Bone Density Distribution in the Cervical Spine

Abstract

Study Design

Retrospective cohort study

Objective

Given changes in bone density induced by degenerative disease, general measures of bone health (ie DEXA) are inadequate to evaluate bone density in surgical areas of interest. Regional differences in HU in the cervical spine may influence surgical strategies. The purposes of our study were to determine whether cervical Hounsfield units (HU) vary by level, examine their relationship with age, comorbidities, and alignment, and propose a technique to measure HU in the lateral masses.

Methods

Two hundred twenty-four patients with degenerative spine pathology with a cervical computed tomography were included (2015-2019). Measurements were performed in each vertebral body (C2-T1; mid-axial, anterior-axial, posterior-axial, mid-coronal, and mid-sagittal) and 2 regions of the lateral masses (C3-C6; mid-cor, mid-sag). To evaluate reliability, 6 observers each measured 355 HU values, inter-relater reliability assessed with intraclass correlation coefficients Correlations of HU with age, BMI, comorbidities, and cervical alignment were evaluated.

Results

Bone density differed by level, with the lowest HU scores in the lower cervical spine (C6-T1) (P < .001). No correlations were found between LM HU and age, BMI, CCI, or alignment (P > .05). Increased kyphosis was weakly correlated with VB HU, while age and CCI showed moderate correlations with VB HU at all levels (P<.001). ICC for HU measurements were good to excellent for the VBs, but poor to moderate for the LMs.

Conclusion

Bone is least dense in the lower cervical spine. HU scoring is not reliable in the lateral masses. We recommend that a level-specific approach to bone density is considered in surgical planning.

Introduction

The density of tissue on computed tomography (CT) reconstructions can be measured in Hounsfield units (HU), which are a measure of tissue attenuation. The clinical prevalence of CT scans has fostered interest in the use of HU as a marker of bone density. ¹ Certain authors have advocated for the use of cervical or lumbar CT scans as an “opportunistic” manner of screening for osteoporosis. ² However, while the relationship between osteoporosis (diagnosed with DEXA) and low HU values on CT scan is well-established,^1-3 the patient’s general bone health may not correlate with the bone density in the surgical zone of interest. In accordance with Wolff’s law, bone remodels under stress. Thus, patients with degenerative lumbar spine disease have been shown to have higher bone density in the lumbar spine, despite having femoral neck DEXA T-scores consistent with osteoporosis.^4,5 In this context, the measurement of HU in the surgical region of interest (ie the levels for which a surgical procedure is planned) may be a more accurate method of evaluating regional bone density and predicting complications. Indeed, several authors have found that HU in the surgical region of interest correlate with mechanical and clinical outcomes, such as pseudarthrosis, cage subsidence, and proximal junctional kyphosis.^6-8

In the lumbar spine, HU have been shown to vary by vertebral level. ³ Similar studies have not been performed for the cervical spine. Understanding trends in bone density of the cervical spine (as measured by HU scores) may be of clinical importance given the associations between HU and mechanical complications. ⁹ Furthermore, HU scores have not been attempted in the lateral masses, which may be of interest for posterior cervical fusion constructs. Thus, the purposes of this analysis were to (1) define trends in bone density of the degenerative cervical spine, (2) examine the relationship between cervical bone density, age, BMI, comorbidities, and alignment, and (3) develop a reliable technique to measure HU scores in the lateral masses.

Methods

Patient Sample

This study was approved by the institutional review board. A retrospective chart review was performed to identify all consecutive adult patients (>18 years old) who underwent a cervical spine CT by one of six spine surgeons (May 2015 – December 2019) for evaluation of cervical-spine related complaints. Patients with outside cervical spine CTs, previous cervical spine instrumentation, pathophysiologic cervical spine conditions (eg, Klippel-Feil), osteoporosis (defined as a medical history of osteoporosis or current or past treatment with osteoporosis medications [excluding supplements]) were excluded.

Data Collection

Demographics and medical history were retrieved from the electronic medical record. All components of the Charlson Comorbidity Index (CCI) were collected from the medical record and used to calculate the CCI for each patient as a composite measure of their chronic health status. ¹⁰ The American Society of Anesthesiologists Class was collected for each patient as an additional composite measure of pre-anesthesia medical co-morbidities. ¹¹

Each patient underwent presurgical AP and lateral cervical spine radiographs in neutral alignment position, from which the C2-C7 sagittal cobb angle, the T1 slope [T1S], and the C2-C7 sagittal vertical axis [cSVA] were measured by an independent research technician using a dedicated measurement software (Surgimap, Nemaris Inc, New York, NY). ¹²

CT scans were performed using a 16-MDCT scanner (MX8000 Philips Healthcare, Andover, MA). For the assessment of bone mineral density, we modified a previously published protocol by Schreiber et al.^1,6 As described by this protocol, HU scores were obtained on circular areas of medullary bone that excluded cortical bone or areas of sclerosis on three thin-cut (1.25 mm) axial CT slices. The protocol was developed in the lumbar spine and consisted of measurements on three axial CT slices (inferior, mid, and superior). Given that the vertebral bodies in the cervical spine are smaller and shorter, we modified this protocol by obtaining HU scores from axial, coronal, and sagittal multiplanar reconstructions. ³ Furthermore, the assessment of trabeculae microstructure over three axes gives a more accurate picture of the overall bone density.^13-15

Hounsfield units were measured in 5 regions of each vertebral body (C2-T1; mid-axial, anterior-axial, posterior-axial, mid-coronal, and mid-sagittal). Mid-axial was defined as the CT slice at the region halfway between the inferior and superior endplates. Mid-coronal was defined as the CT slice halfway between the anterior and posterior aspects of the vertebral body, and mid-sagittal was defined as the slice halfway between the lateral aspects of the vertebral body (Figure 1). Similarly, the lateral masses (left and right, C3-C6) were measured at two regions: mid-coronal and mid-sagittal. The mid-coronal region was defined using the CT slice halfway between the anterior and posterior aspect of the lateral mass, along an anterior-posterior axis in line with the inferior and superior facet joints. The mid-sagittal region was defined as the slice halfway between the medial and lateral aspects of the lateral mass (Figure 2). Notably, mid-axial LM measurements were excluded as the relative thickness of the medial/lateral and anterior/posterior cortical walls of the LM made it near-impossible to measure the HU values at these regions without including cortical or sclerotic bone.

Figure 1.

Measurement regions for vertebral bodies.

Figure 2.

Measurement regions for the lateral masses.

The intraobserver reliability for HU measurements using the described technique has already been well established.^1,6 Measurements were performed by six members of the research team (orthopaedic residents and spine surgery fellows). Six persons were tasked with performing measurements as we aimed to find an HU method with high interrater reliability, especially with regards to the lateral masses. Thus, before all CTs were assessed, the first five patients were independently measured by the six observers, consisting 355 independently measured HU regions (71 regions per patient, 5 patients). After reliability was established, the rest of the CT scans were assessed.

Statistical Analysis

Given that a major purpose of this study was to determine level-specific average HU scores in the cervical spine, we chose to eliminate outliers. For each set of measurements, the top and bottom five values were excluded and analyses were performed. The 3 measured regions in each VB and the 2 measured regions in each LM were combined and averaged to obtain the “Total VB” and “Total LM” HU values, respectively (ie, Total C2, Total LM C5, etc.). These means were utilized for correlation and comparison analyses. One-way ANOVA was used to compare HU measurements between vertebral levels and to compare HU means across age groups. Post-hoc pairwise comparisons were also performed between HU values between vertebral levels, with Bonferroni corrections applied. Correlations of HU values with age, body mass index, radiographic alignment, and medical comorbidities were assessed using Spearman’s coefficients. Inter-observer reliability for the measurement technique was assessed using two-way random-effects models to calculate intraclass correlation coefficients (ICCs). A value of P < .05 was considered significant. Portions of the data were managed using REDCap (Research Electronic Data Capture) hosted by [BLINDED] Medicine Clinical and Translational Science Center under the following grant: [BLINDED]. ¹⁶ Statistical analyses were performed using IBM SPSS Version 25.0 (Armonk, NY).

Results

Patient Sample

201 patients were included in the final analysis (Figure 3 & Table 1). The sample had a mean C2-C7 sagittal cobb angle of −4.88±14.2 (negative values correspond to lordosis), cSVA 26.2±15.6 mm, and mean T1S of 28.2±10.5 (positive values correspond to forward tilt).

Figure 3.

Patient selection diagram.

Table 1.

Description of patient characteristics. BMI, body mass index; ASA, American Society of Anesthesiologists; COPD, chronic obstructive pulmonary disease; CKD, chronic kidney disease (Stage 3 or greater); CCI, Charlson Comorbidity Index.

Demographics and Comorbidities
n=201	Mean/N	Sd/%
Age (years)	56.7	12.3
BMI (kg/m 2 )	28.6	5.6
Gender
Male	129	63.5%
Female	74	36.5%
Race
white	183	90.1%
Black	4	2.0%
Asian	3	1.5%
Other	13	6.4%
Visit diagnosis
Myelopathy	77	37.9%
Radiculopathy	93	45.8%
Facet arthropathy	94	46.3%
Other	6	3.0%
ASA class
I	15	7.5%
II	163	81.1%
III	23	11.4%
Comorbidities
Diabetes	19	9.5%
Current smoker	18	9.0%
Rheumatologic disease	7	3.5%
Copd	18	9.0%
Ckd	3	1.5%
CCI	1.9	1.7

Reliability of Measurement Method

Measurement of vertebral bodies was performed with moderate to excellent inter-relater reliability. ¹⁷ The intraclass correlation coefficients for the three VB regions as follows: .82 mid-ax (good), .88 mid-cor (excellent), .88 mid-sag (excellent). In contrast, the reliability of the lateral mass measurements was fair to good: .46 mid-sag (poor), .61 mid-cor (moderate).

Bone Density Trends in the Cervical Spine

Mean bone density was different at each level (P < .001, Figure 4). While HU values were relatively close from C2-C5 (ranging between 370.2 [C2]to 389.0 [C4]), density gradually decreased through T1, which had the minimum average HU (232.3). Post-hoc comparisons showed that C2 VB was significantly different (P < .05 with Bonferroni correction applied) compared to C6-T1 and all LMs, C3 was significantly different compared to C6-T1 and LM3-5, C4 significantly differently compared to C6-T1 and all LMs, C5 significantly differently compared to C6-T1 and LM3-5. The lower VBs (C6-T1) were significantly different compared to all VBs and LMs.

Figure 4.

Total HU scores of vertebral bodies. HU scores are significantly different, with a maximum score at C4 and minimum score at T1 (P<.001). Post-hoc comparisons showed that C2 VB was significantly different (P<.05 with Bonferroni correction applied) compared to C6-T1 and all LMs, C3 was significantly different compared to C6-T1 and LM3-5, C4 significantly differently compared to C6-T1 and all LMs, C5 significantly differently compared to C6-T1 and LM3-5. The lower VBs (C6-T1) were significantly different compared to all VBs and LMs. LM3 was significantly different compared to LM4 and LM6, LM4 significant different compared to all LMs, LM5 significantly different compared to LM4, and LM6 was significantly differently compared to LM3 and 4. Standard deviations represented by error bars.

Regarding the lateral masses, the mean HU differed by level but fell within a narrow range (480.0-383.3) (Figure 5). On one-way ANOVA, the lateral masses had significantly higher HU scores compared to the vertebral bodies (P < .001). Post-hoc comparisons showed that LM3 was significantly different compared to LM4 and LM6, LM4 significant different compared to all LMs, LM5 significantly different compared to LM4, and LM6 was significantly differently compared to LM3 and 4

Figure 5.

Total HU scores of lateral masses. The HU scores of the lateral masses were higher than those of the vertebral bodies (P<.001).

Relationship Between Clinical Factors and Bone Density

Significant negative correlations were found between VB HU, age, and CCI (P < .05) (Table 2). The largest negative correlations between age and VB HU were found in the uppermost and lowermost portions of the cervical spine. A similar correlation pattern was found between VB HU and CCI (Table 2). When analyzed categorically, a significant decrease in mean bone density was seen with age for all vertebral bodies, with the exception of C4 (P < .05 for all other comparisons). Increasing ASA class was also associated with a significant decrease in bone density with the exception of T1 (P < .05 for all other comparisons).

Table 2.

Correlation coefficients between HU scores, demographics, CCI, and alignment. Bolded values indicate significance (P<.05).

			Spearman Coefficient
	Mean	SD	T1 Slope	C2-7 SVA	C2-C7 Cobb	Age	BMI	CCI
TotalC2	368.0	72.0	−.07	−.09	.15	−.28	.10	−.28
TotalC3	372.6	64.4	−.09	−.01	.16	−.26	.05	−.19
TotalC4	383.6	71.9	−.11	−.03	.17	−.20	−.03	−.17
TotalC5	371.3	85.1	−.10	−.03	.24	−.22	−.07	−.19
TotalC6	329.0	82.0	−.09	−.06	.23	−.24	−.15	−.19
TotalC7	277.1	56.7	−.13	−.07	.25	−.30	−.13	−.28
TotalT1	232.3	46.5	−.12	−.09	.10	−.35	−.05	−.28
LatMassC3	437.9	95.8	−.04	.09	.14	.02	.10	.06
LatMassC4	480.0	103.2	−.05	.001	.11	−.02	.01	−.05
LatMassC5	423.1	98.8	−.05	−.038	.08	−.07	−.01	−.04
LatMassC6	398.3	98.2	−.10	−.15	−.02	−.13	−.05	−.11

Bone density in the lateral masses (at any level) was not associated with age, BMI, CCI or ASA class on correlation or categorical analyses.

Relationship Between Alignment and Bone Density

A more kyphotic cervical spine was weakly correlated with higher bone density in the mid-cervical spine (Table 2). ] T1S and cSVA were not correlated with bone density. Alignment was not correlated with HU scores in the lateral masses (Table 2).

Discussion

Our retrospective review of 201 CT scans in patients with symptomatic cervical spine pathology found that vertebral body HU score varies by level, with the lowest scores in the inferior cervical spine. Age and comorbidity burden were inversely correlated with the density of the vertebral bodies, especially in the upper and lower cervical spine. Regarding alignment, we found that a more kyphotic cervical spine was associated with higher HU scores, with the strongest correlations in the mid-lower cervical spine. In contrast, HU scores in the lateral masses were not correlated with any factor, including alignment. Finally, we showed that the density of the lateral masses is higher than the that of the vertebral bodies, however, the lateral mass analyses must be interpreted cautiously in the context of poor inter-observer reliability.

We used a novel technique in our measurement of HU scores in cervical vertebral bodies, which was found to have similar inter-observer reliability compared to traditional measurement techniques. ⁴ While the majority of HU investigations have exclusively used axial cuts,^1,2,6,9 we incorporated sagittal, coronal, and regional axial HU scores into our protocol. Given that bony trabeculae form a complex, three-dimensional network, we believe that a three-dimensional assessment gives the most accurate picture of local bone quality.^13-15 Previous authors have attempted similar analyses in the lumbar spine, finding that at certain levels, sagittal HU scores were more strongly correlated to DEXA than axial scores (though the differences in correlation were minimal). ³ For the purposes of measuring HU scores, whether one CT reformat is superior to another remains to be seen.

Our HU values are similar to previous investigations of HU scores in the cervical spine.^2,7 Wang et al ⁷ investigated a smaller cohort (91 patients) who underwent one-level ACDF, reporting mean C3-C7 HU values remarkably similar to those reported in our study (C2 and T1 were not measured). Colantonio and colleagues examined 149 cervical spine CTs in the United States Department of Defense database, finding a mean C4 axial HU value of 452±116 in healthy subjects, compared to 320±82 in those with osteoporosis (as diagnosed by DEXA). Notably, they performed their analyses in all patients who had a previous cervical CT scan, not necessarily patients presenting with cervical spine pathology. Thus, we believe that our investigation currently represents the best possible “reference values” for HU in patients with degenerative cervical spine pathology. However, there is a caveat to this statement—given that the primary purpose of our study was to establish average level-specific HU values in the cervical spine, we chose to eliminate the top 5 and bottom 5 measures of bone density. While this may have made our average HU values more accurate, it also increased our inter-observer reliability. To this end, we caution readers that very high or low HU scores (ie several standard deviations lower or higher than the average values we report) may not be reliable. ¹

A similar statement can be made about using HU scoring for the lateral masses. Ours is the first study to attempt HU scoring of the lateral masses. Anecdotally, we found these measurements to be quite difficult given that the high ratio of cortical to medullary bone in the lateral masses. The difficulty in assessing lateral mass bone density was evident in the poor to moderate inter-observer reliability of our measurements. Thus, we do not believe HU scores should be used to evaluate the density of the lateral masses. However, there is still a clinical need to evaluate local bone quality in the lateral masses (eg to prevent instrumentation failure). ¹⁸ This may be especially relevant for long posterior fusion constructs extending into the cervical spine. Future avenues of research could analyze new modes of assessing bone quality, such as cortical thickness or cortical to medullary ratios. The bone quality of the cervical pedicle may also be an area of research to allow for comparisons of cervical posterior fixation methods.

In our study, we found that the least dense bone was found in the lower cervical spine. Notably, the relationship of HU scores with mechanical outcomes is a well-researched topic, with studies showing correlations between HU scores and cage subsidence in ACDF and lateral lumbar fusions,^7,9 loss of pedicle screw fixation,^18,19 proximal junctional kyphosis,^6,20 and lumbar pseudarthrosis. ⁸ Thus, the fact that HU scores vary by level has several potential clinical consequences. First, anterior constructs ending in the lower cervical spine (the majority of 3- and 4- level ACDFs) are seated in the least dense region of bone. This may help to explain the higher rate of pseudarthrosis in multilevel constructs,^21,22 and especially why multilevel fusions in the lower cervical spine demonstrate a larger loss of lordosis than those in the mid-upper cervical spine. ²³ Second, while age and a higher comorbidity burden were correlated with decreased bone density at nearly all cervical levels, the fact that density varied shows that a level-specific (and not necessarily patient-specific) approach should be taken when determining the risk for mechanical complications.^7,24 For example, the C6-T1 region in a healthy 50 year-old may have similar HU scores as the C3-C5 region in a sick 70 year old, which may in turn influence surgical strategy or the risk-benefit discussion. We recommend that surgeons evaluate the bone density in the surgical regional of interest when planning fusions of the cervical spine.

These considerations become even more complex when the effect of alignment is taken into account. Wolff’s law states that bone adapts to stress, organizing its trabeculae to support load. In this sense, the fact that posterior-axial HU scores were higher than anterior-axial HU scores in our population (who had an average lordotic alignment of 4°) should not be surprising. Wolff’s law could also explain why patients with kyphosis began to show a reversal of this relationship, with similar HU values in the anterior-axial and posterior-axial regions of C5 and C6. Kyphotic cervical spines were also associated with higher HU scores overall—correlations strongest in the C5-C7 regions, where degenerative disc disease is most likely to occur. ²⁵ This finding only strengthens our recommendation that surgeons consider the HU scores in the surgical region of interest, and avoid assuming that bone density throughout the spine is the same.

This study has several limitations. First, we did not have gold-standard measures of bone density (DEXA or quantitative CT scans) available for all patients, which precludes our ability to relate cervical spine HU scores with these metrics. However, several lumbar spine studies have shown that DEXA may overestimate the bone density at the surgical region of interest. ⁴ Thus, we do not believe that the inclusion of DEXA scores in our analysis would affect our conclusions. Second, while we excluded patients with a clinical history of osteoporosis or history of taking osteoporosis medications, we could not account for patients with undiagnosed osteoporosis. Regardless, our patient population is a real-world example of patients presenting for evaluation by a spine surgeon, many of who will present without a diagnosis of osteoporosis. Third, while the purpose of this study was not to evaluate risk factors for decreased bone density in the cervical spine, the small number of smokers, patients with rheumatologic disease, and non-white subjects precluded meaningful analysis of these risk factors. Thus, our conclusions may not apply to patients dissimilar to our cohort. Fourth, as stated above, given that we eliminated outliers in order to obtain the most accurate HU values, we caution readers that very high or low HU scores (ie several standard deviations lower or higher than the average values we report) may not be reliable. Fifth, our findings only apply to patients with degenerative spine pathology presenting for cervical-spine related complaints, we cannot comment on HU values or trends in patients with healthy cervical spines. Furthermore, we did not include the “degree of degeneration” in the cervical spine. While we did note that kyphosis was associated with an increase in HU, this is an imperfect surrogate for disc degeneration. Further study on the specific relationship between HU values and disc degeneration is merited.

In conclusion, our investigation of 201 CT scans in patients with cervical pathology demonstrates that cervical HU scores vary by vertebral level. The lowest density was found in the cervical spine, which may partially explain the high rate of mechanical complications in multilevel anterior fusions. Further investigation on preventing these complications should incorporate HU scoring. Age, comorbidities, and alignment were all found to influence HU scores of the vertebral bodies, which demonstrates that a level-specific approach to bone density should be considered when evaluating a patient for surgery. Future studies will be needed to determine if such an approach will influence clinical and mechanical outcomes. Finally, we encourage continued research on methods to measure regional bone density in the lateral masses, as HU scoring may not be appropriate.

ORCID iDs

Francis Lovecchio, MD https://orcid.org/0000-0001-5236-1420

Bryan Ang https://orcid.org/0000-0002-4174-997X

Philip K. Louie https://orcid.org/0000-0002-4787-1538

Virginie Lafage, PhD https://orcid.org/0000-0002-0119-7111