Risk Factor Analysis for Predicting Kyphosis Reoccurrence of Thoracolumbar Burst Fracture Patients Treated with Posterior Short-Segment Fixation

Background: The thresholds of risk factors of kyphosis recurrence in thoracolumbar burst fracture patients were still controversial. The aim of this multi-center study was to identify these thresholds. Methods: 169 patients were included in this study. Upper intervertebral angle (UIVA), lower intervertebral angle (LIVA), Cobb angle (CA), anterior vertebral height ratio (AVH%), regional angle (RA), posterior vertebral height ratio (PVH%), vertebral wedge angle (VWA), anteroposterior ratio (A/P%), Clinical assessment included Load Sharing Classication (LSC) score, Thoracolumbar Injury Classication and Severity (TLICS) score, Visual Analogue Scale (VAS), and Body mass index (BMI) were perioperatively evaluated. Patients were divided into KR group and none KR (NKR) group according to whether the loss of CA correction was less than 15˚ or not. The risk factors of KR before or after implant removal were analyzed, respectively. Result: There were signicant improvements in postoperative parameters compared with preoperative parameters, such as AVH%, A/P%, VAS, CA, VWA, PVH% (P < 0.001, respectively), and UIVA (P = 0.02). Age (AUC = 0.828) and BMI (AUC = 0.846) were good predictors of KR before implant removal. BMI (AUC = 0.871) was a good predictor of KR after implant removal. Conclusion: There were signicant differences in risk factors of KR at different postoperative follow-up stages: age > 49 years, BMI > 24 were risk factors of KR before implant removal; BMI > 25.17 was a risk factor of KR. SSPI-f = short-segment pedicle instrumentation with screw insertion in the fractured vertebra, KR = the recurrence of kyphosis, AVH% = anterior vertebra height ratio, ROC = curve: receiver operating characteristic curve, TLICS score = Thoracolumbar Injury Classication and Severity score, BMD = bone mineral density, CT = computed tomography, NKR = none recurrence of kyphosis, CA = Cobb angle, RA = regional angle, VWA = vertebral wedge angle, AVH% = anterior vertebral height ratio, AVH0 = anterior height of the fractured vertebra, AVH1 = anterior height of the upper adjacent vertebra, AVH2 = anterior vertebral height of the lower adjacent vertebra, PVH% = posterior vertebral height ratio, PVH0 = percentage of posterior height ratio of the fractured vertebra, PVH1 = posterior height of the upper adjacent vertebra, PVH2 = posterior height of the lower adjacent vertebra, A/P% = anteroposterior ratio, UIVA = upper intervertebral angle, LIVA = lower intervertebral angle, VAS = Visual Analogue Scale, LSC score = Load Sharing Classication score, BMI = body mass index, OR = odds ration, CI = condence interval, AUC = area under the curve, SE = superior endplate, IE = inferior endplate, UAV = upper adjacent vertebra, LAV = lower adjacent vertebra.


Introduction
Thoracolumbar spine (T11-L2) is the most frequent site of vertebral fractures; the transition from the less mobile thoracic spine with support from the ribs and sternum to the more dynamic lumbar spine makes this junction a biomechanical weak point [1]. Therefore, most spinal injuries affect the thoracolumbar region, 17% of which are burst fractures [2].
The short-segment pedicle instrumentation with the screw insertion in the fractured vertebra (SSPI-f) usually provides su cient xation with good clinical outcomes [3]. However, it is also associated with common complications, such as failure of the instrumentation, the recurrence of kyphosis (KR) and sagittal spinal imbalance [4,5]. Jang et al. [6] reported that predictors of vertebra re-collapse for thoracolumbar burst fracture were age and the loss of preoperative body height. Nevertheless, they combined long-segment pedicle instrumentation with short-segment pedicle instrumentation in the study.
Because the follow-up time of their study is only one year, they did not fully discuss the in uence of implant removal on spinal correction. In addition, they failed to discuss the KR after SSPI-f, which seemed to be of greater clinical signi cance than vertebral body re-collapse. Chen et al. [3] studied the effect of implant removal on KR after implant removal on KR after SSPI-f, and found that the loss of correction was associated with anterior vertebra height ratio (AVH%) < 50%, anteroposterior ratio and age. However, they did not fully analyze the factors in uencing KR before implant removal. The thresholds of risk factors in their study were based on experience and the results of previous studies; however, these were not analyzed by receiver operating characteristic (ROC) curve. Therefore, their research methods need to be improved.
In order to solve all the problems mentioned above, we analyzed the risk factors of KR before implant removal and after implant removal respectively in the present study. To identify the thresholds of the risk factors, we used logistic regression model and ROC curve in the analysis. Cases gathered from three health centers were studied in this research, which would make the results more convincing.

Inclusion and Exclusion Criteria
We retrospectively reviewed 224 consecutive thoracolumbar burst fracture patients presenting to the hospitals (Xuanwu Hospital Capital Medical University; Beijing Chaoyang Hospital, Capital Medical University; Chui Yang Liu Hospital a liated to Tsinghua University) between January 1, 2014 and January 1, 2016. All patients received SSPI-f within 3 days. The patients were older than 18 years. The follow-up time was two years. We removed the xation device with about 12 months after SSPI-f. The inclusion criteria were as follows: Thoracolumbar Injury Classi cation and Severity (TLICS) score was more than 4; kyphosis was more than 15°; anterior vertebral height was less than 50%. The exclusion criteria were as follows: Multiple contiguous fractures or non-contiguous fractures; bone mineral density (BMD) of the thoracolumbar spine was at least 2.5 SD below the mean of young normal men; pathological fracture such as tumor, ankylosing spondylitis, in ammatory arthritis; posterior fusion or posterolateral fusion, and laminectomy in the operation; neurological de cit, obsolete thoracolumbar fracture, previous history of spinal surgery. Finally, 169 (90 men and 79 women) patients were included in this study.

Radiological Measurements
Whole-spine X-ray (Philips Digital Diagnost) and computed tomography (CT) (Philips Medical Systems) were used to evaluate the patients. Superior endplate (SE) fracture in the fractured vertebra was evaluated on CT images. Radiographic parameters were obtained, preoperatively or postoperatively. In addition, the parameters were collected before implant removal and at last follow-up. The following radiological parameters [3,6,7] were evaluated: Cobb angle (CA) was measured between the inferior endplate (IE) of the lower adjacent vertebra (LAV) and the SE of the upper adjacent vertebra (UAV); vertebral wedge angle (VWA) was the angle between the IE and the SE of fractured vertebra; regional angle (RA) was de ned as the angle between the IE of the UAV and the SE of the LAV; posterior vertebral height ratio (PVH%) was calculated by the ratio of posterior height of the fractured vertebra (PVH0) to the average between posterior height of the UAV (PVH1) and posterior height of the LAV (PVH2); anterior vertebral height ratio (AVH%) was calculated by the ratio of anterior height of the fractured vertebra (AVH0) to the average of anterior height of the UAV (AVH1) and anterior vertebral height of the LAV (AVH2); anteroposterior ratio (A/P%) was the ratio of AVH0 to PVH0; upper intervertebral angle (UIVA) was the angle between the IE of UAV and the SE of fractured vertebra; lower intervertebral angle (LIVA) was the angle between the SE of LAV and the IE of fractured vertebra.

Clinical Assessments
The extent of anterior column comminution was assessed in accordance with Load Sharing Classi cation (LSC) score [8]. Back pain was evaluated according to a Visual Analogue Scale (VAS).
Thoracolumbar Injury Classi cation and Severity (TLICS) score [9] was used to weight the severity of fracture. Body mass index (BMI) was used to evaluate the body size of the patients. Whether the patients got diabetes or not was also recorded in this research.

Grouping Methods
Patients with the CA less than 15° were included in the KR group; however, patients with the CA more than 15° were included in the none KR (NKR) group [10]. The risk factors of KR before or after implant removal were analyzed, respectively. SPSS 17.0 (SPSS Inc, Richmond, CA, USA) was used to perform the statistical analysis. Mean ± standard deviations were used to represent continuous variables. Kolmogorov-Smirnov test was performed to the normal distribution of the data. Normally distributed values were analyzed with independent Student t test. Skewed values were analyzed with Kruskal-Walllist test. Categorical variables were reported as the number of cases and were evaluated in Pearson's χ 2 test. Candidate variables were identi ed for multivariate logistic regression model via univariate screening to identify the risk factors. ROC (receiver operating characteristic) curve was calculated by the results of multivariate logistic regression analysis. It was used to determine the predictability of the risk factors for KR. P value < 0.05 was considered of statistical signi cance.
Results of multivariable logistic regression analysis showed two signi cant risk factors of KR after implant removal: BMI (P < 0.001) and AVH% (P = 0.008). However, preoperative AVH% was a protective factor for KR ( Table 2).

Discussion
During daily activities, spinal posture and thoracic kyphosis in uence the load distribution in vertebral body; repetitive cyclic loading will cause collapse of fractured vertebral body; these can lead to thoracolumbar kyphosis [11]. It has been reported postoperative KR may be of greater signi cance than postoperative vertebral body re-collapse [12,13]. Therefore, our study analyzed the risk factors of postoperative KR. Our study showed that correction loss progressed most rapidly within 12 months after surgery; correction loss progressed much slowly after implant removal. Therefore, it's necessary to separately discuss the risk factors of KR before or after implant removal.
Previous studies showed that age will primarily affect bone remodeling toward resorption; this was because vertebral body contained numerous trabecular bone [14,15]. All these studies supported our results that age > 49 years was a risk factor of KR before implant removal. There was a lack of studies on the correlation between BMI and KR. Our study showed that BMI > 24.0 was a risk factor of KR before implant removal; BMI > 25.17 was a risk factor of KR after implant removal; these implied that BMI works differently at different stages after surgery. Bone healing of fractured vertebra has not been fully achieved before implant removal [16]. Therefore, a high BMI will increase the load on instruments and even lead to the deformation of implants. The fractured vertebra has healed after implant removal; therefore, the KR after implant removal may be mainly caused be the loss of UIVA correction; in this period, a higher BMI with longer action time is needed to result in space collapse, reduction of buffering capacity and degeneration of upper intervertebral discs [17]. Chung et al. [18] reported that diabetes mellitus could increase the risk of fragility fractures, which was often associated with osteoporosis. Our study showed that diabetes mellitus was not a risk factor of KR. This was because the most thoracolumbar fractures of patients with diabetes mellitus involved more than one vertebra, which was much severer than singlelevel thoracolumbar burst fractures. According to the inclusion and exclusion criteria, our study included a small number of patients with diabetes mellitus, which might in uence our results.
Fractured endplates and nucleus pulposus were intruded in to fractured vertebral body during burst fracture; this could result in collapse of the cancellous framework in the vertebral body [12]. During SSPI-f, annulus brosus of the collapsed intervertebral disc was distracted; then the wall of fractured vertebral body was reduced by the annulus brosus attached to the periphery of the endplate; however, because the annulus brosus was not su ciently attached to the center of the endplate, the fragmented nucleus, endplates and collapsed cancellous framework were not su ciently reduced [12]. Similarly, our study showed that AVH% > 49.5%, PVH% > 85.5% were protective factors of KR before implant removal. This was because low AVH% and low PVH% usually meant a severer vertebral body compression. These would lead to large bone defect created inside the fractured vertebra after reduction, which had been reported as an important reason of internal xation failure and KR [19]. Additionally, the results in our study showed that AVH% > 61.5% was a protective factor of KR after implant removal, which implied that the loss of anterior vertebra height had a long-term impact on therapeutic effect. Preoperative CA, RA and VWA were not the risk factors of KR in our study. The possible explanation might be that thoracolumbar burst fractures usually involved anterior and posterior columns, both anterior and posterior walls of vertebra were fractured; therefore, compared with vertebra height, the variation of vertebra angle could not accurately the severity of thoracolumbar burst fractures [20].
The fractured endplates and nucleus pulposus were intruded into the fractured vertebral body during burst fracture, which would result in collapse of the cancellous framework in the vertebral body [21]. Similarly, Oner et al. [22] reported that even with the fracture of end plates, the upper discs were intact in most of the patients. It could be seen from the trend of parameters in our study that the loss of correction caused by the collapse of upper intervertebral disc of fractured vertebra was a slow process, which was not due to the adjacent disc injured by direct high-energy violence. Therefore, superior endplate fracture, the decrease of preoperative UIVA and preoperative LIVA were not the risk factors of KR.
Our results showed that preoperative VAS score could not be used to predict the risk of KR. The reason was that VAS score had strong subjectivity with many in uence factors, which made its accuracy limited [3,6]. Pellise et al. [23] pointed out that LSC score was positively correlated with correction loss. However, our study showed that LSC score was not a risk factor of KR. The possible explanation might be that most of the previous studies discussed the curative effect of skip-level xation without inserting additional screws into the fractured vertebra. Our study analyzed the treatment effect of SSPI-f, which could increase stiffness and reduce the failure rat of instrumentation with the insertion of two screws in the fractured vertebra; this seemed to reduce the in uence of LSC score on the treatment effect [19]. TLICS was comprehensive not only in determining injury prognosis, but also in guiding treatment [24].
Our study found that TLICS was not a risk factor of KR. The reason might be that the mean TLICS was more than 4 points in our study; the fractured vertebras of patients with higher TLICS were highly unstable, which was usually treated with long-segment xation and should be excluded from our research.
There were limitations in our study. First, it was a retrospective study that developed the detailed prediction model; however, our study was a multi-center study with strict inclusion and exclusion criteria, which was more meaningful compared with prior risk factor analysis. Second, magnetic resonance imaging (MRI) was not fully discussed in our study, which made the evaluation of soft tissue around fractured vertebra very di cult; this might lead to the omission of signi cant risk factors.

Conclusion
Our study proved that the effectiveness of SSPI-f in the treatment of thoracolumbar burst fractures was good; the correction loss was mainly found within 12 months after surgery. Therefore, patients should be carefully followed up within 1 year after surgery. There were signi cant differences in the risk factors of KR at different postoperative follow-up stages. Patients who have older age (> 49 years), higher BMI (> 24), lower AVH% (< 49.5%) and lower PVH% (< 85.5%) were more likely to suffer from KR before implant removal. These patients should be properly treated with conservative methods, such as waring braces and doing rehabilitation exercises, to actively prevent the occurrence of KR before implant removal. KR before implant removal is more likely to occur in patients with higher BMI (> 25.17) and lower AVH% (> 61.5%). In such patients, the removal time of internal xation should be extended. Patients with poor physical conditions should maintain the internal xation without removal. Among all these predictive factors, age and BMI were the most accurate factors.

Abbreviations
SSPI-f = short-segment pedicle instrumentation with screw insertion in the fractured vertebra, KR = the recurrence of kyphosis, AVH% = anterior vertebra height ratio, ROC = curve: receiver operating characteristic curve, TLICS score = Thoracolumbar Injury Classi cation and Severity score, BMD = bone mineral density, CT = computed tomography, NKR = none recurrence of kyphosis, CA = Cobb angle, RA The trend of clinical and radiological parameters in patients with thoracolumbar burst fracture