Preoperative beta blockers and other drugs in relation to anastomotic leakage after anterior resection for rectal cancer

Previous research has indicated that preoperative beta blocker therapy is associated with a decreased risk of complications after surgery for rectal cancer. This is thought to arise because of the anti‐inflammatory activity of the drug. These results need to be reproduced and analyses extended to other drugs with such properties, as this information might be useful in clinical decision‐making. The main aim of this work was to replicate previous findings of beta blocker use as a prognostic marker for postoperative leakage. We also investigated whether drug exposure might induce anastomotic leaks.


INTRODUC TI ON
Anastomotic leakage after anterior resection for rectal cancer is a common postoperative complication, ranging from 10% to 20% in population-based cohorts [1,2].Leaks cause morbidity [3,4] and mortality [2] and do not seem to decrease with time [5].Several risk factors for leakage are well established, such as a low anastomosis, male sex, comorbidity, smoking and preoperative radiotherapy [6,7].
The pathophysiology of anastomotic leakage is likely to be multifactorial in nature, while the failed process of wound healing necessarily involves inflammation [8].It is therefore interesting to note a recent registry-based nationwide study suggesting that preoperatively prescribed beta blockers, putatively reducing inflammatory activity by decreasing the activity of the sympathetic nervous system [9], might be associated with a substantial reduction in postoperative anastomotic leaks after rectal cancer surgery [10].While not necessarily causal in nature, such a relationship might prove to be a valuable prognostic marker for anastomotic leakage if the results can be reliably reproduced.To replicate this novel finding, we have conducted a multicentre cohort study with detailed data on anastomotic leakage, thus alleviating misclassification from registry data [11] and also considering delayed leaks.We hypothesize that preoperative use of beta blockers, as well as other common medications potentially affecting inflammatory pathways, are predictive of a lower anastomotic leak rate after anterior resection for rectal cancer.

Checklist for the reporting of observational studies
This article was written in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) checklist for the reporting of observational studies [12].

Study design
This is a retrospective multicentre cohort study conducted in 11 centres in Sweden.Patients operated on with anterior resection for rectal cancer during the years 2014-2018 were identified from theatre lists at each hospital.Chart review was done at each site and data on relevant perioperative variables, including preoperative medication and occurrence of postoperative anastomotic leakage, were entered into the RedCap system (Vanderbilt University, Tennessee, USA).The RedCap system is a secure web application for building and managing online surveys and databases [13].
Linkage was made between the RedCap database and the Swedish Colorectal Cancer Registry (SCRCR), ensuring that all included patients were operated on for rectal cancer as well as providing additional information such as demographics, tumour stage and height, preoperative treatment and other perioperative data.The SCRCR is a national registry established in 1995, including all hospitals that operate for rectal cancer in Sweden [14].The registry defines rectal cancer as an adenocarcinoma of the large bowel within 15 cm of the anal verge, as measured by rigid sigmoidoscopy.The SCRCR has been validated with an average completeness of 99% and overall agreement between registry and re-abstracted variables at 90% for the period 2008-2015 [15].
Nevertheless, some variables such as complications may be underreported to a larger degree [16].This holds true for anastomotic leakage in particular [11], as there is no formal definition of leakage and only leaks diagnosed within 30 days or in hospital are mandatory to register in the SCRCR.

Exposure
The primary exposure was regular preoperative use of any beta blocker at any dose up to the date of surgery.Secondary exposures were preoperative use of metformin, any statin or any angiotensinconverting enzyme inhibitor or angiotensin receptor blocker (ACE/ARB).Postoperative exposure to the above drugs was not ascertained.

Outcome
The primary outcome was any anastomotic leakage within 12 months of the index surgery.Leakage was defined according to the international consensus definition provided by the International Study Group of Rectal Cancer [17].In summary, leakage is defined as a defect of the integrity of the intestinal wall leading to a communication between the intra-and extraluminal compartments.
A pelvic abscess close to the anastomosis also denotes a leakage.
The leaks are further classified: grade A leaks require no change in patient management, grade B leaks require treatment other than laparotomy and grade C leaks require laparotomy [17].In this study, relaparoscopy was treated in the same manner as relaparotomy when assigning leak grade.In accordance with previous research [1], leakage was also classified as early or late, corresponding to a diagnosis within 30 days of surgery and more than 30 days after surgery, respectively.

What does this paper add to the literature?
Previous registry-based data suggest that preoperative beta blocker use might decrease anastomotic leakage after anterior resection for rectal cancer.In this multicentre cohort study with more detailed data, no such association could be detected despite high leak rates, prompting other avenues of research to reduce anastomotic leakage.

Statistical analysis
Frequency tables concerning patient characteristics, tumour stage and operative details were constructed.Continuous variables were described using the median along with the interquartile range (IQR).
The association between beta blocker use and anastomotic leakage within 12 months was analysed using logistic and Poisson regression (with robust standard errors) models, with the main aim of replicating a previous strong association.Covariate set A consisted of the same covariates as the index study [10] and thus included age To estimate the effects of preoperative medication (beta blocker, ACE/ARB, statin or metformin) on anastomotic leakage (including sensitivity analyses for early or late leaks) we also used the targeted maximum likelihood estimation (TMLE) approach [18,19].To determine covariates considered as potential confounders to estimate a total effect, a causal diagram (Figure 1) was drawn [20] from which the following covariates were included in the modelling: CCI, hospital volume, sex, current smoking status and operation year; in sensitivity analyses, CCI was replaced with ischaemic heart disease, heart failure and peripheral vascular disease when evaluating beta blockers, while CCI was replaced with diabetes when evaluating metformin use.TMLE is a two-step procedure: first, an initial regression to fit the expected value of the outcome adjusting for the potential confounders is run; second, the initial outcome regression is updated relying on a fit of the propensity score, i.e. regressing the exposure on the potential confounders, to obtain an optimal biasvariance trade-off for the parameter of interest, i.e. the effect of exposure on outcome.This procedure was repeated for each exposure-outcome pair.The outcome and propensity score regression models were fitted using lasso regression [21] and influence-curvebased standard errors [18] were used to calculate CIs.The results are expressed as relative risks (RRs) together with their 95% CIs.
Multivariate imputation by chained equations was used to handle missing values [22] and the estimates from 10 imputed data sets were pooled according to Rubin's rules [23].All analyses were performed using R 4.1.3statistical software [24].The R-packages tmle [25], glmnet [26] and mice [27] were used for analysis and imputation.Before analysis of data, a power calculation was performed.
Assuming a significance level at 5% and 80% statistical power as well as a 20% and 12% leak rate in patients without and with beta blockers, respectively, and an assumption that a third of all patients would have preoperative beta blockers, a total of 759 patients had to be included.These assumptions were based on a previous registry-based study, albeit with a lower leak rate, although the presumed leakage ratio was retained.In the current study, we assumed that leak rates F I G U R E 1 A directed acyclic graph depicting the assumed relationships between variables involved in the development of anastomotic leakage.Preoperative medication denotes exposure, while leakage denotes outcome.A minimal adjustment set to derive a total effect on the outcome from the exposure consists of the variables sex, smoking, CCI, operation year and hospital (ASA, American Society of Anesthesiologists fitness grade; CCI, Charlson Comorbidity Index; c/pTNM, clinical/pathological tumour-node-metastasis). would be substantially higher, considering a longer follow-up and known underreporting in registry data [11].

RE SULTS Patients
Some 1162 patients underwent anterior resection between 2014 and 2018 and were entered into the RedCap database.Of these, 15 patients were not registered in the SCRCR (missing registrations or no adenocarcinoma), 12 patients were operated on for sigmoid cancer and nine patients underwent transanal total mesorectal excision; these were thus excluded.In the final analysis, a total of 1126 patients remained.In comparison, 1207 patients were registered in the SCRCR at the participating hospitals, and the study cohort thus represented 93% of the potentially eligible population.A study flowchart is presented in Figure 2.
The demographic and clinical variables are presented in Table 1, stratified by the occurrence of preoperative beta blocker exposure; stratification by the other drugs is presented in Appendix A (Tables A1-A3).A total of 255 patients (22.7%) used beta blockers preoperatively.There were more men than women in the beta blocker group.The beta blocker users were also older, had a higher ASA fitness grade and a higher CCI.In addition, current smokers were almost twice as common in the beta blocker group.However, a notably higher share of this group underwent partial mesorectal excision, accompanied by a lower proportion with neoadjuvant therapy and a higher rate of end-to-end anastomosis.Tumour stage, operating time, intraoperative bleeding and BMI were similarly distributed between exposure groups.

Primary exposure: preoperative beta blockers
The anastomotic leakage rate was similar in the group using beta blockers (20.8%) compared with the group without such use (20.5%).
In the main replication analysis when using the same covariate set as the index study [10], neither the logistic (OR 0.94, 95% Cl 0.64-1.38)nor the Poisson (IRR 0.95, 95% CI 0.68-1.33)model resulted in a significant association between preoperative beta blocker use and anastomotic leakage within 12 months.When this covariate set was replaced with other presumably important covariates the results were similar (Table 2).Moreover, the causal analysis also did not

Main results
In this retrospective cohort study of patients subjected to anterior resection for rectal cancer we could not confirm the previous finding [10] of a relationship between preoperative exposure to beta blockers and anastomotic leakage.In a secondary analysis, there was an indication that metformin use might increase the risk of leakage.

Strengths and weaknesses
As this is a multicentre study including patients from county hospitals as well as university hospitals selection bias is limited, although this study is not truly population based.The collaborative nature of the study enabled us to include a fairly large sample size, while still retaining accuracy for the key data points through chart review; the latter is especially important for anastomotic leakage, as this is underreported to a substantial degree [11] and a sizable share of leaks occur outside of the usual 30-day window captured in registry data [28].
There are numerous limitations to this study.As the data collection was initiated using theatre lists instead of identification through the SCRCR there was some under-ascertainment.However, previous reports have suggested that there is a certain degree (4.3%) of incorrect registration of the operative procedure itself in the registry, where permanent stoma procedures are sometimes registered as anterior resections [29]; it is therefore likely that many patients not captured by chart review in this study did not in fact have an anterior resection.The retrospective design precluded uniform detection and diagnostic protocols of anastomotic leakage, and it was not possible to gather detailed data on drug exposure before admission for surgery; moreover, some of the drug exposure could have been discontinued postoperatively, potentially no longer affecting anastomotic breakdown.Withdrawal of at least beta blockers is not recommended postoperatively [30], and most of the participating centres adhered to this principle.However, we do not know the discontinuation rate of each medication postoperatively, although one might assume that beta blockers were more seldom withdrawn compared with the other drugs as, for example, ACE/ARBs and metformin are commonly temporarily discontinued after surgery.Such a policy was common among the participating hospitals, albeit patientlevel data could not be ascertained.
While our main aim was to replicate previous findings of beta blocker use as a prognostic marker for postoperative leakage, we also investigated whether drug exposure might induce anastomotic leaks.When employing observational data, the latter type of analysis assumes that the causal mechanisms can be accurately depicted, which was attempted using a directed acyclic graph [20].The causal analyses must thus be interpreted cautiously as we cannot exclude influences from unknown mechanisms inducing residual confounding.Nevertheless, the causal analyses resulted in similar estimates to the analyses using covariates selected beforehand.

Literature review
This is the second retrospective study to investigate the relationship between beta blockers and anastomotic leakage after anterior resection for rectal cancer [10].Strikingly, the present study had a much higher leak rate than the earlier one (20.6% vs. 4.8%).The study by Ahl et al. captured leaks registered within 30 days or in-hospital registered in the SCRCR.In the present study, however, leak rates within 30 days, termed early leaks, were also considerably higher (14.3%).In comparison, in publicly available data from the SCRCR for the years 2007-2016, included in Ahl et al, a leak rate of 8.5% is reported [31].

Biological mechanism/implications
Earlier research has indicated that operative trauma triggers a catecholamine response [32,33], while beta blockers have been shown to downregulate catecholamine-induced cytokine production and attenuate proinflammatory responses [34,35].Although the healing of a colorectal anastomosis is not fully understood, it necessarily involves an inflammatory phase.Thus, there is a theoretical possibility that beta blockers could reduce leak rates by reducing the inflammatory response, but we could not find any corroboration for this hypothesis in our data.In a secondary analysis, a relationship between metformin use and increased risk of leakage was found.To the authors' knowledge, no other such studies have been conducted and this might represent a novel finding.Metformin is thought to have anti-inflammatory properties independent of glycaemic control and decreases cardiovascular disease in diabetic patients [36].
Any impact on leak rates is nevertheless speculative, and the apparent effect may rather be explained by the underlying indication for the drug, namely diabetes mellitus, which has consistently been shown to be associated with anastomotic leakage [37,38].
Judging from the research available thus far on preoperative drugs in general and beta blockers in particular, there does not seem to be enough evidence to state that any investigated drug either predicts or causes anastomotic leakage after anterior resection for rectal cancer.More research is certainly warranted, ideally including more detailed data on the exposure, such as type of substance, dosage, length of exposure and indication for medication use; moreover, perioperative withdrawal patterns need to be studied as well.

CON CLUS ION
While previous research suggested that preoperative beta blocker use could be prognostic of anastomotic leakage, this study could not detect any such association.On the contrary, these results indicate that preoperative beta blocker use neither predicts nor causes anastomotic leakage after anterior resection for rectal cancer.

(
continuous), body mass index (BMI; continuous), sex (male or female), the American Society of Anesthesiologists (ASA) fitness grade (I, II or III-IV), surgical technique (open or minimally invasive), clinical tumour stage (I, II, III or IV), pathological tumour category(T1-2, T3 or T4), pathological node category (N0, N1-2 or NX) and clinical metastasis category (M0 or M1).Covariate set B consisted of covariates chosen beforehand, based on the available literature on anastomotic leakage, and included the Charlson Comorbidity Index (CCI; 0, 1, 2 or 3 and above), individual hospital (categorical), sex, current smoker (no or yes), age, ASA fitness grade, BMI, neoadjuvant therapy (none, radiotherapy or chemoradiotherapy), defunctioning stoma (no or yes), clinical tumour stage, type of mesorectal excision (total or partial) and operation year (categorical).In a post hoc analysis, this set of covariates replaced CCI with ischaemic heart disease (yes or no), heart failure (yes or no) and peripheral vascular disease (yes or no), to form a covariate set C. Results are reported as odds ratio (OR) and incidence rate ratio (IRR) with corresponding 95% confidence intervals (CIs).
show an effect of beta blocker use on anastomotic leak (RR 0.98, F I G U R E 2 Study flowchart (SCRCR, Swedish Colorectal Cancer Registry).TA B L E 1 Clinical and demographic data for 1126 patients (patients with missing covariate values included), stratified on preoperative beta blocker use.Anastomotic leakage, n (%) Corresponding data for 2014-2018 from the SCRCR showed a nationwide leak rate of 8.4%, while the 11 hospitals in the present study reported a rate of 9.3% in the registry.The discrepancy between the present study, Ahl et al. and the SCRCR data might partially be explained by differences in study design, where the chart review and the longer follow-up in the present study reduce misclassification and capture more leaks, respectively.It has also been shown that the SCRCR seems to underreport anastomotic leaks by 41% when not utilizing all available auxiliary data and by 29% when such data are used[11].The other major difference concerns the capture of drug exposure, where registry data were used to ascertain any prescribed drug within a 12-month period before surgery in Ahl et al. while the current study employed chart data, typically from the admission just before surgery; nevertheless, beta blocker use was quite similar (29% vs. 23%).The main drawback of the present study relates to study size, as it is about a quarter of the size of the previous one; however, there was still ample statistical power to prove a difference as large as was shown before, as the outcome anastomotic leakage was decidedly more common in the current cohort.
Association between preoperative beta blocker use and anastomotic leakage within 12 months.Effect of medication on anastomotic leakage in a causally oriented analysis (based on 10 imputed datasets).
[10] L E 1 (Continued)TA B L E 2Note: Logistic and Poisson regression results (based on 10 imputed datasets) for n = 1126 patients, showing odds ratios (ORs) and incidence rate ratios (IRRs) with 95% confidence intervals (CIs).Analysis similar to that in Ahl et al.[10], with three different covariate sets as follows.Set A: age, body mass index, sex, American Society of Anesthesiologists fitness grade, surgical technique, clinical tumour stage, pathological tumour category, pathological node category and pathological metastasis category.Set B: Charlson comorbidity index group, American Society of Anesthesiologists fitness grade, individual hospital, sex, current smoking status, age, body mass index, neoadjuvant therapy, defunctioning stoma, clinical tumour stage, type of mesorectal excision, operation year.Set C: ischaemic heart disease, heart failure, peripheral vascular disease, American Society of Anesthesiologists fitness grade, individual hospital, sex, current smoking status, age, body mass index, neoadjuvant therapy, defunctioning stoma, clinical tumour stage, type of mesorectal excision, operation year.TA B L E 3Note: targeted maximum likelihood estimation with propensity score and outcome models estimated by lasso regression, showing relative risks (RRs) with 95% confidence intervals (CIs).Abbreviation: ACE/ARB, angiotensin converting enzyme inhibitor/angiotensin receptor blocker.a Adjustment for confounding using Charlson Comorbidity Index, hospital volume, sex, current smoking and operation year.
Clinical and demographic data for 1126 patients (patients with missing covariate values included), stratified on metformin.
TA B L E A 2 (Continued) TA B L E A 2 (Continued) TA B L E A 3 node category, n (%) TA B L E A 3 (Continued)