Level-Specific Volumetric BMD Threshold Values for the Prediction of Incident Vertebral Fractures Using Opportunistic QCT: A Case-Control Study

Michael Dieckmeyer, Maximilian T. Löffler, Malek El Husseini, Anjany Sekuboyina, Bjoern Menze, Nico Sollmann, Maria Wostrack, Claus Zimmer, Thomas Baum, Jan S. KirschkeFrontiers in Endocrinology 2022 · DOI

TL;DR

This study introduces a method for predicting incident vertebral fractures (VFs) using opportunistic quantitative computed tomography (QCT) scans, a common byproduct of routine medical imaging. By developing level-specific volumetric bone mineral density (vBMD) thresholds across the entire thoracolumbar spine, the research offers a more comprehensive and accessible approach to osteoporosis screening and fracture risk assessment, particularly for elderly and oncologic patients.

What Problem Does This Paper Address?

Osteoporosis is a major public health concern, leading to debilitating vertebral fractures (VFs). Current screening methods, like DXA, have limitations in predicting VFs, and the gold standard for bone density measurement, QCT, is not always readily available or utilized opportunistically. While QCT can be performed on routine CT scans (opportunistic QCT), a standardized approach for risk assessment across the entire thoracolumbar spine is lacking. Specifically, there's a need for level-specific vBMD thresholds that can reliably predict incident VFs, moving beyond the current focus on the lumbar spine and enabling broader application of opportunistic screening.

Methods

The researchers employed a case-control study design, comparing 35 patients with incident VFs to 70 age- and sex-matched controls. They utilized routine multi-detector CT (MDCT) scans of the thoracolumbar spine. A key methodological innovation was the use of an automated pipeline for extracting trabecular vBMD. This pipeline involved:

  • Vertebral Segmentation: An automated framework using convolutional neural networks (CNNs) to identify and segment individual vertebral bodies.
  • Subsegmentation: Further segmentation of vertebral bodies into compartments like trabecular bone.
  • Asynchronous Calibration: A method to convert CT attenuation values (Hounsfield Units) to vBMD, accounting for different scanners and tube voltages, and importantly, correcting for the presence of intravenous contrast agents. This asynchronous approach allows for the use of scans acquired with or without contrast, significantly broadening the applicability of opportunistic QCT.
  • Threshold Determination: Receiver operating characteristic (ROC) curve analysis was used to determine optimal vBMD threshold values for predicting incident VFs. These thresholds were calculated based on the Youden index (maximizing sensitivity and specificity) and for fixed sensitivity levels (60-85%).

Datasets

The study utilized retrospective MDCT scans from the hospital's digital picture archiving and communication system (PACS). The scans were acquired between June 2007 and May 2017. The dataset comprised 35 incident VF cases and 70 controls. No new benchmark datasets were introduced; the study relied on existing clinical data.

Key Insights and Ideas

  • Thoracolumbar Spine as a Whole: The study demonstrates that vBMD across the entire thoracolumbar spine, not just the lumbar region, is a significant predictor of incident VFs. This broadens the scope of opportunistic screening.
  • Automated Pipeline is Key: The development and application of an automated pipeline for vBMD extraction, including contrast correction, is crucial for making opportunistic QCT practical and scalable in a clinical setting. This reduces manual effort and potential for error.
  • Level-Specific Thresholds Matter: The research highlights that optimal vBMD thresholds for VF prediction vary across different vertebral levels, generally increasing from the lumbar to the thoracic spine. This nuanced approach allows for more accurate risk stratification.
  • Opportunistic QCT's Potential: The findings strongly support the utility of opportunistic QCT, leveraging existing CT scans, as a powerful tool for early risk assessment of imminent VFs, especially in patient populations already undergoing CT scans for other reasons (e.g., oncologic staging).

Key Results

  • Association with VFs: Lower vBMD at every single level of the thoracolumbar spine was significantly associated with an increased risk of incident VFs, even after adjusting for age, sex, and prevalent VFs. The odds ratio (OR) per standard deviation decrease in vBMD ranged from 3.28 to 3.98 across different spinal regions.
  • Predictive Performance: The area under the curve (AUC) for predicting incident VFs showed no significant difference between vertebral levels, with the highest AUC (0.75) observed at the thoracolumbar junction (T11-L2) and for combinations of vertebral levels (e.g., L1-L4).
  • Threshold Values: Optimal vBMD threshold values ranged from 40.9 mg/cm³ (L4) to 83.1 mg/cm³ (T1) for single vertebrae, and from 45.8 mg/cm³ to 69.3 mg/cm³ for combinations of four vertebrae. A threshold of 80.0 mg/cm³ (a commonly used threshold for osteoporosis diagnosis) showed sensitivities between 76-88% and specificities between 41-61% for incident VFs at the thoracolumbar junction and lumbar spine.
  • Sensitivity vs. Specificity Trade-off: The study demonstrates a clear trade-off: higher sensitivities (detecting more VF cases) required higher vBMD thresholds, leading to lower specificities (more false positives).

Limitations and Future Work

  • Study Population Bias: The study population was predominantly elderly and Caucasian with oncologic conditions. This limits the generalizability of the findings to other demographics and patient groups.
  • Low VF Case Count: The relatively low number of VF cases might restrict the robustness of the findings and the transferability of thresholds.
  • Short Follow-up Period: The median follow-up of 1.5 years is shorter than the typical 10-year period used for clinical fracture risk assessment, potentially impacting specificity.
  • Potential Confounders: While efforts were made to control for confounders, the prevalence of oncologic disease and variations in medical history (e.g., pharmacotherapy, chemotherapy) could still introduce bias.

Future work could focus on validating these findings in larger, more diverse cohorts. Further research could also explore the prediction of fracture location and refine threshold values for specific patient subgroups. The authors also suggest that the level-specific thresholds should be considered in the context of clinical decision-making, potentially with patient-specific adjustments.

Summary updated

Abstract

At all levels of the thoracolumbar spine, lower vBMD was associated with incident VFs in an elderly, predominantly oncologic patient population. Automated opportunistic osteoporosis screening of vBMD along the entire thoracolumbar spine allows for risk assessment of imminent VFs. We propose level-specific vBMD threshold at the thoracolumbar spine to identify individuals at high fracture risk.
BibTeX
@inproceedings{dieckmeyer2022level,
  author = {Michael Dieckmeyer and Maximilian T. Löffler and Malek El Husseini and Anjany Sekuboyina and Bjoern Menze and Nico Sollmann and Maria Wostrack and Claus Zimmer and Thomas Baum and Jan S. Kirschke},
  title = {{Level-Specific Volumetric BMD Threshold Values for the Prediction of Incident Vertebral Fractures Using Opportunistic QCT: A Case-Control Study}},
  booktitle = {Frontiers in Endocrinology},
  year = {2022},
  doi = {10.3389/fendo.2022.882163},
}
Figure