The prognostic value of vertebral fractures on chest CT

Abstract

Osteoporosis is a large and growing problem in the aging world population, causing bone fractures. Preventative treatments for osteoporosis are available but a cure is not. This makes early detection important in managing the disease. Vertebral fractures and vertebral deformities are an early marker of osseous deterioration. Vertebral fractures have also been associated with all-cause mortality and cardiovascular diseases. We investigated vertebral fractures in a large multicenter cohort of chest computed tomography (CT) scans from routine radiological practice; the PROVIDI study. To demarcate the scope of the problem we investigated the prevalence of clinically relevant unrequested findings on cardiac CT. We found nineteen radiological studies comprising 12922 patients. The pooled prevalence of clinically relevant unrequested extra-cardiac was high, at 13%, but with a large variation between the different studies. We then set out to investigate the reliability and reproducibility of vertebral fracture assessment on CT. We found acceptable reproducibility interobserver Cohen's kappa's of 0.56-0.81 for the presence of fracture. We found good interobserver (74-88%) agreement, and excellent interobserver reliability with square-weighted kappa's of 0.84-0.94 for the severity of vertebral fractures. We used the PROVIDI cohort to investigate whether vertebral are associated with future hip fractures. Patients with any vertebral fracture had a roughly tripled the risk of future hip fracture. We found an age- and gender-adjusted hazard ratio of 3.1 (95% CI 2.1-4.7). This hazard ratio rose to 3.8 (CI 2.6-5.6) if mild fractures were discounted, as they sometimes are in the literature. The PROVIDI cohort further showed that prevalent vertebral fractures conferred an elevated risk of cardiovascular events with an adjusted HR of 1.28 (CI 1.07 - 1.54). This effect remained moderate after correction for cardiovascular calcifications (HR 1.20, CI 0.99 to 1.44). However, in terms of discrimination, vertebral fractures did not have substantial incremental prognostic value on top other available predictors (C-index was 0.683 versus 0.682 for models with and without vertebral fractures respectively). Recent work has shown that vertebral density on CT correlates well with DXA defined bone mineral density. We performed an external validation study of the diagnostic performance of CT for DXA scores and we also assessed the relation of prevalent vertebral fracture with vertebral density and DXA score. The diagnostic performance for vertebral HU measurements was substantial, with an AUC of 0.74 (0.68 – 0.80). This confirmed the diagnostic value of vertebral density for low BMD, albeit with a lower AUC than the 0.83 [95% CI, 0.81–0.85]) that had been previously reported. To investigate the link between all-cause mortality, vertebral density and vertebral fracture status we turned to the NELSON cohort, a lung cancer screening study conducted amongst smokers. Vertebral fractures were independently predictive of mortality, with an adjusted hazard ratio of 2.04 (1.43–2.92). Finally we conducted a simulation study to investigate different approaches towards analyzing case-cohort data (the study design used for the PROVIDI cohort), in the context of prognostic modelling. We found that various pragmatic and user-friendly adaptations of the analysis techniques yielded equally valid results as established techniques.