Improved estimates of strength and stiffness in pathologic vertebrae with bone metastases using CT-derived bone density compared with radiographic bone lesion quality classification

Abstract

OBJECTIVE

The aim of this study was to compare the ability of 1) CT-derived bone lesion quality (classification of vertebral bone metastases [BM]) and 2) computed CT-measured volumetric bone mineral density (vBMD) for evaluating the strength and stiffness of cadaver vertebrae from donors with metastatic spinal disease.

METHODS

Forty-five thoracic and lumbar vertebrae were obtained from cadaver spines of 11 donors with breast, esophageal, kidney, lung, or prostate cancer. Each vertebra was imaged using microCT (21.4 μm), vBMD, and bone volume to total volume were computed, and compressive strength and stiffness experimentally measured. The microCT images were reconstructed at 1-mm voxel size to simulate axial and sagittal clinical CT images. Five expert clinicians blindly classified the images according to bone lesion quality (osteolytic, osteoblastic, mixed, or healthy). Fleiss’ kappa test was used to test agreement among 5 clinical raters for classifying bone lesion quality. Kruskal-Wallis ANOVA was used to test the difference in vertebral strength and stiffness based on bone lesion quality. Multivariable regression analysis was used to test the independent contribution of bone lesion quality, computed vBMD, age, gender, and race for predicting vertebral strength and stiffness.

RESULTS

A low interrater agreement was found for bone lesion quality (κ = 0.19). Although the osteoblastic vertebrae showed significantly higher strength than osteolytic vertebrae (p = 0.0148), the multivariable analysis showed that bone lesion quality explained 19% of the variability in vertebral strength and 13% in vertebral stiffness. The computed vBMD explained 75% of vertebral strength (p < 0.0001) and 48% of stiffness (p < 0.0001) variability. The type of BM affected vBMD-based estimates of vertebral strength, explaining 75% of strength variability in osteoblastic vertebrae (R2 = 0.75, p < 0.0001) but only 41% in vertebrae with mixed bone metastasis (R2 = 0.41, p = 0.0168), and 39% in osteolytic vertebrae (R2 = 0.39, p = 0.0381). For vertebral stiffness, vBMD was only associated with that of osteoblastic vertebrae (R2 = 0.44, p = 0.0024). Age and race inconsistently affected the model’s strength and stiffness predictions.

CONCLUSIONS

Pathologic vertebral fracture occurs when the metastatic lesion degrades vertebral strength, rendering it unable to carry daily loads. This study demonstrated the limitation of qualitative clinical classification of bone lesion quality for predicting pathologic vertebral strength and stiffness. Computed CT-derived vBMD more reliably estimated vertebral strength and stiffness. Replacing the qualitative clinical classification with computed vBMD estimates may improve the prediction of vertebral fracture risk.

Introduction

Advanced cancer stages, 30%–50% of patients with cancer present with bone metastases (BM) in the spine. ¹ Although MRI and CT are routinely used to evaluate the impact of spinal BM on the anatomical integrity of the vertebral column, ^{2; 3} quantifying the accompanying risk for clinically significant spinal failure remains challenging ^{4; 5}. Pathologic vertebral fractures (PVFs), which occur when BM burden has affected the vertebra’s structural and anatomical integrity rendering it unable to withstand daily loads, afflict 15%–20% of patients with spinal BM ⁶. The risk of PVF is a critical determinant in managing cancer patients with metastatic spine disease ^{7; 8}. BM type, vertebral body destruction, and/ or pedicle destruction, age, pain, anatomical location, activity levels, radiographic alignment, previous vertebral collapses, and radiotherapy were identified as PVF risk factors ⁹. The Spine Instability Neoplastic Score (SINS) ⁵, which categorizes the degree of spinal instability based on some of these factors, is widely used to indicate the need for surgical stabilization in this patient population. However, the prognostic utility of the SINS for predicting ^{10; 11} remains imprecise ¹². Given that the management of pathologic spinal column fractures is a critical aspect of the care of these patients, it is imperative to develop a reliable prediction of PVF risk to allow appropriate treatment.

The vertebral bone invasion by metastatic tumors disrupts the bone’s cellular homeostasis ¹³, affecting the microarchitecture and composition of the bone tissue ^{14; 15}. On CT, vertebral BM are classified based on their radiological appearance ¹⁶. Osteolytic (bone-destroying) BM cause rarefication of the trabecular network and lower apparent bone density and may include lytic foci of varying sizes. Osteoblastic (bone-forming) BM, generally accepted as less aggressive with slow tumor growth and prolonged patient survival, typic ally manifest a sclerotic phenotype involving extensive deposition of new bone with markedly higher apparent bone density. Although forming the “bone lesion quality” classification in SINS, these qualitative, subjective classifications have been shown to suffer from poor reproducibility between clinical interobservers ^{5; 17; 18}. Notably, although clinicians may routinely employ bone lesion quality as an aid for qualitative inference of vertebral strength,⁸ this inference has not been confirmed in the laboratory setting. Thus, with current clinical prognostic protocols for predicting PVF remaining imprecise, managing the spine’s mechanical instability in patients at risk of posttreatment morbidity or existing fractures remains challenging ^{9; 19}.

This study investigated the performance of the qualitative clinical classification of bone lesion quality (lytic, mixed, blastic, and healthy) in predicting the strength and stiffness of human cadaver vertebrae from donors with breast, kidney, lung, or prostate cancer. We compared this performance to the prediction of vertebral strength and stiffness by using computed CT-derived volumetric bone mineral density (vBMD). We hypothesized that the computed vBMD outperforms the qualitative clinical classification for predicting the measured strength and structural stiffness of the pathologic vertebrae.

METHODS

Sample Selection

Eleven cadaver spines (3 female and 8 male, age range 49–71 years, mean 54 years) were obtained (Anatomy Gifts Registry) as part of a previous study ²⁰ from donors with a history of breast, lung, prostate, kidney, or esophageal cancer. Forty-five vertebrae confirmed to present lytic, blastic, or mixed metastatic bone were selected based on the radiological review. Table 1 details each specimen’s demographic properties.

Table 1.

Details of each study specimen’s demographic, image-derived, and experimentally measured parameters.

ID	VL	Ca	BM	Age (Y)	R	H (In)	W (Lb)	Sex	vBMD [mgHA/cm3]	BV/TV (%)	Strength (kN)	Stiffness (kN/mm)
1	T11	B	Ost	59	C	67	175	F	133.69	0.13	4.91	21.68
1	L3	B	Ost	59	C	67	175	F	150.57	0.15	7.57	34.49
1	L5	B	BL	59	C	67	175	F	155.98	0.16	7.46	17.72
1	T10	B	M	59	C	67	175	F	116.33	0.12	4.43	14.74
1	L1	B	M	59	C	67	175	F	134.62	0.14	5.79	28.09
1	L2	B	M	59	C	67	175	F	138.48	0.14	6.09	26.19
2	T11	K	Ost	71	C	67	120	M	91.3	0.09	1.89	13.62
3	L3	L	Ost	49	B	70	150	M	153.32	0.16	5.94	18.26
3	T10	L	Ost	49	B	70	150	M	131.11	0.13	4.16	26.05
3	T9	L	Ost	49	B	70	150	M	124.89	0.12	3.42	14.36
4	L1	K	BL	56	C	71	175	M	157.74	0.15	4.98	19.73
4	T10	K	BL	56	C	71	175	M	126.94	0.13	4.64	23.90
4	T6	K	Ost	56	C	71	175	M	112.85	0.13	4.82	18.86
5	L1	L	BL	60	B	72	180	M	186.05	0.18	5.8	20.50
5	T12	L	BL	60	B	72	180	M	202.64	0.19	5.9	25.25
5	T9	L	BL	60	B	72	180	M	197.64	0.19	5.24	13.36
6	L2	L	BL	53	C	69	189	M	181.83	0.19	8.21	28.11
6	T6	L	N	53	C	69	189	M	172.74	0.19	6.27	32.95
6	T10	L	BL	53	C	69	189	M	186.34	0.2	7.49	27.47
6	L4	L	BL	53	C	69	189	M	194.83	0.21	9.61	18.76
6	T11	L	M	53	C	69	189	M	198.81	0.21	7.69	13.95
6	L5	L	N	53	C	69	189	M	241.87	0.27	12.99	51.30
6	L3	L	Ost	53	C	69	189	M	186.86	0.2	9.08	22.47
7	L5	E	M	52	C	67	280	M	214.28	0.24	13.78	29.56
7	T5	E	Ost	52	C	67	280	M	203.2	0.23	5.84	16.14
7	L4	E	BL	52	C	67	280	M	166.59	0.18	12.01	49.64
8	L1	P	BL	55	C	72	150	M	339.18	0.36	15	37.01
8	T9	P	BL	55	C	72	150	M	372.2	0.41	15	17.69
8	T10	P	BL	55	C	72	150	M	387.83	0.44	15	46.58
8	L4	P	BL	55	C	72	150	M	423.93	0.47	15	74.13
9	L1	B	BL	60	C	65	150	F	377.79	0.41	15	52.85
9	L5	B	BL	60	C	65	150	F	333.91	0.38	15	50.52
9	L4	B	M	60	C	65	150	F	298.33	0.33	11.5	33.27
9	T9	B	M	60	C	65	150	F	259.67	0.25	5.73	14.41
9	T12	B	BL	60	C	65	150	F	326.96	0.34	11.56	38.88
9	T8	B	BL	60	C	65	150	F	240.47	0.21	5.32	11.70
10	T11	P	BL	71	C	74	150	M	250.01	0.23	8.48	25.55
10	T12	P	BL	71	C	74	150	M	140.97	0.14	7.08	28.81
10	T7	P	BL	71	C	74	150	M	146.66	0.14	4.33	15.79
10	T10	P	BL	71	C	74	150	M	166.87	0.16	5.59	23.18
11	L3	B	BL	60	C	64	90	F	308.17	0.37	10.85	49.61
11	L1	B	M	60	C	64	90	F	324.68	0.38	10.83	41.01
11	T12	B	BL	60	C	64	90	F	325.25	0.39	12.27	55.44
11	L4	B	BL	60	C	64	90	F	325.22	0.39	15	58.49
11	T11	B	BL	60	C	64	90	F	318.3	0.39	11.07	35.47

BM (bone lesion–quality grouping based on the clinician’s classification): BL = osteoblastic; M = mixed; N = normal; Ost = osteolytic. Ht = height; Wt = weight.

Primary cancer: B = breast; E = esophageal; K = kidney; L = lung; P = prostate.

Race: B = Black; C = Caucasian.

Stiffness: vertebral structural stiffness (kN/mm).

Strength: vertebral strength (kN).

Sample Preparation and MicroCT Imaging

As part of a previous study, ²⁰ and per our laboratory protocol, ²¹ a diamond blade saw (Exakt 300, EXAKT Technologies) was used to section the posterior elements proximal to the vertebral body along its coronal plane and both endplates were removed under constant water irrigation. The resulting plane-parallel specimen was imaged in a microCT scanner (μCT 100, Scanco Medical) at 24.5-μm isotropic voxel size (see Appendix).

CT-Based Bone Lesion Quality Classification of BM

For each vertebra, the microCT data were reformatted to simulate clinical axial and midsagittal CT images (voxel size = 1 mm). For this study, the CT images were blindly reviewed by two orthopedic spine surgeons, a neurosurgery spine surgeon, and two radiologists, all of whom routinely treat or diagnose patients with metastatic spine disease. Each reviewer was asked to classify the images for the type of vertebral BM as osteolytic, osteoblastic, mixed, or healthy (i.e., no radiological evidence of metastasis). Each vertebra was assigned a bone lesion quality classification based on the majority score among the 5 reviewers. Two segments lacked a majority vote (i.e., no classification gained more than 2 votes) and were not included in the analysis.

vBMD

Per our laboratory protocol, ²¹ we applied standard hydroxyapatite (HA) phantom calibration data (Scanco Medical) to map the microCT image data (Hounsfield units) to BMD values. An adaptive threshold algorithm ^{22; 23} was used to segment the vertebrae with negative vBMD values, caused by air bubbles within the trabecular space, set to zero. The segmented images were used to compute overall bone volume (BV; i.e., cortical + trabecular bone tissue). This value, divided by the segment’s overall total volume (TV), provided the segment’s BV/TV value expressed as a percentage. Volumetric density (expressed as g/cm3) was computed by dividing the tissue bone mineral content, computed from the microCT HA phantom data and expressed in grams, by the overall BV excluding the vertebral cortex, expressed in cm3.

Mechanical Testing

As part of a previous study, ²⁰ the imaged vertebral segments underwent mechanical testing to measure their strength and stiffness. The test commenced by applying a tare load of 25 N, followed by a compressive displacement at a 5 mm/min rate. Failure was defined as the maximum compressive force recorded on the displacement-force curve. For this study, a linear regression model was fitted to the linear portion of the test load displacement, and the vertebral stiffness was computed from the coefficient of the linear regression model (i.e., slope of the curve).

Data Analysis

Statistical analysis was performed in JMP Pro (version 14.3, SAS Institute, Inc.). Descriptive statistics were used to summarize the specimens’ demographic, bone, and mechanical properties. Fleiss’ kappa ²⁴ was used to test the agreement among clinical observers in classifying bone lesion quality. Univariate analysis was used to test for normality (Shapiro-Wilk test) of the computed vBMD, BV/TV, and vertebral strength and stiffness. Based on the normality test results, we applied the nonparametric Kruskal-Wallis 1-way ANOVA to test the effect of bone lesion quality classification on the difference in vertebral vBMD, BV/TV, vertebral strength, and stiffness. Post hoc comparisons were performed using the Steel-Dwass non-parametric test.

The sampling of multiple vertebrae per spine can introduce clustering (nonindependence) of the data. We fitted linear mixed-effects models (LMMs) under different assumptions about the correlation structure among segments from the same spines to test this effect. According to the Akaike information criterion with small sample correction (AICc), the independence structure best fits the data ²⁵. Based on this finding, we applied multivariable linear regression to test the independent contribution of bone lesion quality, computed vBMD, age, gender, and race to the model’s prediction of vertebral strength and stiffness. We repeated these analyses to test whether the model’s prediction (strength or stiffness) was affected by BM type. Statistical significance was set at the 5% level.

Results

Table 1 details the specimens’ CT-derived vBMD values and their strength and stiffness values measured via mechanical testing.

Radiological Classification of Bone Lesion Quality

Based on the clinician’s review, 26 vertebrae were qualitatively classified as osteoblastic, 9 as osteolytic, and 8 as mixed BM. We found a poor interobserver agreement for classification of bone lesion quality (κ = 0.19).

vBMD and Fraction (BV/TV)

The univariate analysis revealed vBMD (range 91.30–423.93 mgHA/cm³) and BV/TV (range 0.09%–0.47%) values to be nonnormally distributed (p = 0.0026 and p = 0.0002, respectively; Shapiro-Wilk test). Bone lesion quality affected these distributions. Both vBMD and BV/TV were nonnormally distributed in osteoblastic vertebrae (p= 0.0179 and p= 0.0017, respectively; Shapiro-Wilk test). In contrast, both parameters were normally distributed in vertebrae with mixed and osteolytic BM.

Based on the finding of nonnormality, application of the nonparametric Kruskal-Wallis ANOVA analysis revealed significant differences between the bone lesion quality group for vBMD (p = 0.0090) and BV/TV (p = 0.0176; Fig. 1). Post hoc analysis found that osteoblastic vertebrae showed higher vBMD (median 221.6 mgHA/ cm³, interquartile range [IQR] 166.9–328.7 mgHA/cm³) and BV/TV (median 21.0%, IQR 17.5%–39.0%) than osteolytic vertebrae (median 133.7 mgHA/cm³, IQR 118.9–179.1 mgHA/cm³, p = 0.0056, and median 0.13%, IQR 0.13%–0.18%, p = 0.0066, respectively; Fig. 1). No significant differences were observed between vertebrae with mixed BM (median 206.5 mgHA/cm³, IQR 135.6–288.7 mgHA/cm³, and median 0.23%, IQR 0.14%–0.31%, respectively) and either the osteolytic or osteoblastic vertebrae (Fig. 1).

FIG. 1.

Box-and-whisker plot of the computed vBMD (A) and bone fraction (B; BV/TV) of the vertebral specimens grouped by the clinician classification of bone lesion quality. Vertebrae classified as osteoblastic and mixed show marked variation in vBMD (35.1% and 33.4%, respectively) and BV/TV (33.3% and 37.7%, respectively) compared to vertebrae classified as osteolytic (14.7% and 15.4%, respectively).

Mechanical Testing

Seven of the vertebrae exceeded the maximum safe force for the materials testing system built-in load (15 kN, model no. 662.20D-04, Instron20) and were assigned a strength of 15 kN. The test showed vertebrae having median values of 7.49 kN (IQR 5.46–11.79 kN) and 26.05 kN/mm (IQR 17.99–37.94 kN/mm) for vertebral strength and stiffness, respectively.

Bone Metastasis Affects the Relationships Between Strength and Stiffness

Tobit regression (performed to account for the truncated strength data, i.e., vertebrae whose strength is > 15 kN) showed vertebral stiffness was significantly associated with vertebral strength (strength [N] = 1.435 + 0.256 × stiffness (kN/mm), p < 0.001, R2 = 0.61). However, as illustrated in Fig. 2, BM type affected this association. In osteoblastic vertebrae, vertebral stiffness was significantly associated with higher vertebral strength (p < 0.0001, R2 = 0.58; Fig. 3A). However, no such statistically significant associations were observed in vertebrae with mixed (Fig. 3B) or osteolytic (Fig. 3C) BM.

FIG. 2.

Experimental load-displacement example curves for vertebral bodies containing osteoblastic, mixed, and osteolytic BM. The corresponding CT axial image for each of the tested vertebrae is presented. Compared to the osteoblastic and mixed BM vertebrae, the osteolytic vertebrae show reduced strength (vertebral strength, computed as the maximal force value recorded for the test) and stiffness (vertebral stiffness, computed from the coefficient of the linear regression model fitted to the linear portion of the load-displacement curve). Although the osteoblastic vertebrae demonstrated markedly higher vertebral strength than the mixed BM vertebrae, there is little difference in their respective vertebral stiffness values.

FIG. 3.

Regression analysis demonstrates the BM effect on the association of vertebral compressive strength with stiffness for each bone lesion category, with the association statistically significant for osteoblastic vertebrae only (A). Regression models and 95% confidence curves are presented for each bone metastasis type.

Utility of Qualitative Bone Lesion Quality in Evaluating Strength and Stiffness of Pathologic Vertebrae

The comparison of the vertebrae based on classification of the bone lesion quality showed a statistically significant difference in vertebral strength (p = 0.0148, Fig. 4A). Post hoc analysis showed this difference was due to the higher strength of the osteoblastic vertebrae (median 9.0 N, IQR 5.75–15.00 N) compared to the osteolytic vertebrae (median 4.9 N, IQR 3.8–6.8 N). However, no statistically significant difference in vertebral strength was observed between vertebrae with mixed BM (median 6.9 N, IQR 5.8–11.3 N) and either the osteolytic or osteoblastic vertebrae (Fig. 4A). The comparison of the vertebrae based on bone lesion quality classification found no statistically significant differences in vertebral stiffness (Fig. 4B). Regression analysis showed that bone lesion quality explained 19% of the vertebral strength variability (p = 0.0063), with the blastic bone classification a significant independent correlate of strength (p = 0.0016). Bone lesion quality was not a significant correlate of vertebral stiffness (R2 = 0.13, p = 0.0553).

FIG. 4.

Box-and-whisker plot of the vertebral strength (A) and stiffness (B) grouped by the bone lesion quality classification. The ability of bone lesion quality to evaluate vertebral strength was limited to only the difference between osteoblastic and osteolytic vertebrae.

Computed vBMD and Estimates of Strength and Stiffness of the Pathologic Vertebrae

As previously described in the Methods section, LMM analysis revealed the sampling of multiple vertebrae to have no significant effect on the association between vBMD and vertebral strength (p = 0.2913) or stiffness (p= 0.0815). The multivariable model incorporating vBMD, age, race, and gender (as independent variables) explained 75% of the variation in vertebral strength (p < 0.0001); vBMD, age, and race (Black) were significant independent correlates of strength (Table 2). The model explained 48% of the variation in vertebral stiffness (p < 0.0001) using vBMD, a significant independent correlate of stiffness (Table 2). Univariate regression showed vBMD explained 70% of the variation in vertebral strength (p < 0.0001, Fig. 5A) and 46% of the variation in vertebral stiffness (p < 0.0001, Fig. 5B).

Table 2.

Summary of the significant independent correlates in the multivariable model for predicting vertebral strength and stiffness.

Independent Correlate	Estimate	SE	95% CI	Test	p-Value
Strength (kN)
vBMD (mgHA/cm3)	0.041	0.004	0.033, 0.048	108.398#	<.0001
Age (years)	−0.142	0.053	−0.247, −0.039	7.151#	0.0175
Race (Caucasian, Black)	−2.340 (B)	0.906	−4.116, −0.564	6.670#	0.0298
Stiffness (kN/mm)
vBMD (mgHA/cm3)	0.113	0.018	0.077, 0.149	37.966$	<.0001

SE = estimated precision of the coefficients (standard error).

Computed vBMD, age (years), and race (Caucasian, Black) are explanatory variables in the model.

^*Wald chi-square for the tobit regression.

^†t-test for the ordinary least-squares regression multivariable models.

FIG. 5.

vBMD is a predictor of vertebral strength (A) and vertebral stiffness (B). Within individual BM types (C), the strength of prediction was maintained for osteoblastic vertebrae, while the prediction of strength was lower for vertebrae with mixed and osteolytic BM. For vertebral stiffness (D), although the prediction strength was maintained for osteoblastic and mixed BM vertebrae, bone density was weakly associated with osteolytic vertebral stiffness, suggesting that other mechanisms other than bone tissue density affect the structural response of osteolytic vertebrae.

BM Type and vBMD Association With Vertebral Strength and Stiffness

Application of the linear regression analysis revealed BM type affected the vBMD association with both vertebral strength and stiffness (Table 3). The tobit multivariable model, which included age, race, and gender as independent variables, explained 80% of the variation in osteoblastic vertebral strength (p < 0.0001) and 50% of the variation in osteoblastic vertebral stiffness (p = 0.0012, Table 3). vBMD and race were significant independent correlates of strength, and vBMD was also a significant independent correlate for vertebral stiffness (Table 3). Univariable regression found vBMD explained 79% of the variation in osteoblastic vertebral strength (p < 0.0001, Fig. 5C), and 50% of the variation in osteoblastic vertebral stiffness (p = 0.0024, Fig. 5D).

Table 3.

Summary of the significant independent correlations in the multivariable model for predicting vertebral strength and stiffness for the different bone lesion quality classification types.

Variable	Estimate	SE	95% CI	Test	p-Value
Osteoblastic vertebrae
Strength (kN)
Race (Black)	4.516	4.516	−5.389, −0.543	5.758*	0.0364
vBMD (mgHA/cm3)	−2.966	−2.966	0.0349, 0.065	42.983*	<0.0001
Stiffness (kN/mm)
vBMD (mgHA/cm3)	0.109	0.032	0.047, 0.172	11.847†	0.0024
Mixed vertebrae
Strength (kN)
Age (Years)	−5.766	1.543	−10.049, −1.483	−3.74†	0.0202
Gender (F)	18.347	5.401	3.353, 33.341	3.4†	0.0373
vBMD (mgHA/cm3)	0.060	0.0105	0.031, 0.089	5.72†	0.0046
Osteolytic vertebrae
Strength Fv (kN)
vBMD (mgHA/cm3)	0.047	0.015	0.012, 0.082	3.18†	0.0381

Computed vBMD, age (years), gender (female), and race (Black) are explanatory variables in the model.

^*Wald chi-square for the tobit regression.

^†t-test for the ordinary least-squares regression multivariable models.

In the mixed vertebrae group, the multivariable model (including age, race, and gender as independent variables) explained 48% of the variation in vertebral strength (p= 0.0118), with vBMD, gender, and age as significant independent correlates of strength (Table 3). Although the model explained 39% of the variation in vertebral stiffness (p = 0.1628), no parameter was a significant independent correlate of stiffness. Univariable regression found vBMD explained 41% of the variation in mixed metastases vertebral strength (p = 0.0168, Fig. 5C). No statistically significant association was found for mixed metastases vertebral stiffness (Fig. 5D).

The multivariable model explained 47% of the variation in osteolytic vertebral strength (p = 0.0154), with vBMD a significant independent correlate of osteolytic vertebral strength (Table 3). Although the model explained 49% of the variation in osteolytic vertebral stiffness (p = 0.5133), no parameter was a significant independent correlate of osteolytic vertebral stiffness. Univariable regression found vBMD explained 39% of the variation in osteolytic vertebral strength (p = 0.0381, Fig. 5C). No significant association was found for osteolytic vertebral stiffness (Fig. 5D).

Performance of vBMD Compared With Bone Lesion Quality in Estimating Strength and Stiffness of the Pathologic Vertebrae

We employed the AICc to compare the regression models based on bone lesion quality and the computed vBMD for estimating vertebral strength and stiffness (the preferred model being the one with the minimum AICc value). The computed vBMD-based models showed higher performance than the bone lesion quality–based model for estimating strength (AICc = 174.48 vs 227.74) and stiffness (AICc = 330.76 vs 355.16).

DISCUSSION

The current study compared an approach based on the qualitative radiological bone lesion quality classification with an approach based on computed CT-derived vBMD for evaluating the pathologic strength and stiffness of human vertebral bodies. In agreement with the clinical literature, we found a poor agreement among the clinical raters for classifying the BM. Our primary finding is that bone lesion quality performs poorly for estimating vertebral compressive strength or stiffness, with the estimates being particularly low for vertebrae containing mixed and osteolytic metastases. Our study demonstrated that the approach based on computed CT-derived vBMD offers significantly improved estimation of the strength and stiffness of pathologic vertebrae. However, vBMD prediction for vertebral strength and structural stiffness was less reliable for vertebrae with mixed or osteolytic BM, suggesting a direction for additional research to improve the prediction of mechanical performance in pathologic vertebrae.

Given the current absence of a quantitative fracture risk assessment, ⁸ clinicians currently employ qualitative image-based classification of metastatic bone lesions to estimate vertebral strength. Unfortunately, our study shows the subjective clinical classification suffers from a poor interobserver agreement (κ = 0.19). Consistent with our findings, a study by the Spine Oncology Study Group, which developed the SINS protocol, reported an interobserver kappa of 0.244 for classifying BM (bone lesion quality rubric)⁵. Similarly, in a systematic review and meta-analysis of this topic, Pennington et al. ¹⁷ reported an interobserver kappa of 0.29 for bone lesion quality. In contrast, Fisher et al., ¹⁸ using the SINS protocol, reported an interobserver kappa of 0.55 for classifying BM in a study among radiation oncologists. Although our findings appear to agree with those in these reports, direct comparison is problematic as these studies often do not report the distribution of BM among lytic, mixed, and blastic categories and that some used a binary (lytic or not) classification.

The limitation of applying subjective clinical classification resulting in mischaracterizing vertebral strength is shown in Figs. 3 and 4. Although, as a group, vertebrae classified as osteoblastic showed significantly higher vertebral strength than those classified as osteolytic, 53% of the osteoblastic vertebrae and 63% of the mixed BM vertebrae had strength values within the range measured for osteolytic vertebrae. This qualitative classification is even less reliable for structural stiffness, with up to 69% of osteoblastic and 87.5% of mixed BM vertebrae having structural stiffness values within the range measured for osteolytic vertebrae. Notably, based on bone lesion quality, we could not differentiate vertebrae classified as mixed BM from either the osteoblastic or osteolytic vertebrae based on vertebral strength values. This situation mirrors the increased uncertainty of clinicians when required to evaluate the effect of metastases on the degradation of vertebral strength (“instability”) in vertebrae with mixed metastases. Although osteoblastic vertebrae are likely to be more robust than mixed or osteolytic vertebrae, our findings highlight the importance of osteolytic regions within the osteoblastic bone network for determining the failure of osteoblastic vertebrae (see below). These findings highlight the limitations of applying the subjective “bone lesion quality” ²⁶ as a surrogate measure for evaluating vertebral strength or stiffness, both of which form an essential determinant of mechanical “instability” in pathological vertebrae ⁹.

Previous studies reported a strong association between microCT-derived vBMD (computed as vBMD × BV/TV, R2 = 75%) ²⁷, or ash density (R²=89%) ²⁸ values with the strength of cancellous bone samples obtained from pathologic bone biopsies from the spinal or proximal femur. Consistent with these findings, the computed CT-derived vBMD was strongly predictive of vertebral strength (R2= 75%). Significantly, based on the measure of the goodness of fit (AICc value), the quantitative approach (vBMD) provided a superior approach for estimating the vertebral strength and stiffness than the qualitative (bone lesion quality) approach.

However, our finding did reveal that the vBMD-based mode offered lower performance in estimating the strength and stiffness of vertebrae with mixed or osteolytic BM. This finding appears to contrast with the finding in the Nazarian et al. study, ²⁷ which reported that the association between pathologic bone density and compressive strength follows a constant relationship regardless of underlying BM. The structural interaction between the cancellous bone and vertebral cortex is critical in determining the structural response and the onset of the failure of the complete vertebral body ^29-31. Of significance, the Nazarian study tested isolated pathologic bone cores with the bone samples obtained from femurs and vertebra specimens with no differentiation between the two sample populations. These differences further highlight the care required when extrapolating the mechanical response obtained for isolated bone samples to the more complex failure process of the pathologic vertebral body. Our laboratory is actively investigating the different BM’s role in affecting the interaction between the cancellous bone and vertebral cortex as determinants of vertebral failure.

The mechanical strength of vertebral bone is related to its intrinsic material properties, its apparent density, and the structural efficiency of its architecture ³². Consistent with previous studies ^{14; 27; 28}, our finding indicates the higher osteoblastic vertebral strength and, conversely, the lower strength of osteolytic vertebrae may, to no small extent, be affected by BM-mediated changes to the bone architecture, as reflected by changes in vBMD and BV/TV (see Appendix). We did not find consistent trends regarding the effect of age or race on vertebral strength. This finding agrees with a recent retrospective study in a patient cohort treated with radiotherapy for metastatic spine disease, ³³ which reported gender, age, and type of primary cancer were not significantly associated with the development of skeletal adverse events.

Our data demonstrated vBMD to be associated with pooled data for vertebral stiffness. Significantly, although the vBMD-based prediction of stiffness was similar between osteoblastic and mixed BM vertebrae, it was drastically inferior for predicting the stiffness of osteolytic vertebrae. The hallmark of osteolytic BM is the rarefication of bone architecture caused by the loss of number and thickness of bone trabeculae ¹⁴. Recent preclinical micro finite-element computational studies ^{34; 35} demonstrated osteolytic bone metastasis affects a heterogonous strain distribution within the remaining vertebral bone architecture ^{34; 36}. Significantly, these studies revealed the extent of lytic regions within the vertebral body affect the variation in local structural behavior of the bone ^{34; 36}. This mechanism may explain our finding of the similarity in the osteoblastic and mixed BM’s measured stiffness shown in Fig. 2, suggest-ng that the extent of osteolytic regions introduces local areas of structural weakness causing a higher risk of localized failure within the bone network, determining the failure of lytic and osteoblastic vertebrae. Due to the scalar nature of the vBMD or BV/TV information, these image-based measurements offer limited ability to detect such regions within the spatial organization of the vertebral bone network. Better understanding the effects of a specific BM type on the mechanical and structural synergism between the vertebral bone and cortex, shown to affect the state of stress and strain ³⁷ and failure ³⁸ of vertebrae with osteolytic foci, will improve our ability to predict pathologic compression fractures.

Limitations

This study has several limitations. This is a cadaver study and may not have modeled the clinical situation with complete fidelity. Our findings are limited to middle-aged and elderly spines, reflecting the demographic in which BM are most commonly observed. Precise measurement of vertebral strength and stiffness involves destructive testing, which precludes an in vivo study. However, having been obtained fresh frozen from a cohort of cancer patients with verified BM, the material and mechanical properties of the study vertebrae should closely emulate the in vivo vertebrae. To focus on the BM effect on vertebral strength and stiffness, we employed our standardized mechanical test for single vertebral bodies,²¹ providing highly controlled loading. This protocol necessitated removing the intervertebral disc, endplates, and posterior elements, retaining the vertebral body. Although removing the intervertebral disc and endplates can modify the body’s internal loading under applied loading, which may alter its strength,²⁹ retaining the intervertebral discs introduces a complex interaction between the disc and vertebral endplates yielding significant errors in predicting vertebral strength.³⁹ The posterior elements play an essential role in affecting the stability of the pathological spine.⁹ However, given the lack of information about the effect of BM type on the structural integrity of the posterior elements, we elected to focus on the vertebral body. Although our test configuration cannot fully capture the structural response of the entire spine unit, it is fully justified given the study’s aims. The computation of vBMD requires segmentation of the vertebral body from the CT data. Although it can be readily performed using freely available imaging platforms such as 3D Slicer (https://www.slicer.org), it is not currently available as part of the clinical, routinely available radiological study.

Conclusions

The low performance of qualitative bone lesion quality classification for estimating metastatic vertebrae strength and stiffness, critical components of the pathologic spine biomechanical stability, and risk of fracture9 highlight the limitations of this subjective classification. We have demonstrated the paradigm of computed CT-derived vBMD to offer improved estimates of the effect of BM on the mechanical performance of pathologic vertebrae compared to the paradigm based on radiographic bone lesion quality. The application of this paradigm will allow clinicians to evaluate patients with BM and provide the ability to initiate treatment before the vertebral column’s structural deficiency leads to a devastating spinal cord injury. Equally important, it will allow more stable patients to have their spinal column defects followed expectantly and their systemic therapy to proceed without interruption. However, the model’s performance was not uniform across BM type, suggesting that additional research is required. As our model of vertebral strength and risk of PVF evolves, we aim to replace the reliance of the current protocol on subjective, qualitative judgments of vertebral strength with precise quantitative estimates based on image-derived engineering analysis estimates of the strength of the pathologic vertebrae. This development will allow clinicians to understand fracture risk in these patients with increased precision and confidence.

Supplementary Material

Fig A.1

Fig. A.1. Independent of BM type, BV/TV is a significant predictor of vertebral strength (A) and vertebral stiffness (B). Within individual bone metastases types (C), the strength of prediction was maintained for osteoblastic vertebrae, while the prediction of strength was lower for vertebrae with mixed and osteolytic BM. For structural stiffness (D), the prediction strength was maintained mainly for the osteoblastic and mixed BM. However, we observed poor prediction of structural stiffness for osteolytic bone metastases, suggesting that other mechanisms other than bone tissue fraction affect the structural response of osteolytic vertebrae.