Surgeon Specialty Effect on Early Outcomes of Elective Posterior/Transforaminal Lumbar Interbody Fusions: An Updated Propensity-Matched and Subgroup Analysis of 13,072 Patients

Safwan Omar Alomari Johns Hopkins University School of Medicine Ryan Planchard Johns Hopkins University School of Medicine Allan Belzberg Johns Hopkins University School of Medicine Sheng-Fu L Lo Johns Hopkins University School of Medicine Daniel M Sciubba Johns Hopkins University School of Medicine Nicholas Theodore Johns Hopkins University School of Medicine Timothy Witham Johns Hopkins University School of Medicine A Bydon (  abydon1@jhmi.edu ) Johns Hopkins University School of Medicine https://orcid.org/0000-0002-3058-1470


Introduction
Degenerative disease of the spine is one of the common and costly condition in the United States. [29,17] Although instrumented PLF has been historically the most commonly used approach for lumbar fusion, recent developments of interbody fusion techniques, including PLIF/TLIF, have increased in popularity. [12] Spine surgery is performed by specialists credentialed in either neurological or orthopedic surgery. Given the differences in training between neurological and orthopedic surgeons including training duration, case volume, and fellowship requirements, several studies have investigated the impact of specialty training on procedural/patient outcomes of spinal fusion. [18,15,31] However, these studies have been limited by including multiple spinal fusion levels (cervical, thoracic, and lumbar) and multiple surgical technique in the same analysis. [18,15,31] Moreover, the results have been limited by confounding of preoperative functional status (independent versus dependent) and urgency of surgery (elective versus emergent) [18,15]. Both factors affect outcomes in reported literature [13]. One of the previous studies reported unadjusted odds ratios for outcomes, without an adjusted odds ratio accounting for multivariate regression models [15]. Prior studies also excluded post-operative measures associated with patient outcomes including the proportion of patients requiring ICU admission, hospital discharge destination [18,15,31]. This present study used the most recent data the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) database to examine the effect of spine surgeon specialty on early morbidity and mortality after elective PLIF/TLIF utilizing strict propensity score matching analysis and subgroup analysis to minimize the effects of any preoperative characteristic differences between cohorts [2]. We hypothesized that there is no signi cant difference in patient perioperative outcomes between orthopedic and neurological surgeons performing PLIF/TLIF.

Data Acquisition and Patient Selection
The ACS-NSQIP database provides reliable surgical outcomes data submitted from participating hospitals. In 2017, 708 hospitals contributed data on patient demographics, preoperative health status, risk factors, and 30-day postoperative mortality and morbidity outcomes [16,33] . All variables were used as de ned in the ACS-NSQIP User Guide, and outcomes were reported for 30 days after the index surgery.
[16] Data from 2015 through 2018, were analyzed using current procedural terminology (CPT) codes for PLIF/TLIF (22630/ Arthrodesis, posterior interbody technique, including laminectomy and/or discectomy to prepare interspace (other than for decompression), single interspace; lumbar), (22633/ Arthrodesis, combined posterior or posterolateral technique with posterior interbody technique including laminectomy and/or discectomy su cient to prepare interspace (other than for decompression), single interspace and segment; lumbar). The secondary Current Procedure Terminology codes for an additional level (22632/ Arthrodesis, posterior interbody technique, including laminectomy and/or discectomy to prepare interspace (other than for decompression), single interspace; lumbar, each additional interspace), (22634/ Arthrodesis, combined posterior or posterolateral technique with posterior interbody technique including laminectomy and/or discectomy su cient to prepare interspace (other than for decompression), single interspace and segment; lumbar) were then used to stratify the cohort into 2 groups (single/multi-level

Outcome Variables
Thirty-day postoperative morbidity and mortality outcomes were analyzed. Complications included pulmonary embolism, DVT/ thrombophlebitis, perioperative blood transfusion, myocardial infarction, stroke with neurological de cit, unplanned intubation, pneumonia, urinary tract infection, surgical site infection, sepsis, return to operating room, and death. Hospital course data, including operative time, the total length of hospital stay, the proportion of patients requiring ICU admission, the proportion of patients discharged to other than home and discharged after postoperative day 3 were also noted. Readmission data was included. These outcome variables are prede ned in the ACS-NSQIP database [26], except for the discharge event after postoperative day 3, which was considered to be any discharge event occurring after a total hospital stay of 96 hours.

Statistical Analysis
The specialty of spine surgeon (neurosurgery vs orthopedic surgery) performing the procedure was used as group indicator resulting in 2 cohorts of PLIF/TLIF cases. To minimize the effects of any preoperative characteristic differences between the cohorts, propensity score matching was used before evaluating outcomes differences using SPSS Statistics for Windows, Version 26.0 (IBM SPSS Statistics for Windows, Version 26.0. Armonk, NY: IBM Corp)/ Statistics Regression module and the Python Essentials. All preoperative categorical variables were dichotomized to either present or absent. Variables included in propensity score matching were age, gender, ethnicity, race, body mass index, diabetes mellitus, hypertension, smoking status, steroid use, history of chronic obstructive pulmonary disease, history of congestive heart failure, disseminated caner, bleeding disorders, inpatient versus outpatient status, American Society of Anesthesiologists (ASA) class, and pre-operative functional status. Match tolerance was set at 0.01. Pearson chi-squared or Fisher exact tests (when appropriate) were used for categorical variables, and student's t tests were used for continuous variables to measure the statistical differences between the 2 cohorts [38, 3,36]. A p value ≤ 0.05 was considered to be statistically signi cant. The effect of pre-operative functional health status of the patients on outcomes was further identi ed using regression models to explain the difference in the results between our study and the prior published studies. In addition, the association between pre-operative functional health status and surgeon specialty was identi ed through Chi-Square Test or Fisher Exact Test (when appropriate) on the unmatched dataset. 22,176 PLIF/TLIF cases met inclusion criteria and were reviewed. 15,216 cases underwent single-level PLIF/TLIF, and 6,960 cases underwent multilevel PLIF/TLIF. In the single-level group, there were 10,043 patients (66%) and 5,173 patients (34%) in the neurosurgery and orthopedic surgery cohorts, respectively. In the multi-level group, there were 4,315 patients (62%) and 2,645 patients (38%) in the neurosurgery and orthopedic surgery cohorts, respectively. (Table 1). Table 1 Baseline demographics of patients with single/multi-level TLIF/PLIF.

Demographics and Baseline Characteristics
In both single/multi-level groups, patients in the neurosurgery cohort were more likely to be older, have higher BMI, be of white race, have an ASA class 3 or more, be a smoker, have hypertension, and be functionally dependent. In the single-level groups, patients in the neurosurgery cohort were more likely to be male, have COPD and have baseline dyspnea (P < 0.05), while in multi-level groups, patients in the neurosurgery cohort were more likely to have diabetes mellitus.

Effect of functional health status on outcomes
The impact of pre-operative functional health status (independent vs dependent) of the patients on outcomes were further investigated using regression models. Being functionally dependent predicted worse postoperative outcomes, Table 4. In addition, the association between functional health status and surgeon specialty was identi ed through Chi-Square Test or Fisher Exact Test (when appropriate) on the unmatched dataset. Patients operated on by neurosurgeons were signi cantly more likely to be dependent when compared to patients operated on by orthopedic surgeons, Table 1. Table 4 The effect of pre-operative functional health status on thirty-day postoperative outcomes in patients undergoing PLIF/TLIF. Univariate and multivariate regression were used to demonstrate the effect of functional health status as a predictor on each of the outcome measures in the tables. The regression analysis was conducted on the unmatched cohorts with age, ASA class and number of levels (single vs. multiple) adjusted in all regression models. Additional predictors were included for regression analysis for some outcomes if they had relative high prevalence rate.

Discussion
It was estimated that the percentage of patients undergoing lumbar interbody fusion for degenerative spondylolisthesis increased from 13 % to more than 30% between 1999 and 2011. [10] Recently, there has been a signi cant increase in use of lumbar interbody fusion, and this was accompanied by an increase in the age and number of comorbidities in patients undergoing this procedure. [17,22] Therefore, post-operative morbidity and mortality measures of lumbar fusions have been the focus of recent research. [35,18] Specialty training is considered an important provider-side variable driving differences in comparative effectiveness between surgeons [8]. The effect of surgeon specialty on outcomes of spinal fusion surgeries has been examined by several studies [18,31]. A recent study conducted on MarketScan database [15] included patients undergoing lumbar laminectomy or lumbar fusion surgeries, reported that lumbar fusion cases operated by neurosurgeons have slightly higher odds of experiencing any complication (OR, 1.1) and higher revision surgery rate (OR, 1.14), however, this study was limited by reporting the bivariate logistic regression results only and without adjusting for any potential confounding through multivariate logistic regression, despite the fact that the patients in the neurosurgery cohort where signi cantly more comorbid than the orthopedic cohort [15]. Other studies concluded that spine surgeon specialty is not a risk factor for any of the reported postoperative complications in patients undergoing spinal fusions except for the observed higher rate of perioperative blood transfusion [18,31], and slightly higher odds for prolonged length of stay among orthopedic surgeons [31]. Similar to previous reports in the spine literature [18,31], our analysis showed that PLIF/TLIF patients operated by orthopedic surgeons were more likely to receive perioperative blood transfusion when compared to similar patients operated by neurosurgeons. Moreover, while a previous study [31] reported that patients undergoing spinal fusions have slightly higher odds for prolonged length of stay among orthopedic surgeons, our study added that PLIF/TLIF patients operated by orthopedic surgeons were more likely to have higher return to operating room rates within the same admission, be discharged to destination other than home, be discharged after postoperative day 3, and higher readmission rates.
The differences in results between our current study and previous studies might be explained by several reasons. It is worth mention that previous studies have been limited by signi cant heterogeneity since they included different spinal fusion levels (cervical, thoracic and lumbar), number of operative levels, and surgical techniques in the same analysis [18,31]. In contrast, we included a homogenous cohort of patients undergoing PLIF/TLIF for degenerative spine disease and we further strati ed the cohort into single-and multi-level groups to account for these limitations in previous reports. In addition, previous studies did not control for the preoperative functional health status (independent / dependent), and surgery status (elective/emergency) [18] which several studies showed that they have signi cant effect on the outcomes [13]. In fact, our analysis on patients undergoing PLIF/TLIF showed that patients operated on by neurosurgeons were signi cantly more likely to be dependent and that being dependent predicts worse outcomes, Tables (1,4). Therefore, adjusting for such confounding variables are important to avoid biased results. We also limited our cohort to elective surgeries to avoid the confounding of this variable that was present in previous reports.
These differences between the cohorts might be attributed to different trends in spine surgery training during neurological and orthopedic surgery residencies. Although both residency specialties are exposed to subspecialty spine training, the length of training as well as the level of exposure of spinal pathology itself have been found to vary greatly. [5,14,21,34,30] A recent study [20] included a ten-year analysis of ACGME case log data and found that case volume of spine surgery procedures is signi cantly larger for neurological surgery residencies when compared to the orthopedic counterparts. Moreover, they found that this discrepancy in case volume is enlarging over time, [20] which might explain that these differences between the 2 cohorts were apparent in our updated analysis, while they were not present in previous similar studies. Despite that case volume alone cannot solely determine the quality of training, it is considered one of the key measures to assess opportunities to develop optimal surgical education [20].
Another study [6], which evaluated self-assessed surgical competence of senior neurosurgery and orthopedic residents by mail-out questionnaire, concluded that neurosurgery residents graduate with a signi cantly higher level of con dence to perform spine surgery (which included cervical and lumbar fusions), while orthopedic residents report signi cantly higher need for additional training in spine surgery.
Moreover, our study showed that nationally, nearly 2 times as many PLIF/TLIF procedures are performed by neurosurgeons than orthopedic surgeons. One possibility is that ACS-NSQIP database includes cases from larger or academic institutions [18,19] that might employ more neurosurgeons who perform spine procedures than orthopedic surgeons.
Although the reasons behind these differences between the two cohorts remain largely unknown and might be beyond the scope of this study, it is worth mentioning that these outcome measures might have signi cant clinical and cost implications for patients, physicians, program directors, hospitals and payors [28,37,27,1]. The higher incidence of blood transfusion among orthopedic cohort is of interest since it may represent a potentially modi able practice. Purvis et al reported that higher perioperative blood transfusion might be associated with increased morbidity in patients undergoing spine surgery and that modi cation of transfusion practice may be a potential area for improving patient outcomes and reducing costs [25,24]. This is concordant with a study by Corwin et al which concluded that packed red blood cell transfusion is an independent risk factor for increase in patient morbidity, mortality, and length of stay [4]. In addition, a more recent institutional study showed more than $2 million savings when packed red blood cell transfusions were decreased by one third. [23] Shields et al [32] recently demonstrated that each hour of decreased length of hospital stay following lumbar fusion directly correlated with cost savings. Moreover, Hydrick et al found that 90-day readmissions were associated with an average of $96 152 in increased hospital costs per patient following lumbar fusion [7].
Overall, focused studies are encouraged in the future to address potential reasons and possible solutions for these differences between the two cohorts.

Limitations:
The retrospective design of the study is a major limitation. In addition, selection bias, confounding and reliability of data collection remain potential concerns. However, a strict propensity score matching (match tolerance of 1%) was applied to minimize bias and confounding of different variables, yielding similar, matched cohorts, and improving the internal validity of the paper. The ACS-NSQIP database provides data with high rates of inter-rater reliability and validation sets performed. Per NSQIP, audit reports have revealed an overall disagreement rate of less than 2% [9,33,11]. The ACS-NSQIP database does not include data on skill level of individual surgeons, such as years in practice, which may have an effect on the outcome. Moreover, the ACS-NSQIP database does not include several variables of interest speci c to neurosurgery, including postoperative symptom relief, rates of neurological complications, such as postoperative sensory loss, weakness, or cerebrospinal uid leak, and outcomes beyond the 30day postoperative period. Finally, although ACS-NSQIP covers a broad base of hospitals -more than 700 hospitals-, only a fraction of hospitals participates in ACS-NSQIP, which might limit the (external validity) generalizability of these data.

Conclusion
This study represents an updated analysis of the surgeon specialty effect on a wide spectrum of early outcomes of PLIF/TLIF. We found that nearly two times as many PLIFs/TLIFs are performed by neurosurgeons than orthopedic surgeons. Although PLIFs/TLIFs patients operated by neurosurgeons had signi cantly more preoperative comorbidities, our analysis showed that, in both groups (single/multi-level PLIF/TLIF), patients undergoing elective PLIF/TLIF for degenerative spine disease by neurosurgeons were more likely to have longer operation time, shorter total hospital stay, lower return to operating room rates within the same admission, lower non-home discharge, lower discharge rate after postoperative day 3, lower readmission rates and lower perioperative blood transfusion rate. These outcome measures might have signi cant clinical and cost implications for patients, payors, hospitals and physicians. Focused studies are encouraged to address potential reasons and possible solutions for these differences between the cohorts.

Declarations
Con icts of interest/Competing interests: On behalf of all authors, the corresponding author states that there is no con ict of interest.