesthesia and outcomes. As a post hoc sensitivity analysis, we performed Cox regression analyses in propensity score–matched patients and instrumental variable analyses using the proportion of total intravenous anesthesia in each of 47 prefectures as another instrumental variable.