Improved prediction of incident vertebral fractures using opportunistic QCT compared to DXA

Osteoporosis is a metabolic bone disease leading to reduced bone strength and manifesting in low-energy fractures [1]. Resulting pain and disability pose a huge burden on patients and society [2, 3]. Effective prevention and medical treatment for osteoporosis exist [4], but are not initiated in many patients [3, 5], partly because bone densitometry is under-utilized [6, 7]. Bone mineral density (BMD) as the single most important parameter accounts for approximately 70% of bone strength [8].

For the diagnosis of osteoporosis, the up-to-date reference standard in clinical bone densitometry is dual energy X-ray absorptiometry (DXA) [9]. This projectional technique is performed at the spine and hip in order to formulate a diagnosis based on a WHO normative population [10]. However, the role of DXA in the diagnosis of osteoporosis can be put into question, given that in a large population-based study, less than half of women (44%) and even fewer men (21%) of all individuals with prevalent osteoporotic fractures were correctly diagnosed with osteoporosis by DXA [11].

Quantitative CT (QCT) is a notable alternative to DXA with at least the same ability to predict vertebral fractures in women, although it is not officially approved to diagnose osteoporosis, since the WHO classification relies on T-scores derived by DXA [12]. QCT is a non-projectional technique performed on clinical CT scanners to measure volumetric BMD. Due to its three-dimensional characteristic, QCT is largely independent of degenerative changes in the spine and can differentiate between cortical and trabecular bone. Trabecular bone is about eight times more metabolically active than cortical bone and therefore prone to changes in osteoporosis [12]. Osteodensitometry in routine CT scans, which have been acquired for other purposes, can distinguish osteoporotic from healthy individuals [13] and bears a huge potential of opportunistic screening [14]. Accordingly, densitometry based on non-dedicated CT scans is named opportunistic QCT. In the following, “BMD” will refer to volumetric BMD as assessed by opportunistic QCT—not DXA measured areal BMD—as previously encouraged [12]. Wherever helpful for the reader to avoid confusion, we explicitly identify BMD_QCT as being derived from CT measurements.

The comparative potential of QCT and DXA to discriminate between patients with and without prevalent vertebral fractures has been investigated in many cross-sectional studies [15‐18]. Recently, the risk of future vertebral fractures has been investigated in opportunistic CT data [19]; however, DXA data was not included. In the present study, we investigate the association between the risk of future osteoporotic vertebral fractures and opportunistic BMD measurements in routine CT scans acquired for other purposes compared to measurements of the reference standard DXA.

Over a median follow-up of 2.6 years (interquartile range 1.7–3.6 years), 16 of 84 patients (19%) sustained an incident vertebral fracture (Table 1). Patients with incident vertebral fracture were significantly older with a mean age of 73.9 ± 7.4 years and had a lower mean BMD_QCT of 56.7 ± 31.6 mg/cm³ than patients without fracture with a mean age of 67.7 ± 8.6 years (p = 0.01) and a mean BMD_QCT of 93.3 ± 41.7 mg/cm³ (p = 0.001). The number of patients defined as osteoporotic by BMD_QCT differed significantly between patients with and without incident vertebral fractures (p = 0.004). However, there was no significant difference in DXA T-score between patients with and without incident vertebral fractures (p = 0.179). Seven of 16 patients (44%) with incident vertebral fractures were correctly diagnosed with osteoporosis according to DXA, whereas 13 of 16 (81%) were categorized having osteoporotic BMD_QCT.

There was a statistically significant association between the risk of incident vertebral fractures and unadjusted trabecular BMD with a HR of 4.07 per SD decrease in BMD_QCT (CI, 1.98–8.38); there was no significant association with unadjusted DXA T-score (Table 2, Figs. 3 and 4). The HR for incident vertebral fractures per SD decrease in BMD_QCT varied, but remained statistically significant, after adjusting for age (3.60; CI, 1.70–7.64), for age and sex (4.02; CI, 1.83–8.82), and for age, sex, and prevalent fractures (2.54; CI, 1.09–5.90). Only after adjusting for age and sex, a statistically significant increase in HR per SD decrease in DXA T-score was observed (1.57; CI, 1.04–2.38). BMD_QCT was a significant classifier to predict incident vertebral fractures (AUC = 0.76; CI, 0.64–0.89), DXA T-score was not (Table 3 and Fig. 5). BMD_QCT values greater than or equal to 79.6 mg/cm³ could predict incident vertebral fracture with a specificity of 81% and a sensitivity of 59%.

Long-term scanner stability was good for all three investigated MDCT scanners. Linear drift was SL_iCT =  − 0.55 HU per year over an observation period of 5.33 years with a CV_iCT = 1.1% for Philips iCT 256, SL_AS =  − 2.29 HU per year over an observation period of 4 years with a CV_AS = 1% for Siemens Somatom Definition AS, and SL_C64 =  − 0.81 HU per year over an observation period of 4.09 years with a CV_C64 = 0.7% for Siemens Somatom Sensation Cardiac 64 (Table 4).

In this retrospective study, trabecular BMD assessed by opportunistic QCT showed a high association with the risk of incident vertebral fractures in a mixed population of mainly neurosurgical and oncologic patients. In contrast, the association of T-scores measures by DXA was non-significant. Only after adjusting for age and sex, T-scores were associated with the risk of incident vertebral fracture. Furthermore, more than every second patient (56%) who developed a new osteoporotic vertebral fracture was not diagnosed with osteoporosis according to DXA, whereas the rate of false-negative diagnosis by opportunistic QCT was much lower (19%).

Many cross-sectional studies compared the capability of DXA and conventional QCT to discriminate between patients with and without prevalent spinal fractures [15‐18, 31, 32]. A better ability of opportunistic QCT than DXA to classify these patients was suggested as a secondary result in a study, in which 22 out of 37 patients with a prevalent vertebral fracture (59%) had non-osteoporotic DXA T-scores [33]. There are further reports when DXA struggled to correctly diagnose approximately every second patient with manifest osteoporosis [34, 35]. The influence of degenerative changes of the spine on the results of DXA is a long known issue [36‐38], that can be mostly overcome by QCT [34, 39, 40]. It seems plausible that in our study population the diagnosis of osteoporosis by DXA was less accurate than by QCT, because there was a majority of elderly neurosurgical patients presenting themselves with back pain and most likely showing a degree of spinal degeneration above average.

Longitudinal studies reporting future vertebral fractures are rare, mostly using dedicated quantitative or biomechanical CT in prospective cohorts [41‐43], and/or lacking reference DXA scans of the spine [19, 43]. To the best of our knowledge, no study has been conducted comparing non-dedicated (=opportunistic) QCT with DXA regarding the association with the risk of future vertebral fractures. Our results are in accordance with previous findings of similar longitudinal studies. Analyzing the prospective database of men aged 65 years and older (MrOS), a higher association with the risk of new clinical vertebral fractures was found for integral BMD measured by QCT than for areal BMD measured by DXA [41]. The age-adjusted relative hazard for new clinical vertebral fractures increased by 5.7 per SD decrease in integral BMD at the spine, and by 3.2 and 1.8 per SD decrease in areal BMD at the lumbar spine and femoral neck, respectively. Of note, integral volumetric BMD summarizes trabecular and cortical bone in a similar way to areal BMD, but stays independent of bone size and degenerative alterations. A study on the same prospective cohort (MrOS) found an age- and race-adjusted HR of 3.69 for the prediction of clinical fractures of the spine by trabecular BMD by QCT [42]. Areal BMD by DXA at the lumbar spine had also a strong association with the risk of these fractures (HR = 3.57), but DXA at the femoral neck performed similar to our results (HR = 1.95). In ROC analysis, trabecular BMD performed also better (AUC = 0.79) than areal BMD at the spine (AUC = 0.72). In our study, more diagnoses (58%) were based on DXA measurements at the lumbar spine. This could explain why the predictive performance of DXA was substantially worse in both aforementioned statistical measures (age- and sex-adjusted HR = 1.57, AUC = 0.63). As discussed above, this might be due to a selection bias towards neurosurgical patients with above-average spinal degeneration limiting the capabilities of DXA in our population. Recently, the first longitudinal study using opportunistic CT data of multiple scanners without calibration found that L1 vertebral trabecular attenuation blow 90 HU was a significant indicator of decreased fracture-free survival [19].

A BMD_QCT cutoff predictive for incident vertebral fractures with 81% specificity (59% sensitivity), found in our data, closely matched the threshold suggested to be equivalent to the WHO diagnostic category for osteoporosis (< 80 mg/cm³) [44, 29]. Predefined thresholds can only be used if CT attenuation values are calibrated to a density-reference phantom usually with known hydroxyapatite (HA) concentrations. Otherwise, validated machine-specific cutoff values have to be determined [23]. Opportunistic screening for osteoporosis becomes increasingly popular [14, 45]. In contrast to numerous studies of opportunistic screening, where HU values in thoracic or lumbar vertebra were reported [33‐35, 46‐50], we used asynchronous calibration to obtain lumbar trabecular BMD. Synchronous calibration with an in-scan phantom as in conventional QCT can be replaced by asynchronous calibration as in the opportunistic setting, if scanner stability is maintained [23].

We performed asynchronous calibration of five different MDCT scanners to allow for opportunistic BMD screening in routine CT exams. In this opportunistic setting, the benefits of conventional QCT can be appreciated without its disadvantages of additional radiation and costs compared to DXA. We developed a protocol for dedicated calibration scans using the same parameters (tube voltage, average tube current, and reconstruction algorithm) and creating a similar geometrical setup of the scanned slice (anthropomorphic abdomen phantom with obesity extension rings and central inserts of known HA concentrations close to the position of the lumbar spine) as in routine scans. In case of the two already decommissioned MDCT scanners, we had to rely on routine scans with an in-plane density-reference phantom, yet we were able to exploit a huge number of scans averaging over a period of more than 4 years. We deemed the eccentric position of the in-plane phantom underneath the patient tolerable, as it was still close to the spine of the patient in supine position. This difference in position probably explains the additional intercept of approximately 15 units in the conversion equations of the two decommissioned Siemens scanners compared to the still operational Siemens scanner (Table 4).

Long-term scanner stability could be shown for three CT devices of two major manufacturers. Good short-term precision and low precision errors of intra-observer [51] and inter-observer reproducibility [26] of asynchronous QCT have been shown before. How to correct for intravenous contrast-enhanced scans in opportunistic QCT is still under debate [14, 45]. We used correction offsets for arterial and portal venous contrast phase from a previous study [28], although these were calculated for a different CT scanner. These minor corrections might be negligible, because they did not seem to affect the accuracy of CT measurements [34] and the overall performance for predicting osteoporosis was similar [52], in previous studies. Moreover, in our data, contrast-enhanced scans were equally distributed between patients with and without incident vertebral fractures (Table 1), thus unlikely to bias the results.

There are limitations to this retrospective observational study. As only a limited cohort of mainly neurosurgical and oncologic patients were analyzed, the results might not be applicable to other patient populations. Additionally, all patients received both MDCT and DXA; thus, osteoporosis was already suggested. This may introduce a selection bias; however, this is exactly the patient population where omitting an additional DXA scan could save time and costs. Loss of follow-up is a possible confounding factor, though independent of the employed densitometric technique.