Risk of Postoperative Surgical Site Infections after Vedolizumab versus Anti-Tumor Necrosis Factor Therapy: A Propensity-Score Matching Analysis in Inflammatory Bowel Disease

Abstract

Background

Perioperative vedolizumab (VDZ) and anti-tumor necrosis factor (TNFi) therapies are implicated in causing postoperative complications in inflammatory bowel disease (IBD).

Aim

To compare the risk of surgical site infections (SSIs) between VDZ- and TNFi-treated IBD patients in propensity-matched cohorts.

Methods

The Optum® Research Database was used to identify IBD patients who received VDZ or TNFi within 30 days prior to abdominal surgery between January 2015 and December 2016. The date of IBD-related abdominal surgery was defined as the index date. SSIs were determined by ICD-9/10 and CPT codes related to superficial wound infections or deep organ space infections after surgery. Propensity score 1:1 matching established comparable cohorts based on VDZ or TNFi exposure before surgery based on evidence-based risk-modifiers.

Results

The propensity-matched sample included 186 patients who received preoperative biologic therapy (VDZ, n=94; TNFi, n=92). VDZ and TNFi cohorts were similar based on age, gender, IBD type, concomitant immunomodulator exposure, chronic opioid or corticosteroid therapy, Charlson Comorbidity Index, and malnutrition. VDZ patients were more likely to undergo an open bowel resection with ostomy. After propensity score matching, there was no significant difference in postoperative SSIs (TNFi 12.0% vs. VDZ 14.9%, P=0.56). Multivariable analysis indicated that malnutrition was the sole risk factor for developing SSI (OR 3.1, 95% CI 1.11, 8.71) regardless of the type of biologic exposure.

Conclusion

In the largest, risk-adjusted cohort analysis to date, perioperative exposure to VDZ therapy was not associated with a significantly higher risk of developing an SSI compared to TNFi therapy.

INTRODUCTION

Anti-tumor necrosis factor-α (TNFi) agents have traditionally been the biologic therapies of choice for moderate-to-severe inflammatory bowel diseases (IBD), consisting of ulcerative colitis (UC) and Crohn’s disease (CD). While the use of TNFi has effectively improved health outcomes such as corticosteroid dependency in patients with IBD, the risk of IBD-related abdominal surgery remains high in patients with medically refractory disease and those who lose their response to TNFi agents.¹ Since the natural history of IBD is that of relapse and remittance, a significant proportion of patients with IBD rely on complementary or curative surgery as a cornerstone treatment option for CD or UC.^2,3,4,5

Considering the need to optimize surgical outcomes in select patients, decisions about preoperative pharmacotherapy require judicious medical care and evidence-based recommendations. Although there is no clear consensus⁶ regarding TNFi exposure prior to major abdominal surgery,⁷ research supports the relationship between preoperative TNFi use and increased risk for surgical complications.^8,9,10

Similar to TNFi, preoperative use of vedolizumab (VDZ) has more recently been implicated as a significant predictor of worse postoperative outcomes, including surgical site infections (SSI).^{11,12,13,14,15,16,17} VDZ is a fully humanized monoclonal antagonist targeting α4-β7 integrin receptors of the gastrointestinal tract.^18,19 It received FDA-approval in May 2014 for moderate-to-severe UC and CD.^20,21,22 Considering VDZ’s gut selectivity, the safety profile has been shown to be favorable.²³ One study using single-center data showed that VDZ exposure was not associated with postoperative complications,²⁴ including a recent meta-analysis.²⁵ However, since VDZ’s mechanism of action blocks leukocyte migration into the intestinal lining,²⁶ wound and anastomotic healing after IBD-related surgery remains a potential concern.²⁷ If differences in mechanism of action for various biologics affect postoperative outcomes in IBD, methods to optimize pharmacotherapy options and surgical strategies remain a major area for future study.

Patients with IBD are at higher risk for infectious postoperative complications because of their underlying disease, malnutrition, and immunosuppression. SSIs include superficial wound infections, deep wound infections affecting the fascia, and organ space infections including abdominal and pelvic abscesses and anastomotic leaks²⁸. The sequelae of these infections can be quite severe. In the short term, patients may develop life-threatening sepsis, require reoperations²⁹ or procedures for abscess drainage,^30,31 suffer from prolonged ileus or prolonged hospital stay, and require hospital readmission.³² Long-term effects include the need for an ostomy, development of hernias,³³ and poor quality of life due to poor bowel function.^34,35

Considering the devastating consequences of SSIs in this population, judicious and evidence-based selection of pharmacotherapy for patients at high risk for surgical intervention is critical. Given the clinical equipoise from published studies and the mechanistic concerns for postoperative SSI after VDZ exposure, we aimed to compare the risk of postoperative SSI between VDZ- and TNFi-treated patients in propensity-matched cohorts using a population-based dataset.

METHODS

Data Source

We performed a longitudinal retrospective cohort analysis of IBD patients in the Optum Research Database from years 2015 through 2016. The Optum database contains de-identified inpatient, outpatient, and pharmaceutical claims data as well as laboratory data from approximately 12–14 million privately insured patients per year across 50 states. The database includes demographic and socioeconomic characteristics (e.g., age, sex, geographic region, race, household income), encounter data (i.e., hospital admissions, outpatient visits, and associated procedures), pharmaceutical data (i.e., filled pharmaceutical claims, days supply, dose dispensed, strength, administration method), financial data (i.e., total cost, copayment, deductibles), and lab results (e.g., test description, result number and unit). The Stanford University Institutional Review Board (IRB) determined this study does not meet the definition of human subject research as defined in federal regulations 45 CFR 46.102. The study was therefore exempt from IRB review.

Cohort Identification and Classification

We selected a cohort of patients with: (1) at least one IBD-related inpatient or outpatient visit between 2015 and 2016, (2) at least one claim for VDZ or TNFi in 2015 or 2016 within 30 days preceding an abdominal surgery, and (3) continuous enrollment in Optum for at least six months preceding surgery. To classify each patient as CD, UC, or indeterminate colitis (IC), the total number of distinct encounters including CD and UC International Classification of Disease (ICD-9 and ICD-10) codes were summed (555.xx and K50.xx for Crohn’s disease, 556.xx and K51.xx for ulcerative colitis). Patients were classified as CD or UC if at least 80% of the ICD codes were 555.xx/K50.xx or 556.xx/K51.xx, respectively. The patient was classified as having IC if neither of the CD nor UC conditions were fulfilled. Patients were classified as receiving either VDZ or TNFi based on outpatient pharmaceutical claims or Healthcare Common Procedure Coding System (HCPCS) J-codes. TNFi included infliximab, adalimumab, certolizumab, and golimumab. Patients in the VDZ group and TNFi group were excluded if they had additionally taken a TNFi or VDZ, respectively, in the thirty days preceding surgery (3.4% for VDZ group and none for TNFi group). Surgical procedures were characterized as either laparoscopic or open, and by whether a primary bowel anastomosis or end ostomy was performed. Charlson comorbidity score was calculated using inpatient and outpatient claims from six months preceding surgery.³⁶ Use of corticosteroids was identified by at least three corticosteroid refills covering at least 100 days in 2016.

Outcomes

The primary outcome of interest was the occurrence of postoperative SSI, including superficial wound infections and deep organ space infections, within 30 days after abdominal surgery (CPT codes are included in Supplementary Table 1). The primary predictor of interest was VDZ infusion within 30 days before surgery compared to TNFi prescription within 30 days before surgery. This 30 day preoperative window defining “drug exposure” is consistent with the published pharmacokinetics of VDZ.³⁷ Clinically, this time frame is consistent with patients’ real-world TNFi or VDZ maintenance dosing frequency (every four weeks) if they were not responding to biologic therapy prior to surgery.

Propensity Score Matching

As patients prescribed VDZ may have already failed first-line biologic therapies (suggestive of more severe disease than patients prescribed anti-TNF agents alone), we performed propensity score matching to establish comparable cohorts using risk-modifying clinical determinants. Propensity scoring helps adjust for confounders resulting from baseline characteristics of the patients. Patients were matched by a Mahalanobis matching algorithm,³⁸ based on age, gender, IBD type (UC, CD, IC), Charlson comorbidity score, diagnosed malnutrition (as captured by ICD-9 and ICD-10), and use of immunomodulators (i.e., small molecules used in severe colitis including azathioprine, 6-mercaptopurine, methotrexate sodium, ciclosporin, or tacrolimus within 30 days of surgery), corticosteroids, and opioids within 30 days before surgery. We utilized 1:1 matching with no replacement, and a maximum caliper of 0.007. Propensity score matching was implemented using the psmatch2 module in Stata/SE 14.2 (College Station, TX). Comparability of baseline characteristics between matched groups was assessed using chi-square tests for categorical variables and student’s independent sample t-tests for continuous variables.

Mulitvariable Model

A multivariable logistic regression model was calculated to estimate the association between preoperative VDZ use and SSIs. Covariates included age, sex, type of surgery performed, immunomodulator use, corticosteroid dependency (≥100 days of corticosteroid use³⁹ per year), and malnutrition. Statistical significance was assessed at the level of α=0.05. All analyses were performed using Stata/SE 14.2 (College Station, TX).

RESULTS

Patient Cohort Identification & Characteristics

The data for 1,933 and 522,837 IBD patients receiving VDZ or TNFi, respectively, were retrieved from intpatient and outpatient claims dataset in Optum. Of these, 573,484 patients were excluded as they did not undergo surgery or had VDZ or TNFi exposure outside the 30-day abdominal surgery date. Ultimately, 94 VDZ and 1,725 TNFi IBD patients met criteria for study inclusion. After propensity score matching, we identified and compared 92 VDZ-exposed patients with 94 TNFi-exposed patients, creating a cohort of 186 patients. Figure 1 illustrates the patient cohort identification process.

Figure 1.

Patient Cohort Identification

Table 1 displays the patient characteristics of VDZ and TNFi patients. After propensity matching, there were no statistically significant differences in patient demographics, operative characteristics, and predictors of post-operative complications. The proportion of laparoscopic bowel resections with ostomies and the mean number of days of cumulative steroid use in 2016 were greater among patients using VDZ. The proportion of patients with open bowel resection with ostomy, laparoscopic bowel resection with anastomosis, open bowel resection with anastomosis, and the use of corticosteroids within 30 days of surgery were greater among patients using TNFi. These differences were not statistically significant.

Table 1.

Comparison of Anti-TNF vs VDZ Cohorts after Propensity Score Matching

Variables	Anti-TNF Patients	Vedolizumab Patients	P-value
Age, years (mean)	37.8	39.0	0.569
Sex, female (%)	41.3%	43.6%	0.751
Disease type
Ulcerative Colitis (%)	41.3%	43.6%	0.828
Crohn’s Disease (%)	55.4%	48.9%
Indeterminate Colitis (%)	3.3%	7.5%
Operative characteristics
Laparoscopic bowel resection with ostomy	21.7%	30.9%	0.160
Open bowel resection with ostomy	32.6%	30.9%	0.798
Laparoscopic bowel resection with anastomosis	10.9%	7.5%	0.421
Open bowel resection with anastomosis	34.8%	30.9%	0.570
Systemic corticosteroids
Within 30 days surgery (%)	52.1%	45.8%	0.383
Cumulative steroid days (mean)	158.7	167.5	0.743
Immunomodulator (%)	10.9%	13.3%	0.542
Chronic Opioid Use (%)*	63.0%	61.6%	0.737
Charlson Comorbidity Index (mean, median)	1.4, 1	1.7, 1	0.257
Malnutrition (%)	50.0%	52.1%	0.773
Hemoglobin (g/dL) (mean) [median]	12.7 [12.6]	12.3 [12.4]	0.454
Albumin (g/dL) (mean) [median]	3.8 [4.0]	3.5 [3.9]	0.503

^*Chronic opioid use was defined ≥3 separate opioid drug claims on distinct dates within a 2-year rolling window, as previously shown.⁴⁹

Comparison of SSI Risk Based on TNFi vs VDZ Exposure

The probability of SSI within 30 days post-operatively was not statistically different between the VDZ versus the TNFi groups (VDZ 14.9% versus TNFi 12.0%, P=0.56) (Figure 2). The risk of postoperative SSI in the first 10 days was lower in the TNFi group, but the difference was not statistically significant. Of those who ultimately developed SSI within 30 days, 10 of the 14 VDZ patients (71.4%) compared to 5 of the 11 (45.5%) TNFi patients had SSIs in the first 10 days (Figure 3). There was no significant difference in the rates of deep organ space versus superficial SSIs between VDZ and TNFi groups (p=0.17).

Figure 2.

Comparison of SSI Risk Between TNFi vs VDZ

Figure 3.

KM Plot Comparing SSI Risk Over 30-days

Malnutrition is an Independent Risk Factor for SSI

Multivariable analysis showed malnutrition as the sole significant predictor for post-operative SSI (OR 3.1, 95% CI 1.1–8.7, P= 0.03) (Table 2). None of the other variables, including VDZ exposure (OR 1.8, 95% CI 0.7–3.7, P= 0.25), being female (OR 1.1, 95% CI 0.4–3.1, P= 0.8), or having open surgery (OR 1.4, 95% CI 0.5–4.2, P= 0.51), were associated with increased risk for SSI. Exposure to systemic corticosteroids prior to surgery trended towards increased risk but was not statistically significantly associated with SSI (OR 1.9, 95% CI 0.7–5.0, P= 0.2). In a post-hoc analysis, we modified the model to evaluate CD and UC separately. VDZ exposure did not increase SSI in either UC or CD patients, while malnutrition remained a significant predictor of SSI in UC and corticosteroids became a significant predictor of SSI in CD.

Table 2.

Assessment of Individual Risk Factors for SSI

Variables	OR	95% CI	P-value
Vedolizumab	1.785	[0.671, 4.747]	0.246
Age	0.995	[0.965, 1.027]	0.776
Sex, female	1.141	[0.425, 3.058]	0.794
Operative characteristics
Open (vs. laparoscopic)	1.441	[0.493, 4.216]	0.505
Ostomy (vs. anastomosis)	0.881	[0.321, 2.415]	0.805
Concomitant Immunomodulators	0.765	[0.184, 3.185]	0.713
Corticosteroid dependency*	1.914	[0.728, 5.029]	0.188
Malnutrition	3.110	[1.110, 8.713]	0.031*

^*≥100 days of corticosteroid use in 1 year

DISCUSSION

Postoperative infectious complications in patients undergoing surgery for IBD-related indications are a major cause of morbidity and increased healthcare resource utilization.^31–33 The effects of TNFi or VDZ exposure perioperatively on the risk of infectious complications remains unclear.^6–14 Understanding the differences in operative risk posed by different biologic agents has become increasingly important, given the numerous competing therapy options after first-line treatment failure. In the largest risk-adjusted cohort analysis to date, we sought to answer this clinically relevant question.

Through propensity-score matching using potential predictors for postoperative infectious complications, we found that SSI risk was not statistically different between TNFi and VDZ exposed patients. Furthermore, in multivariable analysis, malnutrition was the only significant predictor of SSI among multiple well-established risk factors including age, surgical approach, presence of an ostomy, concomitant immunomodulatory use, and corticosteroid exposure. Patients who were classified as having malnutrition had over a three-fold increased risk of SSI as a complication.

Our study findings have significant clinical implications, demonstrating the need to optimize nutrition in patients with IBD, who are often chronically malnourished during states of prolonged disease exacerbation. Although the prevalence of malnutrition in IBD by ICD coding from a representative inpatient cohort of IBD patients is 5–10%,⁴⁰ the probability of malnutrition just prior to IBD-related surgery is likely much higher.⁴¹ Previous literature suggests that more than half of all patients may meet criteria for malnutriton at the time of IBD diagnosis.⁴² Furthermore, it is likely that malnutrition especially in acute care settings may be under-reported and not diagnosed in patients with IBD.^43,44 This may be more apparent in patients who have been hospitalized for surgical management of a primary diagnosis.^45,46 Therefore, we recommend preoperative measures to optimize nutrition including supportive enteral feeds or parenteral nutrition as indicated.

Exposure to corticosteroids in the perioperative period is a potential risk factor for surgical site infections, although there is some debate in the literature about its effect.^47,48 In this study, we found that perioperative corticosteroid exposure was associated with a greater risk for surgical site infections in patients with Crohn’s disease. Patients exposed to corticosteroids in the perioperative period are more likely to have steroid-refractory disease, so this may also be an indicator of greater clinical disease severity.

The limitations of this study are associated with the inherent limitations of insurance claims data. Ascertaining clinical status such as disease severity requires the use of surrogate markers of illness such as presence of diagnosis codes for malnutrition, lab values for hemoglobin and albumin, and the need for surgical intervention. Underreporting of comorbid conditions such as malnutrition could lead to falsely increase the effect on SSI risk. A potential weakness of our retrospective claims analysis is the reliance on data which may not adequately capture variables like malnutrition. Regardless, propensity matching served as a method of likely balancing any over- or under-estimation of malnutrition in the two groups. Another limitation of the dataset is the ability to determine cause of deep organ space infections (intra-abdominal abscesses). These infections may be due to anastomotic leaks, leaks from a Hartmann’s pouch (rectal stump), inadvertent injury to bowel, or contamination from the original operation or underlying condition, but these details are not available in insurance claims data. Finally, although we report on the largest risk-adjusted cohort to date comparing VDZ to TNFi exposure on post-operative complications, our final cohort size remains relatively small. As vedolizumab use becomes increasingly common in clinical practice, smaller differences undetected in this study may become apparent.

In conclusion, preoperative vedolizumab use does not increase postoperative SSI risk as compared to preoperative TNFi treatment. However, a chart diagnosis of malnutrition is a significant predictor of SSI. While a diagnosis of malnutrition certainly indicates severe overall disease, further prospective research is needed to determine whether this risk can be modified with perioperative interventions to address nutritional status.

Supplementary Material

Supp TableS1

Supplementary Table 1. Superficial Wound Infections and Deep Organ Space Infections