Neer Classification of Proximal Humerus Fractures: Parts, Displacement Rules and Interobserver Reality

Every displaced proximal humerus fracture is described in parts: "a three-part fracture in a 70-year-old", "a four-part with a head split". The Neer classification supplies that language, and it is the system examiners expect. The question is how much weight a part count can carry. The short answer: Neer's four-segment idea explains how these fractures displace and groups them usefully, but observers agree on the category only moderately, so the count should start the treatment discussion, not settle it. This post covers the parts, the displacement rule, the reliability data and the decisions that follow. The clinical overview is on the proximal humerus fractures topic page.

What are the four parts?

Neer built on an observation of Codman's: every proximal humerus fracture is made up of four segments, the articular segment (the head), the greater tuberosity, the lesser tuberosity and the shaft. Neer's 1970 paper drew on the radiographs and operative findings of 300 fractures he treated in New York between 1953 and 1967 (Carofino and Leopold's summary of the original).

The segments matter because of what is attached to them. The muscle attachments predict the direction of displacement, which is how you read a film:

• Greater tuberosity: supraspinatus, infraspinatus and teres minor; supraspinatus pulls it superiorly.
• Lesser tuberosity: subscapularis, which pulls it into internal rotation.
• Shaft: pectoralis major pulls it medially, and the deltoid may displace it laterally.
• Head: no muscle of its own; its position follows the impact or the direction of dislocation, and it carries the blood-supply question.

How does the Neer classification count parts?

A segment is a separate part only if it is displaced by more than 1 cm or angulated by more than 45°. You count displaced parts, not fracture lines or fragments. A shattered-looking film in which nothing passes the threshold is a one-part fracture, and a comminuted surgical neck with one displaced greater tuberosity is two-part.

The full system also has separate groups for fracture-dislocations (anterior and posterior) and articular-surface fractures (head-splitting and impression fractures). Neer separated them because they carry a different prognosis and treatment: dislocations are more severe injuries, with more avascular necrosis and heterotopic ossification. With those groups, the scheme has 16 categories, as numbered by Sidor and colleagues.

The three-part rotation. In a three-part fracture the tuberosity left attached to the head keeps its cuff, so the head is pulled into a rotational deformity by whichever tendons remain. That is why a three-part fracture can look odd on an AP alone, and why the shoulder fracture-dislocation and articular groups need the axillary view before anyone counts parts.

Where did 1 cm and 45° come from?

Not from outcome data. When the paper went to the Journal of Bone and Joint Surgery, the editor, Thornton Brown, insisted on defined criteria for when a segment counts as displaced. Neer set 1 cm of separation and 45° of angulation in response, and he later acknowledged that the numbers were somewhat arbitrary while accepting that some definition was needed for the system to be used at all. The thresholds have become the most quoted part of the classification. In 2002 Neer revisited the system in a paper titled "Four-segment classification of proximal humeral fractures: purpose and reliable use", which is where the valgus-impacted four-part category was added.

Measuring is harder than it sounds. The PROFHER investigators, preparing to classify 250 trial radiographs, concluded that a centimetre cannot be measured accurately on a plain film without a calibration object at the fracture, and that the PACS scale is lost when images are anonymised. They noted that the stems of the standard L and R cassette markers are 9 mm long, a rough ruler on some films. Angulation depends on the projection and on how clearly the fragment edges show.

The greater tuberosity: the exception to the rule

The 1 cm rule defines a displaced part; it is not the threshold for fixing an isolated greater tuberosity. The topic page gives the fixation threshold as over 5 mm, and over 3 mm in an overhead athlete or manual worker, because even small superior displacement causes subacromial impingement and loss of external rotation. A greater tuberosity displaced 7 mm is a one-part fracture by Neer's definition and still a candidate for screws or sutures. Calling it undisplaced is the commonly missed error.

How reliable is the Neer classification?

Moderately at best, and the finding has been repeated for three decades. Kappa runs from 0 (agreement no better than chance) to 1 (complete agreement); the studies below describe values around 0.4–0.5 as fair to moderate.

Intraobserver agreement is better but still imperfect. In Sidor's study it ranged from 0.83 for the shoulder specialist to 0.50 for the radiologist, with a mean of 0.66.

The four-part boundary is the weak point. In PROFHER the two independent raters disagreed most on the four-part category: one called 30 fractures four-part and the other 4. After a consensus meeting they settled on 11 (4.4%), while recruiting surgeons had recorded 64 (25.6%) as involving both tuberosities. "Both tuberosities involved" and "four-part" are not the same claim, and the gap shows how loosely the term is used in clinic.

Does CT, a simpler scheme or experience fix it?

Mostly not.

• CT. In Bernstein's study, adding CT raised intraobserver reliability slightly (0.64 to 0.72) but not interobserver agreement (0.52 to 0.50). The authors concluded that even experts cannot uniformly agree on which fragments are fractured. Majed's group gave senior surgeons physical models printed from CT, and Neer agreement was still only 0.33.
• Fewer categories. Sidor collapsed the 16 categories to six, and agreement did not significantly improve. Bernstein ignored displacement and scored only which segments were fractured, and interobserver kappa moved from 0.52 to 0.56.
• Experience. The systematic review found that selecting experienced observers did not improve agreement, and the authors could not support the belief that experts disagree less than non-experts. Individual studies show at most a small edge: in Sidor's series attendings scored 0.52 against 0.48 for residents.

Training does. In a randomised trial, Brorson and colleagues gave 14 doctors either two training sessions or none before they classified 42 pairs of radiographs. In the trained group mean kappa rose from 0.27 to 0.62, and among specialists from 0.30 to 0.79. One trial of 42 cases is a small base, but it was the only randomised test of an intervention the systematic review found, and it worked.

If surgeons disagree on the category, does it matter?

Less than the kappa values suggest. Brorson's 2012 study asked five experienced shoulder surgeons to classify 193 consecutive fractures twice, three months apart, and to recommend non-operative treatment, a locking plate or hemiarthroplasty for each. Agreement on the treatment (kappa 0.48 and 0.52) was significantly higher than on the Neer category (0.33 and 0.36). Observers changed category in 36% of repeat readings, but changed their treatment recommendation in only 34% of those. Agreement on treatment still did not exceed moderate.

The PROFHER analysis points the same way. Neither of the trial's subgroup analyses, by tuberosity involvement or by Neer category, supported choosing surgery on the basis of fracture type for the Oxford Shoulder Score; the trial itself is covered in our post on the PROFHER trial. The practical reading: surgeons decide on a fuller picture than the part count, and the count is a summary of that picture rather than its basis.

Does the part count predict avascular necrosis?

Neer expected it to. He reasoned that in four-part fractures the blood supply to the head had been severed, reported avascular necrosis in more than half of them and advocated arthroplasty (as summarised by Carofino and Leopold). Later work has not confirmed that a four-part pattern on its own predicts necrosis.

Two findings explain the gap.

• The head's main supply. The textbook vessel, the arcuate artery from the anterior circumflex humeral artery, is disrupted in about 80% of proximal humerus fractures, yet osteonecrosis is uncommon. Cadaver MRI work by Hettrich and colleagues, in uninjured shoulders, found that the posterior circumflex humeral artery supplies about 64% of the head, which may explain why a head can survive an obviously disrupted arcuate artery. The detail is on the proximal humerus fractures page.
• What predicts ischaemia. Hertel assessed perfusion at operation in 100 intracapsular fractures. The best predictors were a calcar (dorsomedial metaphyseal head extension) shorter than 8 mm (accuracy 0.84), a disrupted medial hinge (0.79) and the basic fracture pattern, particularly the anatomical neck. Neer's own variables did worse: a four-fragment fracture had an accuracy of 0.67, head angulation over 45° 0.62 and tuberosity displacement over 10 mm 0.61. Combining anatomical neck, short calcar and disrupted hinge gave positive predictive values of up to 97%.

Hold Hertel loosely as well. The perfusion test was drill-hole backflow at surgery, the series came from one institution, and ischaemia is not necrosis: many ischaemic heads revascularise, and Carofino and Leopold note that these markers of poor perfusion did not predict later necrosis either. Read the calcar and the hinge as reasons to plan for possible replacement, as the humeral head fractures page puts it, not as a prediction that the head will collapse.

Neer or AO/OTA?

The AO/OTA system is more comprehensive. In its 2018 revision (opens in a new tab) it absorbed Neer's part count: 11-A is extra-articular, unifocal and two-part, 11-B extra-articular, bifocal and three-part, and 11-C articular or four-part. It is not clearly more reproducible. Siebenrock and Gerber found the earlier AO system slightly ahead at the main level (kappa 0.53 against 0.40 for Neer), but called both fair or poor, and agreement worsened as AO groups and subgroups were added. Majed's model study gave AO the lowest agreement of the four systems tested (0.11). Know both. Use Neer to talk about displacement and treatment, and give the AO code when a trial or registry asks for it.

How does the classification feed the decision?

The part count is weighed against age, functional demand, comorbidity and bone quality, and the pattern details Neer does not grade: calcar length, the medial hinge, varus or valgus head position and articular involvement. Get a CT for three- and four-part fractures and any head-split pattern. The topic page's algorithm runs as follows.

Two cautions travel with the table. PROFHER randomised adults with displaced fractures involving the surgical neck; fracture-dislocations and open fractures were excluded, and the Cochrane review records insufficient evidence for patients under 60, high-energy injuries, two-part tuberosity fractures, fracture-dislocations and articular-surface fractures. And surgery in PROFHER meant fixation or humeral head replacement, not reverse arthroplasty, so the preference for a reverse prosthesis in older patients rests on other evidence. The fixation technique is in the locking plate operative guide, and the arthroplasty side on the reverse total shoulder arthroplasty topic page.

In the exam

Examiners want the parts counted correctly, the threshold stated, the weakness acknowledged and a decision that does not rest on the count alone. For learning the categories, see how to memorise orthopaedic classifications.