Preoperative Versus Extended Postoperative Antimicrobial Prophylaxis of Surgical Site Infection During Spinal Surgery: A Comprehensive Systematic Review and Meta-Analysis

Guidelines over the optimal timing of perioperative AMP in spinal surgery have been subject to clinical debate for over half a century. Prior to publication of Burke’s [53] landmark bovine study in 1961, the prophylactic efficacy of perioperative antibiotic utilization was poor secondary to administration that was commonly confined to the postoperative window [54, 55]. Succeeding publications, including Classen et al.’s [30] seminal 1992 observational study, substantiated Burke’s findings over the next three decades [29, 56‐63] and highlighted the importance of administering preoperative AMP just prior to first incision. Horwitz and Curtin [6] are often credited with first demonstrating the efficacy of initiating AMP preoperatively in spinal surgery in 1975 and, ultimately, these authors recommended extended prophylaxis. Despite this evidence, RCTs [64‐69] had so far failed to demonstrate clinical efficacy of perioperative AMP in preventing spinal SSI. Barker [19] performed a meta-analysis of these RCTs in 2002 and found results that robustly reinforced administration of AMP with at least a single preoperative dose during spinal surgery, even when anticipated rates of SSI without AMP are low.

Although research has heavily focused on when to start AMP in spinal surgery, questions over its overall duration remain. In 2003, Dobzyniak et al. [52] and Beiner et al. [70] were the first to call attention to the lack of consistent recommendations over how long, if at all, to continue AMP postoperatively. Clinical studies have since not only assessed the efficacy of extended postoperative therapy to preoperative AMP alone [27, 28, 33, 51, 52] but have also compared the therapeutic benefit of extended AMP regimens to others that differ in postoperative duration [32]. The NASS [24, 25] has made recommendations for optimal antibiotic duration for both uncomplicated and complicated spinal surgeries after qualitatively reviewing these studies in 2013, but a meta-analysis of comparative clinical trials has still not yet been conducted.

The broad confidence interval of the risk ratio and the insignificant test statistic imply that the meta-analysis is likely underpowered to detect small differences in the risk of SSI after spinal surgery between these two AMP regimens. However, the above finding is consistent with previously published studies. As mentioned previously, Barker performed a meta-analysis of six RCTs in order to evaluate the role of AMP in spinal surgery. A subgroup analysis of these studies failed to demonstrate a greater benefit of prophylaxis when postoperative doses were added to a single preoperative dose [19]. Furthermore, a large meta-analysis of 28 RCTs comparing preoperative and extended AMP in the setting of major surgery also failed to demonstrate superiority of extended AMP over preoperative only regimens [20]. Although the findings of this meta-analysis and the previous literature cannot definitively answer the question of whether extended postoperative AMP provides additional benefit over preoperative only AMP during spinal surgery, it contributes significantly to the existing state of knowledge by providing more evidence that preoperative AMP alone may be sufficient to decrease the risk of SSI.

This is a valuable observation related to the high costs of SSI, the association of antibiotic use with increasing antimicrobial resistance, and related morbidity such as C. difficile colitis. Attributable costs per spinal SSI range from US$15,000 to upwards of US$100,000 [7, 11, 71‐78]. Notably, the costs incurred from SSI after spinal surgery are greater than the reported additional expenditures required for the average surgical patient who experiences SSI [79, 80].

In terms of SSI and the potential for AMR, most infections reported in the included studies were, unsurprisingly, due to organisms related to skin flora, and included infections with S. aureus and coagulase-negative Staphylococcus species. Infections with E. faecalis and gram-positive bacillus were also observed. Interestingly, among those studies that commented on antimicrobial patterns, resistant organisms, such as MRSA, were cultured in six of fifteen patients treated with extended AMP and in zero of four patients who only received preoperative AMP. Clearly, the numbers of patients in these two groups are low, and, thus, meaningful statistical inference is impossible. Furthermore, the patients were not randomized and it is possible that patients with increased risk for developing infection from resistant organisms may have also been more likely to receive extended AMP. However, despite these limitations, this observation is concerning given emerging antibiotic resistance patterns. The potential to reduce AMR infections by minimizing the duration of perioperative AMP should continue to encourage additional, high-quality research.

It is incumbent upon all surgeons to endeavor to prevent SSI and AMR emergence by exercising conservative caution when using perioperative AMP. To date, surgeon adherence to recommended guidelines often deviates significantly in practice, especially with regard to how long AMP is extended into the postoperative window [81‐83]. This variation in prophylactic care is secondary to numerous factors, including instrumentation use, surgical complexity, and patient comorbidity status [9, 16, 52], but can also result from institutional guidelines or surgeon discretion [28, 52]. This lack of homogeneity in AMP in spinal surgery is unsurprising given the historically contradicting recommendations made regarding extension of AMP into the postoperative window [6, 16, 29, 52, 53, 67], and highlights the need for this systematic review and meta-analysis. It is important to weigh the real risk of antimicrobial resistance against a perceived, yet scientifically unsupported, benefit of continuing antibiotics postoperatively. These considerations aided the authors in developing the recommendation that AMP should not be continued postoperatively in uncomplicated, adult patients during spinal surgery.

However, it is difficult to generalize the results of this meta-analysis to all spinal surgery patients because of the complex interaction between patient-related risk factors, surgical complexity, and other unmeasured factors, particularly in complicated cases. In order to help address this uncertainty, we performed stratified meta-analyses. Most of the studies did not report the rates of SSI among various related medical comorbidities; therefore, the types of stratified analyses that could be carried out were limited. However, stratification was possible according to whether instrumentation was implanted. Patients undergoing instrumented spinal fusion are subjected to a higher risk of postoperative infection due to longer operative time, increased procedure complexity, and the implantation of a foreign body [15, 16, 61]. As expected, the rate of SSI in patients following instrumented spinal surgery was nearly twice that observed in patients following non-instrumented spinal surgery. However, we did not find a significant benefit of extended AMP over preoperative only AMP in patients either undergoing instrumented spinal surgery or in those undergoing non-instrumented spinal surgery.

Overall, the strengths of this study make its findings noteworthy for spine surgeons. First, the inclusion criteria concerning the study design were strict. In particular, each study was required to have a near equivalent comparison group which allowed for more accurate effect estimates between the two AMP protocols. This is highlighted in Kanayama et al.’s [51] cohort study in which the investigators identified a potential self-imposed limitation to their study: the definition of SSI required surgical treatment of the wound. Since the majority of SSIs were superficial and managed with conservative therapy, the reported rate of SSI for both treatment arms may be underestimated. However, since the same definition was utilized for both treatment arms, the relative rate of SSI between the two groups may actually be accurately reflected, despite the conservative nature of the definition itself. But, this may not necessarily be the case. Preoperative AMP may have a different effect on SSI requiring surgical intervention when compared to extended postoperative AMP, and this presents an opportunity for bias in this meta-analysis. Also, since Kanayama et al. provided data for the primary analysis and both subgroup meta-analyses, this confounding could extend throughout the study.

The internal validity of this review was high secondary to assessment with funnel plots and Egger’s test, which both failed to identify publication bias. Since this meta-analysis found a non-significant result, the publication of further non-significant studies on this topic would only lend greater support to our current findings, making publication bias of little concern to the authors. Additionally, there was no heterogeneity present among the five studies. Taken together, these results support the internal validity of the results reached in this meta-analysis. With that said, however, the external validity of this study may be called into question. It is important to note that weaknesses of this meta-analysis and its methodology are mutually exclusive to the limitations of the included studies. One weakness of the present study is the generalizability of results. This study aimed to report on the efficacy of temporally variable perioperative AMP in spinal surgery at any level. However, 96.2% (2717/2824) of patients underwent spinal operations confined to the lumbar spine. Stratification of the 107 patients undergoing either cervical or thoracic operations was not possible because of the small sample size. Although unlikely, it is possible that cervical, thoracic, or coccygeal procedures experience different SSI rates under different perioperative AMP protocols. Additionally, there was not enough data provided to conduct further subgroup analyses on patient comorbidity status, which may impact SSI outcomes concerning AMP duration. We were able to stratify based on instrumentation use, but these subgroup analyses were hindered by the same limitations detracting from the primary meta-analysis. There were also differences between studies in the length of AMP prescribed based on individual surgeon preference. Thus, although there was no evidence supporting the extensions of AMP postoperatively for either instrumented or non-instrumented patient subgroups, decisions regarding a patient’s risk of SSI and whether antibiotic prophylaxis is needed following surgery must be made on an individual basis until more data becomes available.

There were several limitations of the studies included in this review and meta-analysis that should be considered separately from the meta-analysis itself. The GRADE quality of evidence score for the overall included data ranged from low [48] to moderate [46] quality, which was primarily related to their observational and often retrospective nature [10, 19‐21], as the authors were unable to control for unmeasured confounding and bias. Since the RCT [27] was not blinded, investigators were aware of the AMP protocol administered to each patient and bias could arise from increased observation and surgical performance. The majority of included studies were underpowered with relatively small sample sizes, increasing the likelihood of type II error and limiting the ability to detect significant differences in SSI rates between the different AMP protocols. Most studies failed to include explicit and objective definitions of SSI, including those for superficial and deep SSIs; the criteria in which SSI was diagnosed; or who was making these decisions. One study followed guidelines set forth by the CDC [28]. Another study required positive wound cultures and/or the attending surgeon’s clinical impression of typical infectious signs that led to additional surgical intervention [51]. Meanwhile, others did not mention their criteria for defining and diagnosing SSI [33]. Although some studies stated what individuals were making these clinical decisions, possible sources of bias, especially for the prospectively collected databases and RCT, could arise. In particular, if the prospective studies failed to have a third-party record their data, observer bias could occur if the investigators knew which patients received which AMP protocol. At the same time, the retrospective studies could be biased since the authors had no control over data collection. Similarly, the follow-up observational period for SSI identification demonstrated considerable variation among studies and ranged from within 30 days [47] to 6 months [50] and longer [48] to unspecified follow-up times [50]. This review did not exclude studies based on antibiotic type used or changes in antibiotic administered during treatment. The timing of initial preoperative AMP administration and the length of postoperative AMP duration in the extended postoperative groups also demonstrated significant temporal heterogeneity both within and between studies, ranging from within 30 min prior to skin incision to 10 days postoperatively [27, 28, 51]. The timing of perioperative AMP was either unknown or not reported by Dobzyniak et al. and Khan et al. [33, 52]. Hellbusch et al. [27] statistically assessed the impact of demographic factors, comorbidities and past surgical history, and operative variables on the outcome of SSI development, finding no associations of statistical significance. No multivariate analyses were conducted in any of the included studies. The considerable variability in defining, diagnosing, and reporting SSI; the inconsistency in the postoperative observational period; and the inconsistency in preoperative and extended postoperative extended AMP somewhat diminish the validity of this meta-analysis.

These strengths and weaknesses were succinctly enumerated and tallied by the GRADE Working Group scoring system in the results section. As a result of the relative novelty of the GRADE approach and the very low to moderate quality of evidence scores awarded to the included studies, the GRADE scoring system deserves further mention here. The GRADE approach is currently regarded as one of the most effective and unequivocal scoring methods to systematically connect evidence-quality evaluations to degree of efficacy for clinical recommendations [44]. This underlies one of the primary reasons why the GRADE approach is valuable in improving clinical practice: its scoring system transparently separates decisions regarding evidence quality from decisions evaluating recommendation strength [45]. In other words, studies with high quality evidence do not always indicate strong clinical recommendations, and strong clinical recommendations can sometimes be drawn from studies with low quality evidence [45]. Therefore, despite the average low quality of evidence scoring assigned to the included studies through the GRADE system in this review, the statistically null result reported in the present analysis and its associated clinical recommendation to provide spinal surgery patients with preoperative AMP only should give pause to spinal surgeons, nonetheless.

The authors also believe that certain qualities of the GRADE scoring system led to artificially low scores among the included articles. Firstly, the GRADE scoring approach is partial towards studies that demonstrate statistical significance. All included studies failed to demonstrate statistically significant results, and this contributed to their overall low quality of evidence scores through poor scoring in two of the five categories of the GRADE scoring system: effect size and consistency. Effect size awards increasing points when every included study reports large, unidirectional magnitudes of effect. Meanwhile, consistency assigns points to studies that are characterized by dose response patterns, either between or within studies. Magnitude of effect and dose response are two of Hill’s criteria for causality [84] and are desirable if, in reality, there is a difference in effect between treatment arms. However, their inclusion in the GRADE scoring methodology assumes that the true and desirable outcome is always statistically significant. It is pertinent to remember that statistical significance and clinical significance are distinguishable, and statistically insignificant results can impact care in clinically meaningful ways. Given the clinical significance of our null result and its potential to reduce unnecessary medical expenditures and decrease risk of AMR development, it is appropriate here to highlight these shortcomings.

An additional issue brought on by the GRADE approach that contributes to the overall low quality of evidence scores is the explicit value placement on study design. Specifically, the GRADE scoring system allocates four points to RCTs and two points to observational studies, reflecting the old adage that RCTs are the “gold standard” of research. However, the sample size necessary for an RCT to establish statistical significance is prohibitively large in SSI studies in spinal surgery. For example, the investigators [27] who conducted the only RCT included in this meta-analysis hesitated to make clinical recommendations secondary to their small sample size of 233 patients. Utilizing the low SSI rates experienced by patients in their study and the small effect size difference in proportions (ES = 0.074), they performed a power calculation to determine the sample size that would have been required to obtain statistically significant results. They found that 1400 patients, a study population six times greater than their original sample size, would be required to establish statistical significance in comparing SSI rates in patients receiving preoperative AMP to those receiving extended postoperative therapy at the desired power. Additionally, Dimick et al. [26] further justified the impracticality of conducting adequately powered RCTs on SSI development and perioperative AMP in spinal surgery and the notable risk of making a type II error when not enough patients are enrolled. They demonstrated that 474, 988, and 2518 patients would be required to detect baseline SSI rates of 10%, 5%, and 2%, in each treatment group, respectively, of an RCT with 80% power to detect a 50% reduction in SSI through perioperative AMP in spinal surgery. Therefore, it was highly unlikely that RCTs would have been available for this meta-analysis because of the exorbitant costs required to enroll enough patients to ensure adequate power. Thus, the relative ease of identifying patients for a large observational study may, in the case of investigating SSI rates in spinal surgery, make observational data more meaningful than data from small RCTs. However, this is not reflected in the GRADE approach, which broadly penalizes observational studies when, in this clinical application, they are actually the best option for finding statistically significant results, especially given the low incidence of SSI in spinal surgery [85]. This meta-analysis benefited from the relatively large sample sizes provided by the majority of the included observational studies. The GRADE scoring system aims to objectively evaluate a research process in which investigators must invariably make subjective decisions. Researchers and practitioners must consider the need for flexibility in the interpretation of such evaluative methods in order to adapt to the unique challenges of conducting research in particular medical and surgical specialties.

In light of these deficiencies, the authors of this review endorse the validity of the present analysis despite the low GRADE evidence quality scores assigned to the included studies. This is evidenced by the strong strength of recommendation GRADE scores awarded to four of the five included studies, emphasizing the importance of the research context and the discrepancy between statistical and clinical significance. Therefore, the evidence utilized for this review and analysis, although evaluated to be of moderate to poor overall quality, still provides useful, evidence-based clinical recommendations that have the potential to impact current practices as spinal surgeons still commonly extend perioperative AMP based on personal discretion. Therefore, the recommendation set forth by this systematic review and meta-analysis is strong according to the GRADE scoring system.