A clinical decision support tool for accurate hip fracture prediction: A nationwide cohort study
PLoS Medicine · 27 Aug 2026
† Impact is the editorial-priority score. All scores are out of 100. Clinical is a stored clinical-importance score from the earlier method; it has not been recalculated.
Why it matters
Orthopaedic surgeons and osteoporosis clinics may eventually use registry-based screening tools like this to identify at-risk patients earlier for fracture prevention, changing how referrals into osteoporosis/fall-prevention pathways are triggered.
The evidence
Researchers used Swedish national registry data on 3.5 million adults aged 50+ to build a machine-learning model (FRACTURE-ML) that predicts hip fracture risk using only routinely collected data, no patient interview or exam needed. In a held-out validation cohort, the model achieved an AUC of 0.89 at 1 year and 0.85 at 5 years, and identified about seven times more at-risk people than the current fracture liaison service (secondary prevention) approach at 2 years (sensitivity 0.84 vs 0.12), with only a modest drop in specificity (0.79 vs 0.98).
A coherent primary effect estimate is unavailable. Clinical relevance does not estimate treatment benefit.
Why this rigour score?
Retrospective case seriesDesign starting point: 8.8
Raised the score
- Adequately powered+6
- Competing risks handled+4.3
Lowered the score
No negative adjustments recorded.
Methods at a glance
Recorded assessment- Confounding adjustmentNot applicable
- Covariate balance confirmedNot applicable
- Outcome assessors blindedNot applicable
- Comparable start of follow-upMet
- Adequate statistical powerMet
- Prospective registrationNot met
Not reported does not mean not done. Checks have different weights; this is not a count of boxes ticked.
See the full score breakdown
- Study design starting point
- 8.8
- Adequately powered
- +6
- Competing risks handled
- +4.3
- Journal prior (not an observed method)
- +1.5
- Final rigour score rounded
- 20 / 100
Study design sets the starting point and a ceiling of 42. Recorded methods then adjust the score. Points are shown to one decimal place; the final score uses unrounded values.
Checklist from the stored appraisal. “Not met” describes the recorded criterion; it does not always produce a deduction. This overview supports reading the full study.
For journal club
1.If a tool like FRACTURE-ML flagged many more patients as high-risk than current fracture liaison service criteria, how should orthopaedic and primary care teams handle the resulting increase in referrals for DXA and osteoporosis workup?
2.Given that the model has not been externally validated outside Sweden or tested in real clinical workflows, would you support using such a risk score to guide bone density testing or osteoporosis treatment decisions in your practice today?
Original publication title
A clinical decision support tool for accurate hip fracture prediction: A nationwide cohort study