Skip to main content
. 2021 Apr 24;14:101070. doi: 10.1016/j.bonr.2021.101070

Fig. 2.

Fig. 2

Data selection flowchart. The OSTPRE study originally involved 14,220 elderly females from Kuopio region. 2497 participants from the OSTPRE study and 614 non-overlapping participants from a side-study OSTPRE-FPS had DXA scans taken between 2003 and 2007. These subsets were combined and linked with the fracture data from postal enquiries in the years 1994, 1999, 2004, 2009, and 2014. The fractures reported by the participants were verified from the national patient registers. Some participants were excluded due to lost or corrupted measurement data. The DXA images of the final dataset were labelled for scoliosis and unreliability by two radiologists. This resulted in 2949 labelled DXA-images in the OSTPRE dataset. An external non-overlapping dataset of 574 random patients was gathered from Kuopio University Hospital (KUH) and labelled for scoliosis and unreliability by three radiologists. The number and proportion of positive (1) and negative (0) samples in each case are shown in the final dataset boxes.