Advanced glenohumeral osteoarthritis: the relationship between radiographic pathoanatomy and clinical presentation asks "does the x-ray tell us how the patient is doing?" The authors studied 280 shoulders with advanced glenohumeral osteoarthritis and an intact cuff, all of which went on to arthroplasty: 147 anatomic total shoulders, 81 reverses, and 52 ream and runs [1]. Every shoulder was graded before surgery by three classifications: Samilson-Prieto, Kellgren-Lawrence, and the Walch system as modified for three-dimensional imaging [3,4,5]. The authors also measured critical shoulder angle, humeral head medialization, humeral head flattening, the length of the inferior humeral neck spur, posterior decentering, glenoid version, and glenoid inclination. Then they asked whether any of these predicted how the shoulder moved, how comfortable it was, or how the patient rated his or her health.
With two exceptions, it did not. No clinically meaningful association between Samilson-Prieto grade, Kellgren-Lawrence grade, Walch type, critical shoulder angle, medialization, version, or inclination and any patient-reported outcome or quality-of-life score. The exceptions were both on the humeral side: greater flattening of the head and a longer humeral neck spur were associated with less motion.
This replicates what we reported in 2019 in 544 shoulders [6], what Kohan and colleagues reported in 256 [7], and what Kircher and colleagues reported earlier still [13]. Four groups, four cohorts, one answer: the radiographic severity of the arthritis does not tell us what the patient is experiencing.
The second paper ---Reverse and anatomic total shoulder arthroplasty for glenohumeral osteoarthritis: a propensity-matched comparison at early and midterm follow-up ‚--- asks "does the implant choice change the result?" From a single high-volume surgeon's practice, Leinweber and colleagues matched 61 anatomic total shoulders to 61 reverses, one to one, on age, sex, body mass index, preoperative ASES, preoperative forward elevation, and Walch glenoid type [2]. Notably, they matched on the very pathoanatomy that the first paper found does not predict much. All shoulders had osteoarthritis with an intact cuff. All were seen early (about two years) and at midterm (about five years).
Both groups improved a great deal, and by the same amount. More than 96% of patients in both groups reached the minimal clinically important difference for the ASES at both time points. The anatomic patients had better external rotation at the early visit (63 degrees vs. 57 degrees) and better internal rotation as well; by five years the internal rotation advantage was gone and the external rotation difference had narrowed. Complications were 3.3% in each group.
Consider what each study holds constant and what it lets vary.
The second paper holds the surgeon constant and varies the implant. It finds no meaningful difference in what the patient reports.
The first paper considers the variable pathoanatomy across 280 shoulders and finds that it explains almost nothing about how the patient presents.
Neither the glenoid nor the prosthesis seems to matter. (The two measures that did survive are on the humerus, and we will come to them a bit later.) If the explanation for the differences among our patients' outcomes is not in the glenoid we spend our time classifying and not in the implant we spend our time choosing, the two possibilities left standing are the patient and the surgeon.
One incidental note: the two papers used different MCID values for the same score ‚--- 16 points for the ASES in the first [8], 10.4 points in the second [9]. Both are defensible and both are published. Whether a result is "clinically important" can depend on which threshold the authors selected.
Eight radiographic parameters and six classification terms were each tested against eleven outcomes: 154 comparisons, with no correction for multiplicity. At the conventional threshold, roughly eight false positives are expected by chance alone. Seventeen associations were reported as significant ‚ --- more than chance alone would produce. Of those seventeen, the authors' own MCID screen disqualified nearly every one. The findings are not false; they are true but too small to act on.
Four associations survived as "clinically relevant":
Forward elevation and humeral head flattening. A coefficient of 0.56 degrees per percentage point of humeral head flattening, against an MCID of 17.1 degrees [8], requires a change in flattening of about 30 percentage points: nearly five standard deviations, and roughly three-quarters of the entire observed range.
External rotation and head flattening. About 24 percentage points, close to four standard deviations.
Neither is a quantity that could exist in a patient.
Internal rotation and head flattening. About 10 percentage points, 1.6 standard deviations, entirely attainable.
External rotation and humeral neck spur. A difference in spur length of about 22 mm, well within the observed range of 0 to 48 mm. This result replicates Kircher [13].
Those last two are real, they are reachable, and they are both on the humeral side ‚ ---where nobody has been looking ‚ --- rather than on the glenoid, where everyone has been looking. In other words the strongest signals in 280 shoulders are from the humerus; the glenoid---the object of twenty years of classification---apparently contributes little.
A limitation of the analysis is that every shoulder in this series went on to arthroplasty. That is a range-restricted sample: each of these patients had already crossed somebody's operative threshold, so the variance in symptoms is truncated at the low end, and truncation attenuates correlation. Some of the null result is built into the sampling. The claim is not "radiographs never relate to symptoms." The actual finding is: among patients already being considered for shoulder arthroplasty, the images doe not tell you who hurts, how much, or how far the shoulder moves.
In the subgroup analysis, the concentric (Walch A) glenoids had worse pain, worse DASH, and worse ASES than the eccentric (Walch B) glenoids. By the standard we have just applied to the rest of the paper, these differences fall below the MCID and we should not make much of them‚---so we will only state: the direction runs opposite to the expectation of most surgeons. The eccentric, retroverted, posteriorly decentered glenoid is the one that most often prompts a surgeon from an anatomic reconstruction to a reverse. In this cohort those shoulders were, if anything, the more comfortable ones. That is not what much current thinking about the B glenoid would predict, and it is worth a study designed to test it rather than a subgroup that happened to find it.
The surgeon performed 1,310 reverses and 546 anatomic total shoulders in the study window. Of the reverses, 133 (only 10% of the original cohort) had complete clinical and outcome follow-up at both time points; 61 entered the matched analysis. Of the anatomics, 129 (only 24% of the original cohort) qualified; 61 entered the matched analysis. So the survivorship bias and the matching dramatically reduced the studied sample size, making it less relevant to the original group of patients.
Note the inclusion rule was complete outcome scores and clinical follow-up at both visits. That rule removes the patients who were revised elsewhere, who stopped answering, who were dissatisfied and did not come back, and who died. The paper then reports a 0% revision rate for the reverse and a 1.6% acromial stress fracture rate. These are not accurate incidence estimates. A revision rate cannot be calculated in a cohort whose definition requires having completed follow-up; it requires the whole exposed group and a time-to-event analysis.
Substantial clinical benefit was reached by 95.1% of anatomic patients and 80.3% of reverse patients early. That 14.8-point gap has a P value of .027 and is discussed as a real difference. At five years the gap was 13.1 points (91.8% vs. 78.7%), with a P value of .074, and was described as no significant difference.
The two gaps are nearly the same size. What changed is which side of .05 the P value fell on, in a study that states it could not perform a power analysis. A 13-point difference in substantial clinical benefit at five years has not been shown to be absent; it has merely not been shown to be present. Those are different statements, and only the second one is supported here. The gap favors the anatomic shoulder at both time points, and it deserves a larger study rather than a rounding to "similar."
When more than 96% of both groups clear the MCID, the MCID has stopped discriminating. It tells us that both operations work, which is worth knowing and is the correct headline. It cannot tell us whether they work equally well.
One anatomic shoulder was revised; no reverse was. That appears in the abstract as "aTSA had more revisions." A failed anatomic shoulder has somewhere to go, --- conversion to a reverse. A failed reverse does not, and the threshold for taking a painful reverse back to the operating room is far higher, because the surgeon has less to offer. A revision count compares the availability of a salvage operation as much as it compares the durability of an implant. Counting one against zero and reporting it as a durability signal asks a single patient to carry the argument.
The same asymmetry appears in how radiographic failure was judged. A reverse with baseplate lucencies and broken screws was recorded as a non-failure because the patient was comfortable. Asymptomatic glenoid lucencies after anatomic arthroplasty, graded by the Lazarus system [10], were simultaneously presented as the durability liability. One standard should apply to both.
The whole argument for using a reverse in a cuff-intact arthritic shoulder is long-term durability: glenoid loosening and late cuff failure at ten, fifteen, twenty years. Five years samples precisely the window in which the anatomic shoulder is known to do well [11], and the one series that has followed eccentric-wear shoulders past that window to a minimum of seven years found the anatomic reconstruction holding up [12]. This study cannot address the question that prompted it.
Put the two preoperative cohorts next to each other. Both are advanced osteoarthritis with an intact cuff, in the same country, in the same era. The shoulders moved almost identically before surgery: mean forward elevation 96.6 degrees in the first series, 97.5 degrees and 98.2 degrees in the second.
The patients, however, were not in the same condition. Mean preoperative ASES was 29.9 in the first series and 41.4 and 38.9 in the second. Mean pain was 7.4 out of 10 versus 5.3 and 5.5. The patients in one practice arrived at the operating room roughly ten ASES points and two pain points better off than the patients in the other.
There is more than one way to get such a gap. These are different practices with different referral streams, different payer mixes, and different geography; the first cohort also includes 52 ream-and-runs, a self-selected group that has no counterpart in the second practice. So the difference may reflect who walks through the door as much as when the surgeon decides to operate. But that distinction does not rescue the number. Referral pattern and operative threshold are both properties of the practice, not of the shoulder. Either way, what varies between these two cohorts is the surgeon's context, not the patient's pathoanatomy ‚ --- and it varies by more than the pathoanatomy does. This is the kind of unwanted variation that Kahneman called noise.
If you are a patient looking at your own x-ray or CT scan, the first paper is reassuring: the severity of what you see on the film does not predict how much your shoulder will hurt, how far it will move, or how you will feel about your life. Some very ugly-looking shoulders belong to comfortable people, and some mild-looking ones hurt badly. The x-ray describes the joint. It does not describe you.
If you are a patient choosing between an anatomic and a reverse replacement for arthritis with an intact cuff, the second paper says that at five years, in the hands of one experienced surgeon, both do well. The anatomic shoulders showed somewhat better rotation early, though by the standard applied above that early difference is at the edge of what a patient would notice. What happens after five years is not yet known, and that is the question that actually separates these two operations.
If you are a surgeon, the two papers together take away two of the things you were counting on. The images you spend the most time studying did not explain how the patient presented. The implant you spent the most time deciding on did not explain how the patient ended up. What is left is the judgment that sits between them: whom you offer an operation to, when in the course of the disease you offer it, what you tell the patient to expect, and how well you execute. Those are the variables neither study measured, and the ten-point preoperative ASES gap between these two practices is a direct measurement of how much they vary from one surgeon to the next.
We keep looking for the explanation in the imaging and in the implant because those are the things we can see and buy. The evidence keeps pointing somewhere else.
Both of these are careful papers by serious groups, and both are more honest in their limitations sections than most. The first calls itself a pilot and names its multiplicity problem. The second calls itself hypothesis-generating, names the risk of type II error, and says that longer follow-up is needed.
Which leaves the question: if the picture does not explain the patient's presentation and the implant does not explain the result, what does?
Follow on twitter/X: https://x.com/RickMatsen
Follow on facebook: https://www.facebook.com/shoulder.arthritis
Follow on LinkedIn: https://www.linkedin.com/in/rick-matsen-88b1a8133//
[1] Covarrubias O, Luther L, Portnoff B, Levins J, Hoffman R, Molla V, Toavs T, Molino J, Paxton ES, Green A. Advanced glenohumeral osteoarthritis: the relationship between radiographic pathoanatomy and clinical presentation. J Shoulder Elbow Surg 2026;35:1620-1631. https://doi.org/10.1016/j.jse.2026.01.007
[2] Leinweber KA, Bowler AR, Diestel DR, McDonald-Stahl M, Arnold RP, Le K, Dunn WR, Kirsch JM, Jawa A. Reverse and anatomic total shoulder arthroplasty for glenohumeral osteoarthritis: a propensity-matched comparison at early and midterm follow-up. J Shoulder Elbow Surg 2026;35:1632-1641. https://doi.org/10.1016/j.jse.2025.12.008
[3] Samilson RL, Prieto V. Dislocation arthropathy of the shoulder. J Bone Joint Surg Am 1983;65:456-460.
[4] Kellgren JH, Lawrence JS. Radiological assessment of osteo-arthrosis. Ann Rheum Dis 1957;16:494-502.
[5] Bercik MJ, Kruse K 2nd, Yalizis M, Gauci MO, Chaoui J, Walch G. A modification to the Walch classification of the glenoid in primary glenohumeral osteoarthritis using three-dimensional imaging. J Shoulder Elbow Surg 2016;25:1601-1606. https://doi.org/10.1016/j.jse.2016.03.010
[6] Matsen FA 3rd, Whitson A, Hsu JE, Stankovic NK, Neradilek MB, Somerson JS. Prearthroplasty glenohumeral pathoanatomy and its relationship to patient's sex, age, diagnosis, and self-assessed shoulder comfort and function. J Shoulder Elbow Surg 2019;28:2290-2300. https://doi.org/10.1016/j.jse.2019.04.043
[7] Kohan EM, Hill JR, Lamplot JD, Aleem AW, Keener JD, Chamberlain AM. Severity of glenohumeral osteoarthritis does not correlate with patient-reported outcomes. J Shoulder Elb Arthroplast 2020;4:2471549220901873. https://doi.org/10.1177/2471549220901873
[8] Simovitch RW, Elwell J, Colasanti CA, Hao KA, Friedman RJ, Flurin P, et al. Stratification of the minimal clinically important difference, substantial clinical benefit, and patient acceptable symptomatic state after total shoulder arthroplasty by implant type, preoperative diagnosis, and sex. J Shoulder Elbow Surg 2024;33:e492-e506. https://doi.org/10.1016/j.jse.2024.01.040
[9] Levy JC, Everding NG, Gil CC Jr, Stephens S, Giveans MR. Speed of recovery after shoulder arthroplasty: a comparison of reverse and anatomic total shoulder arthroplasty. J Shoulder Elbow Surg 2014;23:1872-1881. https://doi.org/10.1016/j.jse.2014.04.014
[10] Lazarus MD, Jensen KL, Southworth C, Matsen FA 3rd. The radiographic evaluation of keeled and pegged glenoid component insertion. J Bone Joint Surg Am 2002;84:1174-1182. https://doi.org/10.2106/00004623-200207000-00013
[11] Kirsch JM, Puzzitiello RN, Swanson D, Le K, Hart PA, Churchill R, et al. Outcomes after anatomic and reverse shoulder arthroplasty for the treatment of glenohumeral osteoarthritis: a propensity score-matched analysis. J Bone Joint Surg Am 2022;104:1362-1369. https://doi.org/10.2106/JBJS.21.00982
[12] Cuff DJ, Simon P, Patel JS, Munassi SD. Anatomic shoulder arthroplasty with high side reaming versus reverse shoulder arthroplasty for eccentric glenoid wear patterns with an intact rotator cuff: comparing early versus midterm outcomes with minimum 7 years of follow-up. J Shoulder Elbow Surg 2023;32:972-979. https://doi.org/10.1016/j.jse.2022.10.017
[13] Kircher J, Morhard M, Magosch P, Ebinger N, Lichtenberg S, Habermeyer P. How much are radiological parameters related to clinical symptoms and function in osteoarthritis of the shoulder? Int Orthop 2010;34:677-681. https://doi.org/10.1007/s00264-009-0846-6
