Objectives The impact of immunosuppression on postoperative outcomes has primarily been studied in patients undergoing joint replacement surgery. We aimed to evaluate the impact of biologics and glucocorticoids on outcomes after other major surgeries.
Methods This retrospective cohort study used Medicare data 2006–2015 to identified adults with rheumatoid arthritis undergoing hip fracture repair, abdominopelvic surgery (cholecystectomy, hysterectomy, hernia, appendectomy, colectomy) or cardiac surgery (coronary artery bypass graft, mitral/aortic valve). Logistic regression with propensity-score-based inverse probability weighting compared 90-day mortality and 30-day readmission in patients receiving methotrexate (without a biologic or targeted synthetic disease-modifying antirheumatic drug (tsDMARD)), a tumour necrosis factor inhibitor (TNFi) or a non-TNFi biologic/tsDMARD <8 weeks before surgery. Similar analyses evaluated associations between glucocorticoids and outcomes.
Results We identified 10 777 eligible surgeries: 3585 hip fracture, 5025 abdominopelvic and 2167 cardiac surgeries. Compared with patients receiving methotrexate, there was no increase in the risk of 90-day mortality or 30-day readmission among patients receiving a TNFi (mortality adjusted OR (aOR) 0.83 (0.67 to 1.02), readmission aOR 0.86 (0.75 to 0.993)) or non-TNFi biologic/tsDMARD (mortality aOR 0.78 (0.49 to 1.22), readmission aOR 1.02 (0.78 to 1.33)). Analyses stratified by surgery category were similar. Risk of mortality and readmission was higher with 5–10 mg/day of glucocorticoids (mortality aOR 1.41 (1.08 to 1.82), readmission aOR 1.26 (1.05 to 1.52)) or >10 mg/day (mortality aOR 1.64 (1.02 to 2.64), readmission aOR 1.60 (1.15 to 2.24)) versus no glucocorticoids, although results varied when stratifying by surgery category.
Conclusions Recent biologic or tsDMARD use was not associated with a greater risk of mortality or readmission after hip fracture, abdominopelvic or cardiac surgery compared with methotrexate. Higher dose glucocorticoids were associated with greater risk.
- rheumatoid arthritis
- DMARDs (biologic)
- orthopaedic surgery
Statistics from Altmetric.com
What is already known about this subject?
Studies of the risk of immunosuppression in patients undergoing surgery most commonly involve patients undergoing elective arthroplasty, but few studies have evaluated other major surgery.
What does this study add?
In this study, patients with rheumatoid arthritis receiving biologics did not have a greater risk of postoperative infection after hip fracture, abdominopelvic or cardiac surgery compared with patients receiving methotrexate without a biologic.
Glucocorticoids were associated with a dose-dependent increase in the risk of postoperative infection.
How might this impact clinical practice or future developments?
Prolonged interruptions in biologics before major surgery is likely not required, but minimising glucocorticoids before surgery should be a focus of perioperative management.
Patients with rheumatoid arthritis (RA) and other autoimmune conditions are frequently treated with immunosuppression, which can increase the risk of infection. Infections are of particular concern in patients undergoing major surgery, in which postoperative complications such as pneumonia and wound infections can have substantial morbidity.
Perioperative immunosuppression management has been studied primarily in patients with RA undergoing arthroplasty. Patients with RA are at higher risk of postoperative infections after arthroplasty, with disease activity, comorbidities and immunosuppression all potentially contributing.1–5 Conventional disease-modifying antirheumatic drugs (csDMARDs) such as methotrexate have not been associated with increased postoperative risk.6 The risk with biological therapies remains uncertain, although studies have suggested that stopping a biologic before arthroplasty does not reduce infection risk and have shown no differences in risk across different biologics.7–9 Several studies, however, have demonstrated increased infection risk and readmission after arthroplasty with even modest glucocorticoid doses.1 4 10–12
Few studies, however, have evaluated the contribution of specific immunosuppressive therapies to outcomes after other types of surgery.13 14 These surgeries have different postoperative risks than arthroplasty and are often non-elective, with no opportunity for holding medications before surgery.
The goal of this study was to evaluate associations between biologic use and adverse outcomes in patients with RA undergoing common major surgical operations other than arthroplasty, including hip fracture, abdominopelvic and cardiac surgeries. A secondary goal was to evaluate associations between glucocorticoids and adverse outcomes.
Using Medicare claims data 2006–2015, we identified adults≥18 years old with RA based on ≥2 International Classifications of Diseases (ICD-9) provider diagnoses ≥7 days apart, use of a DMARD, and absence of psoriatic arthritis, ankylosing spondylitis and inflammatory bowel disease diagnoses in the year before surgery. Included patients underwent one of the following surgeries between 1 January 2007 and 31 August 2015 based on primary ICD-9 procedure codes from inpatient hospitalisations using established definitions: hip fracture repair, abdominopelvic surgeries (cholecystectomy, hysterectomy, hernia surgery, appendectomy, colectomy for diverticular disease) and cardiac surgeries (coronary artery bypass graft (CABG), mitral or aortic valve surgery). We required that surgery occur by hospital day 3 for hip fracture, hysterectomy, hernia surgery and appendectomy and hospital day 7 for the other surgeries to avoid surgeries in which an outcome may have occurred prior to surgery or in which the surgical procedure might be the results of a complicated hospitalisation, rather than the main indication for hospitalisation. More than 90% of surgeries fell within these windows. We also applied other restrictions to ensure that surgeries were performed for specific indications and were not part of a larger procedure (online supplementary table S1). We further subdivided certain applicable surgeries (cholecystectomy, hernia, colectomy, CABG and valve surgery) into elective and non-elective surgery, as postoperative risks are different after elective and non-elective surgery. All elective surgeries were required to occur by hospital day 2 with no admission through the emergency department, transfer from another acute care hospital or admission status of ‘emergent’. Hip fracture repair and appendectomy were all considered non-elective, and all included hysterectomies were elective.
We required 1 year of continuous enrolment in Medicare A, B and D prior to surgery (baseline; online supplementary figure S1). We excluded patients with a diagnosis of malignancy (except non-melanoma skin cancer) or HIV in the past year, patients receiving multiple different biologics/targeted synthetic DMARDs (tsDMARDs) within 6 months before surgery and patients with another major surgery within 6 months. Patients could contribute multiple surgeries if >6 months apart.
To evaluate associations between immunosuppression use and postoperative outcomes, we classified patients in mutually exclusive groups: (1) methotrexate prescription fill <8 weeks before surgery with no biologic/tsDMARD <6 months before surgery (to avoid recent biologic discontinuations), (2) TNFi prescription/infusion<8 weeks before surgery with or without methotrexate, or (3) non-TNFi biologic prescription/infusion or tsDMARD prescription fill <8 weeks before surgery (16 weeks for rituximab), with or without methotrexate. All groups could receive other csDMARDs (hydroxychloroquine, sulfasalazine or leflunomide).
We separately evaluated associations between glucocorticoids and outcomes. Glucocorticoid dose in the 90 days prior to surgery was based on oral prescriptions for prednisone, prednisolone and methylprednisolone, using prescribed dose in prednisone equivalents and days supply to determine each daily dose and truncating prescriptions if a new prescription was filled before the prescription end date. To account for differences between prescription instructions and patient behaviour (eg, a 30-day prescription taken over 90 days), we averaged the daily dose over 90 days, categorising the average daily dose as ≤5 mg, >5 to 10 mg or >10 mg/day).
In secondary analyses, rather than compare biologic users to non-users, we directly evaluated the timing of biologics before surgery similar to previous studies.7 8 Here, we evaluated only patients receiving the two most common infusions (infliximab or intravenous abatacept), since the timing of treatment for infusions can be precisely determined in this data source. We evaluated whether the time between the last infusion and surgery (<4 weeks, ≥4 to <8 weeks or ≥8 weeks to 6 months) was associated with different risk of postoperative outcomes.
The primary outcomes of interest were (1) mortality within 90 days after surgery (including the index hospitalisation) or (2) readmission to an acute care hospital within 30 days after discharge. Readmission was identified among patients discharged to home, home healthcare, acute rehabilitation or a skilled nursing facility, considering rehospitalisations within 1 day of discharge to be continuations of the initial hospitalisation and not counting hospitalisations with a primary diagnosis indicating rehabilitation.15 Reasons for readmission were based on primary discharge diagnosis codes classified using Chronic Conditions Warehouse software.16 17
Secondary outcomes included postoperative pneumonia and wound complications18 based on diagnosis codes from the index hospitalisation or any acute care hospitalisations within 30 days after surgery (online supplementary table S1) because of specific concerns about these outcomes with biologics and glucocorticoids.19–21
Covariates measured at baseline included demographics, zip code-based median household income from the American Community Survey 2009–2013,22 disability status, skilled nursing facility residence, surgery type, admission type (elective, urgent, emergent) and hospital day of surgery. We identified comorbid conditions, an adaption of the Charlson Comorbidity Index,23 and measures of healthcare utilisation (outpatient visits, emergency department visits, hospitalisations and hospitalised infection) during the baseline period. Medication use based on filled prescriptions≤90 before surgery days included non-steroidal anti-inflammatories, opioids, non-methotrexate csDMARDs and antibiotics. Number of previous biologics was determined using all data prior to the surgery date (mean 4 years).
Associations between exposures and outcomes were evaluated using logistic regression using propensity score-derived inverse probability weights to balance all measured covariates across exposure groups, as described below. Cluster robust standard errors accounted for patients contributing multiple surgeries.24 Analyses were first conducted in the overall population and then reanalysed (with recalculated propensity scores) separately for each surgery category (hip fracture, abdominopelvic and cardiac).
Propensity scores were generated based on the probability of being in each treatment group (methotrexate, TNFi, non-TNFi biologic/tsDMARD) using multinomial logistic regression models with all covariates of interest including a squared term for age (to account for nonlinearity) and average glucocorticoid dose before surgery (included variables in online supplementary table S2).10 25 26 Propensity scores were used to create stabilised inverse probability treatment weights truncated at the 1st and 99th percentile.26–29 Balance of covariates across treatment categories was assessed compared with the reference group (methotrexate), with standardised mean difference ≤0.1 indicating good balance. Unbalanced covariates were added as covariates to weighted models. Propensity scores and stabilised inverse probability treatment weights were recalculated in analyses with readmission as the outcome and in analyses of specific surgery categories. Discharge status was considered an intermediate variable and was not included in readmission propensity scores. To address the possibility that methotrexate-treated patients with previous biologic treatment >6 months before surgery were more ill, we conducted a sensitivity analysis excluding patients with previous biologic use.
The same method was used to generate inverse probability treatment weights in glucocorticoid analyses, also including methotrexate use, biologic use and time of last biologic or methotrexate before surgery in propensity score models.
In secondary analyses, associations between infliximab timing or intravenous abatacept timing and outcomes were assessed using multivariable logistic regression models with potential confounders included as covariates.
Analyses were performed using Stata V.15.1 (StataCorp). Preliminary results were previously presented.30
We identified 10 777 surgeries among 10 483 patients meeting inclusion and exclusion criteria (figure 1). Mean age was 72 and 80% of patients were female (table 1, online supplementary table S2). RA treatment before surgery was methotrexate without a biologic/tsDMARD in 6194 (57%), TNFi in 3523 (33%) (709 adalimumab, 88 certolizumab, 972 etanercept, 46 golimumab, 1708 infliximab), and non-TNFi biologic/tsDMARD in 1060 (10%) (701 abatacept, 203 rituximab, 127 tocilizumab, 29 tofacitinib). Surgery type was hip fracture repair in 33%, abdominopelvic surgery in 47% (most commonly cholecystectomy) and cardiac surgery in 20%. Patients receiving a TNFi or non-TNFi biologic/tsDMARD were younger, had more outpatient visits, were more likely to have received prior biologics or be disabled and were less likely to have been in a skilled nursing facility. After applying inverse probability weights, patient characteristics were well balanced between methotrexate and TNFi groups with only slight residual imbalance in the non-TNFi biologic/tsDMARD group (online supplementary table S2).
Death occurred within 90 days of surgery in 577 (5.4%) patients and readmission within 30 days of discharge in 1282 (12.8%) of 9977 surgeries with qualifying discharge. Outcome rates varied substantially by surgery type and were highest after non-elective colectomy or valve surgery and lowest after hysterectomy (table 2). The most common reasons for readmission were complication of the procedure (18.6%), septicaemia (6.6%) and congestive heart failure (6.0%; online supplementary table S3).
There was no significant difference in 90-day mortality among patients receiving a TNFi (aOR 0.83 (0.67 to 1.02)) or non-TNFi biologic/tsDMARD (aOR 0.78 (0.49 to 1.22)) versus patients receiving methotrexate without a biologic/tsDMARD (table 3). Compared with methotrexate, the risk of 30-day readmission was somewhat lower in patients receiving a TNFi (aOR 0.86 (0.75 to 0.993)) and was similar in those receiving a non-TNFi/tsDMARD (aOR 1.02 (0.78 to 1.33)). Results were similar in sensitivity analyses excluding patients with previous biologic use (online supplementary table S4).
Unadjusted 90-day mortality and 30-day readmission were greater for patients receiving higher doses of glucocorticoids (table 3). Results from adjusted analyses were attenuated but still showed significantly higher risk of 90-day mortality among patients receiving an average of 5–10 mg/day (aOR 1.41 (1.08 to 1.82)) or >10 mg/day of glucocorticoids (aOR 1.64 (1.02 to 2.64)), and significantly higher risk of 30-day readmission among patients receiving 5–10 mg/day (aOR 1.26 (1.05 to 1.52)) or >10 mg/day of glucocorticoids (aOR 1.60 (1.15 to 2.24)) versus no glucocorticoids.
When hip, abdominopelvic and cardiac surgeries were evaluated independently, we found no significant differences in 90-day mortality or 30-day readmission with a TNFi or non-TNFi biologic/tsDMARD versus methotrexate for any of the surgery categories (figure 2). Associations between glucocorticoid use and adverse outcomes varied by surgery category (figure 3). Associations with mortality were most pronounced among patients undergoing cardiac surgery although 30-day readmission was not significantly different across glucocorticoid doses in these patients. In contrast, patients undergoing hip fracture repair had no significant difference in 90-day mortality across glucocorticoid doses, but did have greater risk of 30-day readmission with higher glucocorticoid doses.
We found no significant difference in the risk of secondary outcomes of hospitalised pneumonia or hospitalised wound complication within 30 days of surgery in patients receiving a TNFi or non-TNFi biologic/tsDMARD versus patients receiving methotrexate (online supplementary table S5). Higher dose glucocorticoids were associated with greater risk of wound complications (aOR 1.72 (1.24 to 2.39) for 5–10 mg/day, aOR 1.68 (0.97 to 2.94) for >10 mg/day).
Among 2208 patients who received infliximab and 827 who received intravenous abatacept within 6 months of surgery, there was no significant difference in the risk of mortality or readmission in patients who received their last infusion <4 weeks before surgery compared with those who received the last infusion 4 to <8 weeks or ≥8 weeks before surgery after adjusting for potential confounders (online supplementary table S6).
In this large observational study of patients with RA undergoing hip fracture, abdominopelvic or cardiac surgery, we found no increased risk of 90-day mortality, 30-day readmission, wound complications or postoperative pneumonia versus patients receiving methotrexate without a biologic/tsDMARD. Conversely, even low-dose glucocorticoid use (5–10 mg/day) was associated with greater risk of adverse outcomes.
Although previous studies of perioperative immunosuppression in patients undergoing the surgeries we examined are sparse,14 our results showing no significant increase in adverse outcomes in patients receiving a biologic are in line with data from studies of joint replacement surgery. These studies showed no difference in postoperative risk in patients who stopped versus continued biologics before surgery.7–9 Interestingly, in our study, biologic-treated patients had lower rates of adverse outcomes than patients receiving methotrexate without a biologic before adjusting for potential confounders. Although biologic-treated patients may have more severe RA (in our study biologic-treated patients were more likely to be disabled), biologic-treated patients also tend to be younger and have fewer comorbidities,31 as we found here. The small differences in risk favouring biologic-treated patients after adjusting for confounders likely reflect some residual confounding rather than risks associated with methotrexate or protective effects of biologics—reassuringly, we found no difference in outcomes in patients receiving a biologic with versus without methotrexate (data not shown).
As we previously observed for arthroplasty,7 8 the lack of association between biologic use and adverse outcomes in this study suggests that holding biologics before the surgeries we examined may not be necessary. We required that all biologic-treated patients receive an infusion or fill a prescription for their biologic within 8 weeks of surgery (16 weeks of rituximab), excluding patients without evidence of recent biologic exposure. Although some patients may still have stopped their biologic before surgery, the majority of our surgeries were non-elective surgeries, in which patients would not have the opportunity to stop treatment. Additionally, we directly evaluated biologic timing among patients receiving infliximab or intravenous abatacept, for whom we could precisely identify timing based on infusion dates, and found no evidence that patients who received infliximab or intravenous abatacept within 4 weeks of surgery had greater risk of mortality or readmission compared with patients with a longer time between infusion and surgery. Nonetheless, our results should be interpreted cautiously since we cannot rule out the possibility of small benefits with biologic interruptions on the outcomes we studied or impacts on the frequency or severity of other postoperative outcomes.
Our results suggest that necessary non-elective surgeries need not be delayed in patients receiving biological therapy, since delaying these surgeries has its own consequences. For elective surgery, a common practice has been to hold biologics before surgery. While our results show no obvious benefit to treatment interruptions, we also did not find obvious risks with short interruptions. Prolonged interruptions beyond one interval are likely not warranted. When deciding on shorter interruptions (eg, one dosing interval), clinicians should consider the risks and benefits, including the potential need for glucocorticoids if patients have a disease flare.
In analyses with all surgery types combined, we found a dose-dependent association between glucocorticoid use and mortality, readmission and wound complications. The difference between unadjusted and adjusted results did suggest substantial confounding; we cannot definitively establish causality and the potential for residual confounding remains. Additionally, results were more variable when stratified by surgery type. Whether these results reflect differences in glucocorticoid risk in different surgery types or whether these results are due to chance in these smaller subgroups is uncertain. Glucocorticoids have been associated with poor wound healing32 and previous studies demonstrated a dose-dependent association between glucocorticoid use and adverse outcomes in the non-operative setting33–35 and after orthopaedic surgery.7 10 We averaged glucocorticoid dose in the 90 days prior to surgery and could not evaluate dose changes immediately before surgery or doses during the hospitalisation, but our results suggest that glucocorticoid use in the months before surgery, especially >10 mg/day, should remain a concern in patients undergoing major surgery and may be a more important contributor to infection than more targeted immunosuppression.
Several limitations are inherent to this type of observational study. Outcomes from claims data may miss or misclassify specific postoperative complications36; we instead focused on mortality and readmission which can be accurately identified in claims data. We evaluated postoperative pneumonia and wound complications but could not fully assess morbidity or minor complications (eg, delayed wound healing) not leading to hospitalisation. Although typically biologic users would be expected to have more severe disease, we could not measure disease activity, which could be less well controlled in some non-biologic treated patients. We also excluded patients with recent changes in therapy who might have the highest disease activity. We cannot rule out the possibility that patients receiving biologics were selected differently for surgery (especially elective surgery) or were treated or monitored differently after surgery. Additionally, biologics may be avoided completely in patients who are frail or have comorbidities31 or survivorship bias could lead to a healthier biologic group, although we controlled for multiple comorbidities and measures of healthcare utilisation and attained good balance for measured covariates between exposure groups. Glucocorticoid results were more variable and there is even greater potential for unmeasured or residual confounding in these analyses. Some contributors to postoperative outcomes such as surgeon experience, severity of the presentation and difficulty of the surgery could not be directly measured, although these factors would not be expected to differ in biologic users versus non-users. Finally, although we analysed surgeries in groups that would be susceptible to similar types of complications, numbers were not sufficient to evaluate specific surgeries (eg, appendectomy, cholecystectomy) separately.
In conclusion, neither use of biologics nor biologic timing before surgery were associated with increased mortality or readmission after hip fracture, abdominopelvic or cardiac surgery. Higher dose glucocorticoids were associated with adverse outcomes after these surgeries, suggesting that minimising glucocorticoids should be a focus of perioperative medicine.
Handling editor Josef S Smolen
Contributors All authors met criteria for authorship, critically revised the manuscript, gave approval of the final version to be published and agree to be accountable for all aspects of the work related to accuracy or integrity of the results. Specific additional roles included: conception and study design (MG, JFB, KLW, JRC), data acquisition (MG, LC, QW, FX), analysis (MG, JFB, KLW, JRC), interpretation (MG, JFB, KLW, SDG, EA, LC, FX, JRC), and drafting the original manuscript (MG, JFB, JRC).
Funding MG is supported by the National Institute of Arthritis and Musculoskeletal and Skin Diseases 1K23AR073931-01. JFB is supported by a VA Clinical Science Research and Development Merit Award (I01 CX001703).
Competing interests MG has received a research grant from Bristol-Myers Squibb and the National Institutes of Health and consulting fees from AbbVie. JFB has received consulting fees from Bristol-Myers Squibb and Gilead. KLW has received research grants from Pfizer and Bristol-Myers Squibb, and consulting fees from Pfizer, AbbVie, UCB, Lilly, Galapagos, GSK, Roche, and Gilead. EA is an employee of Bristol-Myers Squibb. JRC has received research grants from the Patient Centered Outcomes Research Institute and research grants and consulting fees from Bristol-Myers Squibb, Amgen, AbbVie, Corrona, Janssen, Lilly, Myriad, Pfizer, UCB and Regeneron.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Patient consent for publication Not required.
Provenance and peer review Not commissioned; externally peer reviewed.
Data availability statement Data may be obtained from a third party (Centers for Medicare & Medicaid Services) and are not publicly available. Statistical code is available from the authors upon request.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.