In the original publication by Livingstone and colleagues (BMJMED 2022;1:e000316. doi:10.1136/bmjmed-2022-000316, published 25 October 2022),1 the methods stated that the QFracture model evaluated was QFracture-2016 but this is incorrect. The model evaluated in the paper was QFracture-2012, which is the model described in previous peer reviewed publications.2 3 However, QFracture-2016 is the version available in the current online risk predictor provided by QResearch and automatically calculated in some clinical IT systems in UK general practices. The authors accessed the QFracture batch calculator via the QFracture-2016 webpage, and erroneously assumed that the calculator was for QFracture-2016. However, the batch calculator implemented QFracture-2012 not QFracture-2016 (the batch calculator instructions stated the version, so the error is the authors). The published analysis is valid as an external validation of QFracture-2012, but the implications for practice are somewhat different because QFracture-2016 has an important difference to QFracture-2012.
QFracture-2016 derivation and internal validation have not been published as peer reviewed papers but are briefly summarised on the QResearch website,4 and QFracture-2016 has been externally validated in people with chronic obstructive pulmonary disease in the UK, finding poorer discrimination than in internal validation in the whole population.5 The key difference is that QFracture-2016 uses both data recorded by general practitioners and those recorded at hospital discharge for fracture ascertainment, whereas QFracture-2012 only uses fractures recorded by general practitioners. Fracture ascertainment in QFracture-2016 is therefore the same as in the published external validation. The total hip fracture rates observed in QFracture-2016 derivation4 are very similar to the observed hip fracture rates in the published article (for women, QFracture-2016 2.31/1000 person years of follow-up v 2.30/1000 in the external validation1; for men, 0.86/1000 v 0.88/1000). However, observed major osteoporotic fracture rates in QFracture-2016 derivation are still somewhat lower than those in the published article (for women, QFracture-2016 5.27/1000 person years of follow-up v 6.12/1000 in the external validation; for men, 1.92/1000 v 2.26/1000). This difference is likely at least partly explained by QFracture derivation using data from 1998 onwards whereas the paper’s analysis used data from 2004 onwards (observed major osteoporotic fracture rates in QFracture-2016 are lower in the period 1998-2003 than in 2004–15; for women, 4.35 per 1000 person years in 1998–2003 v 5.69/1000 in 2004-15; for men, 1.40/1000 v 2.16/1000). Similar to the published external validation, the QFracture-2016 summary document also shows that QFracture-2012 under-predicts when outcomes are measured using both fractures recorded by general practitioners and at hospital. The observed under-prediction in the published article is therefore likely largely (hip fracture) or mostly (major osteoporotic fracture) corrected by the QFracture-2016 update, which is the version of the tool currently used in the UK.
There is a need to externally evaluate the QFracture-2016 model and (if possible) the FRAX (fracture risk assessment tool) model.
The main text has been changed to state the correct (QFracture-2012) version throughout, except where statements are made that refer to all three versions of the QFracture model (eg, in terms of not accounting for competing mortality). The abstract and discussion, conclusion, and what this study adds sections have been amended to reflect the implications of the differences between the 2012 and 2016 versions of QFracture.
