Vertebral Rotation Discrepancy for Distal Add-on when Selecting the Touched Vertebra as the Lowest Instrumented Vertebra in Thoracic Curves of Adolescent Idiopathic Scoliosis

Abstract

Study Design

Retrospective cohort study.

Objectives

To assess the influencing factors of the distal add-on when selecting the touched vertebra (TV) as the lowest instrumented vertebra (LIV) in thoracic idiopathic scoliosis (IS).

Methods

A total of 96 thoracic IS patients who underwent deformity correction with LIV = TV were retrospectively analyzed to evaluate the distal add-on. Patients were divided into the DA group and the non-DA group. Radiological variables included coronal (Cobb’s angle, coronal balance, sacral slanting angle, and L4 vertebral tilt), sagittal (sagittal vertical axis, thoracic kyphosis, and lumbar lordosis), and axial parameters (vertebral rotation angle).

Results

When selecting the TV as the LIV, the occurrence rate of distal add-on was 20.8% (20 of 96) at a mean follow-up of 3 years. The mean number of remnant motion segments was significantly higher in the DA group (4.7) compared to the non-DA group (3.8) (P < 0.001). The difference in the vertebral rotation angle between the lumbar apex and the TV level was significantly higher in the DA group (6.6°) than in the non-DA group (1.3°) (P < 0.001). Multivariate logistic regression analysis demonstrated that the distal add-on was significantly associated with remnant motion segments (OR = 3.69, 95% CI: 1.03–13.18) and the vertebral rotation discrepancy (OR = 4.09, 95% CI: 1.98–8.43).

Conclusions

Greater vertebral rotation discrepancy between the LIV and lumbar apex was consistently associated with distal add-on when the TV was selected as the LIV, suggesting that three-dimensional rotational mismatch may contribute to residual distal add-on risk.

Introduction

Surgical treatment of adolescent idiopathic scoliosis (AIS) focuses on achieving a globally balanced spine while minimizing the number of fused levels.^1,2 While selective thoracic fusion (STF) in thoracic AIS curves is advantageous for the preservation of motion segments, inappropriate selection of the lowest instrumented vertebra (LIV) can result in the distal add-on.^3-5 This phenomenon is defined as a progressive loss of correction due to an increase in either the vertebral deviation of the lumbar spine or disc angulation below the instrumentation. ⁶ Therefore, the determination of the ideal LIV is critical for balancing the preservation of remnant motion segments with the reduction of LIV-related complication risks. ⁷

Several studies have proposed criteria for the ideal LIV based on radiological parameters and anatomical locations, including the stable vertebra, neutral vertebra, curve flexibility, residual curves, apical translation (AT), L4 tilt, and sacral slanting.^6,8-13 Since the initial introduction of the touched vertebra (TV) concept by Matsumoto et al, the TV rule in the Lenke classification has been recognized as an important landmark for determining the LIV in thoracic AIS.^4,8 While there is controversy regarding the definition of the TV—specifically whether it contacts the pedicle or the vertebral body—the TV is generally defined as the most cephalad thoracolumbar or lumbar vertebra touched by the central sacral vertical line (CSVL).^4,14 The rationale for selecting an LIV that contacts the CSVL is based on the premise that such a vertebra is relatively centered over the sacrum. ⁴

It is widely accepted that selecting the LIV at the TV represents the most balanced landmark between the risk of the distal add-on and the preservation of motion segments.^14,15 However, this does not imply that such a selection guarantees the prevention of the distal add-on. In other words, while fusing at least to the TV represents the ideal minimum level, the LIV may need to be extended more caudally based on additional preoperative factors, such as sacral slanting, AT, and L4 tilt.^8,9,12,16 Furthermore, despite the focus on these coronal and sagittal parameters, the influence of the axial plane—specifically vertebral rotation—remains underestimated.^1,17 Furthermore, the SRS–Lenke–Aubin 3D classification emphasizes the three-dimensional characterization of AIS, including axial rotation as a key parameter. ¹⁷ Given that the distal add-on is not uncommon when the LIV is set at the TV, the role of vertebral rotation discrepancy between the instrumented spine and the remaining lumbar curve remains unclear. Therefore, this study aims to analyze the occurrence of the distal add-on when the LIV is selected as the TV and to identify the influencing factors, with a particular focus on vertebral rotation discrepancy in patients with idiopathic scoliosis (IS).

Methods

Study design and population

After approval of the protocol from Institutional Review Board of our hospital (IRB number: 2026GR0035, approval date: January 26, 2026), this single-center retrospective study was conducted in compliance with the Declaration of Helsinki and the current guidelines of Strengthening the Reporting of Observational Studies in Epidemiology. ¹⁸ The requirement for written informed consent was waived because of the retrospective nature of the study. Data extraction and analysis were performed after IRB approval.

We performed a retrospective review of patients with IS who underwent deformity correction through posterior spinal fusion at a single-center between March 2015 and December 2023. The inclusion criteria was IS patients, including AIS and young adult IS, (1) who underwent deformity correction through posterior spinal fusion between March 2015 and December 2023; (2) with Lenke classification type 1 or 2; (3) who captured whole spine computed tomography (CT) scans, preoperatively. A total of 398 patients were initially screened. Patients were excluded if they had: (1) a history of previous spinal surgery or infection (n = 13); (2) incomplete medical/radiographic data or loss to follow-up before 2 years (n = 11); (3) non-idiopathic scoliosis (n = 2); or (4) an LIV that did not correspond to the TV (LIV ≠ TV) (n = 276). After applying the exclusion criteria, 96 patients were included in the final analysis. Based on the presence of the distal add-on at the last follow-up, patients were categorized into the distal add-on (DA) group (n = 20) or the non-distal add-on (Non-DA) group (n = 76) (Figure 1).

Figure 1.

Study flowchart

Surgical procedures

All surgical procedures were performed by a single senior surgeon with vast experience in deformity correction of AIS. Deformity correction and posterior spinal fusion was performed in all patients by a single senior surgeon. Polyaxial pedicle screws were inserted at all functional segments. After pedicle screw placement, precontoured rods were sequentially inserted in a cephalad-to-caudal direction. The concave rod was contoured to exaggerate the physiologic sagittal profile by approximately one-third, and the convex rod was contoured accordingly. Deformity correction was primarily achieved through rod derotation, followed by appropriate segmental correction to optimize coronal and sagittal alignment. No osteotomy for deformity correction such as posterior column osteotomy, vertebra column resection was performed in any patient included in the present cohort. Limited facet joint preparation such as partial facetectomy was considered part of the routine exposure for pedicle screw instrumentation and fusion. Posterior fusion was completed using a combination of autologous and allogenous bone grafts. Thoracoplasty was selectively performed when clinically indicated, particularly in patients with a prominent rib hump of ≥ 3 cm. The overall instrumentation strategy, rod contouring, correction principles, and indications for adjunctive procedures remained consistent throughout the study period, without meaningful temporal changes in surgical practice.

Measurements

Demographic variables collected comprised sex, age, body mass index (BMI), Risser stage, and follow-up duration. Curve characteristics assessed included Lenke classification (type 1 or 2), the number of fused segments, performance of thoracoplasty, rib resection level, L4 vertebral tilt (left or right), and sacral slanting (defined as a sacral slanting angle ≥ 3°). ¹²

Radiological parameters were measured on standing whole-spine posteroanterior and lateral plain radiographs taken preoperatively, postoperatively, and at the last follow-up. The TV was defined as the most cephalad thoracolumbar or lumbar vertebra where the CSVL touched any part of the vertebral body. The distal add-on was defined as (1) an increase of ≥ 5 mm in the deviation of the first vertebra below the instrumentation from the CSVL, or (2) an increase of ≥ 5° in the angulation of the first disc below the instrumentation compared to the immediate postoperative radiographs. ⁶ These criteria represent a well-established methodology known to yield highly consistent results and are widely utilized in outcome analyses.^6,19,20 Coronal parameters included the Cobb angle of the main thoracic, proximal thoracic, and lumbar curves, curve flexibility, apical translation (AT), coronal balance (CB), LIV tilt, LIV disc angle, and CSVL–LIV distance. Sagittal parameters included the sagittal vertical axis (SVA), thoracic kyphosis (TK, measured as the sagittal Cobb angle between T5 and T12), and lumbar lordosis (LL, measured as the sagittal Cobb angle between L1 and S1).

To assess the relationship between the lumbar apex and the TV level, we investigated the TV level, lumbar apex level, vertebral-level interval between TV and lumbar apex, number of remnant motion segments, and the vertebral rotation angle of both the lumbar apex and the TV. They were independently assessed by two board-certified orthopedic surgeons, each with more than 5 years of clinical experience, who were blinded to the radiological measurements relevant to the present analysis. The vertebral rotation angle was measured using preoperative CT scans according to the method described by Aaro-Dahlborn et al. ²¹ The vertebral rotation discrepancy was calculated as the absolute value of the difference in vertebral rotation angle between the lumbar apex and the TV. Additionally, the number of remnant motion segments was defined as the number of intervertebral disc spaces between the LIV and the sacrum.

Statistical analysis

Data analysis and visualization were performed using Python (version 3.12.3; Python Software Foundation, Wilmington, DE, USA) using scikit-learn (version 1.8.0; for LASSO modelling), statsmodels (version 0.15.0; for conventional and bootstrap logistic regression and calibration analyses), NumPy (version 2.4.4; for Firth penalized regression and reliability analyses), SciPy (version 1.17.; for Firth penalized regression and reliability analyses), and pandas (version 3.0.2) with matplotlib library (version 3.10.8). The distribution of continuous variables was assessed using the Shapiro–Wilk test together with visual inspection of Q–Q plots. Continuous variables were presented as mean ± standard deviation (SD) and were compared using the independent t-test or Mann-Whitney U test, depending on the normality of the distribution. The Kruskal-Wallis test was used to the compare the rotational discrepancy according to the lumbar modifiers. Categorical variables were expressed as counts and percentages and were compared using the Chi-square test or Fisher’s exact test, as appropriate. Pearson’s correlation analysis was used to assess the relationships between the LIV-related parameters and the change of residual AT. Interobserver and intraobserver reliability for vertebral rotation discrepancy were assessed using intraclass correlation coefficients (ICCs).

To identify independent risk factors for the distal add-on while addressing potential multicollinearity and preventing overfitting, we employed the Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression model. The optimal tuning parameter (λ) was determined using 5-fold cross-validation with the one-standard-error rule. Variables selected by the LASSO model was included in the final multivariable logistic regression analysis model, estimating odd ratio (OR) and 95% confidence interval (CI). ²² The final two-predictor model was internally validated using 1,000 bootstrap resamples after LASSO variable selection. Because variable selection was not repeated within each bootstrap sample, the validation was conditional on the selected model and did not account for model-selection uncertainty. Model performance was assessed using the optimism-corrected area under the receiver operating characteristic curve (AUC) and Brier score, with 95% confidence intervals (CIs) obtained from the percentile distribution of the bootstrap optimism-corrected estimates. A fixed random seed of 2026 was used for resampling. Calibration intercept and slope were estimated by applying each bootstrap-fitted model to the original cohort and averaging the resulting calibration estimates across bootstrap resamples. Statistical significance was set at P < 0.05.

Results

Baseline characteristics

The baseline characteristics are summarized in Table 1. Among the 20 patients who developed distal add-on, 12 patients (60.0%) met the distal disc-angle criterion, 17 patients (85.0%) met the CSVL–LIV+1 deviation criterion, and 9 patients (45.0%) met both criteria. The mean age was 16.1 years in the DA group and 17.8 years in the Non-DA group, showing no significant difference (P = 0.139). The cohort consisted of an adolescent group (≤18 years, n = 54) and a young adult group (>18 years, n = 42), with no statistical difference in the occurrence of distal add-on (P = 0.375). The distribution was modifier A in 69 patients (71.9%), B in 14 (14.6%), and C in 13 (13.5%). Distal add-on occurred in 12 of 69 modifier A patients (17.4%), 5 of 14 modifier B patients (35.7%), and 3 of 13 modifier C patients (23.1%), with no significant difference between groups (P = 0.299). There was no statistical difference in the L4 vertebral tilt (P = 0.246) and preoperative sacral slanting (P = 0.069).

Table 1.

Baseline characteristics between two groups

Variables	Cases (n = 96)	DA group (n = 20)	Non-DA group (n = 76)	P
Demographics
Sex (male/female, n)	6/90	1/19	5/71	0.795
Age (mean ± SD, years)	17.4 ± 4.4	16.1 ± 4.1	17.8 ± 4.5	0.139
Adolescent (10-18 years)	54	13	41	0.375
Adult (>18 years)	42	7	35	​
BMI (kg/m2)	18.7 ± 2.2	19.4 ± 2.5	18.6 ± 2.1	0.135
Risser stage (mean ± SD, grade)	3.8 ± 1.5	3.5 ± 1.5	3.9 ± 1.5	0.263
Follow-up duration (mean ± SD, years)	3.0 ± 1.8	2.7 ± 1.8	3.1 ± 1.8	0.402
Curve characteristics
Lenke classification (type 1/type 2, n)	86/10	16/4	70/6	0.115
Lenke lumbar modifier (A/B/C, n)	69/14/13	12/5/3	57/9/10	0.299
Fused segments (mean ± SD, n)	11.7 ± 1.2	11.4 ± 0.8	11.8 ± 1.2	0.090
Thoracoplasty (yes/no, n)	92/4	19/1	73/3	0.834
Rib resection level (mean ± SD, n)	4.2 ± 1.4	4.2 ± 1.3	4.2 ± 1.4	0.832
L4 vertebral tilt (Left/Right, n)	24/72	7/13	17/59	0.246
Preoperative sacral slanting (yes/no, n)	22/74 (22.9%)	8/12 (40.0%)	14/62 (18.4%)	0.069
Sacral slanting in L4-L type	13/11	5/2	8/9	0.386
Sacral slanting in L4-R type	9/63	3/10	6/53	0.349

N, number; DA, distal adding-on; SD, standard deviation; BMI, body mass index; TV, touched vertebra; R, right; L, Left.

Radiological parameters

The mean preoperative main thoracic Cobb angle was 63.1° in the DA group and 57.1° in the Non-DA group, showing a significant difference (P = 0.011). At the last follow-up, the mean main thoracic Cobb angle was significantly greater in the DA group (24.5°) than in the Non-DA group (20.9°) (P = 0.038). The mean preoperative lumbar Cobb angle was 28.8° in the DA group and 21.4° in the Non-DA group, showing a significant difference (P < 0.001). At the last follow-up, the mean lumbar Cobb angle was significantly greater in the DA group (11.0°) than in the Non-DA group (4.8°) (P < 0.001). The mean postoperative AT was significantly greater in the DA group than in the Non-DA group (P = 0.009). A similar trend was observed for at the last follow-up AT (P = 0.001). However, there was no significant difference in AT loss between the two groups (P = 0.674). The mean postoperative (P < 0.001) and last follow-up (P < 0.001) CSVL–LIV distance were significantly greater in the DA group than in the Non-DA group, showing a similar trend in the AT. There was no statistical differences in the last follow-up SVA, TK, and LL (all P > 0.05) (Table 2).

Table 2.

Comparison of the radiological parameters between two groups

Variables	DA group (n = 20)	Non-DA group (n = 76)	P
Main thoracic Cobb angle
Preoperative (mean ± SD, °)	63.1 ± 10.4	57.1 ± 8.9	0.011
Postoperative (mean ± SD, °)	23.6 ± 8.1	20.8 ± 5.9	0.089
Correction rate (mean ± SD, %)	62.9 ± 10.6	63.2 ± 10.3	0.884
Last follow-up (mean ± SD, °)	24.5 ± 9.3	20.9 ± 5.9	0.038
Correction loss (mean ± SD, °)	0.9 ± 4.0	0.2 ± 3.4	0.388
Flexibility (mean ± SD, %)	25.5 ± 16.6	26.4 ± 13.7	0.814
Proximal thoracic Cobb angle (mean ± SD, °)
Preoperative	23.0 ± 11.8	20.1 ± 9.5	0.223
Postoperative	8.3 ± 6.8	8.1 ± 5.8	0.788
Last follow-up	8.5 ± 5.0	8.2 ± 4.3	0.350
Lumbar Cobb angle (mean ± SD, °)
Preoperative	28.8 ± 8.8	21.4 ± 7.5	< 0.001
Postoperative	6.2 ± 7.6	5.2 ± 3.1	0.870
Last follow-up	11.0 ± 7.7	4.8 ± 3.0	< 0.001
Apical translation (mean ± SD, mm)
Preoperative	58.4 ± 11.1	55.6 ± 10.8	0.307
Postoperative	21.8 ± 9.0	15.3 ± 9.7	0.009
Last follow-up	24.8 ± 8.4	17.5 ± 8.1	0.001
Δ (Last follow-up–Postoperative)	3.0 ± 9.1	2.2 ± 7.9	0.674
Coronal balance (mean ± SD, mm)
Preoperative	-7.7 ± 11.8	-13.0 ± 10.5	0.077
Postoperative	5.8 ± 18.2	1.1 ± 16.2	0.259
Last follow-up	3.8 ± 10.8	6.1 ± 9.4	0.331
LIV tilt angle (mean ± SD, °)
Preoperative	25.9 ± 5.4	19.8 ± 5.2	< 0.001
Postoperative	5.6 ± 3.4	4.5 ± 3.4	0.226
Last follow-up	8.6 ± 4.8	4.5 ± 3.4	< 0.001
Δ (Last follow-up–Postoperative)	3.0 ± 4.0	0.1 ± 3.2	< 0.001
LIV disc angle (mean ± SD, °)
Preoperative	6.1 ± 2.9	7.1 ± 3.7	0.284
Postoperative	3.2 ± 1.9	1.9 ± 1.4	0.001
Last follow-up	5.9 ± 3.7	1.3 ± 1.3	< 0.001
Δ (Last follow-up–Postoperative)	2.7 ± 3.9	-0.6 ± 2.0	< 0.001
CSVL–LIV distance (mean ± SD, mm)
Preoperative	11.1 ± 5.9	10.8 ± 4.4	0.788
Postoperative	9.2 ± 9.0	4.0 ± 3.4	< 0.001
Last follow-up	11.9 ± 7.5	3.9 ± 3.0	< 0.001
Δ (Last follow-up–Postoperative)	2.7 ± 7.3	-0.1 ± 3.3	0.012
Sagittal vertical axis (mean ± SD, mm)
Preoperative	-10.8 ± 32.1	-7.6 ± 22.3	0.607
Postoperative	10.3 ± 30.3	14.1 ± 23.8	0.415
Last follow-up	-5.5 ± 22.5	-4.3 ± 21.6	0.502
Thoracic kyphosis for T5-T12 (mean ± SD, °)
Preoperative	15.3 ± 8.4	17.0 ± 10.7	0.502
Postoperative	18.1 ± 7.9	21.8 ± 6.7	0.068
Last follow-up	19.7 ± 8.1	23.4 ± 6.9	0.070
Lumbar lordosis (mean ± SD, °)
Preoperative	47.8 ± 10.8	46.5 ± 10.1	0.610
Postoperative	39.3 ± 8.7	44.8 ± 10.2	0.022
Last follow-up	51.2 ± 11.1	54.4 ± 9.6	0.249

N, number; DA, distal adding-on; SD, standard deviation; LIV, lowest instrumented vertebra; CSVL, central sacral vertical line.

Relationship between lumbar apex and TV level

There was a statistically significant difference in the TV position between the DA and non-DA groups (P < 0.001), whereas there was no significant difference in the lumbar apex level (P = 0.704). The mean remnant motion segment was significantly greater in the DA group (4.7 levels) than in the Non-DA group (3.8 levels) (P < 0.001). The interval of the vertebral-level was significantly greater in the DA group (2.2 ± 0.6 levels) compared to the Non-DA group (1.3 ± 0.6 levels; P < 0.001). Spearman correlation analysis confirmed that this interval was significantly correlated with both rotation discrepancy (ρ = 0.48, P < 0.001) and remnant motion segments (ρ = 0.65, P < 0.001) (Supplementary Table 1). The vertebral rotation angle at the TV was -6.2° in the DA group and -0.6° in the Non-DA group, showing a statistical significance (P = 0.013). The vertebral rotation angle at the lumbar apex was -4.9° in the DA group and -1.3° in the Non-DA group, showing a statistical significance (P = 0.001). The vertebral rotation discrepancy between TV and lumbar apex was significantly greater in the DA group (6.6°) than in the Non-DA group (1.3°) (P < 0.001) (Table 3). Reliability for vertebral rotation discrepancy was good, with an interobserver ICC of 0.87 (95% CI, 0.78–0.91) and an intraobserver ICC of 0.91 (95% CI, 0.85–0.94).

Table 3.

Relationship between lumbar apex and touched vertebra level

Variables	Cases (n = 96)	DA group (n = 20)	Non-DA group (n = 76)	P
TV level (T12/L1/L2/L3, n)	4/19/41/32	2/9/9/0	2/10/32/32	< 0.001
Lumbar apex level (L2/L3/L4/L5, n)	5/40/46/5	2/7/10/1	3/33/36/4	0.704
Vertebral-level interval between TV and lumbar apex (mean ± SD, n)	1.5 ± 0.7	2.2 ± 0.6	1.3 ± 0.6	< 0.001
Remnant motion segment (mean ± SD, n)	4.0 ± 0.8	4.7 ± 0.7	3.8 ± 0.8	< 0.001
Vertebral rotation angle (mean ± SD, °)
TV	-1.8 ± 6.6	-6.2 ± 7.0	-0.6 ± 6.1	0.013
Lumbar apex	-2.1 ± 5.9	-4.9 ± 6.2	-1.3 ± 5.6	0.001
∣Δ (Lumbar apex–TV)∣	2.4 ± 2.8	6.6 ± 3.5	1.3 ± 1.1	< 0.001

N, number; DA, distal adding-on; SD, standard deviation; TV, touched vertebra.

When comparing the adolescent group (≤18 years, n = 54) and the young adult group (>18 years, n = 42), the preoperative main thoracic Cobb angle was significantly smaller in the young adult group (54.8° ± 8.4°) compared to the adolescent group (61.1° ± 9.4°) (P < 0.001); however, there were no significant differences in flexibility (P = 0.479), remnant motion segments (P = 0.242), and vertebral rotation discrepancy (P = 0.494). We further investigated the association between the vertebral rotation discrepancy and the lumbar modifiers. The mean rotational discrepancy significantly increased as the modifier progressed from A to C: 2.03° ± 2.46° in modifier A, 3.11° ± 2.31° in modifier B, and 3.66° ± 4.58° in modifier C (P = 0.046).

Correlation for the LIV-related parameters and the change of residual AT

The change of residual AT from postoperative to last follow-up was significantly correlated with Δ CSVL–LIV distance (R = 0.30, P = 0.003). Δ CSVL–LIV distance was also significantly correlated with Δ LIV disc angle (R = 0.29, P = 0.004). For the vertebral rotation discrepancy, there was no significant correlation with the change of residual AT (R = 0.03, P = 0.779). However, it showed significant correlation with Δ LIV disc (R = 0.25, P = 0.014) and tilt angles (R = 0.31, P = 0.002) (Table 4).

Table 4.

Correlation analysis for the lowest instrumented vertebra-related parameters and the change of residual AT

Variables	ΔAT	Δ(CSVL–LIV)	ΔLIV disc angle	ΔLIV tilt angle	Vertebral rotation discrepancy (TV and lumbar apex)
ΔAT	1	R = 0.30 P = 0.003	R = 0.06 P = 0.586	R = –0.09 P = 0.393	R = 0.03 P = 0.779
Δ (CSVL–LIV)	-	1	R = 0.29 = 0.004	R = 0.11 P = 0.267	R = 0.11 P = 0.303
Δ LIV disc angle	-	-	1	R = 0.25 P = 0.016	R = 0.25 P = 0.014
Δ LIV tilt angle	-	-	-	1	R = 0.31 P = 0.002
Vertebral rotation difference (TV and lumbar apex)	-	-	-	-	1

Δ indicates last follow-up – postoperative value; AT, apical translation; CSVL, central sacral vertical line; LIV, lowest instrumented vertebra.

Multivariate logistic regression model for distal add-on

The LASSO logistic regression model selected vertebral rotation discrepancy and remnant motion segments for inclusion in the final model (Supplementary Figures 1A and 1B). The Multivariate logistic regression model demonstrated that vertebral rotation discrepancy (OR = 4.09, 95% CI: 1.98–8.43) and remnant motion segments (OR = 3.69, 95% CI: 1.03–13.18) were significantly associated with the occurrence of distal add-on (Table 5, Figures 2 and 3). Based on this model with two significant parameters, predicted probability was estimated using 1/(1+e^-Y), Y = -11.43 + 1.41 x (vertebral rotation discrepancy) + 1.31 x (remnant motion segments). The model demonstrated an apparent AUC of 0.97 and a Brier score of 0.03. Conditional bootstrap internal validation of the final two-predictor model yielded an optimism-corrected AUC of 0.97 (95% CI, 0.92–1.00) and an optimism-corrected Brier score of 0.04 (95% CI, 0.01–0.07). The bootstrap-derived calibration intercept and slope were 0.08 and 0.85, respectively (Supplementary Figure 1C). For the illustrative high-risk probability threshold of 0.70, the model showed a sensitivity of 85.0% (95% CI, 62.1%–96.8%) and a specificity of 98.7% (95% CI, 92.9%–100.0%). When the number of remnant motion segments was fixed, a vertebral rotation discrepancy of 5.05° corresponded to the boundary of this high-risk probability (Figure 4A). When vertebral rotation discrepancy was fixed, preservation of six motion segments (LIV = T12) corresponded to a predicted probability of 0.45 (Figure 4B). In the combined model, the vertebral rotation discrepancy corresponding to a predicted probability of 0.70 was 3.15°, 4.08°, 5.01°, and 5.93° for LIV levels of T12, L1, L2, and L3, respectively (Figure 4C).

Table 5.

Multivariate logistic regression analysis for adding-on phenomenon in our cohort

Variables	B	SE	OR	95% CI	P
Intercept	-11.43	3.32	​	​	0.001
Remnant motion segments	1.31	0.65	3.69	1.03-13.18	0.044
Vertebral rotation difference between TV and lumbar apex	1.41	0.37	4.09	1.98-8.43	< 0.001

After selecting the variables using the Least Absolute Shrinkage and Selection Operator logistic regression model to minimize multicollinearity and overfitting, multivariate logistic regression analysis was performed for the analysis of adding-on phenomenon. A multivariate logistic regression analysis was conducted on the final cohort of 96 patients, which included 20 cases of the distal add-on phenomenon. The final model yielded an Akaike Information Criterion (AIC) of 31.14, indicating an optimal balance between model fit and complexity. B, beta; SE, standard error; OR, odds ratio; CI, confidence interval.

Figure 2.

Illustrative case of no vertebral rotation discrepancy between the touched vertebra (TV) and lumbar apex. (A) A 16-year-old girl with thoracic curves of adolescent idiopathic scoliosis presented with a main Cobb angle of 59.5°, with L2 as the TV and L4 as the lumbar apex. (B) Preoperative CT scan showed a vertebral rotation angle of -5.8° at L2. (C) Preoperative CT scan showed a vertebral rotation angle of -3.9° at L4, indicating a vertebral rotation discrepancy of 1.9°. (D) Deformity correction with posterior spinal fusion was performed between T3 and L2, preserving four motion segments with the LIV set at the TV. (E) At the two-year follow-up, no distal add-on was observed, and the central sacral vertical line–LIV distance was maintained

Figure 3.

Illustrative case of vertebral rotation discrepancy between the touched vertebra (TV) and lumbar apex. (A) A 15-year-old girl with thoracic curves of adolescent idiopathic scoliosis presented with a main Cobb angle of 76.8°, with L2 as the TV and L4 as the lumbar apex. (B) Preoperative CT scan showed a vertebral rotation angle of -3.5° at L2. (C) Preoperative CT scan showed a vertebral rotation angle of -13.7° at L4, indicating a vertebral rotation discrepancy of 10.2°. (D) Deformity correction with posterior spinal fusion was performed between T3 and L2, preserving four motion segments with the LIV set at the TV. (E) At the 2.2-year follow-up, distal add-on was observed, characterized by increased residual apical translation and an increased central sacral vertical line–LIV distance

Figure 4.

Predictive probability of the distal add-on phenomenon. (A) In a model fixing the mean number of remnant motion segments at 3.95, a vertebral rotation difference of 5.0° between the touched vertebra (TV) and lumbar apex marked the boundary for a high predictive probability of distal add-on. (B) In a model fixing the mean vertebral rotation difference between the TV and lumbar apex at 2.41°, 6 levels of remnant motion segments reached a 0.45 predictive probability of distal add-on. (C) A combined model incorporating both vertebral rotation discrepancy and remnant motion segments to estimate the predictive probability of the distal add-on. Delta indicates the vertebral rotation difference between the TV and lumbar apex

Sensitivity analysis for multivariate logistic regression model

Considering this vertebral-level interval, normalized rotation discrepancy, defined as rotation discrepancy divided by the TV–lumbar apex level interval, was used for sensitivity analysis. The normalized discrepancy was significantly greater in the DA group (3.30 ± 1.98°/level) compared to the Non-DA group (1.10 ± 0.91°/level) (P < 0.001). When we replaced the original rotation discrepancy with this normalized variable in our multivariable model, the trend remained highly consistent: Normalized rotation discrepancy: OR 7.55, 95% CI 2.80–20.35, P < 0.001, Remnant motion segments: OR 10.89, 95% CI 2.77–42.84, P < 0.001. These results indicate that rotation discrepancy is not merely explained by the phenomenon of a larger rotatory difference resulting from a longer vertebral-level intervals.

Considering the TV position for the number of remnant motion segments, a sensitivity analysis was performed using a Firth penalized logistic regression incorporating the categorical TV level and the original rotation discrepancy. Vertebral rotation discrepancy remained significantly associated with distal add-on (OR, 3.73; 95% CI, 1.88–7.40; P < 0.001), whereas the overall effect of categorical TV level did not reach conventional statistical significance (global P = 0.080). Thus, although the TV position itself may represent the remnant motion segments, rotation discrepancy was not merely a surrogate for TV position.

We performed an additional sensitivity analysis by incorporating follow-up duration into the multivariable logistic regression model. Follow-up duration was not significantly associated with distal add-on (OR = 0.82 per year, 95% CI 0.39–1.72, P = 0.597). Importantly, the associations of vertebral rotation discrepancy (OR = 4.14, 95% CI 2.00–8.58, P < 0.001) and remnant motion segments (OR = 3.54, 95% CI 1.01–12.39, P = 0.048) remained essentially unchanged after adjustment for follow-up duration. (Supplementary Table 2).

Discussion

In the present study, the distal add-on occurred at a rate of 20.8% when selecting the LIV at the TV, yielding results consistent with previous studies. Although the determination of the LIV is a paramount factor in the risk of distal add-on, the LIV=TV rule itself is not the sole determinant. Notably, we identified that the vertebral rotation discrepancy between the LIV and the lumbar apex, along with the number of remnant motion segments, are substantially associated with the increased risk of distal add-on. Our findings indicate that rotational mismatch at the motion segments is associated with an increased risk of the distal add-on, underscoring the three-dimensional nature of AIS deformation.

The selection of the LIV in thoracic AIS remains a subject of ongoing debate, primarily due to the trade-off between preserving motion segments and minimizing the risk of distal add-on. ¹⁵ Matsumoto et al demonstrated that minimizing residual AT and avoiding LIV selection cephalad to the TV are crucial for reducing the risk of the distal add-on in Lenke 1A curves. ⁸ The TV concept has recently become established as a fundamental principle, grounded in the theory that the LIV should be positioned at least centered over the sacrum.^4,14 While selecting the LIV at the TV represents the most efficient choice from a trade-off perspective, this strategy does not guarantee the prevention of the distal add-on.^11,15 Indeed, our results demonstrated that the number of remnant motion segments was also a significant risk factor for distal add-on even with the LIV=TV selection, indicating that various factors beyond the simple LIV=TV rule contribute to the development of the distal add-on.

Most previous studies have focused on 2D parameters such as sacral slanting and L4 tilt to predict distal add-on.^9,11,12 While 3D classification systems considering vertebral rotation—a core characteristic of AIS—have recently been emphasized, the role of axial rotation in determining the distal fusion level remains underestimated.^1,17 Given these evolving concepts and the existing knowledge gap, our study aimed to deepen the understanding of specific risk factors for distal add-on by incorporating the element of axial vertebral rotation into the analysis alongside conventional risk factors. Interestingly, our results suggest that vertebral rotation discrepancy may provide additional information regarding residual distal add-on risk after LIV = TV selection. Because all patients in the present cohort were treated with LIV = TV, our study cannot determine whether selecting an LIV caudal to the TV would reduce distal add-on. This question requires direct comparison of different LIV-selection strategies in future prospective studies.

As patients reach skeletal maturity, the overall flexibility of the spinal column progressively decreases. ²³ However, our findings indicate that 3D rotational discrepancy and coronal flexibility play distinctly different biomechanical roles. While an age-related loss of flexibility increases the overall rigidity of the spine, the distal add-on phenomenon is associated with the residual torsional stress stemming from the uncorrected 3D rotational mismatch, which inevitably transfers down to the unfused remnant motion segments. In our cohort, because the young adult and adolescent groups demonstrated similar main curve flexibility, these findings suggest that this comparable flexibility, coupled with the uniform application of the LIV = TV rule, explains why no significant difference was observed in the incidence of distal add-on according to age or skeletal maturity.

Beauchamp et al demonstrated that selecting the TV as the LIV provides acceptable positioning in thoracic AIS over the long term, particularly when assessed using the CSVL–LIV distance. ⁴ They further noted that modifier A curves appeared more vulnerable in terms of CSVL–LIV distance when fusion stopped cephalad to the TV, implying a relationship between modifiers and CSVL–LIV distance. ⁴ Our study offers a deeper insight into these findings. First, we observed a significant correlation between changes in CSVL–LIV distance and residual AT.^4,8 Moreover, changes in CSVL–LIV distance were significantly correlated with changes in the LIV disc angle. Second, and notably, vertebral rotation discrepancy contributed significantly to changes in LIV disc and tilt angles, independent of residual AT or CSVL–LIV distance. This indicates that, while the determination of remnant motion segments is influenced by residual AT and CSVL–LIV distance, the rotatory factor serves as a distinct determinant in the development of distal add-on in the remnant curve.

In our multivariate logistic regression model, L4 tilt and sacral slanting showed no statistical significance regarding the development of the distal add-on. Regarding L4 tilt, the specific curve patterns differ distinctively; for instance, a Lenke 1A curve with an L4-Left tilt mimics a Lenke 1B pattern, whereas an L4-Right tilt behaves like a King IV overhanging curve.^9,10 Indeed, the King classification meticulously allocated types (King II-V) within what are now classified as Lenke 1 and 2 types, implying that even if the lumbar curve is non-structural, the variation in rotational patterns makes it a non-negligible factor in STF decision-making. ⁵ Joo et al investigated the impact of sacral slanting and L4 tilt on distal add-on, reporting that these factors influenced surgeons’ LIV selection. ¹² Consequently, because L4 tilt and sacral slanting typically drive the LIV decisions in clinical practice, they did not appear as significant influencing factors in our study where the LIV was consistently selected as the TV. ¹² Meanwhile, their study reported that the lumbar curve itself plays a significant role in the development of distal add-on, sharing conceptual focus on rotational discrepancy. Consistent with this selection effect, the LIV = TV group showed greater main-curve flexibility than the LIV ≠ TV group in our cohort despite otherwise comparable baseline characteristics (Supplementary Table 3). Because LIV selection was influenced by preoperative characteristics and surgeon judgment, restricting the analysis to the LIV = TV subgroup may have introduced confounding by indication and potential collider bias; therefore, the lack of significant associations for L4 tilt or sacral slanting should not be interpreted as evidence of clinical irrelevance.

Similarly, previous literature emphasizes that caution is required when performing selective thoracic fusion in patients with modifier C curves because of a potential risk of coronal decompensation.^24,25 Although the modifier type itself did not demonstrate a statistically significant impact on the occurrence of the distal add-on in our cohort, the significantly increasing trend in rotational discrepancy from modifier A to C was observed. This implies that the underlying 3D rotational mismatch could be a relevant factor to consider alongside 2D lumbar modifiers for reducing the risk of mechanical complications following deformity correction in idiopathic scoliosis. ²⁶

Our study has several strengths in the issue of the LIV in the thoracic IS. First, by restricting our cohort to patients with LIV = TV, we aimed to address a specific clinical question within a uniform LIV-selection strategy. This controlled approach allowed for a more expanded interpretation of the distal add-on as a multifactorial issue. ¹ Furthermore, our findings address a novel question: Even when a construct consistently terminates at the TV, does the risk of distal add-on still vary according to how many mobile segments remain below it? Our findings suggest that the number of remnant motion segments may reflect a positional component of distal add-on risk even when LIV = TV. Furthermore, vertebral rotation discrepancy remained consistently associated with distal add-on across the sensitivity analyses, suggesting that rotational mismatch may provide additional information beyond TV position alone. Second, by incorporating the three-dimensional nature of the deformity, we observed that the discrepancy in axial vertebral rotation could be a relevant factor in preoperative planning. ²⁷ The TV rule is foundational, but because not all TVs share the same anatomical and rotational characteristics relative to the unfused curve, the rotational factor must be considered. Our model also illustrates how the estimated risk varies jointly with these two parameters: the rotational discrepancy corresponding to a given predicted probability becomes progressively smaller as the number of remnant motion segments increases. These probability estimates are presented solely as exploratory, cohort-specific illustrations of the relationship between rotation discrepancy and distal add-on and should not be interpreted as clinical thresholds or operative guidance. Future prospective studies should determine whether incorporating rotational discrepancy into LIV selection can provide clinically meaningful guidance for identifying the optimal LIV level. Finally, this study highlights the heterogeneity of non-structural curves in STF candidates, emphasizing the need for meticulous, individualized consideration of curve patterns. ²⁰

This study has several limitations. First, as a single-center retrospective study, it is subject to inherent selection biases. Since the LIV = TV group had greater main-curve flexibility than the LIV ≠ TV group, and comparability across the other measured baseline variables does not remove this selection bias; restriction to the LIV = TV subgroup, which reflects the surgeon’s treatment decision, may therefore introduce residual confounding by indication. Second, we focused on radiographic outcomes and did not include patient-reported outcome measures.^2,28 Third, our study did not include the substantially TV, which remained a subject of debate in the literature. ¹⁴ Fourth, the assessment of 3D vertebral rotation relied on CT imaging, raising valid concerns regarding radiation exposure in a young population. Recently, low-dose CT has been widely utilized to mitigate this concern.^29,30 When CT imaging is indicated, we use a dedicated low-dose supine acquisition at 80 kVp and 50 mAs to minimize this risk, and assess the vertebral rotation using the Aaro-Dahlborn method. ³¹ Magnetic resonance imaging also represents a highly reliable, radiation-free alternative for this measurement. ³² Conversely, while Nash-Moe grading on standing radiographs is convenient, its broad ordinal scale lacks the resolution necessary to detect the subtle rotational discrepancies identified as critical in our analysis. Nevertheless, the development of optimal imaging approaches that guarantee precise rotational measurement while prioritizing radiation safety remains a critical area for future investigation. Additionally, as a study based on static radiographs and CT scans, we could not directly measure dynamic rotational instability or biomechanical torsional stress. Last, although the illustrative high-risk probability threshold of 0.70 demonstrated favorable sensitivity and specificity, these estimates were derived and evaluated within the same relatively small cohort used for model development, and the confidence interval for sensitivity was relatively wide. In addition, bootstrap internal validation was conditional on the final two-predictor model and did not repeat the complete LASSO variable-selection procedure within each resample. Therefore, the reported performance estimates do not incorporate uncertainty arising from model selection and should be interpreted cautiously until externally validated. Therefore, the predictive probability values and proposed threshold remain exploratory and cohort-specific and should not be considered definitive clinical cutoffs without external validation. These factors should be addressed in future research, and a multicenter prospective study is warranted to validate our findings.

Conclusion

Greater vertebral rotation discrepancy between the LIV and lumbar apex was consistently associated with distal add-on when the TV was selected as the LIV, suggesting that three-dimensional rotational mismatch may contribute to residual distal add-on risk.

ORCID iDs

Hong Jin Kim https://orcid.org/0000-0002-8162-9585

Subin Lim https://orcid.org/0000-0002-8720-5484

Hyung Rae Lee https://orcid.org/0000-0003-3125-1470

Jae Hyuk Yang https://orcid.org/0000-0001-6949-6954

Seung Woo Suh https://orcid.org/0000-0002-1536-4611

Ethical Considerations

This retrospective study was approved by the Institutional Review Board of Korea University Guro Hospital (IRB number: 2026GR0035).

Consent to Participate

Informed consent for retrospective analysis of de-identified data was waived.

Data Availability Statement

The datasets generated during and/or analysed during the current study are not publicly available due to ethical issues but are available from the corresponding author on reasonable request.*