Osteonecrosis · Hip · Lower limb

Avascular necrosis of the femoral head: what a normal radiograph does not tell you

A painful hip in a patient of forty on corticosteroids, with a radiograph read as normal. It is the most ordinary presentation of osteonecrosis, and it is also the one that costs the most: at that stage the disease is reversible, and the investigation that sees it is not the one that was requested.

Why should a painful hip with a normal radiograph cause concern?

This chapter sets out the central problem of the disease, which is not therapeutic but diagnostic. Osteonecrosis is distinguished from referred low back pain, early hip osteoarthritis or gluteal tendinopathy neither by the site of the pain nor by its quality. It is distinguished by the background, and by an investigation that is not requested.

The opening pain of osteonecrosis is a mechanical hip pain, most often in the groin, sometimes in the buttock or referred to the front of the thigh and the knee. There is nothing specific about it. What is specific, on the other hand, is the combination of three elements: a patient younger than the osteoarthritic one, an identifiable exposure, and first-line imaging that shows nothing.

What the Chinese database actually counted

The most useful work on this question is not a case series but a case-control study on the diagnosis itself. Li and colleagues extracted 1,471 patients from the China Osteonecrosis of Femoral Head Database, recruited between July 2016 and December 2018, and separated them according to whether or not they had received an incorrect first diagnosis, defined by a different second diagnosis within six months with unchanged symptoms, or by disagreement with the expert group⁸.

Three hundred and three patients, that is 20.6 %, had been mislabelled. The detail of the errors is worth more than a long discussion: 118 as lumbar disc herniation, 86 as hip synovitis, 48 as hip osteoarthritis, 32 as rheumatoid arthritis, 11 as piriformis syndrome, 5 as sciatica, 3 as soft tissue injury⁸.

The risk factor for error is technical, and it has been measured

The multivariate analysis in that same study is the result that should change practice. Having done only a radiograph at the first examination multiplies the risk of getting it wrong by 4.74 (95 % CI 3.16 to 7.12). Conversely, having performed an MRI straight away is protective (OR 0.65, 95 % CI 0.43 to 0.99)⁸.

Two other factors are protective, and they say something about the history: a known history of corticosteroid therapy (OR 0.30) and documented alcohol consumption (OR 0.31)⁸. In other words, when the clinician knows the exposure, they think of the diagnosis. The delay does not come from ignorance of the disease, it comes from a history that has not asked the two questions.

Two factors finally increase the risk of error: subtle symptoms (OR 1.55) and the grade of the practitioner at the first examination, a trainee or registrar rather than a senior (OR 3.32)⁸.

Stage I is defined by a normal radiograph

This is not a nuance, it is the definition. The 2019 revision of the ARCO classification, obtained by a four-round Delphi bringing together 29 international experts with response rates of 93.1 to 100 %, states stage I thus: “the radiograph is normal, but MRI or bone scan is positive”⁹.

A normal radiograph therefore does not reduce the probability of osteonecrosis in a patient at risk: it only excludes stages II to IV. That is exactly the opposite of the spontaneous reasoning, which treats normal imaging as a reassuring element.

The approach, in order

From those three elements (the background, the definition of stage I, and the measured risk factor for error) follows a simple course of action, which adds no unnecessary investigation but moves one.

Who develops osteonecrosis, and from what exposure?

Two exposures dominate, and they do not carry the same weight. This chapter gives the figures that allow the history to be prioritised, and the result that surprises most: in a patient already on corticosteroids, alcohol adds almost nothing.

What the Japanese registry measures

Japan registers non-traumatic osteonecrosis nationally, which produces data that few countries have. Ikeuchi and colleagues examined over three years the records and images of Aichi prefecture, 7.4 million inhabitants. Of 327 registration applications, 285 met the diagnostic criteria : 42 records, that is 12.8 %, were not osteonecroses¹.

In those 285 patients: mean age 50.4, male-to-female ratio of 2.1 to 1. The aetiological breakdown is the figure to remember.

Corticosteroids: a dose relationship, not a bolus effect

The reference demonstration dates from 1987 and has not been contradicted. Felson and Anderson brought together 22 publications giving both the mean corticosteroid dose of a cohort and the percentage of patients in it who developed osteonecrosis, then correlated the two at the level of the studies³.

The result comes down to two lines. The total daily dose is strongly correlated with the osteonecrosis rate (r from 0.61 to 0.80), and the oral dose more so still (r from 0.70 to 0.86). By contrast, the dose given as a bolus is not associated with risk³.

The authors themselves point out why the individual studies saw nothing: within a given centre, cases and controls receive similar protocols and therefore barely differ in dose. Comparison had to be made across studies for the relationship to appear. That is a methodological limitation to keep in mind: this correlation is ecological, it concerns cohort means, not individuals. It establishes that a dose-effect relationship exists; it provides no individual threshold below which a patient would be safe.

The window of the first twelve months

Koo and colleagues reviewed, across four university hospitals, the records of patients whose corticosteroid-induced osteonecrosis had been caught at ARCO stage I, that is with a positive MRI and normal radiographs. Twenty-two patients, aged 17 to 60, mean 33⁴.

The cumulative dose at the time of detection ranged from 1,800 to 15,505 mg of prednisolone equivalent, mean 5,928 mg. The interval between the start of treatment and the MRI diagnosis ranged from 1 to 16 months, mean 5.3 months. And above all: 21 of the 22 patients were diagnosed in the first twelve months⁴.

Alcohol: a continuous dose-response relationship

Yoon and colleagues conducted a dose-response meta-analysis on five Japanese case-control studies, 1,251 subjects, modelling the relationship with restricted cubic splines⁵.

Current consumption is associated with risk: OR 3.63 in the occasional drinker, 5.90 in the daily drinker. And the relationship is indeed graded: risk increases by 35.3 % per 100 g a week (95 % CI 1.24 to 1.47) and by 44.1 % per 500 g-years (95 % CI 1.295 to 1.601). The authors specify that the shape of the relationship is non-linear⁵.

The result that sets the hierarchy: alcohol does not add to corticosteroids

Fukushima and colleagues conducted a Japanese multicentre hospital-based case-control study, 71 cases of idiopathic osteonecrosis against 227 matched controls, to look specifically for an interaction between the two exposures⁶.

The result is clear-cut and counter-intuitive. Across all subjects, consumption of at least 3,032 drink-years gives an OR of 3.93 (95 % CI 1.18 to 13.1). But after stratification: that OR is 11.1 in those who have never taken corticosteroids (95 % CI 1.30 to 95.5), and 1.10 in those who have (95 % CI 0.21 to 4.79). On a two-way table, corticosteroids alone give an OR of 31.5, and the two combined an OR of 31.6, that is, no additional increase. Neither multiplicative interaction (p = 0.19), nor additive (synergy index 0.95)⁶.

What COVID changed, and what it did not

The massive corticosteroid courses used in severe COVID-19 raised fears of a wave of osteonecroses. The most solid data comes from the Japanese multicentre database, which compared 20 patients (32 hips) who developed osteonecrosis after COVID with 693 patients with corticosteroid-associated osteonecrosis not linked to COVID over the same period⁷.

The COVID group had a much shorter course of corticosteroids (2.3 months against 44.9), at a lower maximum dose (28.1 against 53.2 mg a day), and less often as a bolus (15.0 % against 49.4 %). It had more men, and a higher proportion of type C2 in the Japanese classification, that is of laterally extensive lesions⁷.

The authors' conclusion is cautious and deserves to be taken up as it stands: it must be recognised that osteonecrosis can occur even when the corticosteroid course has been brief and moderate, in a patient who has had COVID, and therefore that a musculoskeletal symptom in that context should be investigated early. Twenty patients out of a database of 5,371 do not constitute an epidemic; it is a signal, described as such.

Should screening be done, and in whom?

The question of systematic screening is settled by no trial, and that has to be said from the outset. None of the sources in this base establishes that systematically imaging patients started on corticosteroid therapy improves their outcome. What the data do allow is to delimit a population and a window where the yield of an MRI would be highest if it were decided to request one.

The three elements overlap. The window is that of the first twelve months of corticosteroid therapy, where 21 of the 22 diagnoses made at ARCO stage I are concentrated⁴. The population is that of prolonged high daily doses, the dose-effect relationship being established on cumulative oral dose and not on boluses³. And the expected benefit is concentrated on hips that are still round, since joint preservation is mainly effective before structural collapse¹⁰.

Where the reasoning tightens markedly is in a patient already diagnosed on one side. That is no longer population screening but the investigation of a patient at known risk, and two sources support it: contralateral MRI is explicitly recommended by the 2026 update¹⁰, and the S3 guideline specifies that contralateral risk is concentrated in the first two years and then becomes unlikely¹¹. An MRI at diagnosis and two years of clinical surveillance are therefore enough to cover most of the risk, without repeated imaging indefinitely.

What the disease costs, and why it is poorly counted

Two orders of magnitude, from two national registries, place the problem.

In Japan, the estimated annual incidence is 1.91 per 100,000 inhabitants¹. In South Korea, analysis of the health insurance database over 2013-2022 shows a crude prevalence starting from 74.1 per 100,000 adults, a fall in 2020 attributed by the authors to restricted access to care during the pandemic, then a rise again. Total medical cost went from 26.7 to 41.2 billion won over the period².

Two cautions before transposing. First, these two countries register the disease nationally, which is not the case everywhere: the gap between 1.91 (incidence) and 74.1 (prevalence) does not measure a difference in frequency between Japan and Korea, but two different indicators. Second, no French incidence data exists in this base, and corticosteroid prescribing and alcohol consumption habits vary enough from one country to another to forbid a direct extrapolation.

The figure from the Aichi registry that speaks most to the clinician is elsewhere, and it is humbling: of 327 records submitted for registration, 42 (12.8 %) did not meet the diagnostic criteria¹. The diagnosis is missed in both directions: by default in one patient in five⁸, by excess in one record in eight.

What becomes of an untreated femoral head?

This is the question every decision depends on, and the answer is not single. Some hips collapse inexorably, others stop along the way, and what separates them is measured on the initial MRI. This chapter gives the two series that make sorting possible.

Asymptomatic hips are not quiet hips

An asymptomatic hip with osteonecrosis is most often discovered as the contralateral side of a painful hip. Should it be treated? Mont and colleagues conducted a systematic review to answer: 16 studies, 664 asymptomatic hips¹³.

Overall, 394 hips out of 664, that is 59 %, progressed to symptoms or collapse. But the stratification is what counts: small medial lesions have a prevalence of collapse below 10 %. Conversely, medium and above all large, or lateral, lesions progress in a substantial proportion of patients. The background modulates it too: sickle cell disease gives the most frequent progression, lupus the most benign course¹³.

And collapse can stop

The counterpart is just as important, and it is less well known. Nishii and colleagues followed for at least five years 54 hips with no initial collapse or with a simple crescent sign¹⁴.

Twenty-eight hips, that is 52 %, collapsed. But among those 28, the collapse stopped in 15 cases, that is 54 %, and notably in 8 of the 9 hips with a small necrotic lesion (89 %), that is occupying less than the medial two-thirds of the weight-bearing area. Of those 15 stabilised hips, 11 stayed under 2 mm of collapse and 10 became asymptomatic again with no surgery at all¹⁴.

The Kerboul angle: the best discriminating predictor

Ha and colleagues tested the hypothesis that the combined necrotic angle, measured no longer on radiographs but on mid-coronal and mid-sagittal MRI, predicts collapse. Thirty-seven hips at an early stage in 33 consecutive patients, followed until collapse or for a minimum of five years¹².

The separation is clear-cut. By grade: none of the 5 grade 1 hips (less than 200 degrees) collapsed; 6 of the 9 grade 2 hips did; all 16 grade 3 hips and all 7 grade 4 hips collapsed before 36 months. Expressed as continuous thresholds: none of the 4 hips at 190 degrees or less collapsed, all 25 hips at 240 degrees or more collapsed, and 4 of the 8 hips between the two¹².

The timetable: two years, then the contralateral risk dies away

The German S3 guideline, built on 3,715 publications screened of which 159 were retained under SIGN criteria, gives two useful time markers. Untreated, the subchondral fracture commonly occurs within two years. And it is during that same period that the risk of contralateral involvement is high: beyond it, it becomes unlikely¹¹.

That second piece of information has a direct practical consequence: it bounds the surveillance. A patient whose second hip has stayed healthy two years after the diagnosis of the first should not be imaged indefinitely.

Should you offload? What the evidence really allows you to say

This is the pivotal chapter of this article, and the one where the gap between practice and evidence is greatest. Offloading is prescribed almost systematically before collapse, on mechanical reasoning that seems obvious. It has never been tested.

The reasoning, and why it is not enough

The reasoning is simple: the necrotic femoral head has lost its mechanical strength, weight bearing loads it, so relieving the weight must delay the subchondral fracture. It is coherent. It has nevertheless never been the subject of a trial comparing protocolised offloading with free weight bearing, on hips otherwise treated the same way.

That void is not an oversight of the literature search: it is an observation that must be stated explicitly, because the existing literature is regularly read as though it had answered that question. It answered a different one.

What Mont 1996 measures, and what it does not

The review by Mont, Carbone and Fairbank is the source cited to condemn non-operative treatment. It brings together 42 publications, 2,025 hips : 1,206 treated by core decompression, 819 by non-operative treatment, excluding electrical stimulation and isolated cases¹⁵.

Satisfactory clinical results are 63.5 % for core decompression against 22.7 % for non-operative treatment. Taking only the pre-collapse hips, the gap remains wide: 71 % against 34.5 %. And the authors themselves correct their own result: excluding the four centres that perform the most decompressions and report the best results from them, the surgical success rate falls to 53 %, that of non-operative treatment staying at 22.7 %¹⁵.

What this review does not say deserves to be written down in black and white. “Non-operative treatment” there aggregates heterogeneous protocols gathered over thirty years, of which offloading is the historical mainstay but whose exact nature, duration and adherence are not controlled. It is not a randomised comparison, it is a gathering of series. The authors themselves in fact conclude that only large prospective randomised trials would clarify the effect.

What can therefore be stated honestly: surgical abstention accompanied by offloading gives, historically, fewer than one satisfactory result in four. What cannot be stated: that offloading is responsible for that, nor that it contributes anything at all.

The only randomised trial comparing physiotherapy with surgery

Neumayr and colleagues conducted a prospective multicentre randomised trial in patients with sickle cell disease and osteonecrosis of Steinberg stage I, II or III. Forty-six patients randomised, eight withdrawals, 38 analysed : 17 in core decompression followed by a physiotherapy programme, 21 in physiotherapy alone¹⁶.

At a mean of three years, hip survival is 82 % in the surgical arm and 86 % in the physiotherapy-alone arm. The functional gain, measured with a modified Harris score, is 18.1 points against 15.7. None of those differences is significant. The authors' conclusion is explicit: physiotherapy alone appeared as effective as core decompression followed by physiotherapy in improving function and delaying the need for further intervention¹⁶.

Two reservations, which do not cancel it but bound its scope. The population has sickle cell disease, and the paper is a reminder in its introduction that in that disease collapse occurs in 90 % of patients within five years of diagnosis: it is a particularly unfavourable background. And above all, there was no untreated arm : that trial establishes that physiotherapy holds its own against surgery, it does not establish that it does better than nothing.

What the guidelines say about it

Two synthesis texts decide, and in the same direction.

The German S3 guideline writes that conservative treatment with iloprost and alendronate may be tried, but that the other pharmacological or physical treatments are inappropriate¹¹. It is harsh wording, and it targets physical treatments explicitly.

The JBJS update of 2020 is of the same order, with one exception worth keeping: “non-operative modalities have generally been ineffective at halting progression. Non-operative treatment is therefore not appropriate at the early stages when the aim is to preserve the native joint, except possibly, on rare occasions, for small, medially located lesions, which may heal without surgery”¹⁷.

That exception matches exactly what the two natural history series show: a small medial lesion, fewer than 10 % collapsing in Mont¹³, collapse stopping in 8 cases out of 9 in Nishii¹⁴.

The defensible position in practice

From all this, a position can be drawn that invents nothing.

• Offloading does not preserve the femoral head. No data shows it, and the historical fate of non-operative treatment does not argue in its favour.
• Offloading remains legitimate as an analgesic measure, just as in any painful joint condition, and for as long as it gives relief. It is a symptomatic use, to be presented as such.
• Prolonged offloading has a measurable cost : quadriceps wasting, deconditioning, disuse bone loss, apprehension. That cost is certain where the benefit is hypothetical.
• What decides is the size and site of the lesion, not the number of weeks on crutches. A Kerboul angle above 240 degrees heralds collapse whatever the non-operative approach; a small medial lesion has a good chance of getting through, including while walking.

What you say to the patient, and what you avoid saying

The conversation about crutches often comes down to two sentences, and those two sentences commit months of behaviour. Here is what the data allow.

What can be said. “The crutches are there to relieve your pain. Use them for as long as they relieve you, and put them down when they no longer make any difference.” That wording is accurate: it assigns to offloading the only role the literature grants it, an analgesic role, identical to the one it holds in any painful joint.

What is better avoided. “Do not put your foot on the ground, or your femoral head will crush.” That sentence is very widespread and has no support. It produces three measurable and all harmful effects: a loss of strength that will lengthen the rehabilitation, a fear of movement that will outlive the indication, and adherence that collapses at the first lapse, since the patient who has walked without harm concludes that the advice was false.

What you say about the prognosis. This is where information has real value, because it is quantified and depends on a parameter the patient can understand: the size of the affected area. A patient whose Kerboul angle is below 190 degrees can be told that none of the hips in that category collapsed in the available series¹² ; a patient whose lesion exceeds 240 degrees should know that collapse is very likely and that the surgical question will arise, which is better than a brutal discovery six months later.

What you say about the timetable. Two years. That is the period in which the subchondral fracture commonly occurs, and the one in which the contralateral risk is concentrated¹¹. A patient who knows that limit accepts close surveillance, and accepts it all the better for knowing that it will stop.

What are the drugs and conservative surgery worth?

The physiotherapist prescribes neither bisphosphonate nor core decompression, but does look after patients who receive them, and is the first to be asked what they are worth. Two randomised trials on alendronate give opposite answers: how they differ is instructive.

Two trials, two opposite conclusions

Lai and colleagues, in 2005, randomised 40 patients with osteonecrosis of Steinberg stage II or III with a necrotic area of more than 30 % (class C). Twenty received 70 mg of alendronate a week for 25 weeks, twenty received nothing. Minimum 24-month follow-up. A spectacular result: collapse of 2 heads out of 29 in the alendronate arm against 19 out of 25 in the control arm (p less than 0.001), and one arthroplasty against sixteen (p less than 0.001)¹⁸.

Chen and colleagues, in 2012, redid the trial differently: multicentre, randomised, double-blind, against placebo, over two years. Sixty-four patients included, 52 analysed (65 hips), University of Pennsylvania stages IIC or IIIC. Result: 4 arthroplasties out of 32 hips on alendronate against 5 out of 33 on placebo (p = 0.837). No radiographic, MRI, Harris score or SF-36 difference. The authors' conclusion: “alendronate has no obvious effect”¹⁹.

Two trials of comparable size, on comparable stages, reach opposite conclusions. The most visible difference is methodological: the first had neither placebo nor blinding, the second had both. That is not proof that Lai is wrong, but it is the hierarchy of evidence that critical reading imposes. What should be said to a patient on alendronate for this indication: the evidence is contradictory, and the only double-blind placebo-controlled trial is negative.

Conservative surgery and arthroplasty

The S3 guideline recommends no conservative procedure in particular, but holds that core decompression may be discussed at early stages if the necrosis occupies less than 30 % of the head. At ARCO stages IIIc or IV, total arthroplasty should be considered, with results comparable to those obtained in osteoarthritis. The main factor for revision is young age¹¹.

The JBJS update of 2020 confirms it: joint preservation procedures should be attempted at the early stages, and the results of total arthroplasty in osteonecrosis are excellent, similar to those in osteoarthritis¹⁷. The 2026 update puts the same thing differently: preservation procedures are mainly effective before structural collapse, and total arthroplasty remains the most reliable treatment once it has occurred¹⁰.

That last sentence is the hinge of the whole management: collapse of the head is the boundary. Before it, preservation can be hoped for. After it, you replace. That is why early diagnosis is not an academic nicety.

The table of modalities, and what to say about it to the patient

What place for the physiotherapist, and at what level of evidence?

One has to be exact here, because there are two temptations: overestimating what rehabilitation can do to bone, or giving up on management that held its own against surgery in the only trial available. Neither.

What physiotherapy does not do

It does not revascularise a necrotic sector, does not close a subchondral fracture, does not change the Kerboul angle. No data suggests that an exercise programme changes the structural natural history of the disease, and the S3 guideline goes so far as to judge physical treatments inappropriate in this indication¹¹. Claiming the opposite to a patient means making them miss a surgical window.

What it does, and what documents it

The physiotherapy arm of the Neumayr trial is the best data that exists: 21 patients treated by physiotherapy alone, 86 % hip survival at three years, 15.7 points gained on the modified Harris score, with no significant difference from surgery¹⁶. It is a level I randomised trial by its design, and it is also a trial of 38 patients in a particular population: both things are true.

To that is added a documented clinical case with unusually long follow-up. Jayaseelan and colleagues describe a 56-year-old man with bilateralosteonecrosis, managed with impairment-based manual therapy, therapeutic exercise and education: eight sessions over ten weeks, clinically significant improvements in pain, in the Lower Extremity Functional Scale and in the Harris score, maintained at two years²¹. The authors themselves write that non-surgical interventions in this disease “are not well described”. One case is not proof of efficacy; it is the demonstration that functional management is possible and that it can hold.

The defensible content of a course of care

• Maintain lower limb strength, and particularly the quadriceps and the abductors, which waste quickly when weight bearing is limited. When pain rules out classic external load, strengthening under blood flow restriction is the tool whose indication this precisely is.
• Maintain hip mobility, working within the pain-free ranges. Loss of internal rotation is a marker of joint distress, not a target to force.
• Dose weight bearing on the symptom, not on a protocol. That is the practical translation of the whole previous chapter: offloading is set by what it relieves, and is stopped as soon as it no longer relieves anything.
• Educate about the timetable : the subchondral fracture occurs within two years, the contralateral risk is concentrated in the same period¹¹. A patient who knows this accepts close surveillance better, and its ending too.
• Recognise when to hand back. A clear worsening, a limp that sets in, rest or night pain, an apparent shortening of the limb all call for repeat imaging.

A programme, and how to present it

What follows is not a validated protocol: none exists for this disease, and writing otherwise would be lying. It is a reasoned transposition of what the physiotherapy arm of the only randomised trial contained¹⁶ and of what the case followed for two years describes²¹, presented as such to the patient.

First session: establish the baseline and the contract. Assessment of abductor and quadriceps strength, of ranges, of walking distance, and of a functional score that will serve as a reference. The Harris score and the Lower Extremity Functional Scale are the ones the two available sources use, which allows comparison with them. The contract comes down to one sentence: rehabilitation targets function, surveillance targets structure, and the two run in parallel.

Strengthening, and the question of load. The practical problem is real: the external load that would build strength is often the one that wakes the pain. That is exactly the configuration in which strengthening under blood flow restriction finds its indication, since it produces a strength gain at low intensities. There is no study of that modality in osteonecrosis, and that has to be said to the patient: a tool is being transposed, no evidence is being followed.

Walking and endurance. Walking distance is the parameter most meaningful to the patient and the easiest to follow. It also allows weight bearing to be set by the symptom rather than by a protocol, which is the defensible position set out above.

Education, which is not an extra. Three messages: what physiotherapy can do (function) and cannot do (bone); the analgesic and not protective role of offloading; and the two-year timetable, with surveillance of the contralateral hip. The patient in the published case obtained his gains in eight sessions over ten weeks, and held them for two years²¹ : a modest volume, which suggests that the added value lies as much in the direction taken as in the number of sessions.

What do published clinical cases teach us?

Two published reports, both from the physiotherapy literature, frame the two sides of the subject: the one where the disease escapes diagnosis, and the one where it is managed without surgery. They are cited with their PubMed identifier.

Case 1: the 51-year-old man, or the diagnosis that arrives through worsening

Signorino, Jayaseelan and Brindle report in the Journal of Orthopaedic and Sports Physical Therapy the case of a 51-year-old man presenting to unscheduled care two weeks after the onset of left lateral hip and buttock pain. The radiographs are unhelpful. An orthopaedic surgeon refers him to physiotherapy. Faced with worsening of the clinical picture, he is sent back to the orthopaedic surgeon for further assessment. The MRI then requested shows an abnormal contour of the femoral head and acetabulum, extensive marrow oedema and a complex joint effusion: osteonecrosis of the femoral head²⁰.

What this case teaches. First, that the presentation can be atypical, here lateral and gluteal rather than in the groin, which spontaneously points towards a gluteal tendinopathy. Next, that the physiotherapist occupies a particular position: they are the only clinician who sees the patient several times a week, and it is that repetition that makes the worsening visible. Finally, that the right decision here was not to change the treatment but to send the patient back for imaging.

Case 2: the 56-year-old man, or conservative management that holds for two years

The same authors, in the Journal of Bodywork and Movement Therapies, report a 56-year-old man with bilateral osteonecrosis of the femoral heads. Management combines impairment-based manual therapy, therapeutic exercise and education, over eight sessions spread across ten weeks. The improvements, clinically significant in pain, in the Lower Extremity Functional Scale and in the Harris score, are maintained at two-year follow-up²¹.

What this case teaches, and what it does not. It shows that functional independence can be obtained and held without surgery, on a modest volume of care. It does not show that physiotherapy changed the bone: no follow-up imaging data is reported in the abstract, and a single case allows no causal inference. The authors say so: work is needed to describe the optimal methods.

What these two cases have in common

In both, the physiotherapist sees the patient for a long time and often. In the first, that position lets them detect the failure; in the second, run continuous care. That is exactly the double role the literature allows us to claim: a lookout role, and a functional role. Not a structural role.

How do you apply this in the clinic?

This chapter translates all the above into actions. It comes down to three moments: what you ask at the first assessment, what you write in the letter, and what you monitor afterwards.

The first assessment

Faced with any painful hip, two questions must be asked systematically, because they are the ones that protect against diagnostic error⁸ :

1. Have you taken cortisone, in any form, in the last two years? For what illness, at what dose, for how long. A reminder: the risk is concentrated in the first twelve months⁴.
2. What is your alcohol consumption? In weekly amount, and for how many years. The relationship is graded, there is no reassuring threshold⁵.

To that are added a history of sickle cell disease, of transplantation, of chemotherapy, of pelvic radiotherapy, of lupus, and of severe hospitalised COVID⁷. And age: under 50 with persistent hip pain and normal radiographs, the 2026 update asks that the diagnosis be suspected¹⁰.

What you write in the letter

The wording counts, because it is what triggers the MRI or not. A letter that says “hip pain, physiotherapy with no improvement” gets nothing. A letter that says the following gets one:

Three errors that cost dear

First error: treating the normal radiograph as a reassuring result. It defines stage I. It is the only fully reversible stage, and the one where joint preservation has the best chance¹⁰.

Second error: prescribing long offloading in the belief of protecting the head. The benefit is not established, the functional cost is. Offload to relieve, yes; offload to preserve, no.

Third error: looking at only one hip. Bilateral involvement is frequent, contralateral MRI is recommended, and the risk is concentrated in the first two years¹⁰¹¹.

What distinguishes this picture from the neighbours the site covers

Frequently asked questions

Does a normal radiograph rule out osteonecrosis?

No, and it is the most important point of this article. Stage I of the ARCO 2019 classification is literally defined by a normal radiograph together with a positive MRI or bone scan⁹. In a patient at risk, a normal radiograph does not change the probability of the diagnosis: it only excludes stages II to IV.

Should the hip that does not hurt be imaged?

Yes, in a newly diagnosed patient. The 2026 JBJS update explicitly recommends MRI of the contralateral hip, bilateral involvement being frequent¹⁰. And an asymptomatic hip is not a hip free of progression: 59 % progress in the systematic review of 664 hips¹³.

How long should the second hip be monitored?

The S3 guideline gives the marker: the risk of contralateral involvement is high during the following two years, and becomes unlikely afterwards¹¹. It is also the period in which the subchondral fracture of the affected hip commonly occurs.

Does offloading protect the femoral head?

Nothing demonstrates it. No trial has compared protocolised offloading with free weight bearing. What is known is indirect: the historical non-operative treatment, of which offloading is the mainstay, gives 22.7 % satisfactory results¹⁵, and the S3 guideline judges physical treatments inappropriate in this indication¹¹. Offloading keeps an analgesic value, and that is how it should be presented.

Can osteonecrosis heal without an operation?

Yes, in one precise configuration: small, medially located lesions. They collapse in fewer than 10 % of cases¹³, the collapse stops in 8 cases out of 9 when it occurs¹⁴, and the JBJS update explicitly recognises that exception¹⁷.

From what corticosteroid dose should you be concerned?

There is no individual threshold. Felson and Anderson established a dose-effect relationship at the level of studies, between the mean daily dose of a cohort and its osteonecrosis rate³. In Koo's series, the cumulative dose at detection ranged from 1,800 to 15,505 mg of prednisolone equivalent⁴ : an eight-fold spread says enough that no threshold protects.

Are corticosteroid boluses more dangerous?

The available data says the opposite: in the analysis by Felson and Anderson, the dose given as a bolus was not associated with risk, whereas the oral dose strongly was³. It is the cumulative daily dose that counts.

Should a patient who drinks little but takes corticosteroids stop drinking?

Stopping remains desirable for other reasons, but its benefit on the risk of osteonecrosis is probably small in that configuration: in corticosteroid users, the odds ratio of alcohol falls to 1.10, and the combination of the two exposures gives no more than corticosteroids alone⁶. In a patient who has never received corticosteroids, on the other hand, alcohol is the dominant factor.

What can physiotherapy bring, honestly?

Function, not structure. The only randomised trial available shows that a programme of physiotherapy alone holds its own against core decompression at three years, in 38 sickle cell patients¹⁶, and a published case documents gains maintained at two years after eight sessions²¹. No data shows an effect on the necrosis itself.

Does arthroplasty give poorer results than in osteoarthritis?

No. The 2020 and 2016 updates agree: the results are excellent and comparable to those in osteoarthritis¹⁷. The main factor for revision is not the disease, it is the patient's young age at the time of implantation¹¹. What follows is described in our article on rehabilitation after total hip replacement.

References

Twenty-one references, verified one by one through the NCBI E-utilities: a real PMID, journal, issue year, complete author list, and abstract read to check that the source does establish what is attributed to it. The working base is versioned in the repository, with the list of what it does not support.

This page deals with osteonecrosis of the femoral head as a disease in its own right. The commonest differential diagnosis in the older person is the subject of a separate article: hip osteoarthritis. Bone marrow oedema and its non-necrotic causes are set out in bone bruise. The same disease in another bone is described in Kienböck's disease, avascular necrosis of the lunate. And the end of the trajectory, when the head has collapsed, is dealt with in rehabilitation after total hip replacement.