FRAX Score: 10-Year Fracture Risk Inputs, Country Calibration, and Treatment Thresholds

The FRAX score is not a bone-density measurement and not a diagnosis. It is an absolute-risk calculator: it converts a patient’s age, sex, body build, and a fixed list of clinical risk factors — optionally plus a femoral neck bone mineral density (BMD) value — into two numbers, the 10-year probability of a major osteoporotic fracture and the 10-year probability of a hip fracture. Those two percentages are what treatment-threshold guidance is written against, which is why FRAX shows up in orthopaedic, endocrine and quality-department workflows far more often than in a radiology report. This page covers what FRAX actually computes, its full input set, the country-calibration property that most summaries skip, and — because that is where CASRAI’s audience sits — how the score gets governed, documented and audited inside a hospital rather than how it is completed at the bedside. For the fall-risk counterpart to this kind of structured risk instrument, see CASRAI’s Morse Fall Scale and Johns Hopkins Fall Risk Assessment Tool guides.

What FRAX computes

FRAX returns two probabilities, always over a 10-year horizon, and they are not interchangeable:

• Major osteoporotic fracture (MOF) probability — the 10-year probability of a clinical fracture at any of four sites: clinical spine, forearm, hip, or proximal humerus. Note “clinical”: a vertebral deformity found incidentally on imaging, with no clinical presentation, is not the outcome this figure predicts.
• Hip fracture probability — the 10-year probability of hip fracture specifically, reported separately because hip fracture carries the mortality and cost burden that most intervention thresholds are anchored to.

Both outputs are probabilities, not scores in the point-tally sense that Wells criteria or the Braden Scale produce. There is no total, no cut-point arithmetic a clinician performs, and no “FRAX points.” Charting a bare number (“FRAX 14”) without naming which of the two probabilities it is makes the note unreviewable, and is one of the most common documentation defects a fracture-prevention audit turns up.

The calculation accounts for the competing risk of death, which is why it is age-calibrated rather than a simple risk-factor sum — a 90-year-old and a 55-year-old with identical risk factors do not have proportionally scaled 10-year probabilities.

Provenance and validated age range

FRAX was developed at the University of Sheffield’s WHO Collaborating Centre for Metabolic Bone Diseases, led by John Kanis, and released in 2008. It is commonly called “the WHO fracture risk assessment tool” for that reason, but it is a Sheffield-developed and Sheffield-maintained instrument rather than a WHO publication — a distinction worth getting right in policy documents and patient-facing material, where “WHO-endorsed” overstates the relationship.

FRAX is validated for adults aged 40 to 90 only. The underlying hazard rates are not calibrated outside that band. A tool that silently accepts an age of 35 or 94 and returns a number is returning an extrapolation, not a validated estimate; if your EHR or intranet calculator does not block out-of-range ages, that is a defect worth logging.

The FRAX input set

FRAX takes a fixed, closed list of inputs. Nothing else influences the result — a factor not in this table is not in the model, however clinically relevant it may be.

Input	Form	Notes
Age	Years (40–90)	Or date of birth; outside 40–90 the model is not validated
Sex	Male / female	Selects a different calibration within the country model
Weight	kg	Combined with height into BMI
Height	cm	Combined with weight into BMI
Previous fracture	Yes / no	A prior fragility fracture in adult life, occurring spontaneously or from trauma that would not normally cause a fracture; includes morphometric vertebral fracture
Parent fractured hip	Yes / no	Hip fracture in a mother or father specifically — not any parental fracture, and not a sibling
Current smoking	Yes / no	Current smoker; the model does not grade pack-years
Glucocorticoids	Yes / no	Currently exposed, or previously exposed for more than 3 months at a dose of prednisolone 5 mg/day or equivalent
Rheumatoid arthritis	Yes / no	Confirmed diagnosis; a self-reported or unconfirmed “arthritis” is not the same input
Secondary osteoporosis	Yes / no	A disorder strongly associated with osteoporosis — type 1 diabetes, adult osteogenesis imperfecta, untreated long-standing hyperthyroidism, hypogonadism or premature menopause (<45 years), chronic malnutrition or malabsorption, chronic liver disease
Alcohol, 3 or more units/day	Yes / no	A binary threshold, not a consumption quantity
Femoral neck BMD	Optional — T-score or g/cm²	Femoral neck site only; see below

Three properties of this list drive most of the operational difficulty. First, nearly every clinical risk factor is binary — a patient on 40 mg/day of prednisolone and one on 5 mg/day produce the identical input. Second, several inputs require a confirmed history rather than a patient report, so the quality of a FRAX result is bounded by the quality of the history-taking behind it. Third, the inputs map poorly onto discrete structured EHR fields, which is why FRAX is so often re-entered by hand even in organisations that have automated other risk scores.

BMD is optional — and what changes when you add it

FRAX runs with or without a BMD value. The BMD-free version exists deliberately, so that fracture risk can be assessed in primary care and in settings without densitometry access — it is a screening-tier calculation, and it is legitimate. Adding femoral neck BMD produces a more precise estimate, particularly for hip fracture probability.

Two constraints are frequently violated in practice:

• Femoral neck only. FRAX accepts the femoral neck site. A lumbar spine T-score is not a valid substitute, and entering one produces a number that looks plausible and is wrong. Where a patient’s spine and hip BMD are discordant — common in glucocorticoid exposure, where trabecular bone at the spine is lost disproportionately — FRAX will under-represent the risk, and that discordance needs handling outside the base calculation.
• One BMD, one entry. Entering both a T-score and a density value, or mixing the T-score reference database the scan reported against, introduces error that no downstream check will catch.

Country-specific calibration: FRAX is not one model

This is the property most commonly lost in summary treatments of FRAX, and it has direct governance consequences. FRAX is not a single global algorithm. It is a family of more than 70 country-specific models. The risk-factor weightings derive from meta-analysed international population cohorts, but each country model is then calibrated to that country’s own fracture incidence and mortality data. Several countries additionally publish ethnicity-specific calibrations — the US model, for instance, is stratified rather than single.

The practical consequences:

• The same patient, same inputs, different country model, different probability. Hip fracture incidence varies by an order of magnitude across populations. The output is only meaningful against the model that matches the patient’s population.
• Model selection is a real clinical decision, not a form default. For a patient who has migrated, guidance is to use the model for the country of residence, since fracture risk tracks the current environment more closely than country of origin. Any organisation whose calculator hard-codes one model without exposing that choice has removed a decision the clinician is supposed to make.
• Intervention thresholds are country-specific too, and do not travel with the number. A threshold written for one national guideline is not automatically valid against another country’s model output.
• Countries without a validated model exist. Where no national model has been developed, guidance is to use a surrogate from a country with comparable fracture epidemiology — and to record that a surrogate was used, because the resulting estimate carries an additional layer of assumption.

For a quality department, the auditable question is simple and rarely asked: which country model does our tooling use, who chose it, and is it recorded on the result? A FRAX percentage in a chart with no model attribution is not fully interpretable after the fact.

FRAX and DXA: how the two combine in a treatment decision

FRAX and DXA scanning are frequently presented as alternatives. They are not — they occupy different positions in the same pathway, and the sequencing is what determines whether FRAX changes any decision at all.

The osteopenia gate

The situation FRAX was designed to resolve is the diagnostic middle ground. Where DXA already shows a T-score of −2.5 or below, the patient meets the densitometric definition of osteoporosis and treatment discussion follows from that; FRAX is not what drives the decision. Where BMD is normal, there is generally no decision pending. The band in between — osteopenia, a femoral neck, total hip or lumbar spine T-score between −1.0 and −2.5 — is where a large population sits, where most fragility fractures actually occur, and where BMD alone gives no clear direction. That is FRAX’s job: to separate the osteopenic patients whose overall clinical risk profile justifies pharmacological treatment from those whose does not.

Read the other way: FRAX adds most where DXA is least decisive, and adds least where DXA has already answered the question. A protocol that runs FRAX on every DXA result regardless of T-score generates numbers that do not inform anything, and dilutes attention to the band where the calculation matters.

Intervention thresholds

In the United States, the Bone Health and Osteoporosis Foundation (BHOF, formerly the National Osteoporosis Foundation) sets the widely-used fixed thresholds. In postmenopausal women and men aged 50 and over with osteopenia, pharmacological treatment is recommended where the FRAX-calculated 10-year probability is:

• ≥3% for hip fracture, or
• ≥20% for major osteoporotic fracture

Four qualifications belong with those figures whenever they are quoted:

1. They are US thresholds against the US model. They are not universal. The UK’s National Osteoporosis Guideline Group, for example, uses an age-dependent intervention threshold rather than a single fixed percentage, on the reasoning that a fixed cut-point behaves differently across the age range. Other national guidelines set their own. Do not apply a percentage from one country’s guideline to another country’s model output.
2. They are conditioned on the osteopenic band, calculated with BMD. Applying them to a BMD-free FRAX result, or to a patient already below −2.5, is using them outside the population they were derived for.
3. They are a threshold for consideration, not an automatic order. Crossing 3% or 20% opens a treatment discussion that still involves patient preference, comorbidity, renal function, and the risks of the specific agent.
4. They are a floor, not a ceiling. A patient below threshold with a recent fragility fracture is not thereby low-risk — see the limitations below.

Known limitations that matter in a quality review

FRAX’s limitations are well-characterised, published by its own developers, and they are the source of most of the cases where a FRAX result and clinical judgement diverge. A fracture-prevention programme that treats the number as sufficient will systematically under-treat a predictable set of patients.

• Falls are not an input. FRAX contains no falls variable at all, despite falls history being among the strongest predictors of fracture. This is the single most important gap for hospital patient-safety work, because it means a FRAX result and a fall-risk assessment are answering different questions and neither substitutes for the other. A patient with a low FRAX and a high fall risk still needs the intervention that a fall-prevention programme provides.
• Glucocorticoid dose is not graded. The input is binary, so high-dose and prolonged exposure are under-weighted relative to their real effect, and the BMD-independent effect of glucocorticoids on bone quality is not captured.
• Prior fracture number and recency are not graded. One fracture 30 years ago and three in the last two years produce the same input. Fracture risk is markedly elevated in the period immediately following a fragility fracture — the “imminent risk” window — and FRAX does not represent it.
• Lumbar spine BMD is not used, so spine/hip discordance is invisible to the base calculation.
• Type 2 diabetes is under-represented. It confers fracture risk beyond what its BMD effect predicts, and it does not qualify as “secondary osteoporosis” in the input list (type 1 does).

FRAXplus and arithmetic adjustments

Two documented routes exist for handling these gaps. Published arithmetic adjustments allow a computed FRAX probability to be adjusted upward for high-dose glucocorticoid exposure, discordantly low lumbar spine BMD, type 2 diabetes, and falls history. Separately, FRAXplus is an extension of the tool that incorporates refinements including falls history, glucocorticoid dose, trabecular bone score (TBS), hip axis length, fracture recency, and spine–hip BMD discordance directly.

The governance point for a hospital: if adjustments are being applied, whether they were applied and which ones must be recorded alongside the result. An adjusted and an unadjusted probability for the same patient are different numbers, and a chart that records only the final figure cannot be reconciled later.

Building FRAX into hospital workflow

Where FRAX generates measurable organisational value is not the individual calculation — it is secondary fracture prevention, which is a recognised care gap rather than a knowledge gap. A patient who presents with a fragility fracture is, by that fact, at high risk of the next one, and a substantial proportion are discharged with the fracture treated and the underlying osteoporosis never assessed.

Fracture liaison and case-finding

The standard organisational response is a fracture liaison service (FLS) or equivalent case-finding pathway: systematic identification of patients presenting with fragility fracture, structured risk assessment including FRAX and DXA where indicated, and a defined handoff for treatment initiation and follow-up. The instrument matters less than the systematic identification — FRAX is the assessment step inside the pathway, not the pathway. Design considerations that determine whether it works:

• Trigger definition. Which presentations enter the pathway, from which entry points (ED, inpatient orthopaedics, outpatient), and who owns the identification. A trigger tied only to admitted hip fracture misses the wrist and humerus presentations that predict it.
• Ownership of the assessment. FRAX inputs require history-taking that emergency and orthopaedic workflows do not naturally produce. Assign it explicitly.
• Closing the loop. Case-finding that ends at a calculated probability with no defined route to treatment initiation and follow-up produces documentation, not prevention.

Clinical decision support

Embedding FRAX in decision support raises the same issues as any risk-score CDS implementation, plus one specific to this tool:

• The country model must be explicit and correct, and ideally recorded on the stored result — the point made above, and the failure mode unique to FRAX.
• Age-range enforcement. Block or flag calculation outside 40–90 rather than returning an unvalidated extrapolation.
• Input provenance. Where inputs are auto-populated from problem lists or medication records, the mapping is a clinical decision with error modes — a resolved historical steroid course auto-firing the glucocorticoid flag, or an unconfirmed “arthritis” problem-list entry firing rheumatoid arthritis. Auto-population that cannot be reviewed and overridden at the point of calculation converts a data-quality problem into a risk-assessment error.
• Store inputs, not just the output. A stored probability with no record of the inputs that produced it cannot be re-derived, corrected, or audited.

Documentation standards

A defensible FRAX entry records, at minimum: which probability (MOF or hip, or both), the country model used, whether BMD was included and from which site and scan, whether any adjustment was applied, and the date. That is more than most templates capture, and the gap is exactly what makes retrospective review of fracture-prevention performance difficult. The same discipline applies to any structured risk instrument — see hospital patient safety plan for how instrument governance fits the wider programme, and CASRAI’s patient safety hub for related assessment tools including the Clinical Frailty Scale, Katz ADL Index, and Charlson Comorbidity Index.

Frequently asked questions

What is a normal or good FRAX score?

There is no “normal” FRAX value, because FRAX does not produce a score against a reference range — it produces two probabilities that are interpreted against an intervention threshold, and that threshold is country-specific. In the US, a 10-year hip fracture probability below 3% and a major osteoporotic fracture probability below 20% are, in an osteopenic patient, below the BHOF treatment threshold. That is not the same as “normal,” and a below-threshold result does not mean low risk in a patient with a recent fracture or a high falls risk.

Do you need a DXA scan to use FRAX?

No. FRAX is designed to run without BMD, which is what makes it usable in primary care and in settings without densitometry. Adding femoral neck BMD improves precision, particularly for hip fracture probability. The published US intervention thresholds, however, are framed around osteopenic patients — a determination that itself requires a DXA result.

Is FRAX the same in every country?

No, and this is its most consequential property. FRAX is a family of more than 70 country-specific models, each calibrated to that country’s fracture incidence and mortality data, with some countries further stratified by ethnicity. The same inputs against different country models return different probabilities, and intervention thresholds are set nationally as well.

Does FRAX account for falls?

No. Falls history is not among FRAX’s inputs, despite being a strong fracture predictor. This is a documented limitation and the reason FRAX cannot substitute for a fall-risk assessment. Published arithmetic adjustments and the FRAXplus extension both offer routes to incorporate falls history explicitly.

What age range is FRAX valid for?

40 to 90 years. Outside that range the model’s baseline hazard rates are not calibrated and the result is an extrapolation rather than a validated estimate.

Can FRAX be used to monitor treatment response?

FRAX is a risk-assessment instrument for treatment-initiation decisions, not a treatment-monitoring tool. Its inputs are largely static or slow-moving, and it was not derived or validated to detect change on therapy. Monitoring uses repeat BMD and, where appropriate, bone turnover markers on the schedule the treating guideline specifies.

References

• FRAX Fracture Risk Assessment Tool, Centre for Metabolic Bone Diseases, University of Sheffield — frax.shef.ac.uk (the calculator, country model list, and the developers’ own risk-factor definitions)
• FRAXplus — fraxplus.org (the extension incorporating falls, glucocorticoid dose, TBS, hip axis length, fracture recency, and spine–hip discordance)
• Bone Health and Osteoporosis Foundation, Clinician’s Guide to Prevention and Treatment of Osteoporosis (2022) — source of the US ≥3% hip / ≥20% major osteoporotic fracture intervention thresholds
• International Osteoporosis Foundation, FRAX: clinical uses, adjustments and developments — osteoporosis.foundation
• National Osteoporosis Guideline Group (UK), Section 3: Fracture risk assessment and case finding — nogg.org.uk (the age-dependent intervention threshold approach)