Volatile versus Propofol General Anesthesia and Long-term Survival after Breast Cancer Surgery: A National Registry Retrospective Cohort Study

This is a cohort study based on national registry data in the Swedish PeriOperative Registry and the National Quality Registry for Breast Cancer. The Swedish PeriOperative Registry contains information of the individual surgical procedures, covering the entire perioperative process from preoperative workup to discharge from the postanesthesia care unit, including the anesthetics used (coverage rate, 93.5%). The National Quality Registry for Breast Cancer also contains important supplementary information (e.g., cancer stage and adjuvant treatments with a coverage rate of 99%) in addition to survival data. Both registries are prospectively maintained with several built-in data validation processes. Incorrect and/or inconsistent posts are returned to the user for correction before inclusion in the database. Both registries use the unique social security numbers given to all Swedish citizens. The designation “quality registry,” which is necessary to obtain government funding, has been given to both registries (Supplemental Digital Content 1, http://links.lww.com/ALN/C875). A data analysis and statistical plan was written and filed (https://www.medfarm.uu.se/ckfvasteras/forskning/studieprotokoll; accessed February 21, 2020) before the data were accessed (October 2, 2020).

After ethics approval (Ethics Review Board, Uppsala, Sweden; approval no. 2020-00573), with individual written informed consent waived, all patients with breast cancer who were operated on between 2013 and 2019 were identified in the National Quality Registry for Breast Cancer, and all variables of interest (explained in subsequent paragraphs) were extracted. This data set from the National Quality Registry for Breast Cancer was sent to Uppsala Clinical Research Center, which is responsible for the Swedish PeriOperative Registry. Uppsala Clinical Research Center added its data to the file from the National Quality Registry for Breast Cancer and deidentified the final file (with a key) before it was sent encrypted to the first author (Dr. Enlund).

The independent/causal variable was the drug given for maintenance of general anesthesia, i.e., propofol or a volatile (desflurane, isoflurane, or sevoflurane). Of the dependent/control variables, age at surgery, body mass index, and American Society of Anesthesiologists (ASA) status were regarded as true confounders, since they were expected to be associated with both the choice of anesthesia and overall survival. The hospitals were aggregated into three groups depending on their volume of surgery (less than 100, 100 to 500, or more than 500 surgeries annually) to minimize the confounding effect of surgical volumes on outcomes. Finally, cancer classification (according to tumor/metastasis/node staging), neoadjuvant and/or adjuvant therapy (chemotherapy, radiotherapy, endocrine therapy, and/or antibody therapy), type of procedure (total or partial mastectomy, with or without axillary clearance, complementary breast procedure), and progesterone receptor, estrogen receptor, antigen KI67, and human epidermal growth factor receptor 2 status are all known to be associated with prognosis, but not necessarily with the choice of anesthetics; therefore, these data are included as effect modifiers in the model for propensity score matching.

Continuous variables were expressed as medians with interquartile range, while categorical variables were presented as absolute numbers and percentages. Standardized mean differences were presented between the propofol- and volatile-based anesthesia groups.

Propensity score matching is a method to minimize selection bias between interventional groups when estimating causal intervention effects in nonrandomized studies.²⁹  The treatment groups (propofol- or volatile-maintained anesthesia) were matched on a propensity score. The propensity score is the probability of intervention assignment conditional on the current baseline characteristics. Once the sample was created, the treatment effect could be estimated by directly comparing outcomes between the groups.

We created two propensity score–matched cohorts. In the first propensity score–matched cohort, the propensity scores were developed accounting for all demographical and clinical characteristics summarized in tables 1 and 2, except from body mass index and blood loss (due to high proportion of missing values), while in the second propensity score–matched cohort, only variables with a standardized mean difference higher than 0.1 were included (sensitivity analysis). All individual propensity scores were calculated through logistic regression models,²⁹  and then a 1:1 nearest-neighbor propensity score matching³⁰  with a caliper size of 0.1 was used.

The primary outcome was overall survival between subjects who underwent propofol- or volatile-maintained anesthesia. All analyses of overall survival included only patients with no missing information. Overall survival was defined as from the date of surgery to death from any cause or the end of follow-up (September 15, 2020, defined as censored), whichever came first. Any subject lost to follow-up was excluded from the study population. Overall survival was presented by using the Kaplan–Meier approach with the corresponding log-rank test. In addition, overall mortality was estimated, expressed as hazard ratios with 95% CI between the two groups using Cox regression models. We tested the proportional hazards assumption for all the Cox regressions using the tests based on weighted residuals.

In a first step, the overall survival and mortality in the unmatched cohort were described. In a second step, after the 1:1 propensity score matching, the outcomes were presented. No imputation of missing data was planned, and all analyses were performed on patients who had information on all variables selected for the propensity score matching. Thus, we excluded patients with missing information on any of the variables. We also changed the caliper size (from 0.1 to 0.5) to see the robustness of the findings for the propensity score–matched cohort. As an exploratory analysis, overall survival and mortality was presented between subjects who underwent propofol or volatile anesthesia in the subgroup of triple-negative breast cancer patients.

All tests were two-sided, and statistical significance was considered with a P value less than 0.05. The statistical analyses were performed using R version 3.6.1 (R basis for statistical calculation; Vienna University of Economics and Business, Vienna, Austria) using the packages cobalt, tableone, survival, and MatchIt.

A total of 23,545 subjects were included in the encrypted file from the Swedish PeriOperative Registry. We excluded duplicate subjects (n = 2,041), subjects with incorrect anesthesia codes (n = 2,134), subjects missing follow-up time (n = 2), subjects with no evidence of cancer in the breast (n = 226), subjects with cancer in situ (n = 370), subjects whose primary tumors could not be evaluated (n = 96), and subjects with missing tumor staging information (n = 2; fig. 1). The final study population consisted of 18,674 subjects. A majority, 13,873 patients, were anesthetized with propofol, and 4,801 were exposed to an inhalational agent. None of the patients were given both propofol and a volatile for anesthesia maintenance. The anesthesia registry coverage increased during the first years of the time frame and reached 93.5% coverage at the end of the period.

The two groups differed in several characteristics (table 1). The propofol group was younger and its average body mass index was lower; these patients had lower ASA classifications; and their tumor status was unevenly distributed between the groups, with generally lower grading for the propofol patients. The proportions of triple-negative cancer did not differ statistically between the groups. The median follow-up time was 33 months (interquartile range, 19 to 48 months).

In the full, unmatched cohort, there was a statistically significantly higher overall survival among patients receiving propofol (13,489 of 13,873 [97.2%]) versus inhaled volatile (4,039 of 4,801 [84.1%]; hazard ratio, 0.80; 95% CI, 0.70 to 0.90; P < 0.001; fig. 2; table 3). After propensity score matching for all of the variables summarized in table 1, except for body mass index and blood loss due to high level of missing data, there was no statistically significant difference between the propofol and volatile groups (propofol, 4,284 of 4,658 [92.0%]; inhaled volatiles, 4,288 of 4,658 [92.1%]; P = 0.756; fig. 3) with a hazard ratio of 0.98 (95% CI, 0.85 to 1.13; table 3). The same pattern was also seen in the second propensity score–matched cohort, in which only variables with a standardized mean difference higher than 0.1 were included and in which no statistically significant differences were observed (Supplemental Digital Content 2 [http://links.lww.com/ALN/C876], tables S1 and S2; Supplemental Digital Content 3, fig. S1 [http://links.lww.com/ALN/C877]). In addition, findings in this cohort were not altered by changing the size of the caliper or when body mass index (caliper size, 0.1; P = 0.397; data not shown) and blood loss (caliper size, 0.1; P = 0.510; data not shown) were included in the propensity score matching. The assumption of proportional hazards did not hold in the unmatched cohort (P < 0.001) but did hold for the full propensity score–matched (P = 0.370) and restricted matched cohorts (P = 0.340), respectively. Triple-negative breast cancer propensity score–matched patients indicated no meaningful survival benefit for propofol (hazard ratio, 1.17; 95% CI, 0.79 to 1.72; P = 0.443; table S3, Supplemental Digital Content 3, fig. S2, http://links.lww.com/ALN/C877). The 3- and 5-yr survival rates were calculated as a complement (table 3).

In this national registry–based, propensity score match study with prospectively gathered data from 9,316 patients, the previous findings of several retrospective studies were confirmed. Neither a statistically significant difference nor a clinically meaningful difference could be found between administration of general anesthesia using propofol versus inhaled volatile in long-term survival for patients with breast cancer.

The current results are not consistent with findings from biomarker studies. The results indicate that: (1) propofol, compared with sevoflurane, has a more favorable inhibiting effect on vascular endothelial growth factor C and transforming growth factor β in women undergoing breast cancer surgery³¹ ; (2) the activity of natural killer cells is higher in blood sampled from women anesthetized with propofol (and receiving a paravertebral block instead of parenteral opioids) for breast cancer surgery in comparison with women given a sevoflurane or opioid anesthetic³² ; and (3) cancer cell apoptosis is higher in a propofol or paravertebral group, compared with a group given sevoflurane or opioid for breast cancer surgery.³³  Furthermore, the transcription factor hypoxia-inducible factor, which improves the cancer cells’ adaptation to hypoxia, acidosis, and starvation in a solid cancer, will be upregulated when exposed to a volatile, while the opposite appears to be the case when it is exposed to propofol.^17,34–36  These biomarker studies suggested that an inhaled volatile general anesthetic may increase the risk of a local recurrence or metastasis, while propofol general anesthesia may be neutral or even protective.

The combined effects of propofol and a paravertebral block compared with sevoflurane and opioids were studied in a randomized controlled trial with recurrence as the primary outcome in 2,132 patients with breast cancer.³⁷  The study was ended after a preplanned futility boundary was crossed, and the median follow-up time was 36 months. The recurrence rates were identical: 10% in both groups. While recurrence rate is distinct from a mortality outcome, the result indicates that anesthesia technique choice does not impact cancer recurrence. In addition, this study compared two different combinations of anesthesia: (1) general anesthesia with propofol combined with regional anesthesia and (2) general anesthesia with sevoflurane combined with an opioid. Despite these differences, the current results are consistent with this trial’s observations.

Perioperative factors may also differ between different cancers depending on the complexity of the operation. Breast cancer, as a rather superficially localized cancer, is relatively easy to access, as opposed to intra- or retroabdominal organ tumors or lung and brain tumors. Not only may this affect the risk of seeding of cancer cells during manipulations of the cancer, but it may also affect the duration of anesthesia. If the hypothesis is valid for some other cancers, the time exposed to a volatile would be critical. Breast cancer operations are relatively short procedures, which may favor a null finding.

It is not only for breast cancer that the data do not support the hypothesis of different impact on long-term survival after cancer surgery depending on the choice of anesthetic. For example, in gastric cancer and lung cancer, there is one study for each of the two cancers indicating no difference in long-term survival between propofol and sevoflurane.^38,39  This contrasts with two other studies for gastric cancer and one other for lung cancer that support the hypothesis.^4,12,13  However, a large, recently published study, using a national registry, indicated no difference in long-term survival after gastrointestinal cancer,⁴⁰  whereas another national registry–based study on colorectal cancer found a benefit for propofol.⁴¹ 

Propofol was used for anesthesia induction in all patients, including those whose anesthesia was subsequently maintained with volatiles. It has been proposed in a large observational study that an increasing propofol dose was associated with reduced odds of 1-yr mortality in patients without a solid cancer but not in patients with solid cancer, a finding that was replicated for 5-yr mortality.⁴²  There were significant interactions between propofol dose and breast, colorectal, and hepatobiliary cancer with regard to 1-yr mortality, but the odds of 1-yr mortality increased with higher doses of propofol only in patients with colorectal or hepatobiliary cancer. Therefore, it is unlikely that an induction dose of propofol to the volatile group in our study would have a modifying effect on survival in the volatile group.

A strength of the current study was the availability of tumor characteristics, such as stage, receptors, and prognostic markers. The main limitation of the current study is that despite the use of prospectively collected data, it still is a retrospective observational study, and there is a risk of residual confounding by unmeasured or unknown covariates. For example, the absence of specific comorbidities, clinical provider information, postoperative management, or other intraoperative medication information in the registry is a significant limitation.

It should also be noted that the registry used for this study does not discriminate between volatiles. Sevoflurane dominates the Swedish market, but we cannot state the proportions between the three volatiles used (desflurane, isoflurane, and sevoflurane). This would be of interest if the three volatiles differ in the effects on the immune system, as suggested in a retrospective study of stage III ovarian cancer, in which patients exposed to sevoflurane had a higher rate of cancer recurrence compared with patients receiving desflurane.⁴³  Finally, cancer biology may modulate the response to the anesthetics, so these findings cannot be generalized to other cancers.

This observational study did not show any difference in survival between breast cancer patients receiving propofol general anesthesia compared with those receiving inhaled volatile general anesthesia.

The data presented in this study were drawn from the Swedish PeriOperative Registry (Uppsala, Sweden) and the National Quality Registry for Breast Cancer (Stockholm, Sweden). The registries do not take responsibility for the methods, analysis, and results, and the views expressed in this study may not necessarily reflect those of the registries. The authors express their gratitude to the steering groups for the Swedish PeriOperative Registry and the National Quality Registry for Breast Cancer and to their administrators for their great support, for the data extraction procedures, and for data deliveries, especially Beata Pajak and Camilla Hartmann-Norman, Ph.D., at Uppsala Clinical Research Center (Uppsala, Sweden), administering the Swedish PeriOperative Registry; and John Lövrot, Ph.D., at the National Quality Registry for Breast Cancer.

Support was provided solely from institutional and/or departmental sources.

The authors declare no competing interests.

Information about Swedish Quality Registries, http://links.lww.com/ALN/C875 Supplemental Tables: Restricted Propensity Score Match, http://links.lww.com/ALN/C876 Overall Survival after a Full Propensity Score Match, http://links.lww.com/ALN/C877