Under the revised ASBMR definition the fracture must lie along the femoral diaphysis, from just distal to the lesser trochanter to just proximal to the supracondylar flare, and at least four of five major features must be present. These are: association with minimal or no trauma; a fracture line that originates at the lateral cortex and is substantially transverse, though it may become oblique as it crosses medially; a complete fracture extending through both cortices, possibly with a medial spike, or an incomplete fracture involving only the lateral cortex; no or minimal comminution; and localised periosteal or endosteal thickening of the lateral cortex at the fracture site, often described as beaking or flaring (Shane et al., 2014). Minor features, none of them required, include generalised cortical thickening of the shafts, prodromal pain, bilateral involvement and delayed healing. Femoral neck, intertrochanteric and periprosthetic fractures and pathological fractures through tumour are excluded.

The link with bisphosphonates is supported by consistent observational evidence, although causation in the individual patient cannot be proved. A Swedish national study with radiographic adjudication found a strong relative association between bisphosphonate use and atypical fracture with a small absolute risk (Schilcher et al., 2011). A secondary analysis of the major randomised trials found too few subtrochanteric and diaphyseal fractures to confirm or exclude an effect (Black et al., 2010). Data from a large health maintenance organisation in California showed that incidence rises with duration of use (Dell et al., 2012), and later work indicated that risk falls quickly after the drug is stopped (Schilcher et al., 2014). The proposed mechanism is that prolonged suppression of remodelling allows microdamage to accumulate at the lateral cortex, where tensile stress is highest, but this remains a hypothesis (Shane et al., 2014). Atypical fractures also occur in people who have never taken an antiresorptive drug.

The lesion behaves as a stress fracture. It begins as a focal thickening of the lateral cortex, may develop a transverse radiolucent line, and can progress to a complete fracture after trivial load. Pain in the thigh or groin precedes complete fracture in a large proportion of reported patients (Kwek et al., 2008). Cortical stress lesions with a visible lucent line and accompanying pain appear more likely to progress than painless thickening alone (Koh et al., 2010). The opposite femur is affected in a substantial minority of patients, sometimes without symptoms, which is the reason every published recommendation calls for it to be imaged (Shane et al., 2010).

Any patient who has taken a bisphosphonate or another potent antiresorptive agent for several years and reports a dull or aching pain in the thigh or groin should be regarded as having a possible stress lesion of the femur until imaging shows otherwise. The pain is typically worse on weight bearing and may have been present for weeks. In practice it is commonly ascribed to lumbar spine disease, hip arthritis or trochanteric bursitis. The pathway therefore begins with a prompt in the history: duration and type of osteoporosis treatment should be asked about in every older patient with unexplained thigh pain.

Anteroposterior and lateral radiographs should include the whole length of both femora, from hip to knee. A radiograph of the pelvis or hip alone will miss a diaphyseal lesion. The reader looks for a focal bump on the lateral cortex, with or without a transverse lucent line. If radiographs are normal or equivocal and suspicion remains, magnetic resonance imaging or a radionuclide bone scan can show a stress reaction before cortical change is visible, and the long femur images produced by some dual-energy X-ray absorptiometry scanners have been used for the same purpose (Shane et al., 2014).

The radiographs are assessed against the major features listed above. The step matters because ordinary osteoporotic fractures of the subtrochanteric region and shaft are more common than atypical ones and are spiral or comminuted in pattern, and because metastatic disease must not be mistaken for either. Recording which features are present, in the notes and in the operative record, also makes later audit and reporting to pharmacovigilance schemes possible.

Lesions are placed in one of three groups. A complete fracture needs fixation. An incomplete fracture with pain, and particularly one with a visible lucent line, carries a meaningful risk of completion (Koh et al., 2010). An incomplete fracture without pain, seen as cortical thickening alone, is the least likely to progress. Both femora are stratified separately, since a patient may have a complete fracture on one side and an incomplete lesion on the other.

For complete fractures the task force recommends intramedullary nailing, which protects the whole femur and allows healing by endochondral repair, a process less dependent on the remodelling that antiresorptive drugs suppress than the direct healing sought with rigid plate fixation (Shane et al., 2010). Surgeons should expect technical difficulty from the bowed, brittle femur and a narrow canal, and should counsel patients that union is usually achieved but is slower than after an ordinary femoral shaft fracture (Egol et al., 2014).

For incomplete fractures with pain, the task force advises that prophylactic nailing be considered; where the pain is minimal, a trial of conservative treatment with limited weight bearing is reasonable, with fixation if symptoms and imaging have not improved after two to three months (Shane et al., 2010). A comparative series supports this caution, since most incomplete fractures managed without surgery in that report progressed to complete fracture, and hospital stay was shorter after prophylactic fixation than after fixation of a completed fracture (Banffy et al., 2011). For incomplete fractures without pain, the pathway follows the task force in advising reduced loading and repeat imaging until the lesion has resolved. The choice in each case belongs to the surgeon and the patient, and the evidence behind it consists of case series, not trials.

An orthopaedic pathway should not end at the wound. The task force recommends that potent antiresorptive treatment be stopped once an atypical fracture is diagnosed, that calcium and vitamin D status be assessed and corrected, and that the further management of osteoporosis be reviewed (Shane et al., 2014). These decisions lie with the physician responsible for the patient's bone health, and the pathway's requirement is only that the referral is made and documented before discharge. The use of teriparatide to promote healing has been described in case reports and small series; the task force regarded the evidence as limited and inconclusive, and this article makes no recommendation on it. General guidance on reviewing the need for continued bisphosphonate treatment after several years, including the concept of a drug holiday for patients at lower risk, is given by a separate ASBMR task force (Adler et al., 2016) and draws on the extension of the alendronate trials (Black et al., 2006).