Introduction
Reducing the volume of adults presenting with devastating hip fractures remains a primary public health priority worldwide. Standard operational pathways deploy the Fracture Liaison Service (FLS) model to identify and proactively manage secondary fracture risks in adults presenting to hospitals with major osteoporotic fractures. However, even though existing vertebral fractures (VFs) strongly predict future hip fracture occurrences, the vast majority are entirely missed by standard FLS triage.
Data indicates that over 10% of CT scans performed on patients aged 50 years and older contain a vertebral fracture; yet, these findings are rarely reported or linked to preventative care. To scale case identification, automated AI frameworks have been introduced to parse existing CT data. This document tracks a real-world performance evaluation comparing two commercially available AI solutions within the On-Premise Interface (OPI): the Nanox-AI HealthVCF and the newly advanced HealthOST model.
Key Takeaways
Study Overview: Head-to-Head Test Performance in an Oxford Fracture Prevention Setting
To evaluate real-world triage capabilities, a consecutive series of 431 CT scans capturing the thoracic or lumbar spine was pulled between October 1st and October 3rd, 2025, and processed simultaneously by both AI engines.
- Sample Size: 364 consecutive scans successfully completed analysis across both models and were included in the final dataset.
- Method: Both models utilized balanced sensitivity equivalent settings inside the On-Premise Interface (OPI) to generate patient demographic tables and localized sagittal bounding boxes. While HealthVCF builds sagittal outputs from axial views, HealthOST utilizes primary sagittal images to measure precise vertebral height. An experienced FLS nurse reviewed all flagged positive and negative CT scans in a blinded sequence to establish the clinical ground truth.
- Key Metric: Statistical agreement and predictive values. The clinician-confirmed VF prevalence was recorded at 9.6% with HealthVCF and 12.7% with HealthOST. The models exhibited high structural agreement (94.8% overall agreement; Cohen’s kappa of 0.74; Gwet’s AC of 0.93). HealthOST achieved a Positive Predictive Value (PPV) of 43.1% and a Negative Predictive Value (NPV) of 99.2%, outperforming HealthVCF’s 26.1% PPV and 98.4% NPV.
Why This Matters: Driving Scalable FLS Workflows and Proactive Bone Health Triage
The clinical implementation of the advanced HealthOST model provides a robust solution to the historic problem of undiagnosed bone trauma in aging demographics. By refining diagnostic accuracy parameters, the software minimizes manual administrative overhead while securing patient safety nets.
Real-World Impact Highlights:
- Scalable Clinical Ecosystems: Utilizing automated software validated by trained FLS specialist nurses establishes an operational pipeline for hospitals where musculoskeletal radiologist resources are heavily strained.
- Fewer Unnecessary Overheads: HealthOST limits systemic friction by generating 159 false-positive flags per 1,000 scans, compared to 215 generated by HealthVCF, saving valuable clinical checking hours.
- Preventative Care Integration: Identifying patients with low BMD directly during routine spinal reviews provides an early pathway for secondary metabolic interventions before catastrophic hip fractures occur.
Frequently Asked Questions
HealthVCF processes axial source scans to construct an artificial sagittal image output. Conversely, HealthOST parses primary sagittal data directly to measure true vertebral heights and incorporates a dedicated algorithmic flag to highlight instances of low bone mineral density (BMD).
In a head-to-head comparison of 364 scans, HealthOST demonstrated superior performance across all core fields, raising sensitivity to 95.7% and specificity to 81.7%, compared to HealthVCF’s 88.6% sensitivity and 73.3% specificity.
Vertebral fractures serve as an early clinical predictor for life-altering hip fractures. Over 10% of older adults undergoing generic CT scans harbor these hidden fractures, which go unrecognized and unreported in standard, non-automated clinical environments.
Related Resources
- Nanox-AI On-Premise Interface (OPI): Integrating Bone Metrics Seamlessly
- The Fracture Liaison Service (FLS) Blueprint for Secondary Prevention
- Algorithmic Screening: Early Detection of Low BMD in General Populations
Professor M. K. Javaid and researcher R. Eckert lead specialized secondary fracture prevention initiatives at the Nuffield Orthopaedic Centre (Oxford, UK) and coordinate global insights within the International XLH Alliance. Their work centers on pioneering automated clinical pathways to identify systemic osteoporotic indicators early.
Portail client