Anesthesia and Circulating Tumor Cells in Primary Breast Cancer Patients: A Randomized Controlled Trial

We used the Consolidated Standards of Reporting Trials recommendations for the reporting of randomized trials.²⁴  This trial was approved by the local ethical committee (Zurich, Switzerland, registration number PB_2016-01791) and was registered with ClinicalTrials.gov (NCT02005770, https://clinicaltrials.gov/ct2/show/NCT02005770, principal investigator: Beatrice Beck-Schimmer, registration date: December 9, 2013). The study protocol is available on ClinicalTrials.gov.

This was a parallel-group, randomized, controlled trial conducted at a university hospital (University Hospital of Zurich) and a private clinic (Hirslanden Group, Zurich) in Switzerland. Patients were considered eligible if they were aged 18 to 85 yr, diagnosed with primary preinvasive and invasive breast cancer without distant metastases (stage 0 to III) and scheduled for surgery with or without axillary node dissection. Patients were excluded if they met one of the following criteria: preoperative chemotherapy, possible immune impairment (i.e., autoimmune disease, human immunodeficiency virus, other active cancer, American Society of Anesthesiologists (ASA; Schaumburg, Illinois) Physical Status IV or V), immunosuppressive or chronic opioid therapy, secondary surgery (e.g., for recurrence, reconstruction), or surgery performed under general anesthesia with concomitant regional anesthesia (i.e., epidural catheter, paravertebral blockade, wound infiltration with local anesthetics). Those with a known or suspected hypersensitivity or allergy to anesthetics were considered ineligible. Patients were approached on the day before surgery by research staff, who evaluated eligibility, obtained written informed consent, and enrolled the participants.

Randomization was performed by research staff using a secure Internet-based system (www.randomizer.at; accessed April 10, 2018) that stratified patients according to their ASA status and ensured concealment of random allocation. The patients were randomly assigned in a 1:1 ratio to either intravenous anesthesia (propofol group) or inhalational anesthesia (sevoflurane group). Patients remained blinded to their assignment group (standardized induction in both groups), as was the study personnel involved in circulating tumor cell measurements (i.e., outcome assessors did not have access to patient charts).

Anesthesia induction was standardized in both groups using fentanyl (2 to 3 μg/kg), thiopental (4 to 6 mg/kg), and rocuronium (0.6 mg/kg). Patients requiring a rapid sequence induction received 0.9 mg/kg rocuronium instead of 0.6 mg/kg. Further administration of fentanyl during surgery followed a standardized protocol (i.e., 2 μg/kg; total amount, 5 to 10 μg/kg). In the propofol group, anesthesia was maintained using a target-controlled infusion device providing an intravenous propofol dose adjusted to keep Bispectral Index values between 40 and 60; in the sevoflurane group, sevoflurane was provided to keep Bispectral Index values between 40 and 60. Postoperative nausea and vomiting prophylaxis and perioperative analgesia followed standardized protocols that were applied until hospital discharge.

The primary outcome was the number of circulating tumor cells assessed postoperatively by the CellSearch assay (Menarini Silicon Biosystems Inc., USA). Based on immunomagnetic separation, this detection technique uses a magnetic field to isolate ferrofluid-labeled tumor cells of epithelial origin, such as breast cancer cells.²⁵  This standardized procedure uses antibodies directed against a common molecular signature displayed by circulating tumor cells in breast cancer patients (i.e., the “EpCAM+/CK+/DAPI+/CD45−” signature, where EpCAM indicates epithelial cell adhesion molecule, CK indicates cytokeratin, and DAPI indicates 4′,6-diamidino-2-phenylindole). After staining of the isolated cells, circulating tumor cell identification was confirmed by two independent, specifically trained laboratory technicians that were masked to treatment assignment. Identification of circulating tumor cells followed a predefined set of criteria (i.e., morphological features, compatible staining pattern).

Peripheral blood was collected at four different time points, i.e., before the induction of anesthesia (baseline), after surgery but before extubation (0 h), on day 2 (48 h), and on day 3 (72 h) postoperatively. The last measurement was initially planned on day 4 but was rescheduled to day 3 in January 2016 to avoid data loss due to early hospital discharge. This was the only change made to the original trial design.

Secondary outcomes were defined as the maximal circulating tumor cell count value at any time point after surgery (0, 48, and 72 h); circulating tumor cell counts as a binary outcome (using two different cutoff values, i.e., at least 1 and at least 5 circulating tumor cells/7.5 ml blood); and the association between natural killer cell activity and circulating tumor cell counts (see also “Additional Analyses”). Initially, only a cutoff value of a least 5 circulating tumor cells/7.5 ml blood was considered. We added the threshold of a minimum of 1 cell at the time of analysis, because evidence suggested that values as low as 1 circulating tumor cell/7.5 ml blood were associated with poorer prognosis in primary breast cancer patients.²²  No other changes were made to primary/secondary outcomes definitions over the study period.

Sample size calculation was performed using a method accounting for repeated measurements of count data over time.²⁶  Because evidence on the effect of intravenous or inhalational anesthesia on circulating tumor cell counts was nonexistent, we adopted a conservative approach and assumed that the expected effect size (Cohen’s d) between groups would be small (0.3). Thus, assuming a within-subject correlation of circulating tumor cell counts over time of 0.4 and a dropout rate of 10%, we estimated that a total of 232 patients would be required (209 patients without dropout) to detect a difference between groups corresponding to an effect size of 0.3, with a power of 80%, at a significance level of 5% (two-sided). Because the dropout rate was particularly low, the trial ended after enrolling 217 patients.

All analyses were based on intention to treat. Continuous data were expressed as means and standard deviations or as medians and interquartile ranges if distributions were skewed. The primary analysis used a mixed Poisson model with random intercept per patient to account for repeated measurements over time and thus correlated observations within subjects. We opted for this approach because the Poisson model is appropriate for count data (primary outcome of circulating tumor cell counts). The results of the Poisson models are presented as rate ratios, denoting the comparison of circulating tumor cell counts between the two groups. To avoid assuming a linear development of circulating tumor cells over time, time was alternatively included as a factor variable in our model. We also explored the effect of anesthetics on the maximal circulating tumor cell count value at any time point after surgery in additional Poisson models (0, 48, and 72 h).

Because circulating tumor cell detection is usually reported as a binary outcome (i.e., positive vs. negative endpoint using a cutoff value of at least 1 or at least 5 circulating tumor cells/7.5 ml blood), circulating tumor cell count data were dichotomized and further assessed using a mixed logistic regression model with random intercept per patient. Finally, models were adjusted to account for tumor-related and perioperative factors presumed to affect circulating tumor cell counts (i.e., tumor size, tumor type, and overall opioids consumption, all preplanned).

All statistical analyses were conducted in R, version 3.6.1. Two-sided tests were performed, and a level of significance of 0.05 was used.

Because of the interplay between natural killer cell cytotoxic activity and tumor growth, we also assessed natural killer cell activity (i.e., apoptosis rate induced in tumor cells) in a preplanned, exploratory, in vitro study nested within this trial. Natural killer cell–induced apoptosis was evaluated in a subgroup of patients randomly selected from the study data set. For each patient, natural killer cell activity was assessed at a single, predefined time point, i.e., when circulating tumor cell counts reached their maximal value. The association between natural killer cell–induced apoptosis rate and circulating tumor cell count was then assessed using linear regression analysis.

Natural killer cell–induced apoptosis rate and necrosis rate were determined in vitro by measuring target cell killing of the K562 tumor cell line (human chronic myelogenous leukemia, ATCC, CCL-243).^27,28  Patients blood samples were collected in EDTA-coated vials. Buffy coats (Blutspende Zürich, Switzerland) were used as controls. Peripheral blood mononuclear cells of both patient samples and buffy coats were isolated by Ficoll–Hypaque density gradient centrifugation and stored in liquid nitrogen. For determination of natural killer cell activity, peripheral blood mononuclear cells were thawed and coincubated with K562 for 24 h at 37°C with 5% CO₂ in 10% human serum/RPMI medium. An effector (natural killer cells)–to–target cell (K562 cells) ratio of 1:1 was used. All cells were then washed in phosphate-buffered saline and stained in 2% bovine serum albumin in phosphate-buffered saline for 25 min at 4°C using the following panel: CD3-APC (lymphocyte staining; Biolegend, United Kingdom), dilution of 1:100; CD 56-PE (natural killer cells staining; Biolegend), dilution 1:100; and CD16-FITC (FcγRIIIA staining, which is essential for cellular cytotoxicity, expressed on the surface of a subset of monocytes; Biolegend), dilution 1:200. After a washing step in annexin V binding buffer, the cells were simultaneously stained with annexin–PerCPCy5.5 for staining of apoptotic cells (Biolegend) at a dilution of 1:20 and Zombie-NIR for staining of necrotic cells (Biolegend) at a dilution of 1:500.

Zombie-NIR–stained K562 boiled for 5 min at 80°C or annexin V–stained apoptotic K562 and treated for 24 h with 10 mM benzamide were used as positive controls for cytotoxicity. Unstained K562, unstained patient peripheral blood mononuclear cells, and unstained peripheral blood mononuclear cells from buffy coats served as negative controls. Cell analysis was performed using the spectral analyzer SP6800 (Sony Biotechnology, United Kingdom).²⁹ 

Between March 10, 2014, and April 10, 2018, 586 patients were assessed for eligibility (fig. 1). Of 217 enrolled participants, seven patients withdrew consent after randomization. We eventually included 210 patients in the intention-to-treat analysis (sevoflurane group: n = 107, propofol group: n = 103).

Baseline characteristics are presented in table 1. Demographic and clinical data were balanced between treatment groups. Most participants were middle-aged, modestly morbid patients with an early-stage tumor. Baseline circulating tumor cell counts and positivity (using a cutoff value of at least 1 and at least 5 circulating tumor cells/7.5 ml blood) were similar in both allocation groups. Table 2 depicts the intra- and postoperative characteristics, which were well balanced between groups.

The evolution of circulating tumor cell counts over time is illustrated in figure 2, table 3, and Supplemental Digital Content figure 1, which depicts predicted tumor cell counts using the estimates from the Poisson model, including a linear time variable and baseline circulating tumor cell counts (http://links.lww.com/ALN/C415). Administrating sevoflurane versus propofol did not affect the primary outcome of circulating tumor cell counts over time (rate ratio, 1.27 [95% CI, 0.95 to 1.71]; P = 0.103). This was the case, regardless of whether time was considered as a linear or a factor variable, or whether an interaction term between time and anesthesia was introduced. However, when we explored the effect of anesthetics on the maximal circulating tumor cells value at any time point after surgery, administrating inhalational anesthesia (i.e., sevoflurane) led to a significant increase in maximal circulating tumor cell counts postoperatively (sevoflurane vs. propofol: rate ratio, 1.36 [95% CI, 1.18 to 1.56]; P < 0.0001; i.e., the maximum number of circulating tumor cells increased by a factor of 1.36 (or 36%) when sevoflurane was used compared with propofol).

When circulating tumor cells were analyzed as a binary outcome over time, the type of anesthesia did not have any effect on circulating tumor cell positivity, regardless of the cutoff value considered (cutoff value of at least 1 circulating tumor cell/7.5 ml blood: sevoflurane vs. propofol odds ratio, 1.21 [95% CI, 0.84 to 1.74]; P = 0.309; cutoff value of at least 5 circulating tumor cells/7.5 ml blood: sevoflurane vs. propofol odds ratio, 1.59 [95% CI, 0.86 to 3.01]; P = 0.139). Similar results were obtained when time was considered as a factor variable, and there was no evidence for an interaction between treatment and time.

We performed predefined analyses to explore whether tumor-related and perioperative factors modified the effect of anesthetics on circulating tumor cell counts. Models adjusted for tumor type (DCIS, luminal A, luminal B, triple negative, HER2 positive, other) and tumor size (Tis, T1, T2, T3, T4) did not reveal any relevant effect modification on circulating tumor cell counts over time or positivity (regardless of the cutoff value considered). Similarly, adjusting for opioid consumption did not yield any effect modification. In the exploratory models, however, the effect of inhalational anesthesia on maximal postoperative circulating tumor cells values remained robust (sevoflurane vs. propofol rate ratio, 1.26 [95% CI, 1.09 to 1.47]; P = 0.002; adjustment for tumor type, size, and opioid consumption).

Exploratory in vitro analyses were conducted in a subgroup of 60 patients randomly selected from the study data set (30 in the sevoflurane group and 30 in the propofol group). Similar natural killer cell–induced apoptosis rates were found in both treatment groups (mean apoptosis rate, for sevoflurane group, 34.7%; for propofol group, 35.7%). Overall, the necrosis rate of K562 tumor cells was less than 1%. Linear regression yielded no evidence for an association between apoptosis rates and maximal circulating tumor cell counts (regression coefficient, −0.077; 95% CI, −0.33 to 0.17; fig. 3). This was the case, regardless of treatment group assignment or whether an interaction term between anesthesia type and natural killer cell activity was introduced.

In this randomized controlled trial including 210 participants undergoing surgery for primary breast cancer, the type of anesthesia did not seem to affect circulating tumor cell counts over time or circulating tumor cell positivity. In one secondary analysis, there was a 36% increase in the maximal number of postoperative circulating tumor cells in patients receiving inhalational anesthesia. Additional in vitro analyses in a random selection of 60 patients did not reveal any evidence for an association between natural killer cell–induced apoptosis rates and maximal circulating tumor cell counts.

This trial investigated the effect of anesthesia on perioperative circulating tumor cell counts, an independent prognostic factor for breast cancer. In contrast to previously published randomized trials,^15–17  our study was larger and had an adequate control group, and the issue of long follow-up periods was mitigated by using a prognostic factor.

In our trial, circulating tumor cell counts at baseline were higher than those reported in previous studies. Several reasons may account for this discrepancy. First, all of our patients underwent sentinel lymph node localization 18 to 24 h before baseline circulating tumor cell assessment, and we cannot formally exclude that an injection in the vicinity of the tumor would not lead to any circulating tumor cells release. Second, approximately 30% of our patients had wire-guided localization of the tumor, which implies direct manipulation of the tumor shortly before circulating tumor cell assessment.

Because the identification of circulating tumor cells with the CellSearch assay may imply some degree of subjectivity (i.e., images of potential tumor cell candidates are displayed to trained laboratory technicians and assessed following predefined criteria), we verified all samples with at least 5 tumor cells/7.5 ml blood using the automated software ACCEPT (Supplemental Digital Content fig. 2, illustrating the flow chart of the validation analysis; http://links.lww.com/ALN/C415).³⁰  Overall, the comparison showed a good correlation (Supplemental Digital Content fig. 3 illustrates the correlation between these two methods; http://links.lww.com/ALN/C415). Compared to the ACCEPT software, there was an overestimation of circulating tumor cell counts by 1.66 units with human assessment (Supplemental Digital Content fig. 4 illustrates the agreement between these two methods; http://links.lww.com/ALN/C415). However, in this validation analysis, only samples with high tumor cell counts were considered. This may bias the results toward an overestimation of the difference in means. In other words, if all samples, i.e., including those with 0 to 4 tumor cells/7.5 ml blood, had been included, the difference in means of 1.66 units would have likely been smaller. Second, the overestimation of 1.66 units was nondifferential, i.e., applied to both groups, regardless of treatment assignment.

Apart from one secondary analysis, our findings contrast with numerous previously published studies suggesting better outcomes with the use of intravenous anesthesia. The potential reasons for this disparity are two-fold. First, clinical studies reporting on cancer outcomes were based on retrospective data analyses,^7–14  which are prone to bias and confounding. Second, evidence of a protective effect associated with propofol was partly driven by in vitro studies,^31–35  which may not reflect the delicate interplay between immune and tumor cells observed in vivo. Our findings, however, are consistent with a recently published, large, randomized controlled trial addressing the effect of regional versus general anesthesia on breast cancer recurrence.³⁶  Although this trial was not specifically designed to compare inhalational with intravenous anesthesia, most patients allocated to general anesthesia received sevoflurane, whereas those allocated to regional anesthesia received propofol. In line with our study, this trial failed to show any difference in cancer outcomes.

Our results, however, need to be interpreted with caution. First, we assumed circulating tumor cell counts would be an appropriate prognostic factor to measure the impact of anesthesia on the risk of tumor recurrence, but we did not perform a long-term outcome analysis to confirm this assumption. Although many oncological markers seem to be ideally placed in the causal pathway leading to distant disease, several other factors will eventually be needed to result in metastatic spread, and uncertainty regarding the ability of these prognostic factors to predict “hard endpoints” must be acknowledged.³⁷  A second concern is that the exact meaning of circulating tumor cell changes in the perioperative period remains unclear. In studies investigating the predictive validity of circulating tumor cells changes in primary and metastatic breast cancer, patients converting from “positive” to “negative” status were found to have longer progression-free survival and overall survival than those with a persisting “positive” status.^23,38–42  However, circulating tumor cell detection was performed over many weeks or months, and there is no firm evidence that these findings also apply to the immediate and rather short perioperative period.

Other limitations are inherent to the CellSearch assay itself. Although the pattern EpCAM+/CK+/DAPI+/CD45− is a widely accepted molecular circulating tumor cell signature, other combinations may also occur: it has been argued, for instance, that 7.8 to 10.3% of breast cancers might lack EpCAM expression.^43,44  Further skepticism has been partly related to the fact that for a given tumor, a variety of circulating tumor cells phenotypes seems to exist.⁴⁵  Thus, in some patients included in our study, the ability to detect circulating tumor cells might have been hampered by the technique used. Finally, the in vitro analysis was performed in a sample of 60 patients only, thereby limiting our ability to fully assess the association between natural killer cell–induced apoptosis rates and circulating tumor cell counts. The risk of other sources of bias (such as selection, performance, attrition, and detection bias) was deemed low.

In this randomized controlled trial, we investigated the effect of anesthesia on an independent prognostic factor in primary breast cancer patients. There was no difference in circulating tumor cell counts over time or circulating tumor cell positivity between patients receiving sevoflurane and patients receiving propofol. One secondary analysis suggested a favorable effect of propofol on maximal postoperative circulating tumor cell values. Trials collecting long-term outcomes (NCT02786329, NCT03034096, NCT01975064, and NCT02660411) will bring further evidence regarding the possible effects of anesthesia during cancer surgery.

The authors thank Sabine Kern (Institute of Anesthesiology, University Hospital of Zurich, Zurich, Switzerland) for the coordination of this study, Anja Zabel (Institute of Anesthesiology, University Hospital of Zurich, Zurich, Switzerland) for her help in conducting the laboratory experiments, and Dr. Sarah Haile, Ph.D., (Epidemiology, Biostatistics and Prevention Institute, University of Zurich, Zurich, Switzerland) for her statistical support.

Supported by the Swiss National Science Foundation, the Stiftung zur Krebsbekämpfung (Zurich, Switzerland), the Masikini Foundation (Triesen, Liechtenstein) and the Uniscientia Foundation (Vaduz, Liechtenstein) and by University of Zurich “Protected time for research” and “Filling the Gap” grants (to Dr. Hovaguimian).

Dr. Schläpfer received travel support from Baxter (Volketswil, Switzerland; unrelated to the study). Dr. Dedes received honoraria and consultancies from Roche (Basel, Switzerland), Novartis (Basel, Switzerland), AstraZeneca (Baar, Switzerland), Amgen (Rotkreuz, Switzerland), Tesaro (Zug, Switzerland), PharmaMar (Berlin, Germany), and Daiichi (Thalwil, Switzerland; unrelated to the study). Dr. Tausch received consultancies from Roche (not related to this work). Dr. Beck-Schimmer received a grant from Baxter AG (Deerfield, Illinois; not related to this work), was a participant of an advisory board meeting of Baxter AG (not related to this topic), and received a speaker’s fee from Abbvie (Baar, Switzerland; topic: “Pro/cons of volatile anesthetics”) for a grand round talk in a Swiss Hospital. Dr. Beck-Schimmer also holds Patent 20140100278 for injectable formulation for treatment and protection of patients having an inflammatory reaction or an ischemia-reperfusion event (April 10, 2014) with M. Urner, L. K. Limbach, I. K. Herrmann, and W. J. Stark (applied as Patent Cooperation Treaty internationally, July 2009). The other authors declare no competing interests.

Full protocol available at: beatrice.beckschimmer@uzh.ch. Raw data available at: beatrice.beckschimmer@uzh.ch.