Preoperative Radiographic Prediction Tool for Early Postoperative Segmental and Lumbar Lordosis Alignment After Transforaminal Lumbar Interbody Fusion

Objective Transforaminal lumbar interbody fusion (TLIF) is a common approach and results in varying degrees of lordosis correction. The purpose of this study is to determine preoperative radiographic spinopelvic parameters that predict change in postoperative segmental and lumbar lordosis after TLIF. Materials & Methods This study is a single surgeon retrospective review of one-level and two-level TLIFs from L3-S1. All patients underwent bilateral facetectomies, 10 mm TLIF cage (non-lordotic) insertions, and bilateral pedicle screw-rod construct placements. Pre- and post-operative X-rays were assessed for preoperative segmental lordosis (SL), lumbar lordosis (LL), and pelvic incidence (PI). Univariate and multi-predictor linear regression analyses were performed to determine the relationships between preoperative radiographic findings and change in early postoperative segmental and lumbar lordosis. Results Ninety-seven patients contributing 128 intervertebral segments were examined. The mean change in SL after TLIF was 7.3 (range: 0.10-28.9°, SD 6.39°). The mean change in LL after TLIF was 5.5˚ (range: -14.8-39.2°, standard deviation (SD) 7.16°). Greater preoperative LL predicted less postoperative LL correction, while greater preoperative PI predicted more postoperative SL and LL correction. Greater anterior disk height was noted to be associated with a decreased change in SL (∆SL). An annular tear on preoperative magnetic resonance imaging (MRI) predicted a 2.7° decrease in ∆SL. A Schmorl's node on preoperative MRI predicted a 4.0° decrease in change in LL (∆LL). Conclusions A greater preoperative lordosis and a lower spinopelvic mismatch lessen the potential for an increase in the postoperative SL and LL after a TLIF, which is likely due to a ‘ceiling’ effect of an otherwise optimized spinal alignment. A greater anterior disk height and the presence of an annular tear are associated with decreased ∆SL.

Sagittal imbalance is the primary predictor of disability in adult spinal deformity, and restoration of sagittal alignment with spinopelvic harmony (or matched PI and LL) is known to lead to the best long-term outcomes in patients who undergo surgical correction [14,15,16,17]. Previous studies have cited varying degrees of lordosis correction following TLIFs. These widely range from 2.1 -27.3 degrees, depending on factors such as the use of lordotic cages, the number of levels instrumented, and whether the TLIF was performed in an open versus minimally invasive fashion [8,9,[18][19][20][21][22][23][24][25][26][27][28][29]. Traditionally, radiographic factors, such as bridging osteophytes or the presence of intradiscal vacuum phenomenon, have been used to predict the capacity to restore SL [30]. Additional factors cited in the literature that may contribute to greater lordosis restoration include multilevel fusion, cage size, use of the cantilever TLIF technique, low preoperative SL, and high preoperative spinopelvic mismatch [18,19,24,31,32].
Currently, there are limited studies that use patient-specific imaging characteristics to predict SL correction after TLIF [33]. Therefore, the objective of this analysis is to demonstrate a patient-specific algorithm for the prediction of early postoperative degree of SL and LL correction after TLIF.

Materials And Methods
After obtaining approval from the University of Florida's internal review board (IRB202002814), we conducted a single surgeon retrospective review of all patients that underwent one-and two-level TLIFs from levels L3-S1, between 2010 and 2015. Data was de-identified and therefore, informed consent was not sought. Patients less than 18 years of age were excluded from this study. Each patient included in the study had preoperative 36-inch standing x-rays and post-operative lumbar x-rays. Additionally, 39% had a preoperative lumbar computerized tomography (CT) scan and 95% had a preoperative lumbar MRI. Each patient included in the study underwent a standard open TLIF technique (as previously described) involving a laminectomy with complete bilateral facetectomies, bilateral pedicle screw-rod construct placements, and placement of a 10mm non-lordotic cage (Concorde, Depuy Synthes, Raynham, MA, USA).

Data acquisition
Preoperative and immediate postoperative x-rays were assessed for spinopelvic parameters, including LL, SL and PI. LL was defined as the Cobb angle between the superior endplates of L1 and S1. SL was defined as the Cobb angle between the superior endplate above and the inferior endplate below the operative level. PI was defined as the angle between a line perpendicular to the sacral plate at its midpoint and a line connecting this point to the femoral head axis. The spinopelvic mismatch was calculated as PI-LL. Changes in LL and SL were calculated by determining the difference between pre-and postoperative Cobb angles of the lumbar spine and at the level(s) of operation, respectively. Immediate postoperative x-rays were used to determine the change in the SL and LL to reflect the primary effect of the TLIF procedure on spinal alignment. Additional radiographic assessments of the intervertebral segment(s) at the operative level(s) were performed using preoperative MRI and CT scans, including the anterior disk height, annular tears, presence of a Schmorl's node, Modic type endplate change, bridging osteophytes, and vacuum disk. All preoperative disk characteristics and pre-/postoperative spinal measurements were performed and collected by two individuals (KP and AD), with a calculated discordance of 2.3% when considering all values included in the study.

Statistical Methods
We used the mixed effect linear regression to fit single-and multi-predictor models of change in LL (ΔLL) and SL (ΔSL) [34,35]. Possible non-linearity of continuous predictors was evaluated using restricted cubic splines and co-linearity was evaluated using a variable clustering method and variance inflation factors (VIF) cutoff of 2 [34]. We then applied a backward elimination selection procedure to the remaining predictors [34]. We re-ran the backward elimination process on 100 bootstrap samples of the original dataset and tallied the frequency with which each predictor from our original candidate list was selected over the 100 final model fits. Our final best models consisted of predictors that were selected in > 50% of these model fits.
We used the adjusted R-square to assess the predictive performance of our best multi-predictor linear regression models and validated using Efron's "optimism" [34]. Residual analysis indicated normally distributed errors for both outcome models. Best model regression coefficients are presented along with 95% confidence limits and t statistic p-values for tests of whether the coefficients differed significantly from zero. We identified the region of the ΔLL and ΔSL response surfaces defined by PI and preoperative LL that differed significantly from ΔLL=0 or ΔSL=0, adjusting the confidence region using Scheffe's adjustment for multiple testing [34]. Statistical calculations were performed using SAS Version 9.4 (SAS Institute, Cary NC) and R Version 3.5.0 (R Foundation for Statistical Computing, Vienna, Austria).

Results
A total of 97 patients undergoing TLIFs at 128 levels were examined ( Table 1). Sixty-six patients underwent single level, and 31 two-level TLIFs. The mean age was 62.5 years (standard deviation (SD) 11.63), and 44 (47.3%) of the patients were male, with an average body mass index (BMI) of 30.24 (SD 6.14). The LL analysis consisted of 66 one-level patients with L3-L4, L4-L5, and L5-S1 disk involvements and 27 two-level patients. The SL analysis consisted of 66 disks from one-level patients and 54 disks from two-level patients.

All patients 93
All disks 120

Immediate postoperative change in segmental and lumbar lordoses
All TLIF levels demonstrated an increase in SL (ΔSL), with a mean improvement of 7.33° (range: 0.10-28.9°, SD 6.39°). Univariate analysis revealed that greater preoperative SL negatively impacted postoperative ΔSL (p<0.01, 95% CI -0.40° to -0.12°) (  Using multi-predictor linear regression models of ΔSL, the presence of an annular tear was significantly associated with less ΔSL (p=0.02, 95% confidence interval (CI) -5.34° to -0.38°) ( Table 4). The presence of an annular tear decreased the ΔSL by 3°. The ΔSL in patients without an annular tear was statistically significant (p<0.05) for values of preoperative SL < 34° (Figure 1). ΔSL in patients with an annular tear was statistically significant (p<0.05) for values of preoperative SL < 23.9° (Figure 2). Preoperative LL was not found to be associated with significant ΔSL (p=0.54). An equation for ΔSL prediction based on significant factors is demonstrated below; see associated flowchart ( Figure 3A).     The SL cohort was further stratified into the 5th percentile (3.4°), the mean (16.6°), and the 95th percentile (30.9°). This analysis demonstrated that for a mean preoperative SL with an annular tear, a statistically significant ΔSL was only obtained if the preoperative LL was >35° and the PI was >31°. With a preoperative SL in the 95th percentile, but without an annular tear, a statistically significant ΔSL was only obtained if the preoperative LL was >50° and the PI was >60°. No significant change was observed for those with a preoperative SL in the 95th percentile with an annular tear. Otherwise, all other preoperative spinopelvic parameters were observed to cause significant post-operative SLs.

Discussion
All patients in our study had improvement of segmental lordosis only in the immediate postoperative period, with variation based on characteristics of the intervertebral segment and preoperative spinopelvic parameters. Of the intervertebral segment characteristics we studied, anterior disk height and the presence of an annular tear at least marginally impacted postoperative segmental lordosis change. Patients with a higher preoperative anterior disk height had significantly less ΔSL for a given instrumented level (i.e., there was a 2.8° reduction with each additional 1mm of anterior height). This suggests a decreased capacity for lordosis correction in these patients, which is a "ceiling" effect as these patients may already have maximal, well-preserved segmental lordosis. This effect, however, lost statistical significance when all predictors were considered, likely due to high variability. The presence of an annular tear was associated with a less postoperative change in SL (2.9°). This may be explained by the association of an annular tear with a more severely degenerated and less mobile disk. Alternatively, the small effect size of < 3 degrees may not represent a true clinically relevant finding.
Lower preoperative SL and high preoperative spinopelvic mismatch are both associated in the literature with greater postoperative lordosis restoration [18,19,24]. Our study also found that a smaller preoperative SL and greater spinopelvic mismatch significantly increased postoperative ΔSL and ΔLL. Patients with a lower preoperative SL and LL, and a high mismatch logically have more capacity for restoration of lordosis, and this was demonstrated in our study. It also may suggest that improvement in SL and LL were a primary indication for surgery in these patients, which may have introduced selection bias. It also appears that lordosis restoration depends on both preoperative LL and PI when preoperative LL is < approximately 57°, which correspondingly was the average PI in this study population (i.e. when the LL is low enough to create a mismatch with the PI). When the preoperative LL is >57°, postoperative LL appears to depend only on PI. This may suggest a "ceiling" effect in which an individual with an already optimized preoperative LL cannot achieve additional lordosis correction by undergoing a TLIF. This "ceiling" is dependent on the level of operation and disk characteristics; it decreases with more rostral levels and with the presence of Schmorl's nodes. While Schmorl's nodes had no effect on segmental lordosis, they predicted a 4° decrease in achievable lordotic correction, which may be due to the more global degenerative disease/rigidity that they represent.
This study has several limitations including its retrospective design and inclusion of patients from a single surgeon and a standardized technique for open TLIF, which may impact the generalizability of the findings. Postoperative radiographic assessment was performed using immediate postoperative x-rays, and not at longer-term follow-up. This study was designed to isolate the primary effect of the TLIF procedure on change in LL and SL. It is well understood that delayed subsidence may occur over time which may reduce LL and SL, but this is likely variably affected by individual patient factors (e.g. bone mineral density and BMI). Therefore, the study findings observed here likely represent the maximal extent of lordosis correction at the index surgical procedure and may demonstrate loss of effect in some patients over time. Additionally, our prediction equations require additional power in order to further reduce error.

Conclusions
The TLIF is a valuable procedure for restoration of lordosis, but individualized intervertebral segment and spinal alignment parameters may limit the degree of attainable correction. A greater anterior disk height and an annular tear are associated with a decreased ΔSL. Greater initial lordotic angles and lower spinopelvic mismatch lessen the potential for change in LL and SL due to a 'ceiling' effect of an otherwise well-aligned spine. Alternatively, patients with significant preoperative losses of segmental and overall lumbar lordosis may demonstrate potentially greater corrections of spinal alignment with TLIFs.

Additional Information Disclosures
Human subjects: Consent was obtained or waived by all participants in this study. University of Florida IRB issued approval IRB202002814. Animal subjects: All authors have confirmed that this study did not involve animal subjects or tissue. Conflicts of interest: In compliance with the ICMJE uniform disclosure form, all authors declare the following: Payment/services info: All authors have declared that no financial support was received from any organization for the submitted work. Financial relationships: N/A declare(s) a grant from University of Florida. Research reported in this publication was supported by the University of Florida Clinical and Translational Science Institute, which is supported in part by the NIH National Center for Advancing Translational Sciences under award number UL1TR001427. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Other relationships: All authors have declared that there are no other relationships or activities that could appear to have influenced the submitted work.