Limited Predictive Performance of the Global Alignment and Proportion (GAP) Score for Mechanical Complications in Older Asian Patients With Adult Spinal Deformity: A Multicenter Validation Study

Abstract

Study Design

Retrospective cohort study.

Objectives

Surgery for adult spinal deformity (ASD) is associated with postoperative mechanical complications (MCs), such as proximal junctional kyphosis (PJK). The applicability of the Global Alignment and Proportion (GAP) score for predicting MCs in older Asians remains unclear. This study aimed to evaluate the validity of the GAP score in Asians patients by stratifying them into younger and older groups.

Methods

We studied 274 patients who had undergone multi-level spinal fusion and were followed up for at least 2 years. Patients were divided into a younger group (Under-69, n = 122) and an older group (Over-70, n = 152). MCs were assessed at 2 years postoperatively. Propensity score matching was performed to compare the incidence of MCs across groups. Logistic regression models were used to examine the interaction between GAP score and age.

Results

MC occurred in 93 patients (34%), with no significant differences between groups. In the Under-69 group, GAP scores did not correlate with MC occurrence. In contrast, the in the Over-70 group patients with MCs had significantly lower GAP scores (6.6 ± 3.4 vs. 8.0 ± 3.2, P = 0.022). The interaction between GAP score and age was not significant in the logistic regression models.

Conclusions

The GAP score demonstrated limited predictive ability for MCs in this Asian ASD cohort, particularly in older patients. These findings suggest that alignment-based prediction models alone may be insufficient for MC risk stratification in aging populations.

Introduction

The recent global increase in the older population has resulted in a rising number of patients with adult spinal deformity (ASD). ¹ ASD causes low back pain and neurological symptoms due to malalignment, which significantly impairs patients’ quality of life (QOL).^2,3 Treatment strategies for ASD involve either surgical or conservative therapy, depending on the severity of the condition and the associated risks for the patient. ⁴ Surgery for ASD is associated with a high incidence of postoperative complications, particularly mechanical complications (MCs) which increase the likelihood of reoperation^5-7 In surgical planning for ASD, global alignment parameters are evaluated to achieve target alignment and enhance postoperative QOL and to prevent MCs such as proximal junctional kyphosis (PJK), proximal junctional failure (PJF), and implant fracture. ⁸

Although lumbar lordosis (LL) that harmonizes with the patient’s pelvic incidence (PI) is considered a fundamental indicator of global alignment, relying solely on this measure is insufficient to minimize postoperative MCs. ⁵ To develop a more comprehensive PI-based alignment indicator, Yilgor et al ⁹ proposed the Global Alignment and Proportion (GAP) score system, which serves as a predictive tool for postoperative MCs following ASD surgery. The GAP score assesses the risk of MCs based on patient age (a binary classification of 60 years or older) and four subscales derived from sagittal plane alignment analysis, thereby indicating an optimal target global alignment index for each patient. Despite its rationale, additional factors such as age, ethnicity, and the presence of osteoporosis may influence the predictive performance of the GAP score. ¹⁰ Thus, previous studies have indicated that the original GAP score’s utility as a tool for predicting MCs is limited. ¹¹

Compared with European populations, where the GAP score was developed, Asian populations with ASD tend to be older and demonstrate a higher prevalence of osteoporosis. ¹² Therefore, the applicability of the original GAP score to Asian patients with ASD has been questioned.^13-15 However, this issue remains inadequately addressed. Older patients with ASD frequently present with osteoporosis and sarcopenia, both of which are established risk factors for postoperative MCs, including PJK.^16-18 Critically, the GAP score is constructed exclusively from radiographic alignment parameters and does not incorporate bone mineral density or muscle quality. This structural limitation suggests that the GAP score’s predictive ability may be particularly constrained in aging Asian populations, where the prevalence of these conditions is high. We therefore hypothesized that older age may alter the association between the GAP score and postoperative MCs.

The present study aimed to externally validate the GAP score within a multicenter cohort of Asian patients with ASD and to determine whether age modifies its predictive ability for postoperative MCs. We predefined the threshold of 70 years based on the median age of our cohort and a multicenter propensity score-matched study of Asian ASD patients. ¹⁹ That study demonstrated that surgery in patients aged 70 years or older was associated with a significantly higher risk of postoperative and implant-related complications (odds ratios of 4.0 and 4.9, respectively) compared to patients aged 50–65 years, as well as inferior quality of life and cost-effectiveness; this suggests that this age is not an arbitrary statistical cutoff but a clinically meaningful threshold indicating increased surgical risk. First, first evaluated the GAP score and the incidence of MCs among Asian patients to assess its applicability. Second, patients were classified into relatively younger and older age groups, allowing for comparisons regarding the GAP score, individual alignment parameters, and MC incidence rates.

This study identified age groups for which the optimal values of the original GAP score can be applied to predict MC occurrence during corrective surgery for Asian patients with ASD, thereby facilitating more precise clinical applications of the GAP score. Although several external validation studies of the GAP score have been published, most were performed in Western cohorts, did not formally test age-by-GAP interactions, and did not examine whether the directional relationship between alignment and MCs differs across age strata. Validation in Asian populations — which differ in body habitus, bone quality, and surgical practice — remains limited. To our knowledge, the present study is the first multicenter external validation of the GAP score in an Asian ASD cohort that explicitly stratifies by age, formally tests the age × GAP interaction, and additionally evaluates robustness across multiple age cutoffs and outcome definitions.

Methods

Patient Population and Grouping

This retrospective multicenter study included patients who had undergone multi-level (≥4 vertebrae, including ≥ 1 lumbar vertebra) posterior spinal fusion between 2017 and 2022. Patients were selected from a database established by five affiliated institutions. The inclusion criteria were as follows: 1) age ≥ 30 years; 2) fusion involving four or more vertebrae, including at least one lumbar vertebra; and 3) a minimum follow-up period of 2 years. Exclusion criteria were as follows: 1) fresh vertebral fracture, 2) infection, and 3) tumor. A total of 368 patients were identified in the dataset. After excluding 56 cases with incomplete background data and 38 cases lacking postoperative GAP scores, 274 cases were ultimately included in the analysis.

The cohort was divided into a younger group aged ≤ 69 years (Under-69 group) and an older group aged ≥ 70 years (Over-70 group) based on the median age. Each group was compared based on the presence or absence of MCs (Figure 1). The age threshold of 70 years was selected due to its alignment with the median age of the cohort and its common clinical use to distinguish older patients with ASD who may exhibit increased frailty and osteoporosis risk. Furthermore, sensitivity analyses were conducted by setting the threshold at 60 years (the same as the age score threshold in the GAP score) and 75 years (since a threshold of 80 years would result in too few participants in the older group to allow for meaningful analysis). In addition, an analysis of the older group was performed when the outcome was limited to PJK. The selection of the 70-year threshold is also supported by a previous report in an Asian ASD population, which demonstrated a marked increase in surgical complications above 70 years of age. ¹⁹

Figure 1.

Patient selection flowchart. Flowchart illustrating patient inclusion and exclusion criteria and final cohort allocation. GAP, Global Alignment and Proportion; MC, mechanical complication

This study adhered to the principles outlined in the Declaration of Helsinki and received approval from the Institutional Review Board of the authors’ research institution (No. 2016-0177). As all interventions (including surgical procedures and preoperative/postoperative examinations) were part of routine care, the requirement for written informed consent was waived.

Baseline Demographics and Clinical Outcomes

Based on patient demographics and clinical outcomes, the following factors were compared between the two groups: age, sex, body mass index, number of fused vertebrae, presence or absence of pelvic fixation, and the Japanese Orthopaedic Association (JOA) score at 2 years postoperatively.

Radiographic Measurements

The GAP score was assessed using postoperative whole-spine radiographs. The subdomains included relative pelvic version (RPV), relative LL (RLL), lordosis distribution index (LDI), relative spinopelvic alignment (RSA), and age score. Each subdomain measured the difference between ideal and observed values and was scored from 0 to 3 points. Ideal values were determined using specified coefficients. Severity was classified into three levels: ≤ 2 points (proportioned), ≥ 7 points (severely disproportioned), and moderately disproportioned in the range between.

Vertebral Hounsfield unit (HU) values obtained from computed tomography (CT) horizontal slices were employed as indicators of bone strength. The vertebra selected for measuring vertebral CT values was typically one of the least degenerated vertebrae from T10 to L4. ²⁰

MCs

MCs, including PJK (regardless of fracture presence), distal junctional kyphosis/failure (DJK), rod fracture, and pedicle screw loosening, were assessed using radiographs obtained at 2 years postoperatively. PJK was defined as a progression of kyphosis of ≥ 10°, measured by the sagittal Cobb angle from the upper instrumented vertebra (UIV) to the two levels cephalad to the UIV, ²¹ by comparing the preoperative and 2-year postoperative periods. DJK was defined as implant failure, fracture, or reoperation for these reasons at the lowest instrumented vertebra or the adjacent caudal vertebra, based on comparisons between the immediate postoperative and 2-year postoperative periods.

To address the heterogeneity of MCs, we performed an additional sensitivity analysis in which the outcome was restricted to PJK. Because the main analysis showed no association between the GAP score and MCs in the Under-69 group, and because PJK is the most clinically relevant MC in older patients with ASD, this PJK-focused sensitivity analysis was performed in the matched Over-70 cohort.

Statistical Analysis

Data were presented as the mean ± standard deviation for continuous variables and as the number and percentages for categorical variables. To facilitate subgroup comparisons based on the presence or absence of MC, 1:1 propensity score matching was conducted utilizing variables with P-values < 0.1, as determined by univariate analysis results. This matching employed a nearest-neighbor matching algorithm without replacement. The variables identified as significant for subgroup comparison differed between the Under-69 and Over-70 groups. In the Under-69 group, the covariates included age, the presence of pelvic fixation, and the preoperative sagittal vertical axis (SVA); in the Over-70 group, the covariates were HU values, the presence of pelvic fixation, and the preoperative SVA (Supplemental Tables 1 and 2). To examine the interaction between the GAP score and the patient’s age, we developed logistic regression models for MC occurrence: a main effects model, which assumes no interaction, and a linear interaction model. Furthermore, receiver operating characteristics (ROC) analysis was conducted for MC occurrence using the GAP score alone and in conjunction with the linear interaction models that included both age and the GAP score. Statistical analyses were performed using EZR version 1.68 (Saitama Medical Center, Jichi Medical University, Japan). ²² Comparisons between the two groups were conducted using Welch’s t-test and Fisher’s exact test. Statistical significance was set at P < 0.05. Standardized mean differences (SMDs) before and after matching are reported for all covariates in Supplemental Tables 1 and 2; all post-matching SMDs were below 0.1, indicating adequate balance.

Results

The median age of the total cohort of 274 patients was 70 years (Figure 2). For the purpose of age-based analysis, patients were divided into two groups: the Under-69 group and the Over-70 group. The Under-69 group comprised 122 patients (82 [67%] women, mean age 59.5 ± 9.0 years), while the Over-70 group comprised 152 patients (97 [64%] women, mean age 75.5 ± 3.8 years). The Over-70 group exhibited significantly lower vertebral HU values (148.5 ± 65.3 vs. 119.4 ± 51.8, P < 0.001), significantly higher postoperative GAP scores (6.1 ± 3.2 vs. 7.1 ± 3.2, P = 0.018), and significantly lower 2-year postoperative JOA scores (23.3 ± 4.0 vs. 21.7 ± 4.6, P = 0.006).

Figure 2.

Distribution of patient age. Histogram showing the age distribution of the study cohort. Vertical lines indicate the cutoff used to define the age-based grouping

MCs occurred in 93 cases: 41 cases of PJK, 4 cases of DJK, and 12 cases of rod fracture. PJK occurred in 13 cases (10.7%) in the Under-69 group and in 28 cases (18.4%) in the Over-70 group (P = 0.030). No significant difference was observed in the overall incidence of MCs between the groups (32% vs. 36%, P = 0.54) (Table 1). When comparing the analyzed cohort with the excluded cohort, the analyzed cohort was approximately 2 years younger, exhibited higher vertebral HU values, and included a greater proportion of cases with pelvic fixation (Supplemental Table 3).

Table 1.

Patients’ Demographics

​	Overall, N = 274	Under-69, n = 122	Over-70, n = 152	P value
Age, years	68.4 ± 10.4	59.5 ± 9.0	75.5 ± 3.8	<0.001
Sex, female	179 (65%)	82 (67%)	97 (64%)	0.56
BMI, kg/m2	24.3 ± 4.0	22.9 ± 4.8	24.0 ± 3.5	0.41
Vertebral HU value	132.4 ± 59.8	148.5 ± 65.3	119.4 ± 51.8	< 0.001
Number of fused levels	6.1 ± 3.2	6.4 ± 3.6	5.8 ± 2.9	0.32
Pelvic fixation	121 (45%)	54 (45%)	67 (44%)	0.92
Preop alignment
TK	26.3 ± 15.2	23.6 ± 14.5	28.3 ± 15.4	0.042
LL	26.0 ± 24.6	26.1 ± 20.4	26.0 ± 27.6	0.33
L4–S LL	21.3 ± 12.2	22.2 ± 13.3	20.5 ± 11.3	0.18
PI	51.4 ± 11.5	51.9 ± 11.8	51.0 ± 11.3	0.49
PT	27.4 ± 11.0	27.4 ± 11.4	27.4 ± 10.7	0.87
SS	24.0 ± 10.9	24.6 ± 11.1	23.5 ± 10.8	0.32
PI − LL	25.4 ± 25.1	25.8 ± 20.3	25.0 ± 28.5	0.54
GT	31.5 ± 18.7	30.1 ± 20.1	32.7 ± 17.3	0.15
SVA	77.5 ± 56.8	66.9 ± 56.7	85.8 ± 55.7	0.004
ΔSVA	−35.5 ± 53.1	−31.8 ± 51.6	−38.4 ± 54.3	0.23
GAP score, postop	6.7 ± 3.2	6.1 ± 3.2	7.1 ± 3.2	0.018
RPV	−10.9 ± 8.7	−10.2 ± 8.3	−11.6 ± 9.0	0.39
RLL	−20.6 ± 10.4	−19.4 ± 10.2	−21.6 ± 10.4	0.059
LDI	61.0 ± 18.4	61.4 ± 17.6	60.7 ± 19.1	0.99
RSA	13.0 ± 14.8	11.6 ± 16.4	14.2 ± 13.3	0.029
Age score, 1 pt	227 (83%)	75 (61%)	152 (100%)	< 0.001
JOA score, 2-years postop	22.3 ± 4.4	23.3 ± 4.0	21.7 ± 4.6	0.006
MC	93 (34%)	39 (32%)	54 (36%)	0.54
PJK	41 (15%)	13 (11%)	28 (18%)	0.073
DJK	4 (2%)	2 (2%)	2 (1%)	> 0.99
Rod breakage	12 (4%)	8 (7%)	4 (3%)	0.11
PS loosening	56 (20%)	22 (18%)	34 (22%)	0.38

BMI: body mass index; HU: Hounsfield unit; TK: thoracic kyphosis; LL: lumbar lordosis; PI: pelvic incidence; PT: pelvic tilt; SS: sacral slope; GT: global tilt; SVA: sagittal vertical axis; ΔSVA: postoperative SVA minus preoperative SVA; GAP: global alignment and proportion; RPV: relative pelvic version; RLL: relative lumbar lordosis; LDI: lordosis distribution index: RSA: relative spinopelvic alignment; JOA: Japanese Orhopaedic Association; MC: mechanical complication; PJK: proximal junctional kyphosis; DJK: distal junctional kyphosis; PS: pedicle screw.

Under-69 Group

The MC subgroup comprised 39 patients, whereas the non-MC subgroup comprised 83 patients. Compared to the non-MC subgroup, the MC subgroup exhibited a significantly higher incidence of pelvic fixation cases (62% vs. 37%, P = 0.012) and a significantly lower 2-year postoperative JOA score (22.1 ± 4.0 vs. 23.9 ± 3.8, P = 0.031). No significant difference in GAP scores was observed between the two subgroups (6.5 ± 3.2 vs. 5.9 ± 3.2, P = 0.45). The proportion of patients scoring 1 point for age (≥60 years) was significantly higher in the MC subgroup than in the non-MC subgroup (74% vs. 55%, P = 0.045) (Supplemental Table 4). After matching, no significant differences were observed in GAP scores or subdomains between the subgroups (total score, 6.4 ± 3.2 vs. 6.6 ± 3.2, P = 0.65). Additionally, the 2-year postoperative JOA score was significantly lower in the MC subgroup (22.0 ± 4.0 vs. 24.0 ± 3.5, P = 0.023) (Table 2).

Table 2.

Comparison in Under-69 Group by Occurrence of Mechanical Complications After Matching

​	Under-69, MC	Under-69, non-MC	P value
	n = 36	n = 36
Age, years	61.3 ± 8.8	60.2 ± 7.5	0.24
Sex, female	26 (72%)	26 (72%)	> 0.99
BMI, kg/m2	25.3 ± 5.1	25.2 ± 4.4	0.85
Vertebral HU value	133.1 ± 51.0	143.9 ± 50.2	0.45
Number of fused levels	6.1 ± 3.5	7.5 ± 3.9	0.10
Pelvic fixation	23 (64%)	23 (64%)	> 0.99
Preop alignment
TK	20.0 ± 13.7	25.6 ± 15.1	0.14
LL	24.6 ± 21.0	23.5 ± 19.9	0.86
L4–SLL	20.9 ± 12.7	23.9 ± 12.7	0.51
PI	51.9 ± 11.3	49.9 ± 12.3	0.58
PT	28.6 ± 10.6	26.7 ± 11.3	0.65
SS	23.3 ± 13.6	23.2 ± 10.1	0.86
PI − LL	27.3 ± 17.9	26.4 ± 20.5	0.91
GT	36.2 ± 24.5	30.4 ± 18.6	0.54
SVA	80.2 ± 56.7	70.5 ± 53.9	0.48
ΔSVA	−40.6 ± 53.3	−28.4 ± 48.1	0.32
GAP score, postop	6.4 ± 3.2	6.6 ± 3.2	0.65
RPV	−11.4 ± 8.6	−10.1 ± 6.4	0.37
RLL	−20.5 ± 10.6	−19.8 ± 8.9	0.94
LDI	61.4 ± 15.3	66.9 ± 19.2	0.084
RSA	14.0 ± 21.0	12.3 ± 15.1	0.97
Age score, 1 pt	27 (75%)	22 (61%)	0.21
JOA score, 2-years postop	22.0 ± 4.0	24.0 ± 3.5	0.023
MC	36 (100%)	0 (0%)	<0.001
PJK	12 (33%)	0 (0%)	<0.001
DJK	2 (5.6%)	0 (0%)	0.49
Rod breakage	8 (22%)	0 (0%)	0.005
PS loosening	19 (53%)	0 (0%)	<0.001

BMI: body mass index; HU: Hounsfield unit; TK: thoracic kyphosis; LL: lumbar lordosis; PI: pelvic incidence; PT: pelvic tilt; SS: sacral slope; GT: global tilt; SVA: sagittal vertical axis; ΔSVA: postoperative SVA minus preoperative SVA; GAP: global alignment and proportion; RPV: relative pelvic version; RLL: relative lumbar lordosis; LDI: lordosis distribution index: RSA: relative spinopelvic alignment; JOA: Japanese Orhopaedic Association; MC: mechanical complication; PJK: proximal junctional kyphosis; DJK: distal junctional kyphosis; PS: pedicle screw.

Over-70 Group

The MC subgroup comprised 54 patients, while the non-MC subgroup comprised 98 patients. The postoperative SVA was significantly higher in the MC subgroup than in the non-MC subgroup (96.9 ± 46.7 mm vs. 79.7 ± 59.5 mm, P = 0.012), while the 2-year postoperative JOA score was significantly lower (21.0 ± 4.2 vs. 22.1 ± 4.8, P = 0.044). No significant difference in GAP scores was observed between the two subgroups (6.6 ± 3.4 vs. 7.4 ± 3.1, P = 0.091) (Supplemental Table 5). After matching, the GAP score in the MC subgroup was significantly lower than in the non-MC subgroup, indicating a more proportionate GAP score in the MC subgroup (6.6 ± 3.4 vs. 7.9 ± 3.3, P = 0.039) (Table 3).

Table 3.

Comparison in Over-70 Group by Occurrence of Mechanical Complications After Matching

​	Over-70, MC	Over-70, non-MC	P value
	n = 54	n = 54
Age, years	75.2 ± 3.5	75.8 ± 4.1	0.61
Sex, female	38 (70%)	36 (67%)	0.68
BMI, kg/m2	24.0 ± 3.5	24.1 ± 3.9	0.91
Vertebral HU value	111.4 ± 54.7	116.1 ± 48.7	0.38
Number of fused levels	6.3 ± 3.3	5.8 ± 2.8	0.42
Pelvic fixation	29 (54%)	26 (48%)	0.56
Preop alignment
TK	28.3 ± 18.1	27.0 ± 14.0	0.90
LL	21.8 ± 18.0	22.3 ± 17.1	0.99
L4–SLL	19.0 ± 13.2	20.0 ± 10.7	0.83
PI	51.2 ± 10.9	50.9 ± 12.6	0.75
PT	28.5 ± 12.3	26.8 ± 10.9	0.45
SS	22.7 ± 11.8	24.1 ± 10.6	0.75
PI − LL	29.4 ± 17.7	28.6 ± 18.2	0.89
GT	34.5 ± 18.2	35.1 ± 17.2	0.95
SVA	96.9 ± 46.7	97.2 ± 59.9	0.64
ΔSVA	−52.7 ± 48.9	−44.6 ± 60.3	0.45
GAP score, postop	6.6 ± 3.4	7.9 ± 3.3	0.039
RPV	−10.5 ± 9.7	−12.5 ± 9.1	0.31
RLL	−20.1 ± 8.9	−23.6 ± 11.7	0.067
LDI	59.1 ± 20.4	63.4 ± 21.0	0.33
RSA	11.1 ± 11.5	18.2 ± 13.5	0.007
Age score, 1 pt	54 (100%)	54 (100%)	NA
JOA score, 2-years postop	21.0 ± 4.2	21.4 ± 4.9	0.22
MC	54 (100%)	0 (0%)	<0.001
PJK	28 (52%)	0 (0%)	<0.001
DJK	2 (3.7%)	0 (0%)	0.50
Rod breakage	4 (7.4%)	0 (0%)	0.12
PS loosening	34 (63%)	0 (0%)	<0.001

BMI: body mass index; HU: Hounsfield unit; TK: thoracic kyphosis; LL: lumbar lordosis; PI: pelvic incidence; PT: pelvic tilt; SS: sacral slope; GT: global tilt; SVA: sagittal vertical axis; ΔSVA: postoperative SVA minus preoperative SVA; GAP: global alignment and proportion; RPV: relative pelvic version; RLL: relative lumbar lordosis; LDI: lordosis distribution index: RSA: relative spinopelvic alignment; JOA: Japanese Orhopaedic Association; MC: mechanical complication; PJK: proximal junctional kyphosis; DJK: distal junctional kyphosis; PS: pedicle screw.

Sensitivity Analysis

At a grouping threshold of 60 years, the GAP score of the MC subgroup in the older group after matching was relatively lower than that in the non-MC subgroup (6.6 ± 3.4 vs. 7.2 ± 3.4, P = 0.21) (Supplemental Table 6). When the cutoff was set at 75 years, the GAP score of the MC subgroup in the older group was also relatively lower than that in the non-MC subgroup (7.1 ± 3.2 vs. 7.8 ± 2.9, P = 0.34) (Supplemental Table 7). In an analysis where the MC outcome was limited to PJK, the GAP score of the MC subgroup after matching was relatively lower in the older group than that in the non-MC subgroup (6.3 ± 3.1 vs. 7.6 ± 3.8, P = 0.24) (Supplemental Table 8).

Logistic Regression Models and ROC Analysis

In the main effects model without interaction, as well as in the linear interaction model of GAP score and age, neither the GAP score nor the interaction between GAP score and age was statistically significant (Table 4 and Table 5). Furthermore, the likelihood ratio test for these two models did not indicate significance for the interaction model (P = 0.29). ROC analysis showed that the area under the curve (AUC) for predicting MC based solely on the GAP score was less than 0.5 (AUC, 0.466; 95% confidence interval [CI], 0.392–0.540). However, when accounting for an interaction between the GAP score and age, the AUC improved to 0.570 (95% CI, 0.497–0.643) (Figure 3). Similar results were obtained when the MC outcome was restricted to PJK alone (AUC, 0.465 [95% CI, 0.373–0.556] for GAP score only; AUC, 0.628 [95% CI, 0.540–0.717] for Age + GAP score; Supplemental Figure 1).

Table 4.

Main Effects Model (No Interactions)

Characteristic	OR	95% CI	P value
Age	1.01	−0.02, 0.04	0.6
GAP score	0.96	−0.13, 0.05	0.4
Sex	1.16	−0.44, 0.75	0.6
Vertebral HU value	1.0	−0.01, 0.00	0.054
Pelvic fixation	2.01	0.12, 1.3	0.018
PI − LL	0.99	−0.03, 0.01	0.3
SVA	1.01	0.00, 0.01	0.041

CI: confidence interval; OR: odds ratio; GAP: Global Alignment and Proportion; HU: Hounsfield units; PI: pelvic incidence; LL: lumbar lordosis; SVA: sagittal vertical axis.

Table 5.

Linear Interaction Model Between Age and GAP Score

Characteristic	OR	95% CI	P value
Age	1.04	−0.02, 0.10	0.3
GAP score	0.97	−0.12, 0.06	0.5
Sex	1.17	−0.43, 0.75	0.6
Vertebral HU value	1.0	−0.01, 0.00	0.063
Pelvic fixation	1.91	0.06, 1.2	0.030
PI − LL	0.99	−0.03, 0.01	0.2
SVA	1.01	0.00, 0.01	0.038
Age * GAP score	1.00	−0.01, 0.00	0.3

CI: confidence interval; OR: odds ratio; GAP: Global Alignment and Proportion; HU: Hounsfield units; PI: pelvic incidence; LL: lumbar lordosis; SVA: sagittal vertical axis.

Figure 3.

ROC curves for predicting MCs. ROC curves demonstrating the predictive performance of the GAP score alone and in combination with age for MCs. GAP, Global Alignment and Proportion; ROC, receiver operating characteristic; MC, mechanical complication; AUC, area under the curve

Discussion

In this study, a stratified analysis was conducted to examine the association between GAP scores and the incidence rates of postoperative MC in patients aged ≥ 70 years and older compared to younger patients. No significant difference in MC incidence rates was observed between the Under-69 and Over-70 groups. After adjusting for background factors, no significant difference in GAP scores was detected between the MC and non-MC subgroups in the Under-69 group. Notably, in the Over-70 group, the MC subgroup demonstrated lower GAP scores than the non-MC subgroup, suggesting that patients with more proportioned alignment, as indicated by the GAP score, still experienced MCs. This finding may reflect age-related factors that exert a more dominant influence than alignment parameters in older patients, as discussed below. ¹⁹ In the present cohort, the GAP score was not significantly associated with the occurrence of MCs, indicating its limited predictive performance in this population.

The diminished predictive performance of the GAP score in the Over-70 group can be attributed to three age-specific mechanisms not captured by alignment parameters: 1) osteoporosis, which reduces the bone’s capacity to anchor instrumentation and is an established risk factor for PJK^16,17; 2) sarcopenia, which compromises paraspinal muscle support of the corrected spine and likewise increases PJK risk^18,23; and 3) frailty, which limits soft-tissue compensation for residual malalignment. Together, these factors may render even well-proportioned sagittal alignment mechanically insufficient in older patients. In such patients, even a well-proportioned alignment may impose strain that the deteriorated bone–muscle envelope cannot absorb, resulting in MCs such as PJK or rod fracture. Conversely, patients with residual malalignment may, in some cases, have undergone less aggressive correction or have benefited from compensatory soft-tissue adaptation, reducing mechanical demand on the construct. None of these mechanisms are captured by the GAP score, which is constructed exclusively from radiographic alignment parameters. We emphasize that this hypothesis is not formally supported by the present statistical analyses and warrants prospective evaluation incorporating bone- and muscle-related parameters.

While the MC subgroup showed lower GAP scores than the non-MC subgroup in the Over-70 group, it is important to note that concluding this indicates an inverted relationship—where lower GAP scores actually predict higher MC risk in older patients—would be premature. This is because the differences in GAP scores were minimal, and both subgroups still exhibited moderate alignment imbalances across subscale scores (Figure 4). Moreover, neither the main effects model without interaction nor the linear interaction model between GAP scores and age showed statistically significant results for either GAP scores or the interaction between GAP scores and age. Consequently, it remains unclear whether these results in the Over-70 group represent an inverted relationship or simply indicate that GAP scores are insufficient for assessing MC risk in that population.

Figure 4.

Comparison of GAP subdomain scores between the Under-69 and Over-70 groups. GAP, Global Alignment and Proportion; RPV, relative pelvic version; RLL, relative lumbar lordosis; LDI, lordosis distribution index; RSA, relative spinopelvic alignment

Currently, the SRS-Schwab classification, which emphasizes the relationship between PI and LL, is widely adopted as an indicator of global sagittal alignment in ASD. It is used to understand the pathophysiology of the condition and to enhance postoperative QOL. However, the Schwab classification, which serves as an alignment metric with postoperative QOL as an outcome, is insufficient to effectively prevent MCs, one of the primary postoperative complications. ⁵ In recent years, alternative indicators have been proposed to overcome this limitation. The GAP score is a scoring system developed to prevent postoperative MCs by comprehensively assessing optimal global alignment and lumbar segmental alignment across four subscales while accounting for PI. In the present study, no significant difference in GAP scores was observed between patients with and without MCs in the Under-69 group. Furthermore, in the Over-70 group, no significant correlation was identified between the incidence of MCs and GAP scores.

This raises two important issues. First, the validity of the GAP score may vary significantly based on the characteristics of the population, including ethnicity and age. Given that sagittal spinal alignment varies among different ethnic groups, it is crucial to consider these differences when assessing an individual’s alignment. ²⁴ For instance, the Roussouly classification indicates that Caucasians predominantly exhibit Type 3 alignment, whereas Asians predominantly exhibit Type 2. Similarly, the correlations between PI, pelvic tilt, and LL differ between US Caucasians and Asians.^24,25 These disparities in normal alignment across ethnic groups suggest the need to establish ideal global parameters tailored to each ethnicity. In the current study, even within the relatively younger non-MC subgroup, the mean GAP score was 5.9 (indicating moderate disproportionality), suggesting that the ideal alignment for Asians may differ from that defined in the original GAP score. Furthermore, a previous validation study conducted in a population with an average age approximately 20 years older reported a tendency toward higher GAP scores in the MC-present group but did not identify a strong correlation between GAP scores and MC. ¹¹ Therefore, alignment criteria aimed at minimizing MC should be developed based on ideal values tailored to each age group and ethnicity. ²⁶

Second, as discussed above, bone- and muscle-related factors associated with advanced age may diminish the effectiveness of the GAP score. Incorporating bone mineral density, ¹⁰ or alternative measure such as vertebral HU values given the limitations of BMD,^20,27,28 alongside muscle quality metrics such as erector spinae assessment, ²³ has therefore been proposed to enhance its predictive capability. Therefore, when establishing correction goals to prevent MCs, it is imperative to consider factors beyond the GAP score, including bone strength and muscle mass, tailored to the specific patient population. ²⁹

To our knowledge, this is the first multicenter external validation of the GAP score in an Asian ASD cohort that explicitly stratifies by age, formally tests the age × GAP interaction, and evaluates robustness across both cutoffs and outcome definitions. The descriptively inverse direction of the alignment–MC relationship observed in the older Asian cohort, although not statistically established, has not been previously reported and provides a hypothesis-generating signal for future studies focused on bone- and muscle-related risk factors. In addition, the presence of pelvic fixation was identified as a significant background factor in both age groups, underscoring its clinical relevance in ASD surgery. A dedicated study examining how pelvic fixation type and extension level interact with sagittal alignment parameters to influence MC risk would provide further clinically actionable insights.

In this study cohort, representative vertebral HU values were used as an alternative to BMD measurements. The results showed that representative vertebral HU values had a borderline association with MCs, indicating that bone quality may partially elucidate the occurrence of MCs beyond sagittal alignment parameters. Risk assessment focusing on HU values, particularly local HU values in the UIV region (which may be important for representative MCs such as PJK) will be addressed in our separate study and is therefore not covered here. However, the importance of HU values in MC risk assessment should continue to be explored. Furthermore, we acknowledge that the absence of assessments of muscle mass, muscle quality, and frailty in this cohort represents a significant limitation. Future studies should conduct comprehensive assessments of patient frailty using both basic evaluations, such as grip strength, and indicators such as magnetic resonance imaging analysis and skeletal muscle mass index measurements.

Although we have discussed the limitations of the GAP score within a relatively older Asian demographic, we do not intend to entirely dismiss its utility. The primary significance of the GAP score lies in the “GAP concept,” which delineates focal points for alignment across four primary domains regarding the correction goals for ASD treatment that emphasize the harmony between PI and LL. Although the predictive power of the GAP score for postoperative MCs remains noteworthy, we argue that it is inherently challenging for a single scoring system to encapsulate human diversity and the myriad issues associated with aging. The predictive ability of the GAP score alone is limited (AUC 0.466), suggesting that sagittal alignment parameters alone may not adequately account for the occurrence of MCs following ASD surgery. As previously mentioned, age-specific factors, age-specific factors not captured by radiographic alignment may be involved in the development of MCs. Therefore, by refining the GAP score to incorporate these factors, the GAP concept could evolve into a more effective tool for predicting MCs. ³⁰ This study contributes to that effort.

We should acknowledge some limitations of this study. First, its retrospective design may have introduced selection bias. Additionally, differences in baseline characteristics between the analysis cohort and the excluded cohort further contribute to this bias. Second, the patient groups included in this study may be inherently heterogeneous, even when stratification or matching is applied. Consequently, statistical significance may not necessarily translate directly into clinical significance. Third, the level of the UIV was not included as a covariate in the present analysis; given that UIV selection may influence PJK risk, future studies should incorporate UIV level as a systematic variable. Fourth, analyzing thresholds to determine a decline in the predictive ability of the GAP score proved challenging due to the limited number of cases. Another limitation is that MCs were defined as a composite outcome including PJK, DJK, rod fractures, and screw loosening. These complications have different pathophysiologies, and their combination might have influenced the predictive performance of the GAP score. Finally, this study was not designed with a formal a priori sample size calculation. A post-hoc simulation-based power analysis (5,000 iterations; two-sided α = 0.05) indicated that, the study had approximately 76–83% power to detect a moderate interaction effect (OR = 0.7–1.3 per unit GAP score per age stratum) but only 55–58% power for smaller effects (OR = 0.8–1.2). The non-significant interaction finding should therefore be interpreted with caution. Nevertheless, this study provides valuable insights that enhance our understanding of the relationship between the GAP score and older Asian patients with ASD. Although propensity score matching achieved adequate balance across measured covariates, residual confounding from unmeasured variables — particularly direct frailty measures and detailed bone-quality assessments — cannot be excluded.

Conclusions

In this multicenter external validation of the GAP score in an Asian ASD cohort with explicit age stratification, the GAP score demonstrated limited predictive ability for postoperative MCs. The hypothesized age-modifying effect was not confirmed by logistic regression, and no significant age × GAP interaction was identified. Notably, within the Over-70 group, the MC subgroup paradoxically showed lower (more proportioned) GAP scores than the non-MC subgroup after matching, a pattern that, although not statistically established, suggests that age-related factors such as osteoporosis, sarcopenia, and frailty may exert a more dominant influence on MC risk than sagittal alignment parameters in older Asian patients.

ORCID iDs

Naoki Segi https://orcid.org/0000-0001-9681-2422

Shiro Imagama https://orcid.org/0000-0002-6951-8575

Hiroaki Nakashima https://orcid.org/0000-0002-0039-9678

Ethical Considerations

This study adhered to the guidelines of the Declaration of Helsinki and received approval from the Institutional Review Board of the coordinating institution (No. 2016-0177) and each of the other institutions.

Consent to Participate

As all interventions (including surgical procedures and preoperative/postoperative examinations) were routine, the requirement for obtaining written informed consent was waived.