.
Disclaimers: I have no relationships with any implant company. I do love anatomic total shoulder arthroplasty. The below is my attempt to analyze the available data, but there is a dearth of high-level evidence. I am not a statistician, but here I share my attempt at a deep dive into an important and hotly discussed issue. It is not a quick read, but the topic deserves the depth.
The relative merits of anatomic and reverse total shoulder arthroplasty for osteoarthritis are—and probably will forever be—hotly debated. Many factors contribute to surgeons' views on this topic, including indications, relative difficulty of the procedure, cost, need for preoperative 3D/CT-based planning, need for transfer technologies (robotics, patient-specific instruments, virtual/augmented reality), and the patient's anticipated postoperative comfort, range of motion, and function. In this post, we take a look at the contention that patients whose osteoarthritis is treated with reverse total shoulder arthroplasty have lower rates of revision than those treated with anatomic total shoulder arthroplasty.
The word on the street is that patients having reverse total shoulder arthroplasty (RSA) are less likely to undergo revision than patients having anatomic total shoulder arthroplasty (aTSA). For example, the Australian Orthopaedic Association National Joint Replacement Registry's 2025 Annual Report puts the 10-year cumulative revision rate at 7.4% for aTSA and 5.4% for RSA [1]. The 14-year figures are 9.5% and 6.4%. These numbers are often cited in support of expanding RSA into indications it was not originally designed for—including osteoarthritis with an intact rotator cuff.
Set against those numbers, a finding from O’Malley and colleagues' 2025 National Joint Registry analysis of 21,918 patients should give us pause. RSA had nearly twice the prevalence of unsatisfactory function—Oxford Shoulder Score below 29—as aTSA: 27.0% versus 15.4%. Yet RSA patients with unsatisfactory function were less than half as likely to undergo revision (4.9% vs. 10.6%, p < 0.001) [6]. The implant with the lower revision rate (RSA) is the implant under which a larger proportion of patients are living with poor function.
That paradox is the subject of this post. Two sampling biases and one structural property of the metric itself combine to make revision rate a misleading proxy for clinical success when comparing aTSA to RSA. The first is a confounding mismatch between the patient populations the two operations are performed on. The second is an attrition mechanism that systematically removes the worst outcomes from any cohort with a minimum time to follow-up. The third—and the most important one is that revision rate measures what the surgeon decides to operate on again, not what the patient is living with. Strip these out and the registry-based case for RSA in osteoarthritis with intact cuff is weak.
Two biases and a metric problem
Bias 1: The patient populations are not the same
Restricting to osteoarthritis controls for the obvious confounders of other indications such as cuff tear arthropathy and fracture; even within that restriction, the AOANJRR shows aTSA at 7.7% revision at 10 years and RSA at 5.0% [1]. But the RSA-for-OA cohort is overwhelmingly older: 87.7% of cases are in patients aged 65 or older [1]. When attention is restricted to the demographic where the operations actually compete—younger patients with osteoarthritis—the favorable RSA signal disappears. Figure 1 shows the actual curves from the registry.
Figure 1. Cumulative percent revision from the AOANJRR 2025 Annual Report. RSA for OA in patients aged <65 (n=3,439, Table ST85) versus aTSA with a polyethylene glenoid for OA, all ages (n=5,120, Table ST44). Markers at the registry's reported time points (1, 3, 5, 7, 10, 14 years); shaded bands show 95% confidence intervals. The RSA <65 curve runs above the aTSA curve at every reported time point. At 14 years RSA <65 reaches 13.2% (95% CI 9.6–17.9) versus 9.4% (95% CI 7.8–11.1) for aTSA—the ratio of point estimates is 1.40. The wide 14-year RSA confidence interval reflects the small number of patients still under follow-up at that time point (n=53), an example of the cohort attrition discussed under Bias 2. aTSA n-at-risk approximated from Figure ST2 of the AOANJRR 2025 report (all-diagnosis polyethylene glenoid, n=5,413) scaled to the OA-only cohort. RSA n-at-risk taken directly from Table ST85.
Cautions about this comparison are in order. The under-65 RSA-OA cohort is itself selected: these patients typically have severe pathology not amenable to aTSA, so part of their failure rate may reflect disease severity rather than implant inferiority. And the aTSA comparison group is not age-matched. Thus the registry cannot fully answer the like-for-like question—younger patient with intact cuff, RSA versus aTSA—because surgeons mostly do not choose RSA for that patient, and so the comparison cases barely exist. What the registry can show is that the apparent RSA-OA advantage in the registry headline is overwhelmingly driven by older patients for whom aTSA was never going to be offered.
Bias 2: The cohorts are themselves selectively curated
Every minimum-follow-up cohort is, by construction, a survivor cohort. To be analyzed at minimum 2, minimum 5, or minimum 10 years, a patient must have remained alive, in clinic, and with the original implants in place. Patients are excluded if their shoulder was revised before the minimum follow-up, if they died, if they were lost to follow-up, or if they left the surgeon's practice. This is not random. The mechanism is the immortal time bias formalized by Suissa: when cohort entry depends on a span of time during which the outcome of interest (e.g., survivorship to the minimum follow-up time) could not occur, the estimated event rate is systematically biased toward the experience of survivors [2]. Including only patients with minimum two-year follow-up systematically excludes those revised or deceased before two years; the percentage excluded by this immortal time effect grows for minimum five-year follow-up and grows again for minimum ten-year follow-up. This exclusion leaves a “purified” sample of patients more likely to continue life without revision and does not reflect the overall revision risk of the initial cohort.
Three independent processes drive this attrition. Khan and colleagues used Medicare claims on 108,667 elective shoulder arthroplasty patients aged 65 or older and reported 5-year mortality of 14.9% in elective non-fracture cases [3]. Torrens and colleagues prospectively tracked 251 shoulder arthroplasty patients and reported cumulative loss to follow-up of 18.3% at 2 years, 31.5% at 5 years, and 34.3% at 7 years, with older, sicker, and more obese patients selectively lost (HR 1.05 per year of age, HR 2.44 for severe obesity, HR 1.93 per ASA point) [4]. After adding exclusion for pre-threshold revision, at minimum 10-year follow-up only a minority of the original cohort—roughly a quarter to a third—remains analyzable. The minimum-10-year revision rate commonly quoted from a published series is therefore drawn from the least at-risk fraction of the original patient population.
Figure 2. Approximate share of an original elective shoulder arthroplasty cohort still available for analysis at minimum 2-, 5-, and 10-year follow-up. Mortality components anchored to Khan 2024 [3] (14.9% at 5 years extrapolated to ~30% at 10 years); loss-to-follow-up components anchored to Torrens 2022 [4] (18.3% at 2 years, 31.5% at 5 years, extrapolated to ~36% at 10 years); pre-threshold revision approximated from registry data.
This compounds asymmetrically with the demographic confounding above. RSA cohorts are older and have higher all-cause mortality, so these cohorts lose more patients to death between landmark thresholds. RSA's particular failure modes—acromial fracture, dissatisfaction with limited internal rotation, persistent pain—are also more likely to lead to patients quietly dropping out of follow-up because there is often no good surgical option for them to consider. A published “minimum 5-year RSA-OA revision rate” of 3.5% is best read as 3.5% of the best segment of the original cohort, not the entire cohort. The same dynamic appears directly in Figure 1: at the 14-year time point the RSA <65 cohort retains only 53 of the original 3,439 patients (1.5%), producing the wide confidence interval that runs from 9.6% to 17.9%.
The metric problem: failure is not the same as revision
This is the largest of the three issues, and the one with the most specific mechanism. Revision rate is commonly used as a surrogate for clinical failure. It is a particularly poor surrogate for RSA. Parada and colleagues reported a 10.7% complication rate and 5.6% revision rate for aTSA, versus 8.9% and 2.5% for RSA, in 1,128 cases at mean follow-up of 23 months [5]. The gap between complication and revision was 3.6-fold for RSA versus 1.9-fold for aTSA. The most common RSA complication—acromial or scapular fracture (2.5%)—had a 0% revision rate [5].
O’Malley and colleagues' finding closes the loop: twice the prevalence of unsatisfactory function after RSA, and less than half the rate at which that unsatisfactory function leads to revision [6]. The lower RSA revision rate does not reflect superior implant performance. It reflects a higher threshold for revision in an older, frailer cohort, combined with failure modes that are difficult to address surgically and that surgeon and patient usually choose to accept rather than trying to fix surgically.
Figure 3. Cumulative revision rate (light bars) and cumulative clinical failure rate (dark bars) for aTSA and RSA at minimum 2-, 5-, and 10-year follow-up. Revision rates anchored to AOANJRR registry data [1]; clinical failure rates anchored to Parada complication rates [5] and O’Malley OSS<29 prevalence [6]. The clinical-failure-to-revision ratio is approximately 3-fold for aTSA across all time points and 5- to 6-fold for RSA.
The asymmetry has a specific cause: the failure modes that drive each operation's clinical burden carry very different probabilities of leading to revision. Most of the failure modes unique to RSA produce clinical problems that rarely lead to revision.
A mode-by-mode look
Figure 4. Estimated cumulative incidence of patients clinically affected by each failure mode (light bars) versus the proportion undergoing revision attributable to that mode (dark bars). Constructed from AOANJRR revision-cause distributions [1], Parada complication rates [5], O’Malley OSS<29 prevalence [6], systematic-review data on acromial fracture [7–9], Young secondary cuff dysfunction rates [14], Papadonikolakis glenoid radiolucency rates [10], Olson RSA instability rates [15], and Rojas baseplate loosening rates [16].
Acromial and scapular fracture. This mode is unique to RSA and represents the largest disconnect in the literature. Three independent systematic reviews place the incidence at 2.8% (Mahendraraj 2019) [8], 4.0% (Kim 2019) [9], and 5.0% (Patterson 2020) [7]. Of 208 fractures in Patterson's review, only 9 (4.3%) underwent revision arthroplasty [7]. Function deteriorates after fracture regardless of treatment. The clinical-failure-to-revision ratio is approximately 25:1.
Persistent pain. Pain is a major driver of unsatisfactory function but rarely the documented reason for revision. The OSS<29 prevalence in O’Malley—15.4% for aTSA and 27.0% for RSA—captures pain alongside stiffness, ADL limitation, and overall functional compromise [6]. The cleanest reading is the one O’Malley reports directly: RSA patients with poor function are less than half as likely as aTSA patients with poor function to be revised. Some unknown proportion of persistent pain after shoulder arthroplasty represents occult Cutibacterium infection that goes undiagnosed; Pottinger and colleagues showed that 22% of revisions performed for stiffness, pain, or loosening had positive cultures for Cutibacterium despite being categorized clinically as aseptic [11]. Patients with low-grade unrecognized infection who do not undergo revision are invisible in registry data altogether.
Cuff and subscapularis failure. This mode is aTSA-specific. Cuff insufficiency accounts for roughly a quarter of aTSA-OA revisions in the AOANJRR data [1]. The actual prevalence of post-aTSA cuff failure is substantially higher. Young and colleagues, in a multicenter European study of 518 aTSAs followed for a mean of 8.6 years, reported a 16.8% rate of secondary rotator cuff dysfunction, with Kaplan–Meier survivorship free of secondary cuff dysfunction of 100% at 5 years, 84% at 10 years, and 45% at 15 years [14]. Most of this dysfunction was managed with rehabilitation or accepted as part of the clinical course; only a fraction underwent revision. RSA bypasses the rotator cuff biomechanically, so this mode does not apply to RSA. Importantly, aTSA failure due to cuff or glenoid component issues can be revised to RSA with patient-reported outcome improvements [12,13]—a salvage option that is less successful for some of the major RSA-specific failure modes.
Asymptomatic glenoid radiolucency and loosening. aTSA-specific. Papadonikolakis, Neradilek, and Matsen's 2013 systematic review of 27 articles representing 3,853 aTSAs reported asymptomatic glenoid radiolucent lines accumulating at 7.3% per year, symptomatic loosening at 1.2% per year, and surgical revision for loosening at 0.8% per year [10]. Radiographically apparent loosening occurs approximately nine times more frequently than revision. Most asymptomatic radiolucent lines do not progress to clinical failure within the patient's lifetime, but those that do represent a substantial population not captured in revision rates.
Dislocation and instability. Both operations can fail by this mode, but RSA more often. Olson and colleagues' systematic review of 17 studies including 7,885 RSAs reported a pooled instability rate of 2.5%, ranging from 1–5% in primary RSA cohorts and 1–49% in revision RSA [15]. Across studies, treatment was successful with closed reduction and casting in 28–100% of cases and with revision RSA in 55–100%; recurrent instability often required hemiarthroplasty or resection. The clinical-failure-to-revision ratio for instability is approximately 1.5:1 to 2:1—smaller than for the modes above, but RSA-specific risk factors (subscapularis insufficiency, proximal humerus fracture, fracture sequelae) load this mode disproportionately onto the older RSA-for-fracture population.
Baseplate failure (RSA) and overt infection (both RSA and aTSA). These are the modes with the smallest disconnect. Aseptic glenoid baseplate loosening is rare: Rojas and colleagues' meta-analysis of 103 studies including 6,583 RSAs reported a pooled prevalence of 1.16% (95% CI 0.80–1.69%), with 0.90% in primary RSA and 1.69% in revision RSA [16]. When symptomatic, baseplate loosening almost always progresses to revision. Overt infection is similarly almost always revised. Both modes have clinical-failure-to-revision ratios close to 1:1—but the Pottinger finding indicates that the population of patients with low-grade unrecognized infection may be substantially larger than overt-infection counts suggest, and many of those patients neither receive a revision nor appear in registry infection data [11].
Bottom line
The AOANJRR shows a population-level revision-rate advantage for RSA over aTSA. It does not show a like-for-like advantage. Two sampling biases compound: an age and indication mismatch that places most RSA-for-OA cases in older patients to whom aTSA was never going to be offered, and an immortal time bias that removes most of an original cohort from the analyzed group at the 10-year mark, with the older RSA population hit hardest. The third issue is structural rather than statistical. Revision rate measures the surgeon's and patient's thresholds for re-operation, and does not reflect the patient's clinical outcome. The clinical-failure-to-revision ratio is two to three times larger for RSA than for aTSA.
This is not an artifact. It has a specific mechanism. RSA's distinctive failure modes—acromial fracture, persistent pain, scapular notching, dissatisfaction with internal rotation—are largely uncountable in revision data because there is often no good surgical option to consider. By contrast, aTSA failures from cuff or subscapularis tear and from glenoid component loosening can be revised to RSA with clinically significant outcome improvements [12,13]. Revision rate as a quality metric privileges the operation (RSA) whose failures are less likely to be surgically correctable.
Evidence for restricting RSA to its established indications—cuff arthropathy, irreparable cuff tear, selected fractures—rather than expanding it into osteoarthritis with intact cuff is present in the AOANJRR's data.
avocet
reverse avocet
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/
References
1. Australian Orthopaedic Association National Joint Replacement Registry. Hip, Knee and Shoulder Arthroplasty: 2025 Annual Report. Adelaide: AOA; 2025. Available from: https://aoanjrr.sahmri.com/annual-reports-2025. Data period 1 September 1999 – 31 December 2024.
2. Suissa S. Immortal time bias in pharmacoepidemiology. Am J Epidemiol. 2008;167(4):492–499. doi:10.1093/aje/kwm324. PMID: 18056625.
3. Khan AZ, Zhang X, Macarayan E, et al. Five-year mortality rates following elective shoulder arthroplasty and shoulder arthroplasty for fracture in patients over age 65. JBJS Open Access. 2024;9(2):e23.00133. doi:10.2106/JBJS.OA.23.00133. PMID: 38685966.
4. Torrens C, MartÃnez R, Santana F. Patients lost to follow-up in shoulder arthroplasty: descriptive characteristics and reasons. Clin Orthop Surg. 2022;14(1):112–118. doi:10.4055/cios21034. PMID: 35251548.
5. Parada SA, Flurin PH, Wright TW, Zuckerman JD, Elwell JA, Roche CP, Friedman RJ. Comparison of complication types and rates associated with anatomic and reverse total shoulder arthroplasty. J Shoulder Elbow Surg. 2021;30(4):811–818. doi:10.1016/j.jse.2020.07.028. PMID: 32763380.
6. O’Malley O, Davies A, Rangan A, Sabharwal S, Reilly P. Is there a difference in thresholds for revision between shoulder arthroplasty types? A National Joint Registry study. PLoS One. 2025;20(8):e0330975. doi:10.1371/journal.pone.0330975. PMID: 40857270.
7. Patterson DC, Chi D, Parsons BO, Cagle PJ. Acromial spine fracture after reverse total shoulder arthroplasty: a systematic review. J Shoulder Elbow Surg. 2019;28(4):792–801. doi:10.1016/j.jse.2018.08.033. PMID: 30497925.
8. Mahendraraj KA, Abboud J, Armstrong A, et al. How common are acromial and scapular spine fractures after reverse shoulder arthroplasty? A systematic review. Bone Joint J. 2019;101-B(6):627–634. doi:10.1302/0301-620X.101B6.BJJ-2018-1187.R1. PMID: 31154841.
9. Kim HM, Chung J, Jeong CW, Yoon JR. Is acromial fracture after reverse total shoulder arthroplasty a negligible complication? A systematic review. Clin Orthop Surg. 2019;11(4):427–435. doi:10.4055/cios.2019.11.4.427. PMID: 31788166.
10. Papadonikolakis A, Neradilek MB, Matsen FA 3rd. Failure of the glenoid component in anatomic total shoulder arthroplasty: a systematic review of the English-language literature between 2006 and 2012. J Bone Joint Surg Am. 2013;95(24):2205–2212. doi:10.2106/JBJS.L.00552. PMID: 24352774.
11. Pottinger P, Butler-Wu S, Neradilek MB, Merritt A, Bertelsen A, Jette JL, Warme WJ, Matsen FA 3rd. Prognostic factors for bacterial cultures positive for Propionibacterium acnes and other organisms in a large series of revision shoulder arthroplasties performed for stiffness, pain, or loosening. J Bone Joint Surg Am. 2012;94(22):2075–2083. doi:10.2106/JBJS.K.00861. PMID: 23172325.
12. Al-Asadi M, Rajapaksege N, Abdel Khalik H, Abesteh J, Athwal GS, Khan M. Outcomes and complications of failed anatomic shoulder arthroplasty revised with reverse arthroplasty: a systematic review. J Shoulder Elbow Surg. 2025;34(7):1832–1840.
13. Tobin JG, Thomas SK, Elwell JA, Roche CP, Rogalski BL, Eichinger JF, Friedman RJ. Anatomic total shoulder arthroplasty revised to reverse total shoulder arthroplasty: clinical and radiographic outcomes compared to primary reverse total shoulder arthroplasty. J Shoulder Elbow Surg. 2025;34(6):1525–1531. PMID: 39694226.
14. Young AA, Walch G, Pape G, Gohlke F, Favard L. Secondary rotator cuff dysfunction following total shoulder arthroplasty for primary glenohumeral osteoarthritis: results of a multicenter study with more than five years of follow-up. J Bone Joint Surg Am. 2012;94(8):685–693. doi:10.2106/JBJS.J.00727. PMID: 22419408.
15. Olson JJ, Galetta MD, Keller RE, Oh LS, O’Donnell EA. Systematic review of prevalence, risk factors, and management of instability following reverse shoulder arthroplasty. JSES Rev Rep Tech. 2022;2(3):261–268. doi:10.1016/j.xrrt.2022.02.009. PMID: 37588866.
16. Rojas J, Choi K, Joseph J, Srikumaran U, McFarland EG. Aseptic glenoid baseplate loosening after reverse total shoulder arthroplasty: a systematic review and meta-analysis. JBJS Rev. 2019;7(5):e7. doi:10.2106/JBJS.RVW.18.00132. PMID: 31145263.