Verification of the Semiquantitative Assessment of Vertebral Deformity for Subsequent Vertebral Body Fracture Prediction and Screening for the Initiation of Osteoporosis Treatment: A Case-Control Study Using a Clinical-Based Setting

Abstract

Background/Objectives: Semiquantitative grading of the vertebral body (SQ) is an easy screening method for vertebral body deformation. The validity of SQ as a risk factor and screening tool for incident osteoporotic fractures in the vertebral body (OF) was investigated using retrospective case-control data. Methods: Outpatients with osteoporosis who were followed up for ≥2 years as patients with osteoporosis were recruited. All of them were tested using X-ray images of the lateral thoracolumbar view and other tests at baseline. Patients were classified according to the SQ grade, and potential risk factors were compared for each SQ group. Cox regression analyses were conducted on the incident OFs. Statistical differences in the possible risk factors among the groups and the likelihood of incident OFs in the variables were examined. After propensity score matching (PSM) and inverse probability of treatment weighting (IPTW) for confounding factors, the possibility of incident OFs was compared between the SQ grade groups. Results: In the crude dataset, the probability of incident OF in SQ Grade 3 was significantly higher than in other grade groups. Using a Cox regression analysis in multivariate mode, SQ grade was the only statistically significant factor for incident OF. However, no significant differences were observed between PSM and IPTW. Conclusions: These results suggest that the SQ classification was inappropriate for predicting incident OFs. However, the grading showed a significantly higher risk than that available for screening.

1. Introduction

Osteoporosis is a disease that creates an internal bone environment prone to fractures due to weakened bone strength [1,2]. Risk factors for bone fragility include decreased bone mineral density [3]; prolonged, chronic, and severe lifestyle-related diseases [4,5]; low striated muscle mass [6]; falls made easier by various locomotive diseases [7,8,9,10]; decreased renal function [11], and poor nutritional status [9,12,13]. Typical osteoporotic fracture sites include the vertebral body, proximal femur, distal radius, and proximal humerus, with vertebral fractures being among the most common [9,14]. Many cases of vertebral fracture (VF) result in breaks without a clear mechanism of injury. This is also referred to as a fracture, often associated with the Japanese concept of “fracture before you know it,” where fractures occur without an episodic fall [15]. Vertebral body fractures are a serious consequence of osteoporosis and are widely recognized as one of the most significant risk factors for subsequent osteoporotic fractures [14,16,17]. Therefore, it is essential to confirm the presence or history of VF through screening to prevent secondary vertebral fractures.

The severity of vertebral deformities and the risk of subsequent fractures are often discussed. In spine radiographs, the semiquantitative (SQ) criteria proposed by Genant et al. are commonly used to identify vertebral deformity (VD) in spines T4–L4 [18]. Genant et al. classified the spine as usual by visual inspection of the lateral view of the spine X-ray image.

Due to its simplicity, the SQ method is often used to screen for VD. In Japan, some facilities have adopted it as a method for osteoporosis screening because of its ease of use [19,20,21]. It has also been reported that the evaluation of VD is a valuable predictor of VF [22]. Yoshimura et al. conducted a large cohort study in three rural towns in Japan. They concluded that a prevalent vertebral fracture is a risk factor for subsequent vertebral fractures, even in the Grade 1 minor deformity group [23]. Minor deformities in Grades 2 or 3 were more common among men using the Genant SQ method [23,24]. However, these analyses did not address the influence of confounding factors. Another report conducted a cross-sectional study on the relationship between mild wedge deformity and its prevalence in vertebrae, finding that mild vertebral deformities are likely indicative of diminishing bone metabolism due to aging. These may be deformities caused not by osteoporosis but by age-related spinal diseases [25]. There are numerous confounding factors in assessing the prediction of subsequent VDs using the SQ method. The gap between clinical-based evidence and population-based evidence is different. In this study, we tried to clarify the appropriateness of the SQ grading in the clinical-based study.

Little information exists regarding the association between SQ and osteoporotic non-vertebral fractures (NVF) [26]. However, there should be a reasonable causal relationship if SQ represents vertebral fractures.

Our clinical question is whether the SQ grading is appropriate for predicting incident VF and relevant as a screening tool for osteoporotic treatment judgment. To clarify the questions, we conducted retrospective case-control studies that aim to determine if SQ grading is suitable for predicting incident VFs. We investigated the frequency of incident VFs and the risk weights of factors for incident fractures using small retrospective case-control study data from a rural area in Japan.

2. Materials and Methods

2.1. Patient Inclusion and Exclusion

A lateral view of the thoracolumbar spine X-ray image was taken of patients who consulted outpatients due to musculoskeletal problems, and bone mineral density (BMD) was measured in the lumbar spine and femoral neck using dual-energy X-ray absorptiometry (DXA) with the DPX^® Bravo ME9309 Bone Densitometer (GE Health Care, Chicago, IL, USA: Coefficients of variation; CV: 1.1% (lumbar spine), 0.9% (femoral neck)) from March 2013 to December 2020. The date of the X-ray image was designated as a baseline, and patients were followed up for at least 24 months unless an incident fracture developed. Blood samples were also collected for testing at baseline. Patients who did not undergo the assessments used in this study were excluded. Patients lost to follow-up before 2 years due to reasons unrelated to fractures, such as long-term admission because of severe comorbidities, relocation, admission to a nursing home, or censored death, were also excluded. Patients with Scheuermann’s deformity and hypophosphatemia rickets were excluded as well. In the patients included, the follow-up periods were determined from baseline to the last consultation date. Patients with confirmed incident vertebral fractures, regardless of whether there was an apparent injury episode, were examined through the lateral view of the spine X-ray; the date of the fracture was set as the final date. The SQ method was not used in these cases, but changes from the baseline X-ray image were adopted. For patients who died, were admitted due to severe comorbidities, or were lost to follow-up for unknown reasons after 2 years of follow-up, the last consultation date at our institute was set as the most recent.

2.2. Primary Vertebral Deformity Defining Using the SQ Method

Primary vertebral deformities were classified using the semiquantitative (SQ) method based on the lateral view of the thoracolumbar spine X-ray image. We arranged the vertebral deformity following the identification of Genant’s SQ classification by assessing vertebral height, specifically the compression rate from Grade 0 to Grade 3 (G-0, compression up to 5% in height; G-1, compression up to 25% or a 10–20% reduction in area; G-2, compression from 26% to 40% in height or a 20–40% reduction in area; G-3, compression > 40% in height and area). Two of the authors, who are physicians, conducted the assessments. If their evaluations did not align, a third assessor was involved, and a majority vote determined the final decision. When multiple fractures were observed on the X-ray images, the more severe grade was assigned. Cohen’s kappa score was calculated.

2.3. Other Factors Extraction

Prevalent clinical VF and NVF, extracted from medical records, current smoke and alcohol habits, and parents’ fracture history, were gathered from the baseline interview. Positive findings of vertebral endplate collapse (VEC), identified using an algorithm-based quantitative method (ABQ) and observed in X-ray images, were defined from the lateral view of the thoracolumbar X-ray image [21]. The presence of abdominal aortic calcification (AAC) was also determined from the lateral view of the thoracolumbar radiograph. Comorbidities related to lifestyle diseases, such as diabetes mellitus (DM), chronic obstructive pulmonary disease (COPD), hypertension (HT), hyperlipidemia (HL), chronic heart failure (CHF), insomnia, and cognitive impairment (C-I), were diagnosed and extracted by a physician certified by the Japanese Society of Internal Medicine. Neuromuscular comorbidities, including musculoskeletal ambulation disability complex (MADS), rheumatoid arthritis (RA), osteoarthritis of the lower extremities (OA), joint contractures of the trunk or lower extremities (contractures), disuse syndrome (disuse), and neuromuscular disorders (Parkinsonism), were also diagnosed and extracted by a physician certified by the Japanese Orthopedic Association and the Japanese College of Rheumatology.

The patients’ potential risk factors at baseline were extracted from several sources. These include the patients’ T-scores for the lumbar spine. The femoral neck measurement was derived from the DXA examination, along with medical records detailing glucocorticoids (GCs), vitamin D (V-D), anti-osteoporotic drug (OPD) administration, and polypharmacy involving no fewer than seven types of tablets or capsules. Furthermore, blood tests provided additional data, such as estimated glomerular filtration rate with cystatin C (eGFR_CysC), albumin (ALB), calcium (Ca), procollagen type 1 amino-terminal propeptide (P1NP), tartrate-resistant acid phosphatase-5b (TRACP-5b), hemoglobin (Hgb), monocyte count (Mono), and lymphocyte count (Lymph). These factors were accessed from 1 September 2023 to 15 April 2024.

2.4. Statistical Analyses

2.4.1. In the Crude Dataset

Group Comparison

The probability of incident clinical VF for each grade group, using the SQ classification method at baseline, was compared with the ANOVA Scheffé test. The primary endpoint of this study was the development of incident vertebral fractures. The mean values of the background clinical data for each grade class were also compared using the ANOVA Scheffé test.

Hazard ratios of the VF in the SQ grade groups were compared using the Kaplan-Meier curve analysis (K-M).

Risk Factor Analysis

Cox regression analyses were performed for these patients. The development of incident clinical VF was established as the dependent factor (binary: yes/no). The follow-up period length in months (continuous number) was designated as the survival time. Independent factors included the patient’s sex (binary; male/female), age (continuous number), current smoking status (binary; yes/no), alcohol consumption (binary; yes/no), parent’s history of fractures (binary; yes/no), prevalent VF (binary; yes/no), prevalent NVF (binary; yes/no), positivity for VEC (binary; yes/no), presence of AAC (binary; yes/no), presence of each comorbidity (binary; yes/no), T-score in the lumbar spine and femoral neck (LS and FN) (continuous number), medication factors such as GCs, V-D, OPD, polypharmacy (binary; yes/no), and blood factors including eGFR_CysC, ALB, Ca, P1NP, TRACP-5b, Hgb, and Mono (continuous number), which were established as independent factors. Vertebral deformities classified as SQ at baseline were also identified as independent factors. This index was categorical data, but each grade was converted to an integer for statistical analysis.

First, these factors were examined using univariate models, followed by a multivariate model for those demonstrating a significant risk ratio. When SQ was identified as a crucial risk factor in the multivariate model, the study progressed to the next step. If not, the association between the significant factor and the SQ grade was assessed.

Differences in the Variables Among the SQ Grade Groups and the Likelihood of the Incident Clinical VFs in These Variables Analysis

We statistically compared variables among the SQ grade groups using Kruskal–Wallis ANOVA. Furthermore, we analyzed these variables for their likelihood of incident clinical OFs separated into VFs using the K-M method. Before testing the K-M, a receiver operating characteristic (ROC) analysis was performed to determine the cut-off index (COI) for continuous or stepwise numbers. We assessed when a variable was statistically significant for both tests; this variable served as the confounding factor for the incidence of VF.

Cox Regression Analysis When the SQ Grading Is Modified, Whether Grade 3 or Not

We conducted another study as a post hoc analysis. A Cox regression analysis was conducted for incident VFs with risk factors, including the SQ grading, but the SQ grading was modified to binary data, whether Grade 3 or not. As described in the first Cox regression analysis, factors were examined using univariate models, followed by a multivariate model for those demonstrating a significant risk ratio.

2.4.2. In the Dataset After PSM

Group Comparison After PSM

As a next step, the probability of incident VF for each SQ grade group was compared using an ANOVA Scheffé test after leveling the confounding factors in the crude dataset with the PSM technique. Furthermore, if SQ grading was not listed for any of the incident osteoporotic fractures, the probability of the other incident fractures for each grade group after PSM was also compared using the ANOVA Scheffé test. The mean values of the background clinical data for each grade group were similarly compared using the Scheffé test.

Hazard ratios of the grade groups were also compared using the K-M.

Risk Factor Analysis in the Dataset After PSM

The dataset was analyzed using Cox regression after PSM. The dependent factors were the incident VFs, while the independent factors consisted of the same candidate risk factors used in the crude dataset. Initially, these factors were assessed using univariate models; subsequently, a multivariate model was applied to the factors that exhibited a significant risk ratio.

Cox Regression Analysis When the SQ Grading Is Modified, Whether Grade 3 or Not

As already conducted in the crude dataset, a Cox regression analysis, including the SQ grading modified to whether it was Grade 3 or not, was also conducted in the dataset after PSM. Factors were examined using univariate models, followed by a multivariate model for those demonstrating a significant risk ratio.

2.4.3. Association Between the Significant Factor and the SQ Grade

The association between the significant factor and SQ grade was examined using binary regression analysis for binary data and linear regression analysis for continuous data in the crude dataset. The probabilities of incident VF for each SQ grade group were compared for each significant factor, separated by COI using ROC analysis.

2.4.4. In the Dataset After IPTW

Group Comparison After IPTW

As a third step, the probability of incident VF for each SQ grade group was compared using an ANOVA Scheffé test after leveling the propensity of confounding factors from Grade 0 to Grade 2 in the crude dataset using the IPTW technique [27]. Before applying the IPTW technique, patients younger than 65 years in men and younger than 50 years in women were excluded from the crude dataset. Furthermore, if SQ grading was not listed for any incident osteoporotic fractures, the probability of other incident fractures for each grade group after IPTW was also compared using the ANOVA Scheffé test. The mean values of the background clinical data for each grade group were also compared using the Scheffé test.

Risk Factor Analysis in the Dataset After IPTW

The dataset was examined using Cox regression analysis after IPTW. The dependent factors were the development of the incident VFs, and the independent factors were the same candidate risk factors used in the crude dataset. First, these factors were examined using univariate models; then, a multivariate model was used for factors that demonstrated a significant risk ratio.

A flowchart of the study, including the inclusion and exclusion criteria, is shown in Figure 1.

• Statistical procedures and software

All statistical analyses were performed using StatPlus:mac^® (AnalystSoft Inc., Walnut, CA, USA). Statistical significance was set below 5%.

• Ethical considerations

The Ethics Committee of the Associated Institute approved the study protocol and patient consent requirements. The participants and their families were informed that the personal information obtained in this study was anonymous and would be used solely for analysis.

3. Results

A total of 1045 patients (143 male and 902 female) were included in this study. The mean age was 78.3 years (SD: 10.8), and the mean follow-up period was 49.7 months (SD: 30.1; median: 33). Among these patients, 419 were in SQ G-0, 195 in SQ G-1, 235 in SQ G-2, and 196 in SQ G-3, included in the crude dataset. The Cohen Kappa score was 0.865 (95% CI, 0.841–0.890).

3.1. In the Crude Dataset

3.1.1. Group Comparison

A total of 59 patients had incident VFs in the crude dataset; of these, 13 (3.1%), 9 (4.6%), 13 (5.5%), and 24 (12.2%) in the G-0, G-1, G-2, and G-3 groups, respectively, were included. G-3 had a significantly higher incidence of vertebral fracture rate than the other SQ groups (p < 0.001).

There were 111 incident NVFs in the crude dataset; of these, 22 (5.3%), 25 (12.8%), 27 (11.5%), and 37 (18.9%) were categorized as G-0, G-1, G-2, and G-3, respectively. The likelihood of incident VF in the crude dataset varied significantly among the SQ groups (p < 0.001). Significant differences were noted between G-0 and G-3, G-1 and G-3, and G-2 and G-3 (p < 0.001). In contrast, the likelihood of incident NVF in the crude dataset was significantly higher for G-3 and G-1 compared to G-0 (

Table 1. Demographic Characteristics of the Patients in the Crude Dataset.

	All (N = 1045)	Grade 0 (N = 419)	Grade 1 (N = 195)	Grade 2 (N = 235)	Grade 3 (N = 196)	p-Value in ANOVA	Statistical Significance in Scheffé Test
male:female (%)	13.7:86.3	15.8:85.8	16.4:83.6	12.3:87.7	8.2:91.8	<0.05	n.s.
age, years	78.3 (10.8)	72.6 (11.7)	80.2 (8.4)	82.4 (8.5)	84.0 (7.0)	<0.001	Grade-0 <<< Grade-1 Grade-0 <<< Grade-2 Grade-0 <<< Grade-3 Grade-1 << Grade-3
incident VF	59 (5.6%)	13 (3.1%)	9 (4.6%)	13 (5.5%)	24 (12.2%)	<0.001	Grade-0 <<< Grade-3 Grade-1 < Grade-3 Grade-2 < Grade-3
incident NVF	111 (10.6%)	22 (5.3%)	25 (12.8%)	27 (11.5%)	37 (18.9%)	<0.001	Grade-0 < Grade-1 Grade-0 <<< Grade-3
current smoke	25 (2.4%)	15 (3.6%)	4 (2.1%)	2 (0.9%)	4 (2.0%)	0.16	n.s.
alcohol habitat	14 (1.3%)	8 (1.9%)	2 (1.0%)	3 (1.3%)	1 (0.5%)	0.53	n.s.
parent’s fracture	16 (1.5%)	5 (1.2%)	3 (1.5%)	4 (1.7%)	4 (2.0%)	0.11	n.s.
prevalent VF	217 (20.8%)	38 (9.1%)	28 (17.0%)	65 (24.6%)	86 (43.7%)	<0.001	Grade-0 <<< Grade-2 Grade-0 <<< Grade-3 Grade-1 <<< Grade-3 Grade-2 <<< Grade-3
prevalent NVF	100 (9.6%)	30 (7.2%)	21 (12.7%)	19 (7.2%)	30 (15.2%)	<0.01	Grade-0 < Grade-3 Grade-2 < Grade-3
VEC	329 (31.5%)	38 (9.1%)	31 (15.9%)	64 (27.2%)	196 (100%)	<0.001	Grade-0 <<< Grade-2 Grade-0 <<< Grade-3 Grade-1 << Grade-2 Grade-1 <<< Grade-3 Grade-2 <<< Grade-3
AAC	784 (75.0%)	202 (48.2%)	174 (89.2%)	226 (96.1%)	182 (92.9%)	<0.001	Grade-0 <<< Grade-1 Grade-0 <<< Grade-2 Grade-0 <<< Grade-3
Diabetes mellitus	209 (20.0%)	94 (22.4%)	37 (19.0%)	44 (18.7%)	34 (17.3%)	0.43	n.s.
COPD	79 (7.6%)	26 (6.2%)	8 (4.1%)	24 (10.2%)	21 (10.7%)	<0.05	n.s.
hypertension	467 (44.7%)	160 (38.2%)	81 (41.5%)	113 (48.1%)	113 (57.7%)	<0.001	Grade-0 <<< Grade-3 Grade-1 < Grade-3
hyperlipidemia	247 (23.6%)	99 (23.6%)	49 (25.1%)	54 (23.0%)	45 (23.0%)	0.95	n.s.
insomnia	196 (18.8%)	75 (17.9%)	39 (20.0%)	42 (17.9%)	40 (20.4%)	0.83	n.s.
Cognitive Impairment	146 (14.0%)	25 (6.0%)	24 (12.3%)	47 (20.0%)	50 (25.5%)	<0.001	Grade-0 <<< Grade-2 Grade-0 <<< Grade-3 Grade-1 << Grade-3
MADS	197 (18.9%)	52 (12.4%)	35 (17.9%)	50 (21.3%)	60 (30.6%)	<0.001	Grade-0 <<< Grade-2 Grade-0 <<< Grade-3 Grade-1 << Grade-3
rheumatoid arthritis	303 (29.0%)	156 (37.2%)	60 (30.8%)	48 (20.4%)	39 (19.9%)	<0.001	Grade-0 <<< Grade-2 Grade-0 <<< Grade-3
osteoarthritis	526 (50.3%)	197 (47.0%)	106 (54.4%)	120 (51.1%)	103 (50.1%)	0.32	n.s.
Contracture	91 (8.7%)	37 (8.8%)	20 (10.3%)	19 (8.1%)	15 (7.7%)	0.81	n.s.
Disuse	55 (6.2%)	17 (4.1%)	10 (5.1%)	17 (7.2%)	21 (10.7%)	<0.05	Grade-0 < Grade-3
Parkinsonism	25 (2.4%)	4 (1.0%)	7 (3.6%)	6 (2.6%)	8 (4.1%)	0.06	n.s.
T-score in the LS	−2.3 (1.7)	−1.7 (1.8)	−2.2 (1.6)	−2.7 (1.5)	−3.1 (1.5)	<0.001	Grade-0 >> Grade-1 Grade-0 >>> Grade-2 Grade-0 >>> Grade-3 Grade-1 > Grade-2 Grade-1 >>> Grade-3 Grade-2 > Grade-3
T-score in the FN	−2.0 (1.2)	−1.6 (1.2)	−2.0 (1.0)	−2.3 (1.1)	−2.7 (1.0)	<0.001	Grade-0 >>> Grade-1 Grade-0 >>> Grade-2 Grade-0 >>> Grade-3 Grade-1 >>> Grade-3 Grade-2 >> Grade-3
GCs	180 (17.2%)	73 (17.4%)	46 (23.6%)	31 (13.2%)	30 (15.3%)	<0.05	Grade-1 > Grade-2
V-D	621 (59.4%)	215 (50.0%)	137 (70.3%)	147 (62.6%)	122 (48.6%)	<0.001	Grade-0 <<< Grade-1
OPD †	212 (20.3%)	57 (13.6%)	49 (25.1%)	39 (16.7%)	67 (34.2%)	<0.001	Grade-0 < Grade-1 Grade-0 <<< Grade-3 Grade-2 <<< Grade-3
polypharmacy	168 (16.1%)	64 (15.3%)	33 (16.9%)	38 (16.2%)	33 (16.8%)	0.94	n.s.
eGFR	58.9 (21.2)	67.1 (21.8)	56.6 (18.8)	52.2 (20.2)	49.9 (16.1)	<0.001	Grade-0 >>> Grade-1 Grade-0 >>> Grade-2 Grade-0 >>> Grade-3
ALB	4.0 (0.4)	4.1 (0.4)	4.0 (0.3)	3.9 (0.4)	3.9 (0.4)	<0.001	Grade-0 >>> Grade-2 Grade-0 >>> Grade-3 Grade-1 > Grade-2 Grade-1 >> Grade-3
P1NP	61.9 (54.4)	62.0 (56.4)	57.4 (48.2)	58.8 (49.6)	69.9 (60.5)	0.13	n.s.
TRACP-5b	505.9 (244.2)	491.1 (218.1)	483.0 (196.4)	542.1 (313.7)	516.2 (240.6)	<0.05	n.s.
Hgb	12.2 (1.6)	12.6 (1.5)	12.3 (1.5)	11.9 (1.8)	11.7 (1.4)	<0.001	Grade-0 >>> Grade-2 Grade-0 >>> Grade-3 Grade-1 >> Grade-3
Lymph	1563 (660)	1637 (652)	1636 (686)	1508 (670)	1401 (608)	<0.001	Grade-0 >>> Grade-3 Grade-1 >> Grade-3
Mono	344 (141)	342 (141)	344 (117)	351 (157)	341 (144)	0.88	n.s.

The values are presented as mean (SD) unless indicated otherwise. † anti-osteoporotic drugs included selective estrogen receptor modulators, bisphosphonates, denosumab, teriparatide, and romosozumab. Symbols: >, left is greater than right at a 5% significance threshold; >>, left is greater than right at a 1% significance threshold; >>>, left is greater than right at a 0.1% significance threshold; <, right is greater than left at a 5% significance threshold; <<, right is greater than left at a 1% significance threshold; <<<, right is greater than left at a 0.1% significance threshold. Abbreviations: VF, vertebral fracture; NVF, non-vertebral fracture; VEC, vertebral endplate collapse positivity with algorithm-based quantitative method; AAC, abdominal aortic calcification; COPD, chronic obstructive pulmonary diseases; MADS, musculoskeletal ambulation disability symptoms complex; LS, lumbar spine; FN, femoral neck; GCs, glucocorticoids; V-D, vitamin D; OPD, anti-osteoporotic drugs; eGFR, estimated glomerular filtration ratio based on cystatin C; ALB, albumin; P1NP, procollagen type 1 amino-terminal propeptide; TRACP-5b, tartrate-resistant acid phosphatase-5b; Hgb, hemoglobin; Lymph, lymphocyte count; Mono, monocyte count.

).

Many significant factors were present when comparing the SQ groups. Factors that demonstrated significant differences among the SQ groups were mean age, follow-up length, prevalent VF and NVF, positivity with VEC, presence of AAC, HT, CHF, C-I, MADS, RA, and Disuse; mean T-score in the lumbar spine and femoral neck; administration of GCs, V-D, and OPD at baseline; and mean eGFR_CysC, mean serum ALB level, blood Hgb level, and lymph level. The additional details are provided in Table 1.

In the K-M, the Hazard ratios for incident VFs of G-2/G-0, G-3/G-0, and G-3/G-1 were significantly higher, with 2.13 (95% CI: 1.07 to 4.24), 4.47 (95% CI: 2.18 to 9.15), and 2.76 (95% CI: 1.19 to 6.44), respectively (Figure 2).

3.1.2. Risk Factor Analysis

In the crude dataset, older age, higher than Grade 1 in SQ, presence of AAC, CHF, C-I, MADS, lower T-score in the lumbar spine and femoral neck, lower eGFR_CysC, lower Hgb, and lower lymph had significantly higher risk ratios of incident VF using univariate models. Only those higher than Grade 1 in the SQ had significantly higher risk ratios using multivariate models (

Table 2. Results of a Cox Regression Analysis for Incident VF Development in the Crude Dataset.

	Univariate Mode		Multivariate Mode
	p-Value	Risk Ratio (95% CI)	β-Value	p-Value	Risk Ratio (95% CI)
female	0.18	2.21 (0.69–7.07)
age, years	<0.001	1.05 (1.02–1.07)	−0.014	0.59	0.99 (0.94–1.04)
current smoke	0.78	0.82 (0.20–3.36)
alcohol habitat	0.98	0.00 (0.00–INF)
parent’s fracture	0.98	0.00 (0.00–INF)
SQ Grade	<0.001	1.82 (1.46–2.27)	0.546	<0.01	1.73 (1.22–2.44)
prevalent VF	0.98	2.6 × 108 (0.00–INF)
prevalent NVF	0.37	1.52 (0.60–3.84)
VEC	0.98	3.2 × 107 (0.00–INF)
AAC	<0.05	1.88 (1.01–3.52)	−0.158	0.76	0.85 (0.30–2.40)
Diabetes mellitus	0.28	0.70 (0.37–1.33)
COPD	0.17	0.44 (0.14–1.41)
hypertension	0.08	1.61 (0.94–2.75)
hyperlipidemia	0.98	0.99 (0.58–1.70)
chronic heart failure	<0.01	2.21 (1.28–3.80)	0.407	0.29	1.50 (0.71–3.19)
insomnia	0.37	0.76 (0.41–1.39)
Cognitive Impairment	<0.05	2.01 (1.10–3.68)	0.141	0.76	1.15 (0.47–3.82)
MADS	<0.05	1.96 (1.15–3.34)	−0.204	0.58	0.82 (0.40–1.68)
rheumatoid arthritis	0.06	0.57 (0.32–1.02)
osteoarthritis	0.71	1.11 (0.65–1.90)
Contracture	0.72	1.14 (0.57–2.25)
Disuse	0.15	1.75 (0.82–3.70)
Parkinsonism	0.84	0.82 (0.11–5.94)
T-score in the LS	<0.01	0.78 (0.66–0.92)	0.012	0.93	1.01 (0.77–1.32)
T-score in the FN	<0.001	0.64 (0.51–0.82)	−0.326	0.11	0.72 (0.48–1.08)
GCs	0.24	0.68 (0.35–1.31)
V-D	0.62	1.15 (0.66–1.98)
OPD †	0.65	1.14 (0.66–1.96)
polypharmacy	0.27	0.67 (0.33– 1.36)
eGFR	<0.01	0.97 (0.96–0.99)	−0.017	0.12	0.98 (0.96–1.00)
ALB	0.21	0.60 (0.27–1.33)
Calcium	0.14	1.17 (0.95–1.45)
P1NP	0.23	1.00 (0.99–1.01)
TRACP-5b	0.16	1.00 (1.00–1.00)
Hgb	<0.01	0.78 (0.67–0.92)	−0.195	0.14	0.82 (0.64–1.07)
Lymph	<0.001	1.00 (1.00–1.00)	−0.000	0.99	1.00 (1.00–1.00)
Mono	0.06	1.00 (1.00–1.00)

† anti-osteoporotic drugs included selective estrogen receptor modulators, bisphosphonates, denosumab, teriparatide, and romosozumab. Abbreviations: SQ, semiquantitative classification of vertebral fracture; VF, vertebral fracture; NVF, non-vertebral fracture; VEC, vertebral endplate collapse positivity with algorithm-based quantitative method; AAC, abdominal aortic calcification; COPD, chronic obstructive pulmonary diseases; MADs, musculoskeletal ambulation disability symptoms complex; LS, lumbar spine; FN, femoral neck; GCs, glucocorticoids; V-D, vitamin D; OPD, anti-osteoporotic drugs; eGFR, estimated glomerular filtration ratio based on cystatin C; ALB, albumin; P1NP, procollagen type 1 amino-terminal propeptide; TRACP-5b, tartrate-resistant acid phosphatase-5b; Hgb, hemoglobin; Lymph, lymphocyte count; Mono, monocyte count.

).

3.1.3. Differences in the Variables Among the SQ Grade Groups and the Likelihood of the Incident VFs in These Variables Analysis

Age at baseline, presence of prevalent VF, presence of VEC using ABQ, presence of AAC, HT, CHF, C-I, MADS, RA, SpA, T-score in the lumbar spine and femoral neck, administration of Vitamin D and OPD, eGFR_CysC, serum ALB level, blood HGB level, and lymph count showed significant differences among the SQ grade groups based on the Kruskal–Wallis test. In contrast, the likelihood of incident VFs in these variables exhibited statistical significance in VEC with ABQ, presence of CHF, C-I, MADS, T-score in the lumbar spine, and eGFR_CysC as indicated by K-M. Consequently, these variables were deemed confounding factors for SQ grade regarding incident VFs (

Table 3. Statistical Differences in Variables among SQ Grades Examined with Kruskal–Wallis ANOVA Test Presented with p-values and Likelihood of Incident VFs of the Variables Examined with a Receiver Operating Characteristic (ROC) and Kaplan-Meier Analyses Presented with Cut-Off Index (COI), Hazard Ratio, and p-values.

Variables	Incident VF (ROC and Kaplan-Meier)
	Kruskal–Wallis	ROC	Kaplan-Meier
	p-Value	COI (†)	Positive	Negative	Hazard Ratio	p-Value
gender	0.43	female	56/902 (6.2%)	3/143 (2.1%)	2.80	0.07
age	<0.001	>86 (#)	51/788 (6.5%)	8/249 (3.1%)	1.69	0.16
current smoke	0.95	smokes	2/25 (8.0%)	57/1020 (5.6%)	1.14	0.85
alcohol habitat	0.99	positive	59/1031 (5.7%)	0/14 (0.0%)	INF	n/a
parent’s fracture	0.98	positive	59/1029 (5.7%)	0/16 (0.0%)	INF	n/a
SQ Grade		≥Grade-2 (#)	37/431 (8.6%)	22/614 (3.6%)	3.5	<0.001
prevalent VF	<0.001	positive (#)	59/217 (27.2%)	0/828 (0.0%)	INF	<0.001
prevalent NVF	0.38	positive	16/111 (14.4%)	43/934 (4.6%)	2.60	<0.001
VEC	<0.001	positive (#)	59/329 (17.9%)	0/716 (0.0%)	INF	<0.001
AAC	<0.001	positive (#)	46/784 (5.9%)	13/261 (5.0%)	1.33	0.36
Diabetes mellitus	0.71	positive	12/209 (5.7%)	47/836 (5.6%)	0.87	0.68
COPD	0.57	positive	56/906 (5.8%)	3/79 (3.8%)	1.89	0.27
hypertension	<0.001	positive (#)	34/467 (7.3%)	25/578 (4.3%)	1.43	0.17
hyperlipidemia	0.98	positive	21/247 (8.5%)	38/798 (4.8%)	1.42	0.19
CHF	<0.001	positive (#)	21/209 (10.0%)	38/836 (4.5%)	2.06	<0.01
insomnia	0.94	positive	14/196 (7.1%)	45/849 (5.3%)	1.13	0.69
Cognitive Impairment	<0.001	positive (#)	14/146 (9.6%)	45/899 (5.0%)	1.96	<0.05
MADS	<0.01	positive (#)	21/197 (10.7%)	38/848 (4.5%)	2.09	<0.01
rheumatoid arthritis	<0.001	negative (#)	43/742 (5.8%)	16/303 (5.3%)	1.42	0.22
osteoarthritis	0.47	positive	38/526 (7.2%)	21/519 (4.0%)	1.59	0.08
Contracture	0.97	positive	10/91 (11.0%)	49/954 (5.1%)	1.69	0.13
Disuse	0.59	positive	8/65 (12.3%)	51/980 (5.2%)	1.98	0.07
Parkinsonism	0.92	negative	58/1020 (5.7%)	1/25 (4.0%)	1.50	0.69
T-score in the LS	<0.001	≤−2.3 (#)	41/577 (6.9%)	18/468 (3.8%)	1.97	<0.05
T-score in the FN	<0.001	≤−2.5 (#)	27/421 (6.4%)	32/624 (5.1%)	1.43	0.16
GCs	0.30	positive	11/180 (6.1%)	48/865 (5.5%)	0.88	0.70
V-D	<0.001	positive (#)	39/621 (6.3%)	20/424 (4.7%)	1.30	0.34
OPD †	<0.001	positive (#)	20/212 (9.4%)	39/832 (4.7%)	1.70	0.05
polypharmacy	0.99	negative	50/877 (5.7%)	9/168 (5.4%)	1.22	0.58
eGFR	<0.001	<42.0 (#)	24/237 (10.1%)	35/808 (4.3%)	2.46	<0.01
ALB	<0.001	≤4.3 (#)	56/824 (6.3%)	3/151 (2.0%)	2.89	0.05
P1NP	0.28	≤22.6	17/138 (12.3%)	42/907 (4.6%)	2.32	<0.01
TRACP-5b	0.13	≤421	30/404 (7.4%)	29/641 (4.5%)	1.50	0.12
Hgb	<0.001	≤12.0 (#)	30/437 (6.9%)	29/608 (4.8%)	1.60	0.07
Lymph	<0.001	≥1136 (#)	50/751 (6.5%)	9/270 (3.3%)	1.80	0.10
Mono	0.86	>271	46/704 (6.5%)	13/330 (3.8%)	1.81	0.05

In the ROC rows, (#) shows a statistical significance within 5%. † anti-osteoporotic drugs included selective estrogen receptor modulators, bisphosphonates, denosumab, teriparatide, and romosozumab. Abbreviations: VF, vertebral fracture; NVF, non-vertebral fracture; VEC, vertebral endplate collapse positivity with algorithm-based quantitative method; AAC, abdominal aortic calcification; COPD, chronic obstructive pulmonary diseases; MADS, musculoskeletal ambulation disability symptoms complex; LS, lumbar spine; FN, femoral neck; GCs, glucocorticoids; V-D, vitamin D; OPD, anti-osteoporotic drugs; eGFR, estimated glomerular filtration ratio based on cystatin C; ALB, albumin; P1NP, procollagen type 1 amino-terminal propeptide; TRACP-5b, tartrate-resistant acid phosphatase-5b; Hgb, hemoglobin; Lymph, lymphocyte count; Mono, monocyte count.

).

3.1.4. Cox Regression Analysis When the SQ Grading Is Modified, Whether Grade 3 or Not

In the significant factors using a univariate model, presenting SQ Grade 3 showed significantly higher risk ratios of 3.39 (95% CI; 1.68–6.86) (p < 0.001). Age showed a significantly lower risk ratio of 0.95 (95% CI; 0.91–0.99) (p < 0.05). The other factors showed no significantly high or low risk ratios.

3.2. In the Dataset After PSM

3.2.1. Group Comparison After PSM

Because the SQ grade showed a significantly higher risk ratio only for incident VF, a PSM procedure was conducted for VFs to level the confounding factors. After PSM, the dataset was narrowed to 560 patients, with 140 patients in each grade group. Of these, 43 patients had incident VFs; 11 (7.9%), 9 (6.4%), 8 (5.7%), and 15 (10.7%) in the G-0, G-1, G-2, and G-3 groups, respectively, were included. The confounding factors, such as CHF, C-I, MADS, T-score in the LS, and eGFR_CysC, showed no statistical differences in any pairs of the SQ grade groups. However, VEC could not diminish the statistical difference in any way because 100% of the patients in the G-3 group were positive for VEC. Among the other factors, age at baseline, follow-up length, presence of AAC, DM, HL, administration of OPD, ALB, P1NP, Hgb, and Lymph significantly differed between the groups. The incidence of VF and NVF was not significantly different between any pairs in the SQ grade (

Table 4. Demographic characteristics of the patient groups after the PSM procedures.

	All (N = 560)	Grade 0 (N = 140)	Grade 1 (N = 140)	Grade 2 (N = 140)	Grade 3 (N = 140)	p-Value in ANOVA	Statistical Significance in Scheffé Test
male:female (%)	11.1:88.9	11.7:89.3	12.9:87.1	10.0:90.0	10.7:89.3	0.89	n.s.
age, years	80.6 (9.0)	77.0 (9.9)	80.1 (8.4)	82.0 (9.2)	83.1 (7.2)	<0.001	Grade-0 < Grade-1 Grade-0 <<< Grade-2 Grade-0 <<< Grade-3 Grade-1 < Grade-3
Time length of follow-up, months	45.4 (29.1)	51.8 (32.0)	46.6 (30.0)	42.0 (27.4)	41.0 (26.0)	<0.01	Grade-0 > Grade-3
incident VF	43 (7.7%)	11 (7.9%)	9 (6.4%)	8 (5.7%)	15 (10.7%)	0.41	n.s.
incident NVF	65 (11.6%)	12 (8.6%)	19 (13.6%)	15 (10.7%)	19 (13.6%)	0.49	n.s.
current smoke	11 (2.0%)	6 (4.3%)	2 (1.4%)	1 (0.7%)	2 (1.4%)	0.14	n.s.
alcohol habitat	2 (0.4%)	1 (0.7%)	0 (0.0%)	0 (0.0%)	1 (0.7%)	0.57	n.s.
parent’s fracture	4 (0.7%)	1 (0.7%)	1 (0.7%)	1 (0.7%)	1 (0.7%)	1.00	n.s.
prevalent VF	159 (28.4%)	33 (23.6%)	23 (19.8%)	42 (26.3%)	61 (42.4%)	<0.001	Grade-0 << Grade-3 Grade-1 <<< Grade-3 Grade-2 << Grade-3
prevalent NVF	63 (11.3%)	14 (10.0%)	18 (15.5%)	12 (7.5%)	19 (13.2%)	0.16	n.s.
VEC	241 (38.4%)	33 (23.6%)	31 (22.1%)	37 (26.4%)	140 (100%)	<0.001	Grade-0 <<< Grade-3 Grade-1 <<< Grade-3 Grade-2 <<< Grade-3
AAC	469 (83.8%)	76 (54.3%)	125 (89.3%)	136 (97.1%)	132 (94.3%)	<0.001	Grade-0 <<< Grade-1 Grade-0 <<< Grade-2 Grade-0 <<< Grade-3
Diabetes mellitus	102 (18.2%)	40 (28.6%)	25 (17.9%)	19 (13.6%)	18 (12.9%)	<0.01	Grade-0 > Grade-2 Grade-0 >> Grade-3
COPD	44 (7.9%)	10 (7.1%)	6 (4.3%)	15 (10.7%)	13 (9.3%)	0.21	n.s.
hypertension	241 (43.0%)	66 (47.1%)	54 (38.6%)	50 (35.7%)	71 (50.7%)	<0.05	n.s.
hyperlipidemia	135 (24.1%)	45 (32.1%)	38 (27.1%)	30 (21.4%)	22 (15.7%)	<0.01	Grade-0 < Grade-3
chronic heart failure	102 (18.2%)	24 (17.1%)	22 (15.7%)	23 (16.4%)	33 (23.6%)	0.09	n.s.
insomnia	99 (17.7%)	31 (22.1%)	25 (17.9%)	20 (14.3%)	23 (16.4%)	0.37	n.s.
Cognitive Impairment	66 (11.8%)	17 (12.1%)	17 (12.1%)	16 (11.4%)	16 (11.4%)	1.00	n.s.
MADS	93 (16.6%)	20 (14.3%)	25 (17.9%)	24 (17.1%)	24 (17.1%)	0.86	n.s.
rheumatoid arthritis	143 (25.5%)	46 (32.9%)	40 (28.6%)	30 (21.4%)	27 (19.3%)	<0.05	n.s.
osteoarthritis	299 (53.4%)	74 (52.9%)	77 (55.0%)	76 (54.3%)	72 (51.4%)	0.94	n.s.
Contracture	54 (9.6%)	17 (12.1%)	14 (10.0%)	13 (9.3%)	10 (7.1%)	0.56	n.s.
Disuse	34 (6.1%)	11 (7.9%)	6 (4.3%)	9 (6.4%)	8 (5.7%)	0.65	n.s.
Parkinsonism	10 (1.8%)	0 (0.0%)	5 (3.6%)	0 (0.0%)	5 (3.6%)	<0.05	n.s.
T-score in the LS	−2.8 (1.2)	−2.8 (1.3)	−2.8 (1.1)	−2.7 (1.1)	−2.8 (1.4)	0.96	n.s.
T-score in the FN	−2.3 (1.0)	−2.2 (1.0)	−2.2 (0.9)	−2.3 (0.9)	−2.5 (1.0)	<0.05	n.s.
GCs	95 (17.0%)	24 (17.1%)	31 (22.1%)	20 (14.3%)	20 (14.3%)	0.55	n.s.
V-D	374 (66.8%)	92 (65.7%)	104 (74.3%)	93 (66.4%)	85 (60.7%)	0.11	n.s.
OPD †	128 (22.9%)	28 (20.0%)	37 (26.4%)	21 (15.0%)	42 (30.0%)	<0.05	Grade-2 < Grade-3
polypharmacy	84 (15.0%)	25 (17.9%)	20 (14.3%)	19 (13.6%)	20 (14.3%)	0.75	n.s.
eGFR	56.8 (17.4)	58.4 (15.4)	58.5 (18.4)	57.2 (20.1)	52.2 (151)	0.12	n.s.
ALB	4.0 (0.4)	4.0 (0.4)	4.1 (0.3)	3.9 (0.4)	3.9 (0.4)	<0.01	Grade-1 > Grade-2 Grade-1 >> Grade-3
P1NP	59.8 (51.2)	57.5 (39.8)	50.5 (30.6)	62.0 (57.9)	69.4 (64.3)	<0.05	Grade-1 < Grade-3
TRACP-5b	507.7 (264.3)	523.8 (227.0)	469.8 (186.4)	528.9 (366.5)	508.7 (245.3)	0.25	n.s.
Hgb	12.1 (1.6)	12.3 (1.4)	12.4 (1.4)	12.0 (2.0)	11.8 (1.4)	<0.01	Grade-0 >> Grade-3 Grade-1 >> Grade-3
Lymph	1518 (656)	1510 (619)	1649 (715)	1523 (677)	1390 (590)	<0.05	Grade-1 > Grade-3
Mono	349 (144)	340 (142)	352 (119)	361 (158)	345 (154)	0.66	n.s.

The values are presented as mean (SD) unless indicated otherwise. † anti-osteoporotic drugs included selective estrogen receptor modulators, bisphosphonates, denosumab, teriparatide, and romosozumab. Symbols: >, left is greater than right at a 5% significance threshold; >>, left is greater than right at a 1% significance threshold; <, right is greater than left at a 5% significance threshold; <<, right is greater than left at a 1% significance threshold; <<<, right is greater than left at a 0.1% significance threshold. Abbreviations: PSM, propensity score matching; VF, vertebral fracture; NVF, non-vertebral fracture; VEC, vertebral endplate collapse positivity with algorithm-based quantitative method; AAC, abdominal aortic calcification; COPD, chronic obstructive pulmonary diseases; MADS, musculoskeletal ambulation disability symptoms complex; LS, lumbar spine; FN, femoral neck; GCs, glucocorticoids; V-D, vitamin D; OPD, anti-osteoporotic drugs; eGFR, estimated glomerular filtration ratio based on cystatin C; ALB, albumin; P1NP, procollagen type 1 amino-terminal propeptide; TRACP-5b, tartrate-resistant acid phosphatase-5b; Hgb, hemoglobin; Lymph, lymphocyte count; Mono, monocyte count.

).

3.2.2. Risk Factor Analysis in the Dataset After PSM

Older age and blood Lymph count were significant factors for the likelihood of incident VFs using Cox regression analysis with a univariate model in the dataset after PSM. However, the multivariate model revealed no significant factors (

Table 5. Results of a Cox regression analysis for incident VF in the dataset after PSM.

	Univariate Mode		Multivariate Mode
	p-Value	Risk Ratio (95% CI)	β-Value	p-Value	Risk Ratio (95% CI)
female	0.49	1.51 (0.47–4.88)
age, years	<0.05	0.97 (0.94–1.00)	−0.03	0.08	0.97 (0.94–1.00)
current smoke	0.97	1.04 (0.14–7.55)
alcohol habitat	0.99	0.00 (0.00–INF)
parent’s fracture	0.99	0.00 (0.00–INF)
SQ Grade	0.26	1.16 (0.89–1.52)
prevalent NVF	0.98	1.3 × 107 (0.00–INF)
VEC	0.98	1.4 × 108 (0.00–INF)
AAC	0.63	0.84 (0.40–1.75)
Diabetes mellitus	0.75	0.88 (0.41–1.90)
COPD	0.29	0.47 (0.11–1.93)
hypertension	0.88	1.05 (0.57–1.91)
hyperlipidemia	067	1.15 (0.60–2.20)
chronic heart failure	0.61	1.20 (0.59–2.44)
insomnia	0.71	1.15 (0.56–2.34)
Cognitive Impairment	0.68	1.20 (0.51–2.84)
MADS	0.12	1.70 (0.87–3.32)
rheumatoid arthritis	0.70	0.87 (0.45–1.71)
osteoarthritis	0.56	1.20 (0.65–2.21)
Contracture	0.20	1.65 (0.76–3.59)
Disuse	0.61	1.31 (0.47 –3.67)
Parkinsonism	0.98	0.00 (0.00–INF)
T-score in the LS	0.29	1.14 (0.90–1.44)
T-score in the FN	0.44	1.13 (0.83–1.54)
GCs	0.46	0.74 (0.33–1.66)
V-D	0.75	0.58 (0.58–2.13)
OPD †	0.20	1.52 (0.81–2.85)
polypharmacy	0.91	0.96 (0.43– 2.15)
eGFR	0.16	0.98 (0.96–1.01)
ALB	0.34	1.60 (0.61–4.23)
P1NP	0.54	1.00 (0.99–1.01)
TRACP-5b	0.92	1.00 (1.00–1.00)
Hgb	0.67	0.96 (0.79–1.17)
Lymph	<0.05	1.00 (1.00–1.00)	1.2 × 10−4	0.73	1.00 (1.00–1.00)
Mono	0.84	1.00 (1.00–1.00)

† anti-osteoporotic drugs included selective estrogen receptor modulators, bisphosphonates, denosumab, teriparatide, and romosozumab. Abbreviations: PSM, propensity score matching; SQ, semiquantitative classification of vertebral fracture; NVF, non-vertebral fracture; VEC, vertebral endplate collapse positivity with algorithm-based quantitative method; AAC, abdominal aortic calcification; COPD, chronic obstructive pulmonary diseases; MADS, musculoskeletal ambulation disability symptoms complex; LS, lumbar spine; FN, femoral neck; GCs, glucocorticoids; V-D, vitamin D; OPD, anti-osteoporotic drugs; eGFR, estimated glomerular filtration ratio based on cystatin C; ALB, albumin; P1NP, procollagen type 1 amino-terminal propeptide; TRACP-5b, tartrate-resistant acid phosphatase-5b; Hgb, hemoglobin; Lymph, lymphocyte count; Mono, monocyte count; PNI, prognostic nutritional index.

).

3.2.3. Cox Regression Analysis When the SQ Grading Is Modified, Whether Grade 3 or Not

In the significant factors using a univariate model, the presence of SQ Grade 3 showed a significantly higher risk ratio of 3.20 (95% CI; 1.78–5.75) (p < 0.001). Older age showed a significantly lower risk ratio of 0.97 (95% CI: 0.94–0.99) (p < 0.05). The other factor, such as a greater lymphocyte count, did not show statistical significance.

3.3. Association Between the Significant Factor and the SQ Grade

SQ grade was not a significant factor, but the presence of HT was a significant factor for incident NVF. Therefore, a binary logistic regression analysis was conducted between HT and SQ grades. The SQ grade significantly correlated with the presence of HT (p < 0.001). The probabilities of incident VF were 3.1%, 3.9%, 7.5%, and 15.5%, respectively.

3.4. In the Dataset After IPTW

3.4.1. Group Comparison After IPTW

A total of 1050 patients, 350 patients in each SQ group, were examined. The age distribution was leveled among the groups. Other significant factors using Cox regression analysis in the crude dataset, such as the presence of AAC, C-I, MADS, RA, lower T-score in the lumbar spine and femoral neck, and administration of V-D and OPD, eGFR_CysC, ALB, Hgb, and Lymph were not significantly different among the three SQ groups. However, prevalent VF, prevalent NVF, VEC, the presence of DM, hypertension, insomnia, osteoarthritis, Contracture, Parkinsonism, and polypharmacy were significantly different among the groups. After the IPTW procedure, the prevalence of incident VF was not significantly different between any pairs in the SQ groups (

Table 6. Demographic characteristics of the dataset after the IPTW procedure.

	All (N = 1050)	Grade 0 (N = 350)	Grade 1 (N = 350)	Grade 2 (N = 350)	p-Value in ANOVA	Statistical Significance in Scheffé Test
male:female (%)	12.7:87.3	16.0:84.0	10.9:89.1	11.1:88.9	0.07
age, years	81.8 (6.7)	81.2 (6.3)	81.6 (6.7)	82.4 (7.1)	0.06
Time length of follow-up, months	45.3 (28.8)	49.9 (30.7)	43.4 (26.9)	42.5 (28.2)	<0.001	Grade-0 > Grade-1 Grade-0 >> Grade-2
incident VF	69 (6.6%)	21 (6.0%)	26 (7.4%)	22 (6.3%)	0.72
incident NVF	99 (9.4%)	28 (8.0%)	35 (10.0%)	36 (10.3%)	0.53
current smoke	14 (1.3%)	8 (2.3%)	4 (1.1%)	2 (0.6%)	0.13
alcohol habitat	7 (0.7%)	2 (0.6%)	2 (0.6%)	3 (0.9%)	0.87
parent’s fracture	10 (1.1%)	3 (0.9%)	4 (1.1%)	3 (0.9%)	0.9
prevalent VF	166 (15.8%)	38 (10.9%)	56 (16.0%)	72 (20.6%)	<0.01	Grade-0 < Grade-1 Grade-0 > Grade-2
prevalent NVF	116 (11.0%)	29 (8.3%)	51 (14.6%)	36 (10.3%)	<0.05	Grade-0 < Grade-1
VEC	201 (19.1%)	45 (12.9%)	76 (21.7%)	80 (22.9%)	<0.05	Grade-0 < Grade-2
AAC	828 (78.9%)	268 (76.6%)	274 (78.3%)	286 (81.7%)	0.21
Diabetes mellitus	192 (27.8%)	121 (34.6%)	105 (30.0%)	66 (18.9%)	<0.001	Grade-0 >>> Grade-2 Grade-1 >> Grade-2
COPD	84 (8.0%)	30 (8.6%)	28 (8.0%)	26 (7.4%)	0.86
hypertension	531 (50.6%)	198 (56.6%)	171 (48.9%)	162 (46.3%)	<0.05	Grade-0 > Grade-2 Grade-1 >> Grade-2
hyperlipidemia	341 (32.5%)	119 (34.0%)	99 (28.3%)	123 (35.1%)	0.11
insomnia	281 (26.8%)	117 (33.4%)	103 (29.4%)	61 (17.4%)	<0.001	Grade-0 >> Grade-2 Grade-1 >> Grade-2
Cognitive Impairment	210 (20.0%)	70 (20.0%)	72 (20.6%)	68 (19.4%)	0.93
MADS	225 (21.4%)	76 (21.7%)	81 (23.1%)	68 (19.4%)	0.48
rheumatoid arthritis	201 (19.1%)	75 (21.4%)	71 (21.4%)	55 (15.7%)	0.13
osteoarthritis	570 (54.3%)	207 (59.1%)	203 (58.0%)	160 (45.7%)	<0.001	Grade-0 >> Grade-2 Grade-1 >> Grade-2
Contracture	112 (10.7%)	74 (21.1%)	22 (6.3%)	16 (4.4%)	<0.001	Grade-0 >>> Grade-1 Grade-0 >>> Grade-2
Disuse	57 (6.4%)	30 (8.6%)	18 (5.1%)	19 (5.4%)	0.12
Parkinsonism	41 (3.9%)	3 (0.9%)	32 (9.1%)	6 (1.7%)	<0.001	Grade-0 <<< Grade-1 Grade-1 >>> Grade-2
T-score in the LS	−2.3 (1.6)	−2.1 (1.7)	−2.2 (1.5)	−2.4 (1.6)	0.05
T-score in the FN	−2.0 (1.1)	−2.0 (1.1)	−2.0 (0.9)	−2.0 (1.2)	0.64
GCs	119 (11.3%)	35 (10.0%)	45 (12.9%)	39 (11.1%)	0.49
V-D	701 (66.8%)	228 (65.1%)	243 (69.4%)	230 (65.7%)	0.43
OPD †	264 (25.1%)	96 (27.4%)	88 (25.1%)	80 (22.8%)	0.48
polypharmacy	229 (12.4%)	98 (28.0%)	77 (22.0%)	54 (15.4%)	<0.001	Grade-0 >>> Grade-2
eGFR	52.6 (20.5)	50.6 (20.5)	51.7 (21.5)	55.6 (19.2)	0.05
ALB	4.0 (0.4)	4.0 (0.4)	4.0 (0.3)	4.0 (0.4)	0.10
P1NP	54.3 (44.4)	57.5 (55.5)	51.8 (42.2)	53.4 (30.8)	0.26
TRACP-5b	493.7 (215.1)	489.7 (234.6)	488.6 (189.7)	503.4 (219.2)	0.62
Hgb	12.0 (1.6)	11.9 (1.5)	12.1 (1.7)	11.9 (1.5)	0.21
Lymph	1499 (607)	1440 (586)	1498 (642)	1558 (585)	0.05
Mono	358 (153)	340 (133)	364 (123)	369 (194)	0.05

The values are presented as mean (SD) unless indicated otherwise. † anti-osteoporotic drugs included selective estrogen receptor modulators, bisphosphonates, denosumab, teriparatide, and romosozumab. Symbols: >, left is greater than right at a 5% significance threshold; >>, left is greater than right at a 1% significance threshold; >>>, left is greater than right at a 0.1% significance threshold; <, right is greater than left at a 5% significance threshold; <<<, right is greater than left at a 0.1% significance threshold. Abbreviations: IPTW, inverse probability of treatment weighting; VF, vertebral fracture; NVF, non-vertebral fracture; VEC, vertebral endplate collapse positivity with algorithm-based quantitative method; AAC, abdominal aortic calcification; COPD, chronic obstructive pulmonary diseases; MADS, musculoskeletal ambulation disability symptoms complex; LS, lumbar spine; FN, femoral neck; GCs, glucocorticoids; V-D, vitamin D; OPD, anti-osteoporotic drugs; eGFR, estimated glomerular filtration ratio based on cystatin C; ALB, albumin; P1NP, procollagen type 1 amino-terminal propeptide; TRACP-5b, tartrate-resistant acid phosphatase-5b; Hgb, hemoglobin; Lymph, lymphocyte count; Mono, monocyte count.

).

3.4.2. Risk Factor Analysis in the Dataset After IPTW

After IPTW, elderly age, prevalent NVF, presence of C-I and RA, T-score in the FN, administration of V-D, polypharmacy, and higher Mono had significantly higher risk ratios for incident VFs using a univariate model. Among these, the presence of C-I, administration of V-D, and higher Mono were associated with significantly higher risk ratios. Conversely, higher age, presence of RA, T-score in the FN, and polypharmacy were linked to significantly lower risk ratios in the multivariate model. The presence of HT, C-I, OA, and higher Mono resulted in a significantly higher risk ratio using a multivariate model. The SQ grade was not a significant factor regarding VF and NVF (

Table 7. Results of a Cox regression analysis for incident VF in the dataset after the IPTW.

	Univariate Mode		Multivariate Mode
	p-Value	Risk Ratio (95% CI)	β-Value	p-Value	Risk Ratio (95% CI)
female	0.98	9.9 × 107 (0.00–INF)
age, years	<0.001	1.08 (1.04–1.12)	−0.08	<0.001	0.92 (0.88–0.96)
current smoke	0.87	0.91 (0.29–2.91)
alcohol habitat	0.98	0.00 (0.00–INF)
parent’s fracture	0.88	0.00 (0.00–INF)
SQ Grade	0.69	1.06 (0.80–1.42)
prevalent VF	0.86	9.9 × 107 (0.00–INF)
prevalent NVF	<0.05	2.01 (1.13–3.78)	0.51	0.19	1.67 (0.78–3.57)
VEC	0.98	9.9 × 107 (0.00–INF)
AAC	0.26	0.73 (0.43–1.25)
Diabetes mellitus	0.12	0.48 (0.18–1.23)
COPD	0.98	0.00 (0.00–INF)
hypertension	0.45	1.30 (0.66–2.57)
hyperlipidemia	0.69	0.86 (0.42–1.77)
chronic heart failure	0.40	1.41 (0.63–3.12)
insomnia	0.18	0.56 (0.24–1.30)
Cognitive Impairment	<0.001	4.58 (2.86–7.36)	1.83	<0.001	6.32 (3.48–11.16)
MADS	0.46	1.35 (0.61–2.97)
rheumatoid arthritis	<0.01	0.22 (0.08–0.61)	−1.91	<0.001	0.15 (0.05–0.45)
osteoarthritis	0.47	1.31 (0.64–2.70)
Contracture	0.95	1.03 (0.43–2.50)
Disuse	0.83	1.14 (0.35 –3.75)
Parkinsonism	0.98	0.00 (0.00–INF)
T-score in the LS	0.08	0.87 (0.74–1.02)
T-score in the FN	<0.001	0.59 (0.46–0.75)	−0.41	<0.05	0.67 (0.49–0.92)
GCs	0.06	0.33 (0.11–1.06)
V-D	<0.05	1.90 (1.06–3.41)	1.01	<0.01	2.73 (1.34–5.59)
OPD †	0.92	1.04 (0.50–2.18)
polypharmacy	<0.001	0.09 (0.02– 0.36)	−2.90	<0.001	0.06 (0.01–0.23)
eGFR	0.24	0.99 (0.97–1.01)
ALB	0.95	0.97 (0.33–2.84)
Calcium	0.14	1.23 (0.93–1.63)
P1NP	0.59	1.00 (0.99–1.01)
TRACP-5b	0.69	1.00 (1.00–1.00)
Hgb	0.63	0.94(0.75–1.19)
Lymph	0.17	1.00 (1.00–1.00)
Mono	<0.001	1.00 (1.00–1.01)	0.01	<0.001	1.01 (1.00–1.01)

† anti-osteoporotic drugs included selective estrogen receptor modulators, bisphosphonates, denosumab, teriparatide, and romosozumab. Abbreviations: IPTW, inverted propensity of treatment weighting; SQ, semiquantitative classification of vertebral fracture; VF, vertebral fracture; NVF, non-vertebral fracture; VEC, vertebral endplate collapse positivity with algorithm-based quantitative method; AAC, abdominal aortic calcification; COPD, chronic obstructive pulmonary diseases; MADs, musculoskeletal ambulation disability symptoms complex; LS, lumbar spine; FN, femoral neck; GCs, glucocorticoids; V-D, vitamin D; OPD, anti-osteoporotic drugs; eGFR, estimated glomerular filtration ratio based on cystatin C; ALB, albumin; P1NP, procollagen type 1 amino-terminal propeptide; TRACP-5b, tartrate-resistant acid phosphatase-5b; Hgb, hemoglobin; Lymph, lymphocyte count; Mono, monocyte count.

).

4. Discussion

The presence or absence of a prevalent fracture is one of the most decisive risk factors for subsequent osteoporotic fractures [14]. Reports indicate that the risk of subsequent VF increases by 4.4 times when an incident VF occurs compared to when there is no VF [28]. This is the most crucial factor to consider when evaluating osteoporosis treatment; thus, a prevalent fracture episode is incorporated into FRAX [29,30]. The SQ method is a screening approach for VDs and is widely recognized as an easy screening tool for VD. Additionally, it is frequently used to predict incident VFs. It is regarded as an excellent screening method because it allows for assessment without measuring bone mineral density. However, this classification has often been debated. There is a consensus that SQ Grade 3, which is severe, is a clear risk factor [31]. Still, opinions differ regarding whether SQ Grade 1 and Grade 2, which are mild and moderate, respectively, constitute risk factors. Nevertheless, this consensus is upheld in real-world clinical settings.

The number of incident VFs was lower than that of incident NVFs, which differs from other established reports [31,32,33,34]. However, this study was conducted in a clinical-based population setting, whereas many previous reports have utilized population-based studies. The mean age in our study was relatively high, which is one reason for the smaller number of VFs compared to NVFs, as the prevalence of hip fractures increases with age. Moreover, cases with prevalent VFs were higher than those with prevalent NVFs. Therefore, it provides a reasonable background. Furthermore, our study’s primary endpoint focused solely on symptomatic clinical VFs, which should have limited the number of cases with incident VFs. A global trend of fragility fractures also indicated an overwhelming number of hip fractures relative to VFs [35]. Additionally, the incidence of VFs varies significantly among countries, races, and years in population-based studies. This discrepancy may be attributed to genetic and epigenetic differences [36].

Furthermore, the hypothesized effect size of this study is 137. That is further greater than the actual number of VFs. The study results should be considered with this limitation in mind. Nevertheless, the sample size was sufficient to achieve a significance level of less than 5%, and the results are notable.

In addition to the SQ grade, we considered previously reported potential items for extracting predicted risk factors [37]. We are also regarded as markers of bone metabolism. VEC positivity on ABQ was deemed the most significant factor for vertebral body fractures. In the crude dataset used in the study, all 59 patients who presented with incident VF exhibited VEC positivity using ABQ. The results from the Kruskal–Wallis and K-M analyses indicated that the presence of VEC demonstrated statistically significant relevance regarding SQ grade and the likelihood of incident VFs. The VEC assessed with ABQ is a visual judgment of bone contour discontinuity; thus, positive findings should indicate prevalent vertebral body fractures; however, these findings cannot distinguish whether the fracture is recent, and this judgment does not equate to a clinical fracture, even though clinical fractures and VEC findings overlap significantly. Therefore, VEC positivity with ABQ is a strong predictor of incident vertebral fractures, but it is not definitive.

The probability of incident VFs in Grade 3 was significantly higher than in the other three groups using the ANOVA Scheffé test in the crude dataset; however, the statistical significance diminished with the K-M because there was no significant difference in the likelihood of incident VFs between the two groups of Grade 0 to Grade 2 groups using the ANOVA Scheffé test. We conducted a Cox regression analysis to identify important factors without confounding variables. When a multivariate model was examined using Cox regression analysis, the SQ grade was the only factor with a significantly higher risk ratio. Although the Grade 3 group showed a significantly higher risk ratio using a K-M when the crude dataset is divided into whether Grade 3 or not, numerous confounding factors are in the crude dataset. There are substantial differences in many background factors between the groups, posing a risk of acting as confounding factors. Therefore, we needed to limit the groups using PSM procedures, level the recognized confounding factors, and reduce the influence of these factors.

After the PSM procedure, VEC could not be leveled because 100% of the patients in the G-3 group had a positive VEC. However, the other factors were leveled. In the dataset after PSM, the SQ grade groups showed no significant differences in the probability of incident clinical VFs and NVFs. However, there were substantial differences in background factors, such as age, blood lymphocyte count, and PNI, although these variables lost their statistical significance compared to the multivariate model. This can be ignored, as the crude dataset showed no statistically significantly higher risk ratios for these factors.

As a post hoc analysis, we conducted a Cox regression analysis, dividing the SQ grading into Grade 3 and other groups in both the crude dataset and the dataset after PSM. Results showed that Grade 3 is a significantly higher risk factor for incident VF development, even in the dataset after PSM. These results suggest that a Grade 3 presentation in the SQ is a critical risk factor for predicting incident VF development. The risk ratio magnitudes in the crude dataset and the dataset after PSM are almost similar. The result suggests a consistency of the risk strength in Grade 3.

Although SQ Grade 3 was a significant risk factor in the crude dataset, various confounding factors were present, prompting us to use PSM to address them. Consequently, we found that the SQ grade across all groups posed no significant risk. Even though Grade 3 is a significant risk factor, one drawback of PSM is that it can exclude a considerable amount of essential data. To mitigate this limitation, we employed IPTW to re-evaluate the significance of the SQ grade. Grade 3 was not included in IPTW because it was already known to be significant. We applied the IPTW technique to adjust for significant risk factors, including the SQ grade. This involved adjusting the number of extracted subjects—essentially modifying the weight of each element—without excluding any individuals. The IPTW technique eliminates significant differences in target elements and provides a more objective method for forming groups to compare differences. IPTW resolved substantial discrepancies. In this analysis, prior to initiating IPTW procedures, we compared patients in Grade 3 and those who were too young with the groups in Grade 0 and Grade 2. The mean age in Grade 0 was significantly younger than in Grades 1 and 2, while Grade 3 was notably older, complicating the correction. Thus, the balance of age distribution among the three groups could confound the results. Age is a critical factor for fragility fractures, so differences among the groups should be minimized. Results indicated that factors with significant risk ratios remained prominent even after IPTW; however, their statistical significance was reduced or inversely correlated. The SQ grade diminished its statistical significance for VF (Table 7).

The above results indicate that the SQ grade classification has weak significance as a predictor of incident vertebral body fractures, even including Grade 3. The SQ grade, determined solely by radiographs, requires accuracy. If higher precision is necessary, imaging methods such as high-resolution CT, which can observe the destruction of the posterior wall of the vertebral body and damage to the endplate or other morphological indices in the lumbar spine, are desirable [19,21,38]. However, the SQ grade is suitable for screening because it can be easily determined and is considered an excellent method for the initial evaluation before osteoporosis treatment.

Limitations of the Study

Limitations include that it was a single-center study, the follow-up period was short (averaging less than 4 years), it was a retrospective study, the study protocol did not account for other confounding factors, such as physical activity and social factors, and the survey was statistically weak even though bias adjustment was performed. Another critical point is that this study was not population-based but rather clinical-based, so the sample size needed to be larger for the analysis of these various statistics. However, this can serve as reference information when screening for osteoporosis.

5. Conclusions

Clinical research has been conducted to evaluate the validity of the SQ method as a risk factor for subsequent VF. The results indicated that Grade 3 exhibited a significantly higher risk ratio for an incident VF. However, the risk ratio was influenced by confounding factors. After adjusting for these factors, the risk ratio was recalculated. The SQ grade classification is an effective method for screening initial osteoporotic treatments.