The study was approved by the institutional review board of Chungnam National University Hospital (approval number CNUH 2017-08-018). The requirement for informed consent was waived in view of the retrospective nature of the study. This clinical trial has been registered at Clinical Research Information Service (registration number KCT0004101).
We reviewed the medical records of patients who underwent surgical resection for gastric, lung, liver, colon, or breast cancer from January 2006 to December 2009 in our hospital. Surgeries during the investigation period in which patients received TIVA included general and thoracic surgeries, such as thyroid, breast, colon, hepatobiliary, gastric, and lung cancer surgery. Although a high number of patients had thyroid cancer, the survival rate was sufficiently high that a comparison was not meaningful. In our hospital, thyroidectomy is rapid and it is difficult to manage intravenous catheters for affected patients; accordingly, these patients have received inhalational anesthesia for many years. Therefore, the five major cancers selected for this study were gastric, colon, liver, breast, and lung cancers.
Patients who had undergone emergency surgery, with no follow-up after surgery, patients whose medical records could not be confirmed, patients whose anesthesia was changed during surgery, and patients who died during or immediately after surgery were excluded from the study. Patients who did not fulfill any of the variables examined in the medical record were excluded. Remifentanil with 2% propofol was used via target-controlled infusion for the induction and maintenance of anesthesia in the TIVA group, while remifentanil or nitrous oxide with a volatile anesthetic agent (desflurane, sevoflurane, or isoflurane) was used for the maintenance of anesthesia in the VIA group. At the induction of anesthesia in the VIA group, propofol or etomidate was used, depending on the condition of the patient and the anesthesiologist's preference. Because the benefits of restrictive fluid therapy were not clearly established, liberal fluid therapy was used. The type of anesthesia selected was entirely based on the anesthesiologist’s preference.
Patient factors were age at the time of surgery, sex, body mass index (BMI), and American Society of Anesthesiologists (ASA) class. Surgical and anesthetic factors were the presence of hypertension and diabetes mellitus (DM), total anesthesia time, operation time, type of anesthesia (volatile inhalational anesthesia vs. total intravenous anesthesia), use of nitrous oxide, application of remifentanil infusion, and presence of metastasis at the time of surgery. We also investigated the patient’s total length of hospital stay. We investigated the correlations between each of the factors and 5-year survival. Patients were followed-up only with regard to the primary outcome, i.e., overall survival.
All data related to the surgery were obtained from the hospital statistical records. Data related to anesthesia, metastasis, and deaths were obtained from the hospital electronic medical records. If we could not find an electronic medical record of the patient’s survival at 5 years after surgery, the patient or caregiver was contacted by phone. In such instances, we briefly explained the study and received verbal consent. In addition, the contact information used at this time was not recorded on the case record sheet. If the contact information was unknown, the case was classified as a missed medical record.
Based on the results of a previous study , to achieve a power of 80% and a two-tailed type I error rate of α = 0.05, G*Power 3.1 calculations revealed that at least 495 patients were needed in each matched group. The total number of surgeries per year in our hospital is approximately 10,000; of these surgeries, approximately 600 involve surgical treatments for the five major cancers. Because the ratio between inhalation anesthesia and TIVA was approximately 2:1 during the test period, a 4-year study period was chosen. Patients who underwent surgery between 2006 and 2009 were included because 5 years had already passed at the beginning of the study. After propensity score matching, there were 729 patients in each group, which exceeded the minimum of 495 patients per group.
The sample consisted of all subjects during the study period. All available patients were considered. To adjust for possible selection bias and confounding factors , 1:1 ratio propensity score matching was performed using the MatchIt package in R . The dependent variable was set as a binary response of 0 or 1, and logistic regression analysis was performed by designating the covariate (age, sex, height, weight, BMI, ASA class, hypertension, DM, anesthesia time, operation time, metastasis, transfusion) to be corrected as an independent variable. The survival rate was different for each cancer, and the numbers of anesthetic methods used were different for each cancer. Therefore, we matched for each type of cancer.
Nearest neighbor matching was performed, which matches the absolute differences of the estimated propensity scores of all subjects in both groups from the smallest to the largest difference. Absolute standardized difference (ASD) was calculated to validate the suitability of propensity score matching balance diagnostics between the two groups, with ASD < 0.1 for the covariate indicating that the two groups were sufficiently balanced.
After validating the balance of the matched data, the normality of continuous data was assessed using the Shapiro–Wilk test. If normality was satisfied, comparisons between groups were performed by independent t tests, with the results expressed as means ± standard deviations. If normality was not satisfied, groups were compared using the Mann–Whitney U test, with the results expressed as medians (interquartile ranges). Categorical data were compared using the chi-squared test or Fisher’s exact test, as appropriate, with the results expressed as numbers (%).
Survival outcomes were analyzed by the log-rank test and expressed by the Kaplan–Meier plot. Cox proportional hazards modeling was used for univariate and multivariable analysis of demographic and clinical variables influencing the survival outcomes. The cut points of the continuous variables were obtained using the maxstst package; survival analysis was performed by separating the patients into two categories based on the following cut points: age, 65 years; height, 165 cm; weight, 57 kg; BMI, 19.7; and anesthesia time, 210 min. Only the meaningful variables (P < 0.2) from univariate analysis were included in multivariable analysis. Akaike’s Information Criterion was considered for final model selection by backward elimination. Associations with P < 0.05 were considered statistically significant. All Data were analyzed using R software version 3.5.2 (R Project for Statistical Computing, Vienna, Austria).