Recent Anti-Microbial Exposure Is Associated with More Complications after Elective Surgery

Patients and Methods

The University of Virginia Institutional Review Board for Health Sciences Research approved this study with a waiver of consent. Between August 4, 2015, and August 3, 2016, all patients having elective surgery at the University of Virginia Health System were queried prospectively about recent anti-microbial exposure. Specifically, each patient was asked “Have you taken any antibiotics in the last three months?” while in the pre-operative holding area. Our pre-operative nursing staff collected responses and recorded them in our electronic medical record system as either “Yes,” “No,” or “Unsure.” The three-month prevalence window was chosen by literature review [12,13,19–21]. We thought that patients would be able to recall such an exposure with a high degree of accuracy. However, we believed that patients would be unlikely to recall more detailed information regarding drug class, duration, or indication with the same accuracy. Therefore, our survey was limited to the single question. Only patients who provided a response to this question were included in the final analysis. For the purposes of this study, “Unsure” responses were considered to reflect an exposure.

Anti-microbial exposure responses were matched to our local American College of Surgeons National Surgical Quality Improvement Project (ACS NSQIP) sample from the same time period. Eligible cases were sampled for inclusion in ACS NSQIP according to an eight-day cycle. Demographic and co-morbidity variables were recorded according to the definitions from the ACS NSQIP Risk Calculator [22]. Incision classification also was recorded. Current Procedural Terminology (CPT) codes were classified by procedure type. Primary outcomes were ACS NSQIP-defined serious complication and any complication, as well as any infection and any surgical site infection (SSI). Secondary outcomes recorded were Clostridium difficile infection, length of stay, and 30-day mortality rate. Only elective general surgery and gynecology cases were included in the final test sample. Urgent or emergency cases, as well as those with pre-operative evidence of systemic inflammatory response syndrome (SIRS), sepsis, or septic shock, were excluded.

The ACS NSQIP Participant Use File (PUF) for the year 2014 was used to create risk models for our four primary outcome measures. The 2014 file year was chosen specifically for its temporal proximity to the test data yet without including any of the study participants. The file was narrowed to include only elective cases and CPT codes identified in our test sample. Risk models were created for each primary outcome using multivariable logistic regression including all recorded demographics, co-morbidities, and procedure data a priori. Use of the large PUF file allowed us to create models with more robust estimates of regression coefficients than we would have been able to create using our local sample. These models were then applied to our local test sample to create risk-adjusted odds ratios (ORs) for anti-microbial exposure while further controlling for individual surgeon influence using mixed effects modeling.

Standard univariable techniques were used. The Wilcoxon rank sum test was used for continuous variables, and the χ² and Fisher exact test were used for categorical variables where appropriate. Regression analyses were performed as described above. A post-hoc analysis of all outcomes with and without patients undergoing colorectal surgery, who routinely received anti-microbial drugs pre-operatively, also was conducted. The threshold for statistical significance was set at p < 0.05. All statistical analysis was conducted using SAS software, version 9.3 (SAS Institute, Cary, NC).

Results

There were 2,397 cases in our ACS NSQIP sample during the study period of which 1,645 (68.9%) met our inclusion criteria. Of these, 1,538 (93.5%) provided a response to our anti-microbial use survey with an overall three-month exposure prevalence of 34.1%. Five hundred patients (32.5%) responded “Yes” whereas only 24 patients (1.6%) responded “Unsure.”

Demographics and co-morbidities are listed in Table 1. Exposed patients tended to be slightly older, had higher rates of steroid use and disseminated cancer, and had higher American Society of Anesthesiologists (ASA) scores. Exposed patients were more likely to have a clean-contaminated, contaminated, or dirty/infected incision classification and were more likely to have a normal body mass index (BMI). Procedure groups are listed in Table 2. Patients having colorectal surgery had the highest prevalence of exposure, whereas hernia patients had the lowest. The prevalence of recent anti-microbial exposure in the colorectal surgery population was 71.7%, 25.6 percentage points higher than the next highest exposure group. This finding prompted a post-hoc analysis with and without these patients, as pre-operative bowel preparation, including oral anti-microbial drugs (erythromycin and neomycin), is standard practice at our institution and may have introduced bias in this subgroup, resulting in the elevated response rate [23].

Exposed patients had higher rates of complications, overall infection, and SSI (Table 3). When evaluated by SSI type, exposed patients had higher rates of organ space infection. Exposure was associated with a median one-day longer hospitalization. There was a higher rate of return to the operating room and deep venous thrombosis in the full cohort, whereas the non-colorectal surgery cohort had a slightly higher rate of pneumonia. There were no differences in the Clostridium difficile infection or 30-day mortality rates.

Using 279,469 elective cases identified from the 2014 PUF file, we created risk-adjustment models for our primary outcomes: Any serious complication, any complication, any infection, and any SSI. These models had c-statistics of 0.72, 0.75, 0.75, and 0.75, respectively. The Brier scores were 0.052, 0.054, 0.047, and 0.033, respectively [22]. The Hosmer-Lemeshow test had a p value <0.0001 for all models. See the Supplemental Table (supplementary material is available online at www.liebertpub.com/sur) for full model details. After applying these models to our local ACS NSQIP data and further controlling for surgeon influence, the adjusted odds ratios for anti-microbial exposure were calculated and are listed in Table 4 (c-statistics 0.57–0.70). Anti-microbial drug exposure within the previous three months was independently associated with more complications. When patients having colorectal surgery were excluded, anti-microbial exposure was independently associated with both higher complication and infection rates.

Discussion

We have demonstrated a high prevalence of anti-microbial exposure within the previous three months in the elective surgical population, reflecting our previously reported data [24]. This high prevalence may reflect an overuse of anti-microbial drugs by the general population, and some data suggest that as many as 30% of such use may be inappropriate [6]. We observed that this exposure was independently associated with complications after elective surgery. Unadjusted data additionally suggest higher overall infection and SSI rates and increased length of stay.

Our findings are most similar to those of Dr. Kollef's group, who identified increased length of stay (median five days) as well as a higher mortality rate in gram-negative sepsis after anti-microbial drug exposure in the prior three months [12,13]. Their cohort consisted mostly of intensive care medical patients, resulting in a much higher observed mortality rate than in our study. Both studies demonstrate an increase in the length of stay associated with exposure, although our observed difference is less pronounced, likely attributable to a healthier cohort in general. Their studies identified prior anti-microbial exposure to be associated with a higher rate of inappropriate initial empiric therapy, suggesting that broader or combination empiric coverage may improve outcomes [12,25]. However, the patients in our study were not septic on the day of surgery and were deemed healthy enough to undergo elective surgery. Adherence to surgical prophylaxis choice based on Surgical Care Improvement Project (SCIP) guidelines is in excess of 90% nationally [26,27]. Although organ space SSI rates were higher in the exposed cohort, exposure was not associated independently with SSI in our multivariable models. It is unlikely that additional or alternative anti-microbial drugs at the time of surgery would address the differences in outcomes we observed.

The patients in our exposure cohort experienced a prior presumed infection for which they were prescribed anti-microbial drugs. Similar to our patients, patients who survive sepsis have poor long-term outcomes, including a higher mortality rate [14–16]. Sepsis survivors also may be at higher risk of subsequent infection, similar to the increased infection rate we observed in our univariable data [16,28]. Although the etiology of these associations is unknown, pro-inflammatory profiles at hospital discharge are associated with subsequent poor post-acute outcomes in septic patients [29]. Interestingly, similar pro-inflammatory patterns may be seen in patients with low gut microbiome diversity and may be a key driver of these findings [30]. In the healthy state, a robust and diverse commensal population serves an immune-modulatory role, suppressing inflammation overall [18,31,32]. Anti-microbial drug exposure decreases microbiome diversity with further decreases in diversity immediately after severe stressors [17,32,33]. Excessive anti-microbial duration or aggressive initiation also has been associated with both a decreased mortality rate and a higher subsequent infection rate, further suggesting that these exposures may not be completely benign [34–36]. We are unable to evaluate the reason each patient received anti-microbial drugs or the extent of time the anti-microbial drugs were prescribed, as most of the drugs were provided in the out-patient setting [7,37].

We hypothesize that each of our exposed patients presented on the day of surgery with some degree of anti-microbial-induced dysbiosis and altered ability to respond appropriately to the inflammatory stress of surgery. Although a full analysis of the host microbiome was outside the capabilities of our study, this idea provides an exciting area for future research and innovation. Intra-luminal phosphate administration, fecal microbiota transplantation, and probiotic administration have been suggested as methods of modulating the microbiome and reducing subsequent inflammation [17,38–41]. If this association proves true, one of these interventions may optimize the microbiome prior to surgical intervention. A simple delay of truly elective procedures may achieve the same goal, although the time to recolonization by commensal flora is variable [42,43]. It is possible that in the future, microbiologic optimization will be a common component of pre-surgical preparation.

We are limited in our ability to control for both the degree of prior sepsis and the duration and type of anti-microbial drug administration. Our results may indicate that recent anti-microbial exposure is a marker of increased fragility: That after a recent illness requiring antibiotics, patients may be less capable of handling surgical stress. Dysbiosis may be one explanation. However, the heterogeneity of our co-morbidity data also suggests that exposed and unexposed patients may represent slightly different populations. Because pre-illness co-morbidities may ultimately be the largest drivers of outcomes, appropriate risk adjustment is key [44]. Using the larger PUF file, we were able to create more robust risk-adjustment models to account for this heterogeneity than our sample size would otherwise have allowed. The Brier Score is a model performance assessment score describing both discrimination and calibration, with smaller values representing higher accuracy. It is the model performance tool preferred by ACS NSQIP [22]. We used the ACS NSQIP Risk Calculator measures as the base of our risk adjustment models, adding incision classification and replacing individual CPTs with procedure groupings. Our models performed on a par with the Brier Scores reported for the morbidity and SSI portions of the universal risk calculator, suggesting that they are functioning appropriately [22].

Our study has several limitations worth discussing. First, although our anti-microbial use data were collected prospectively, we were able neither to verify these data nor to capture specific anti-microbial regimens. Most medical and surgical history reported by patients is not verified in practice, giving us no reason to suspect that these data are more or less valid than other patient-reported history items. Data on specific anti-microbial drug regimens would be helpful to include in future studies. Second, data collected retrospectively from ACS NSQIP may be subject to observation bias. However, ACS NSQIP has proved to be a reliable and well-validated tool for risk stratification that strengthens our analysis. Third, our sample size was confined to that included in the ACS NSQIP sample. A larger analysis of the entire cohort without this sampling restriction may yield different results. Additionally, we were unable to assess or control for the degree of prior illness that resulted in anti-microbial drug administration prior to surgery. Finally, although we propose that pre-operative dysbiosis associated with anti-microbial drug exposure might explain our findings, a formal microbiologic evaluation of each patient's microbiota was outside scope of this study.

Conclusion

In this study of patients undergoing elective surgery, we identified that anti-microbial drug exposure within the three months prior to surgery was independently associated with complications, particularly for patients not having colorectal surgery. We propose anti-microbial-induced dysbiosis as one possible mechanism for this finding. Further work is needed to evaluate more thoroughly the complex interactions between anti-microbial drugs, the host microbiome, and responses to surgical stress.