Nomogram that can predict failure of conservative treatment for thoracolumbar burst fracture was established Authors: Le-le Sun, Wang-bao Qiu, Wei Liang, Jia-qi Chu, Bao-qing Shi, Han-bang Wang Published: 2025 ABSTRACT ObjectiveTo identify predictors of failure of conservative management for thoracolumbar burst fracture (TBF) and to develop and validate a nomogram for risk stratification.MethodsWe analyzed 152 patients with TBF treated conservatively at our center from December 2020 to December 2024. Patients were categorized into the Successful Received Conservative Treatment (SRCT) group (n = 120) and the Converted to Surgical Treatment (CST) group (n = 32). Risk factors for conversion were assessed using univariate and multivariate logistic regression. A nomogram was constructed from the final model and internally validated with 1000 bootstrap resamples. Model performance was evaluated by discrimination [area under the receiver operating characteristic curve (AUC)], calibration, and clinical utility [decision curve analysis (DCA) and clinical impact curve (CIC)].ResultsAge, interpedicular distance, canal compromise, and local kyphosis angle were independent predictors of CST (all p < 0.05). The nomogram showed good discrimination (AUC 0.850, 95% CI 0.774–0.926) and calibration (mean absolute error 0.02). DCA indicated clinical benefit across threshold probabilities of 0.05–0.75, and CIC supported its utility for identifying high-risk patients.ConclusionsAge, interpedicular distance, canal compromise, and local kyphosis angle are independent predictors of CST. The internally validated nomogram offers individualized risk estimates and may aid early identification of patients at high risk of surgical conversion; external validation is needed. Introduction Thoracolumbar (TL) injuries most commonly occur between T10 and L2. The TL region, as a transitional zone between the rigid, kyphotic thoracic spine and the dynamic, lordotic lumbar spine, is particularly susceptible to trauma [1]. Thoracolumbar burst fractures (TBF) are the most frequent type of spinal injury, accounting for approximately two-thirds of all spinal fractures. These injuries typically result from falls from height, sports-related trauma, or motor vehicle accidents. Axial loading, with or without accompanying bending forces, can lead to vertebral fractures involving the anterior and middle columns, and in some cases, the posterior column or lamina, potentially causing spinal cord injury [2]. For patients with intact neurological function and an uninjured posterior ligamentous complex (PLC), conservative management is generally recommended according to the thoracolumbar injury classification and severity score (TLICS) [3]. Conservative treatment offers the advantages of reduced medical costs and avoidance of surgical complications, whereas surgical intervention allows for shorter bed rest, prompt correction of deformities, and prevention of potential kyphotic progression [4]. However, recent studies indicate that progressive kyphosis, worsening neurological deficits, and persistent pain during follow-up may necessitate surgical intervention, which can adversely affect quality of life and increase economic burden [5]. The AO Spine classification demonstrates substantial inter- and intraobserver reliability (κ = 0.62) and facilitates comparison of treatments for fractures with similar morphological characteristics [6]. Nevertheless, TL fractures often exhibit heterogeneous radiological patterns, and misclassification is possible, particularly for A4 fractures (κ = 0.19). Although AO classification modifiers can provide more precise fracture analysis, their use in clinical practice and research remains inconsistent [7, 8]. Beyond fracture morphology and modifiers, PLC integrity is crucial for guiding treatment decisions, as PLC injury compromises spinal stability and increases the risk of fracture subsidence [9]. Treatment decisions are further influenced by surgeon experience, education, regional guidelines, resource availability, and patient preferences [10, 11]. Given these factors, identifying risk factors for conservative treatment failure in TL fractures is of significant clinical importance. Previous studies have evaluated variables such as age, sex, smoking, diabetes, visual analog scale (VAS) scores, interpedicular distance (IPD), canal compromise (CC), and local kyphosis angle (LKA) [12–14]; however, prediction models that account for overall patient health have not been established. This study aims to analyze baseline factors associated with conservative treatment failure in TBF and to develop and validate a nomogram to provide personalized guidance for clinical decision-making. Materials and methods A retrospective review of the institutional database from 2020 to 2024 identified 152 patients diagnosed with thoracolumbar burst fractures (TBF) involving the T11–L2 segments.Inclusion criteria were as follows: (1) single-level fracture at T11–L2; (2) age between 18 and 65 years; (3) absence of neurological symptoms; (4) Concomitant severe scoliosis and degenerative changes in the patients compromised the assessment; and (5) a thoracolumbar injury classification and severity score (TLICS) of less than 4 at the initial assessment.Exclusion criteria included patients younger than 18 years, individuals with illiteracy, pathologic or osteoporotic fractures, multiple vertebral fractures, a history of lumbar surgery, or any neurological impairment. The patient selection process is illustrated in Fig. 1. During the initial study design and sample size calculation, the target sample size was estimated based on the reported failure rate of conservative treatment. Previous studies indicate a conservative treatment failure rate of approximately 20%. Using this parameter, the minimum required sample size was calculated to be n ≥ 246 (formula: n = (1.962 × 0.2 × 0.8)/0.052 ≈ 246) [15, 16]. However, the final analyzed cohort consisted of 152 patients, which did not meet our prespecified target of 246 and falls substantially short of the sample size generally recommended for constructing a robust multivariable model.Fig. 1Flowchart for selecting patients Flowchart for selecting patients The discrepancy primarily stems from several practical research limitations: the target patient population in this study is highly specific, resulting in slower-than-anticipated recruitment progress; to maintain cohort homogeneity, we applied relatively stringent inclusion criteria, leading to a significant number of screened individuals not meeting the eligibility requirements; moreover, within the constrained study timeline, it was not feasible to extend the recruitment period indefinitely. The nonoperative management protocol consisted of immobilization, analgesia, and prophylaxis for venous thromboembolism. Before application of a thoracolumbosacral orthosis (TLSO), all patients underwent 3–5 days of strict bed rest. Thereafter, patients were allowed to ambulate while wearing the brace. The TLSO was required to be worn continuously, except when lying flat in bed. Patients wore the brace for 12 weeks, with gradual weaning initiated at the eighth week of follow-up [17–19]. The surgical indications encompassed the following criteria: (1) the development of new or worsening neurological deficits; (2) documented progression of deformity, characterized by an increase in kyphosis exceeding10° on serial imaging; and (3) the presence of refractory pain, specifically severe, activity-limiting pain that did not respond to conservative management. In young or highly active patients considered at elevated risk for long-term complications, surgical intervention was prioritized to enhance spinal stability and facilitate earlier functional recovery. Surgical procedures included subtotal vertebral resection for direct decompression and reduction, followed by autologous bone grafting to reconstruct the anterior and middle spinal columns. Baseline risk factors were extracted from medical records, operative notes, radiological imaging data, and questionnaire surveys. Radiographic assessments were independently performed by two spine surgeons, with discrepancies resolved through a consensus meeting. Conservative treatment was considered effective if no surgical intervention was required during follow-up. Follow-up was terminated once surgical conversion became necessary due to symptom progression. Based on the follow-up results, we divided the patients into Successful Received Conservative Treatment (SRCT) group and Converted to Surgical Treatment (CST) group. At the final follow-up, clinical outcomes were evaluated using the Oswestry Disability Index (ODI), visual analog scale (VAS), local kyphotic angle (LKA), and percentage of anterior height compression (PAHC). Baseline demographic characteristics included age, sex, body mass index (BMI), hypertension, mechanism of injury, diabetes, and smoking status. The following clinical and radiological parameters were analyzed: (1) fracture level; (2) VAS; (3) interpedicular distance (IPD); (4) PAHC; (5) LKA; and (6) canal compromise (CC). CC was calculated as the ratio of the spinal canal area at the fractured vertebra to the average canal area of the two adjacent, intact levels. The methods used to measure these variables are illustrated in Figs. 2, 3, and 4.Fig. 2LKA Local Kyphotic Angle, LKA refers to the angle at which the two endpoints were localized. PAHC Percentage of Anterior Height Compression, AVH Anterior Vertebral Height, PAHC is equal to 100% of AVH/[(AVH* + AVH**)/2], where AVH refers to the anterior vertebral height, and AVH* and AVH** stand for the anterior vertebral heights of the vertebrae above and below the fracture, respectivelyFig. 3IPD Interpedicular Distance, IPD calculated by comparing the enlargement of the pedicles of the broken vertebrae to the mean of equivalent values obtained from the levels above and belowFig. 4CC Canal Compromise, CC was measured on the axial CT image of the vertebral body at the level showing the greatest protrusion of the fracture fragment into the spinal canal, as illustrated in the figure above LKA Local Kyphotic Angle, LKA refers to the angle at which the two endpoints were localized. PAHC Percentage of Anterior Height Compression, AVH Anterior Vertebral Height, PAHC is equal to 100% of AVH/[(AVH* + AVH**)/2], where AVH refers to the anterior vertebral height, and AVH* and AVH** stand for the anterior vertebral heights of the vertebrae above and below the fracture, respectively IPD Interpedicular Distance, IPD calculated by comparing the enlargement of the pedicles of the broken vertebrae to the mean of equivalent values obtained from the levels above and below CC Canal Compromise, CC was measured on the axial CT image of the vertebral body at the level showing the greatest protrusion of the fracture fragment into the spinal canal, as illustrated in the figure above To investigate whether these variables were associated with the failure of nonoperative treatment, a univariate analysis was first performed. The Shapiro–Wilk test was applied to assess the normality of continuous variables. Depending on the distribution, Student’s t-test was used to compare continuous data between groups, while categorical variables were analyzed using the chi-squared test or Fisher’s exact test. Variables with P < 0.05 in the univariate analysis were further included in multivariate logistic regression to identify independent risk factors. Based on the identified predictors, a nomogram was constructed. Statistical analyses were performed using SPSS (version 26), and the nomogram was developed with R (version 4.3.1). Model performance and predictive accuracy were evaluated using calibration curves, clinical impact curves (CIC), receiver operating characteristic (ROC) analysis, area under the ROC curve (AUC), and decision curve analysis (DCA). An AUC value between 0.5 and 1.0 indicated the model’s discriminatory ability. Calibration curves were applied to assess the agreement between predicted probabilities of treatment failure and observed outcomes. CIC was used to evaluate the accuracy, applicability, and clinical value of the predictive scoring system. Furthermore, DCA was conducted to determine the clinical utility of the nomogram by calculating net benefit across a range of threshold probabilities. Finally, internal validation was performed with 1000 bootstrap resamples to assess the robustness and stability of the prognostic nomogram. Results A total of 152 patients were included in this study, with 32 in the CST group and 120 in the SRCT group, yielding a conservative treatment failure rate of 21.05%. At the final follow-up, significant differences were observed between the SRCT and CST groups in ODI, VAS, LKA, and PAHC scores on both clinical and radiological evaluations (p < 0.001), as presented in Table 1 and Fig. 5. Table 1Comparison of the last follow-up results between the two groups (x ± s)VariablesSRCT group (n = 120)CST group (n = 32)tP valueODI2.92 ± 2.7621.66 ± 6.6715.576 < 0.001VAS0.93 ± 0.943.28 ± 1.309.563 < 0.001LKA (°)4.28 ± 2.5411.77 ± 2.1015.319 < 0.001PAHC (%)82.63 ± 8.1760.41 ± 13.638.811 < 0.001ODI Oswestry Disability Index, VAS Visual Analog Scale, LKA Local Kyphotic Angle, PAHC Percentage of Anterior Height CompressionA Student’s t-test was performed for statistical analysis, and a P value < 0.05 was considered statistically significantFig. 5Boxplot of the clinical and imaging outcomes of the two groups’ last follow-up Table 1Comparison of the last follow-up results between the two groups (x ± s)VariablesSRCT group (n = 120)CST group (n = 32)tP valueODI2.92 ± 2.7621.66 ± 6.6715.576 < 0.001VAS0.93 ± 0.943.28 ± 1.309.563 < 0.001LKA (°)4.28 ± 2.5411.77 ± 2.1015.319 < 0.001PAHC (%)82.63 ± 8.1760.41 ± 13.638.811 < 0.001ODI Oswestry Disability Index, VAS Visual Analog Scale, LKA Local Kyphotic Angle, PAHC Percentage of Anterior Height CompressionA Student’s t-test was performed for statistical analysis, and a P value < 0.05 was considered statistically significant Comparison of the last follow-up results between the two groups (x ± s) ODI Oswestry Disability Index, VAS Visual Analog Scale, LKA Local Kyphotic Angle, PAHC Percentage of Anterior Height Compression A Student’s t-test was performed for statistical analysis, and a P value < 0.05 was considered statistically significant Boxplot of the clinical and imaging outcomes of the two groups’ last follow-up Univariate analysis of baseline demographic, clinical, and radiological data (Table 2) revealed significant between-group differences in age (P < 0.001), IPD (P < 0.001), LKA (P = 0.002), CC (P < 0.001), and smoking (P < 0.001), while no significant differences were noted for other variables. Multivariate logistic regression analysis identified age (OR = 1.057, 95% CI 1.008–1.108, P = 0.023), IPD (OR = 1.130, 95% CI 1.006–1.270, P = 0.040), LKA (OR = 1.136, 95% CI 1.027–1.257, P = 0.013), CC (OR = 1.042, 95% CI 1.006–1.079, P = 0.023), and smoking (OR = 0.310, 95% CI 0.107–0.896, P = 0.030) as independent risk factors for conservative treatment failure. In the multivariate logistic regression analysis, after adjusting for the effects of age, IPD, LKA, and CC, smoking showed an independent correlation with a reduced risk of positive outcome (OR = 0.310, 95% CI 0.107–0.896, p = 0.030). This finding contradicts the univariate analysis results, suggesting the presence of complex confounding effects. Given that this result contradicts the general consensus and removing the smoking variable had minimal impact on the model’s AUC, variable selection was ultimately based on statistical significance and contribution to model performance, leading to the exclusion of variables with non-significant contributions. The adjusted results are presented in Table 3. Table 2Univariate analysis of related variables in predicting failure of conservative treatmentVariablesSRCT group (n = 120)CST group (n = 32)χ2/tP valueAge (year)44.10 ± 11.0052.92 ± 11.454.020 < 0.001Gender (n) Male72 (60.0)18 (56.3)0.1470.701 Female48 (40.0)14 (43.7)Fracture classification (AO) A398 (81.7)25 (78.1)0.2050.651 A422 (18.3)7 (21.9)BMI (Kg/m2) < 18.5315.8840.102 18.5–23.9354 24–27.95715 ≥ 282512Diabetes Yes12 (10.0)4 (12.5)0.1680.746 No108 (90.0)28 (87.5)Hypertension Yes20 (16.7)8 (25.0)1.1670.280 No100 (83.3)24 (75.0)Smoking Yes18 (15.0)13 (40.6)10.218 < 0.001 No102 (85.0)19 (59.4)Cause of injury Road traffic33(27.5)7(21.9)3.6560.454 Sport14(11.7)2(6.3) Fall40(33.3)10(31.3) Assault15(12.5)8(25.0)Other18(15.0)5(15.6)Level of vertebra T106 (5.0)2 (6.3)2.2690.686 T1113 (10.8)3 (9.4) T1218 (15.0)4 (12.5) L148 (40.0)17 (53.1) L235 (29.2)6 (18.8) IPD (%)11.93 ± 4.3716.09 ± 4.654.732 < 0.001 LKA (°)13.29 ± 4.1715.94 ± 4.303.1740.002 CC (%)20.71 ± 13.4833.88 ± 13.524.907 < 0.001 PAHC (%)58.53 ± 10.9955.22 ± 8.241.5860.115 VAS7.65 ± 1.047.88 ± 0.831.2840.204BMI Body Mass Index, IPD Interpedicular Distance, LKA Local Kyphotic Angle, CC Canal Compromise, PAHC Percentage of Anterior Height Compression, VAS Visual Analog ScaleStudent’s t-test was used to compare continuous data between groups, while categorical variables were analyzed using the chi-squared test or Fisher’s exact test. A P value of < 0.05 was considered to indicate statistical significanceTable 3Multivariate logistic regression analysis of independent risk factors for failure of conservative treatmentRisk factorsOR95% CIP valueIPD (%)1.1391.015–1.2770.026LKA (°)1.1341.027–1.2510.012CC (%)1.0401.004–1.0750.027Age1.0631.014–1.1140.012IPD Interpedicular Distance, LKA Local Kyphotic Angle, CC Canal Compromise, OR odds ratio, CI confidence interval Table 2Univariate analysis of related variables in predicting failure of conservative treatmentVariablesSRCT group (n = 120)CST group (n = 32)χ2/tP valueAge (year)44.10 ± 11.0052.92 ± 11.454.020 < 0.001Gender (n) Male72 (60.0)18 (56.3)0.1470.701 Female48 (40.0)14 (43.7)Fracture classification (AO) A398 (81.7)25 (78.1)0.2050.651 A422 (18.3)7 (21.9)BMI (Kg/m2) < 18.5315.8840.102 18.5–23.9354 24–27.95715 ≥ 282512Diabetes Yes12 (10.0)4 (12.5)0.1680.746 No108 (90.0)28 (87.5)Hypertension Yes20 (16.7)8 (25.0)1.1670.280 No100 (83.3)24 (75.0)Smoking Yes18 (15.0)13 (40.6)10.218 < 0.001 No102 (85.0)19 (59.4)Cause of injury Road traffic33(27.5)7(21.9)3.6560.454 Sport14(11.7)2(6.3) Fall40(33.3)10(31.3) Assault15(12.5)8(25.0)Other18(15.0)5(15.6)Level of vertebra T106 (5.0)2 (6.3)2.2690.686 T1113 (10.8)3 (9.4) T1218 (15.0)4 (12.5) L148 (40.0)17 (53.1) L235 (29.2)6 (18.8) IPD (%)11.93 ± 4.3716.09 ± 4.654.732 < 0.001 LKA (°)13.29 ± 4.1715.94 ± 4.303.1740.002 CC (%)20.71 ± 13.4833.88 ± 13.524.907 < 0.001 PAHC (%)58.53 ± 10.9955.22 ± 8.241.5860.115 VAS7.65 ± 1.047.88 ± 0.831.2840.204BMI Body Mass Index, IPD Interpedicular Distance, LKA Local Kyphotic Angle, CC Canal Compromise, PAHC Percentage of Anterior Height Compression, VAS Visual Analog ScaleStudent’s t-test was used to compare continuous data between groups, while categorical variables were analyzed using the chi-squared test or Fisher’s exact test. A P value of < 0.05 was considered to indicate statistical significance Univariate analysis of related variables in predicting failure of conservative treatment BMI Body Mass Index, IPD Interpedicular Distance, LKA Local Kyphotic Angle, CC Canal Compromise, PAHC Percentage of Anterior Height Compression, VAS Visual Analog Scale Student’s t-test was used to compare continuous data between groups, while categorical variables were analyzed using the chi-squared test or Fisher’s exact test. A P value of < 0.05 was considered to indicate statistical significance Table 3Multivariate logistic regression analysis of independent risk factors for failure of conservative treatmentRisk factorsOR95% CIP valueIPD (%)1.1391.015–1.2770.026LKA (°)1.1341.027–1.2510.012CC (%)1.0401.004–1.0750.027Age1.0631.014–1.1140.012IPD Interpedicular Distance, LKA Local Kyphotic Angle, CC Canal Compromise, OR odds ratio, CI confidence interval Multivariate logistic regression analysis of independent risk factors for failure of conservative treatment IPD Interpedicular Distance, LKA Local Kyphotic Angle, CC Canal Compromise, OR odds ratio, CI confidence interval The predictive model was constructed with a threshold of 0.589, yielding a specificity of 0.808 and a sensitivity of 0.781. The classification results showed 114 true negatives, 17 true positives, 15 false negatives, and 6 false positives (Table 4). Table 4Confusion matrix: 0 is defined as true negative and 1 is defined as true positiveActual predict010114151617 Table 4Confusion matrix: 0 is defined as true negative and 1 is defined as true positiveActual predict010114151617 Confusion matrix: 0 is defined as true negative and 1 is defined as true positive A nomogram was developed based on the independent predictors from multivariate analysis to estimate the probability of conservative treatment failure (Fig. 6). The individual risk score for each predictor was summed to obtain the overall probability of treatment failure. The nomogram demonstrated excellent predictive performance, with an AUC of 0.850 (95% CI 0.774–0.926), indicating strong discrimination. Internal validation using 1000 bootstrap resamples showed an average absolute error of 0.02 (Figs. 7 and 8). The calibration curve further demonstrated close agreement between the predicted probabilities and the actual observed outcomes (Fig. 8).Fig. 6Predictive nomogram for failure of conservative treatmentFig. 7The discriminative performance of the predictive model was evaluated using a ROC curve. As shown in Fig. 7, the area under the curve (AUC) was 0.850 (95% CI 0.774–0.926), indicating good diagnostic accuracy. The optimal cutoff value, determined by the Youden index, was 0.192, corresponding to a sensitivity of 0.781, a specificity of 0.808, and a Youden index of 0.589. This cutoff provides the best balance between sensitivity and specificity, suggesting that a predicted probability above 0.192 can be considered a positive result, achieving both a relatively high true positive rate (78.1%) and a high true negative rate (80.8%). Furthermore, compared with individual predictors in the model (such as age, IPD, LKA, and CC), the integrated multivariable model demonstrated superior predictive performanceFig. 8The calibration curve depicts the concordance of the model’s predicted probabilities with the observed event rates. The ‘Ideal’ reference line signifies perfect calibration. The bias-corrected curve, obtained through 1000 bootstrap resampling iterations to correct for overoptimism, is nearly superimposed upon the ideal line, which reflects a very well-calibrated model. This visual assessment is supported by a mean absolute error of 0.02 Predictive nomogram for failure of conservative treatment The discriminative performance of the predictive model was evaluated using a ROC curve. As shown in Fig. 7, the area under the curve (AUC) was 0.850 (95% CI 0.774–0.926), indicating good diagnostic accuracy. The optimal cutoff value, determined by the Youden index, was 0.192, corresponding to a sensitivity of 0.781, a specificity of 0.808, and a Youden index of 0.589. This cutoff provides the best balance between sensitivity and specificity, suggesting that a predicted probability above 0.192 can be considered a positive result, achieving both a relatively high true positive rate (78.1%) and a high true negative rate (80.8%). Furthermore, compared with individual predictors in the model (such as age, IPD, LKA, and CC), the integrated multivariable model demonstrated superior predictive performance The calibration curve depicts the concordance of the model’s predicted probabilities with the observed event rates. The ‘Ideal’ reference line signifies perfect calibration. The bias-corrected curve, obtained through 1000 bootstrap resampling iterations to correct for overoptimism, is nearly superimposed upon the ideal line, which reflects a very well-calibrated model. This visual assessment is supported by a mean absolute error of 0.02 Decision curve analysis (DCA) indicated that the nomogram provided substantial net clinical benefit across a wide range of threshold probabilities (Fig. 9). Moreover, the clinical impact curve (CIC) confirmed the utility of the model for identifying high-risk populations at varying risk levels (Fig. 10).Fig. 9The decision curve analysis provides definitive evidence for the clinical value of our model. Its net benefit surpassed that of both universal intervention and non-intervention strategies across a wide threshold probability range of 5–75%. This implies that for clinicians whose intervention threshold lies within this spectrum, utilizing our model to identify high-risk patients represents the optimal strategy, maximizing clinical outcomes by accurately targeting therapyFig. 10The clinical impact curve shows that both the count of patients stratified into the high-risk group and the number of those who subsequently experienced the event were sustained at clinically meaningful levels. The remarkable proximity of these curves substantiates the model’s efficacy in correctly pinpointing patients at genuine risk, concurrently with a low false-positive rate. This evidence, which aligns with and strengthens the conclusions from the decision curve analysis, provides a compelling argument for the model’s utility in enabling effective risk stratification with a favorable benefit-harm profile The decision curve analysis provides definitive evidence for the clinical value of our model. Its net benefit surpassed that of both universal intervention and non-intervention strategies across a wide threshold probability range of 5–75%. This implies that for clinicians whose intervention threshold lies within this spectrum, utilizing our model to identify high-risk patients represents the optimal strategy, maximizing clinical outcomes by accurately targeting therapy The clinical impact curve shows that both the count of patients stratified into the high-risk group and the number of those who subsequently experienced the event were sustained at clinically meaningful levels. The remarkable proximity of these curves substantiates the model’s efficacy in correctly pinpointing patients at genuine risk, concurrently with a low false-positive rate. This evidence, which aligns with and strengthens the conclusions from the decision curve analysis, provides a compelling argument for the model’s utility in enabling effective risk stratification with a favorable benefit-harm profile Discussion In most cases of thoracolumbar burst fracture (TBF), conservative therapy yields satisfactory radiological outcomes and favorable clinical efficacy; therefore, it is generally recommended as a nonsurgical treatment. However, as the number of treated cases increases, the failure rate of conservative management has also been noted to rise. Scholars [12] reviewed data from 67 patients with TBF who received conservative therapy and found that 23.6% eventually failed nonsurgical treatment and required surgery due to persistent neurological symptoms and progressive spinal cord impairment. Similarly, Patrick [4] reported 68 cases of neurologically complete thoracic vertebral burst fractures, among which 18 patients developed pain during conservative treatment, with approximately 25% ultimately experiencing treatment failure. The TLICS scoring system is widely recognized and adopted due to its good intra- and inter-observer reliability, validity, and safety [20]. However, for neurologically intact patients with a complete burst fracture morphology but no PLC injury, the TLICS score is 2 points, and according to the guidelines, they should receive non-surgical management. Yet, multiple studies have shown that a significant proportion of these patients experience failure of non-operative treatment and ultimately require surgical intervention [21, 22]. This highlights the heterogeneity in the pathological features and treatment approaches for burst fractures, influenced by variables such as initial or progressive vertebral height loss, spinal canal stenosis, or kyphotic deformity—none of which are considered in the TLICS system [23]. The AO Spine Thoracolumbar Injury Classification System and the TL AOSIS score were introduced in 2013 and 2016 [24, 25], respectively. Both aim to establish universal classification systems that integrate clinical and radiological data and can provide treatment recommendations. By utilizing detailed CT-based descriptions of fracture morphology, these new systems address the limitations of the simplified fracture classification in TLICS. For instance, a neurologically intact patient with a complete burst fracture scores 2 points under TLICS, indicating non-surgical management, whereas under TL AOSIS, the complete burst morphology (AO subtype A4) scores 5 points, categorizing the patient into the “indeterminate for surgical indication” group. Multiple independent studies have validated the reliability of TL AOSIS and its advantages over TLICS, with the effectiveness of its treatment decision algorithm also being confirmed. However, the more detailed fracture classification may also pose a potential weakness—the proliferation of fracture subtypes increases system complexity and could lead to reduced inter-observer reliability for certain subtypes [26]. The management of burst fractures in neurologically intact patients remains an area of clinical equipoise which demands further research [27]. Therefore, it is necessary to develop predictive scoring systems to identify patients at high risk for conservative treatment failure in this specific fracture type. This study, using multivariate logistic regression analysis, identified age, IPD, LKA, and CC as significant predictors of conservative treatment failure, and constructed a nomogram prediction model based on these factors. Numerous studies have shown that elderly patients often exhibit diminished bone regeneration capacity and poorer bone quality, which may contribute to their increased risk of conservative treatment failure for TBF [12, 28]. In the present study, IPD was identified as another significant (p < 0.05) risk factor for the failure of conservative therapy. An increased IPD may promote the retropulsion of bone fragments into the spinal canal following trauma, thereby heightening the risk of neurological deficits [29]. Caffaro et al. [30] retrospectively analyzed the relationship between IPD and neurological impairment in TBF and concluded that IPD is a reliable indicator for predicting neurological complications. Ehsan Alimohammadi et al. [31] reported that a greater kyphotic angle at admission was significantly associated with the failure of conservative treatment, with the mean Cobb’s angle being 19.02° ± 3.33° in the failure group compared to 11.59° ± 3.26° in the successful group. Consistently, in the present study, patients who eventually required surgical intervention exhibited a significantly higher local Cobb’s angle (15.94° ± 4.30°) than those in the SCT group (13.29° ± 4.17°, p = 0.002), further supporting the association between an increased kyphotic angle and the failure of conservative therapy. Several studies have also shown that an increased kyphotic angle raises the likelihood of developing an intervertebral cleft at the fractured vertebral body. Moreover, many researchers agree that IVC is directly associated with instability, vertebral re-collapse, and persistent back pain [32]. Furthermore, residual canal area has been identified as a highly significant prognostic factor in several studies [33]. The degree of CC reflects the extent of retropulsion and vertebral body involvement, which may contribute to mechanical pain [4, 14]. In the present study, the CC in the SCT group was significantly higher than that in the CST group (33.88 ± 13.52 vs. 20.71 ± 13.48, p < 0.001). Previous studies have suggested that smoking is associated with failure of conservative treatment [31]. However, in our univariate analysis, smoking showed a statistically significant negative association with the outcome (OR = 0.310, p = 0.030), which contradicts established clinical knowledge and suggests the potential presence of unmeasured confounders or specific interaction effects. To rigorously evaluate the necessity of including this variable, we compared two predictive models—one including smoking (full model) and one excluding it (reduced model). The AUC of the full model was 0.852 (95% CI 0.774–0.930), while that of the reduced model was 0.850 (95% CI 0.774–0.926). The difference between the two AUC values was minimal, and their confidence intervals overlapped substantially, indicating that adding smoking history did not significantly improve model discrimination. Given the paradoxical direction of its effect, its negligible contribution to predictive performance, and our goal of developing a parsimonious and clinically applicable tool, we ultimately excluded smoking from the final model. The final model includes four predictors—age, IPD, LKA, and CC—all of which are clinically plausible and statistically significant, and together provide a tool with good discriminatory ability (AUC = 0.850) that is both interpretable and practical for clinical use. This study aimed to develop a nomogram—a novel predictive tool—to estimate the risk of failure of conservative treatment for thoracolumbar (TL) fractures and to assist in determining whether patients should undergo surgical intervention based on their overall physical condition and imaging findings. This predictive strategy has demonstrated reliability and accuracy in several disease models. Currently, the selection of predictors remains inconclusive and largely depends on statistical methods, clinical experience, and evidence from the literature. In this study, predictors with p < 0.05 in univariate logistic regression analysis were included in a multivariate logistic regression model and incorporated into the nomogram as independent risk factors. Model performance was evaluated using calibration curves, receiver operating characteristic (ROC) analysis, decision curve analysis (DCA), and clinical impact curves (CIC). The developed nomogram may allow clinicians to more intuitively, efficiently, and reliably predict the likelihood of conservative treatment failure in TBF. This study has several limitations. First, its retrospective nature inherently limits the ability to fully control for potential confounding factors, including patient compliance, socioeconomic status, and paraspinal muscle conditions. Second, the single-center design and limited sample size may affect the model’s stability. Although internal validation was conducted, the lack of an independent training/validation split and external validation means the results should be considered exploratory. All findings require verification in larger prospective cohorts before any potential clinical translation can be considered. Furthermore, the follow-up period was relatively short (mean 19.09 ± 5.5 months), highlighting the need for studies with longer-term follow-up in the future. Conclusion Age, IPD, CC, and LKA were discovered to be independent risk factors for CST in this study. The nomogram using the four predictors mentioned above may precisely probability the probability of CST.