Patients considering shoulder surgery ask their surgeon, “How will my shoulder be doing years from now?” Not at two years – at ten and twenty years after surgery, when the 50-year-old will be older but still wanting to be able to work, garden, play with the kids and sleep comfortably.
Often the best data we can provide are two-year results from the fraction of patients who provided follow-up after their procedure. An important study from the University of Utah [1] found that even when a very well-motivated surgeon personally tried to contact 1,028 of his surgical patients by telephone, text and email, only 49% provided two-year follow-up. Younger patients and those having instability surgery – the patients with the most years ahead of them – were the most likely to be lost to follow-up.
There is no way to know how patients without followup are doing. Our own study of patients with rotator cuff tears showed how misleading that gap can be [2]. Patients who stopped returning mailed questionnaires could perform far fewer Simple Shoulder Test functions at their last report than those who kept responding (3.3 vs. 7.9 of 12). Even the way we ask changes the answer: responders reported the ability to perform more functions when asked by a phone call from a real person than when asked by mail.
Over ten or twenty years, many things change: implants appear and disappear, surgeons refine their techniques and their selection of patients, some patients die and others move or stop responding, barriers to returning for follow-up grow, and the measures of outcome change. Long-term studies in shoulder arthroplasty have captured as few as one-fifth and at most about one-half of the patients who had the operation [3].
Why does follow-up matter? Long-term follow-up improves the care of patients in two ways. First, it informs our specialty about what works well and what doesn't – which operations, implants and techniques serve patients well over the decades, and which fail late. Second, it informs each surgeon about what works and what doesn't in that surgeon's own hands. The surgeon is the method. A surgeon who asks how each patient is doing, studies every adverse outcome and adjusts their method accordingly is engaged in recursive self-improvement: each cycle makes the next one better. A surgeon who does not know their outcomes is not learning from their failures.
The anatomic glenoid component illustrates why both matter. Glenoid component failures are rarely noted in the first five years after total shoulder arthroplasty, then become progressively more common: in one multicenter series, survival free of radiographic loosening fell from 99% at five years to 34% at fifteen [4]. When we used a now outdated implant system, patients whose glenoid components failed had their self-assessed function fall back to their preoperative levels at a mean of seven years after surgery – long after a two-year study would have been closed [5].
Our challenge
As a specialty, we need a practical way to capture outcomes for as many of our patients as possible, for as long as possible – one that works across the specialty and in the everyday practice of community surgeons. It must be easy for patients, so that taking part is not a burden (for example, it can't require repeated office visits). And it must be easy for surgeons and investigators to request, record and analyze the results. Simple measures that capture most patients for decades are worth more than detailed measures on a few.
What is the simplest question we could ask every patient, every year, that would tell us what we most need to know?
The rest of this post makes the case in detail.
Why follow-up matters for the field
• Informed Consent. Patients should know their chances of being better than before, of further surgery and of complications – over the years that matter to them. Without long-term data from a representative group of patients, we can only guess, and our guesses may well be over-optimistic.
• Late failures declare themselves late. The anatomic glenoid component is the clearest example. In a European multicenter study of 226 total shoulder arthroplasties with a cemented, all-polyethylene, keeled glenoid component, survival free of glenoid revision was 99.1% at five years, 94.5% at ten and 79.4% at fifteen; survival free of radiographic loosening was 99.1%, 80.3% and 33.6% [4]. The authors noted that revision and loosening increased with time, but not until after five years, and recommended that studies of new glenoid designs report follow-up of at least five to ten years. Other late failures are similar: in one long-term series of reverse total shoulder arthroplasty, 10 of the 47 complications occurred at a mean of 8.3 years after surgery [6], and among 359 failed hemiarthroplasties revised at our referral center, 45% were revised two or more years after the index operation and 18% after five or more years [7]. A two-year study sees none of these.
Glenoid component survival after anatomic total shoulder arthroplasty, plotted from Young et al. [4]. The shaded band marks the period covered by a typical two-year study.
• Laboratory promise must be confirmed in patients. In the early 1990s, a new polyethylene, Hylamer, was used for the glenoid component of a widely used shoulder system because laboratory testing showed improved strength and resistance to creep and wear [8]. After implantation, however, components sterilized by gamma irradiation in air oxidized, became brittle and eroded. The problem surfaced only as surgeons following their patients began revising failed components and comparing what they found – cracking, delamination, fragmentation and abnormal wear. In one surgical team's revisions, the mean time from the index arthroplasty was 36.8 months (range, 32 to 99 months). The authors recommended monitoring patients with these components at six-month intervals, and concluded that each new bearing surface should be closely monitored clinically and radiographically to determine whether it functions as well in patients as it did in the laboratory.
• Implants and techniques need surveillance. New implants, modifications and techniques enter practice constantly, and none has a long-term record when it arrives. Only follow-up – by registries and by surgeons – reveals which ones serve patients well and which fail. In registries, though, the long term follow-up data may be on implants that are no longer in use.
• Implant survival is not patient outcome. Revision-free survival is easy to measure, but a retained implant may be painful, unstable or functionless. In the long-term reverse arthroplasty series cited above [6], prosthetic survival was 93% at a mean of more than 12 years, yet clinical scores had declined significantly from the medium-term evaluation. Only asking patients tells us how they are doing.
• Choosing among treatments. Comparing operations – anatomic or reverse arthroplasty, one design or another – requires knowing how each performs over the long term in similar patients. Comparisons built on two-year data, or on the minority of patients who remain in follow-up does not provide the information we need.
Why follow-up matters for the surgeon: recursive self-improvement
As craftsmen, each of us wants to get better at our craft. In surgery, the surgeon is the method: the choice of patient, the choice of operation, the technique and the aftercare all pass through the surgeon's judgment and hands. Improving outcomes for future patients means improving the surgeon's method.
Improvement requires feedback. A surgeon who operates, learns how the patient is doing, studies what went wrong when the outcome is adverse, and adjusts the method for the next patient closes the loop. Each pass through the loop improves the method that will be applied to every subsequent patient – and each improved patient outcomes becomes new information. That is recursive self-improvement, and it is how a surgeon's method becomes progressively better over a career.
Without follow-up, the loop is open. Experience without feedback is repetition, not learning. Worse, the feedback a surgeon does receive may be biased: the patients who keep returning may be disproportionately those who are doing well, while the unhappy patient who moves her care elsewhere, or the shoulder revised by another surgeon, never report back. A surgeon can come to believe that their results are better than they are. The failures that wander off are exactly the ones with the most to teach.
An example from our own practice. Between 1992 and 1997, 115 primary total shoulder arthroplasties for osteoarthritis were performed with a standardized technique with a certain glenoid component, and patients were mailed the Simple Shoulder Test (SST) every six months [5]. In the shoulders later revised by the original surgeon for isolated glenoid component failure, the SST had risen from a mean of 4.4 before surgery to 11.3 after it, then fallen to 4.6 – essentially the preoperative level – at a mean of 7.4 years (range, 2.9 to 11.1 years). Every one showed a drop of at least three SST functions from its peak, and in over half the drop appeared more than a year before the patient came in for evaluation. Periodic self-assessment told the surgeon, in his own practice and with his own technique, which shoulders were failing and when – information that no two-year study and no annual office visit schedule would have provided. As the authors noted, these failures occur beyond the duration of routine postsurgical follow-up, and the need to monitor patients persists throughout the life of the arthroplasty. With this information, use of that glenoid component was discontinued.
Because most of what we do turns out well, the greatest opportunity for improvement lies in the instances in which the outcome was adverse. Learning of those instances – wherever and whenever they occur – is the purpose of follow-up for the surgeon.
The evidence: who is lost, and what we don't know
The Utah study. The authors reviewed 1,028 consecutive patients who had shoulder surgery by a single surgeon between 2017 and 2022 [1]. At two years, every patient was personally contacted by the surgeon by telephone, text and email, followed by automated messages through the electronic medical record; a response by any method counted. Only 507 (49%) completed follow-up. Those who responded were older (mean 53 versus 49 years), more often had an arthroplasty and were more often insured by Medicare; patients having instability surgery and those with Medicaid or Workers' Compensation were more often lost. In the final model, younger age and a missing preoperative pain score independently predicted loss to follow-up, while current alcohol use predicted better follow-up. Neither the Social Deprivation Index nor distance from the hospital predicted follow-up. Importantly, preoperative Subjective Shoulder Value (48 versus 47) and pain scores (4.7 versus 4.6) did not differ between those who responded and those who did not.
This was big-effort study; a 49% response is what the diligent effort produced. But the study tells us who was lost, not how they fared. Similar starting scores tell us that responders and nonresponders began in the same place, not where they ended up.
What happens to the patients who stop answering? More than 25 years ago, we mailed the Simple Shoulder Test (SST) and the SF-36 every six months to 224 patients with full-thickness rotator cuff tears [2]. Sixty-seven (30%) responded regularly, 55 (25%) occasionally, and 102 (46%) stopped responding.
• Those who later stopped had preoperative SST scores similar to those who kept responding (4.69 versus 5.09), though lower SF-36 social function and mental health scores.
• On their last mailed questionnaire, the nonresponders were doing much worse than the responders: mean SST 3.26 versus 7.85.
• The method of asking mattered: 30 randomly chosen responders scored 9.03 by telephone and 7.47 by mail within the same month.
Two lessons follow. A patient's last known status may not be the current status. And scores obtained by different methods cannot simply be pooled.
A little arithmetic. Suppose 50 of 100 patients treated responded and that 80% of the patients who responded reported an improvement exceeding the minimal clinically important difference (MCID). What we know is that 40% of all patients treated reached the MCID. The true figure for the overall percent of patients reaching the MCID could lie anywhere from 40% (if none of the nonresponders improved) to 90% (if all of them did). Reporting 80% clinically significant improvement assumes that the nonresponders did exactly as well as the responders-something we cannot know.
Hypothetical: 100 patients treated, 50 respond | Patients reaching MCID | Reported rate |
Responders (40 of 50 improved) | 40 | 80% of responders |
None of the 50 nonresponders improved | 40 | 40% of all treated |
All of the 50 nonresponders improved | 90 | 90% of all treated |
The patients most often lost are those for whom long-term results matter most. Younger patients and patients having instability surgery were the least likely to respond. They are the ones with the most years ahead, the highest demands on their shoulders and the greatest concern about durability.
Why long-term follow-up is so hard: a lot happens over the long term
• Implants rise and disappear. The humeral and glenoid components a patient received 10 or 20 years ago may no longer be made, and today's implants have no long-term record at all.
• Surgeons change. With experience, surgeons refine their technique and their selection of patients for each operation. Results of a surgeon's early cases may not represent the same surgeon's later ones.
• Frail patients die. Their outcomes are neither successes nor failures, but their deaths leave a healthier group of survivors to be studied.
• Unhappy patients wander off. Dissatisfied patients may move their care to another surgeon or have a revision that ends follow-up of the original operation; patients with complications and revisions tend to be lost.
• Barriers to follow-up grow. Over the years patients move, become ill, find travel harder or costlier, and tire of questionnaires.
• Metrics change. The scores in use change over the decades, and so does how they are presented to the patient – in the office, by mail, online or by telephone – which, as shown above, can change the score itself.
• The sample shrinks. What remains is a small and potentially unrepresentative fraction of the patients who had the operation.
Each of these changes the group being studied between the day of surgery and the final follow-up. The patients reporting at ten years are not the cohort that had the operation, and the operation they had may not be the one being offered today.
Examples. A systematic review of reverse total shoulder arthroplasty with a minimum of 10 years of follow-up combined seven case series and 469 procedures [3]. The weighted mean loss to follow-up was 63% (range 47% to 80%); the studies used different scores (absolute and relative Constant score, Subjective Shoulder Value, ASES, SANE) and different methods of follow-up. Revision-free survivorship was reported as 88% at 10 years, yet quizzically 36% of patients had complications and 23% needed further revision. The 469 procedures were fewer than 1% of the 63,845 reverse arthroplasties performed in the United States in 2017 alone [9] - this small sample is unlikely to be representative of the overall picture.
In one series, 191 reverse arthroplasties were followed for a mean of 150 months [6]. By then, 79 prostheses were in patients who had died and 17 had been lost; clinical data were available for 87 - just over half.
A comparison of pegged and keeled cemented glenoid components from a single practice yielded a matched cohort of 36 shoulders at a mean of about nine years – a small fraction of the more than 600 total shoulder arthroplasties performed in that practice between 2001 and 2015 [10]. Many of the components studied are no longer in common use.
Our own experience is similar. In our study of anatomic total shoulder and ream-and-run arthroplasty with a minimum of 10 years of follow-up, 63 of the 127 eligible patients (50%) responded [11]. In our ream-and-run study at a minimum of five years, 228 of 333 patients agreed to follow-up beyond two years, and data were available for 201 – 60% of the patients who had the operation [12]. The results among those we heard from were good, but they are the results from the responders.
A final caution: comparing the mean score of the patients available at two years with the mean score of the smaller group available at ten years does not tell us whether shoulders change over time, because the two groups are different patients. Change over time is better studied by following the same patients at each interval.
A path forward, part 1: decide what we need to know, and make it easy to collect
Long-term follow-up is worth its cost only if it answers a question that matters today. Before starting, we need to decide what question is to be answered, and how many patients – both the number and the percentage of the original cohort – and how long a follow-up is needed to answer it.
A small core set of information, collected the same way at every interval, may serve better than an elaborate one:
• Is the patient alive, and is the original implant still in place? Deaths and revisions are outcomes, not reasons to drop a patient from the analysis.
• Has there been any further surgery on this shoulder, by anyone, anywhere – when, and why? Asking the patient directly is the most practical way to learn of revisions performed by other surgeons.
• Comfort and function, self-assessed, item by item. "At ten years, 80% could sleep comfortably on the operated side" is information a patient can use.
• The patient's overall judgment. "Would you have the surgery again?" or "Are you satisfied?"
Is "Are you happy you had the surgery?" sufficient? It is a valuable, nearly cost-free summary of the patient's own judgment. But it cannot stand alone. It blends the shoulder with expectations, gratitude, the relationship with the surgeon and how bad things were before; it clusters near the top of the scale; and it does not tell the next patient which functions to expect.
ASES, Constant or SST? The Constant score requires an examination and strength testing – a visit – which raises exactly the barrier that drives attrition. The patient-reported section of the ASES score can be completed remotely, but it yields a single weighted number. The Simple Shoulder Test asks 12 yes-or-no questions about specific functions, takes a couple of minutes, can be completed from anywhere and reports each function separately. Whatever the choice, the essential thing is to choose one measure and keep it – and to keep, or at least record, the method of administration.
Do we need radiographs? Not for every patient at every interval. Radiographs require a visit, and several common radiographic findings have shown weak relationships to patient-reported outcome; in our ream-and-run series, for example, glenoid medialization was not associated with the SST result [13,14]. Radiographs are essential when the question is about structure – how glenoid wear progresses, whether a design loosens – in which case standardized, comparable views (such as the axillary truth view) should be planned for a defined group of patients. They are also needed when a patient develops new symptoms. Whether periodic surveillance radiographs of asymptomatic shoulders improve outcomes by detecting silent failure early is plausible but not shown.
Making it easy to collect:
• Consent every patient for long-term follow-up before surgery, explaining why it matters; obtain a preoperative baseline; and organize automated reminders at fixed intervals – for example 1, 2, 5, 10, 15 and 20 years.
• Follow each patient's own trajectory over time, rather than comparing different groups of patients at different intervals.
• Report the follow-up rate with the characteristics of those lost, and report outcomes both as a percentage of those who responded and as a percentage of all patients treated.
A path forward, part 2: since the surgeon is the method, learning from failure
Even complete long-term follow-up yields averages, and averages from past patients do not predict the outcome for the individual patient in front of us [15]. What we can do is improve the method – ourselves.
Population registries. One of the world's most comprehensive population-based resources, the Australian Orthopaedic Association National Joint Replacement Registry [16], can inform the implants we choose, the implant modifications we adopt and the types of adverse outcome we can expect to encounter. Registries capture revisions wherever they are performed – precisely the failures that individual series miss.
Counterfactual analysis of failures. A complementary approach asks of each failure the counterfactual question: what might have been done differently to lower the risk of this complication [17,18]?
Personal continuous quality improvement. Each of us can learn from our own adverse outcomes, and the approach is straightforward [19]. For each adverse outcome in our own practice:
(1) Identify the mode of failure (for example, a displaced acromial stress fracture after reverse total shoulder arthroplasty).
(2) Identify the important parameters of our treatment (for example, the position of the glenosphere and humeral component).
(3) Compare these parameters in the case with those in similar patients in our own practice who did not have the complication.
(4) Note the differences and consider adjusting our technique in the direction of the controls (for example, making sure that the distance from the glenosphere center of rotation to the acromion is greater than the distance from the glenosphere center of rotation to the greater tuberosity).
This intrapractice analysis does not require a large cohort or decades of follow-up: every adverse outcome is information, whenever it occurs. But it does require that we learn of our failures. The patients who wander off, the revisions performed elsewhere and the shoulders that fail after a study's window has closed are exactly the cases from which we could learn the most.
The bottom line
Follow-up is how we learn which operations serve patients over the decades they care about. For the surgeon, it is the feedback loop through which each failure improves the care of the next patient. When a patient asks what their shoulder will be like in ten or twenty years, the honest answer today too often begins with what we know at two years among the half who answered. We can do better: decide in advance what we need to know, make it easy for patients to tell us, capture revisions and deaths wherever they occur, and study every adverse outcome to improve the method that matters most – ourselves.
Epilogue
Reference [5] demonstrates a possible approach for long-term followup in an individual surgeon's practice. Of 115 total shoulders, 12 had revision by the index surgeon for glenoid component loosening. Simple Shoulder Test scores were solicited annually after surgery. They averaged 4.4 out of 12 before surgery, rose to a mean of 11.3 after surgery and fell to a mean of 4.6 before revision surgery performed at an average of 7 years after the index arthroplasty. The charts below show the individual data for each of the 12 patients. The vertical axis shows the SST score, the horizontal axis shows the time after the index procedure. The last data point for each patient is the SST score prior to revision. What is remarkable about these data is the variability in the post operative SST pattern. Some patients reported good scores up to the point of a dramatic drop, while others had a progressive drop in scores over the years after an initially good response.
This example demonstrates that obtaining a singe score, such as the SST, may be a practical method for following patients after surgery and that a drop in the score may suggest the need for the patient to come back to the office for evaluation and X-ray.
Looking
The author of this post has no financial relationship with any orthopaedic company.
Follow on twitter: https://twitter.com/shoulderarth
Follow on facebook: https://www.facebook.com/frederick.matsen
Follow on LinkedIn: https://www.linkedin.com/in/rick-matsen-88b1a8133/
References
1. Soule S, Smith A, Staten T, Kirkham M, Zhang C, Presson A, Joyce CD, Tashjian RZ, Chalmers PN. Predictors of loss to long-term follow-up after shoulder surgery. J Bone Joint Surg Am. 2026;108(19):1536-42. doi:10.2106/JBJS.25.00767
2. Norquist BM, Goldberg BA, Matsen FA 3rd. Challenges in evaluating patients lost to follow-up in clinical studies of rotator cuff tears. J Bone Joint Surg Am. 2000;82(6):838-42.
3. Biner M, Klotz S, Andronic O, Karczewski DC, Zingg L, Karl W, Kriechling P. Long-term outcomes following reverse total shoulder arthroplasty: a systematic review with a minimum follow-up of 10 years. JBJS Open Access. 2025;10(2):e25.00025. doi:10.2106/JBJS.OA.25.00025
4. Young A, Walch G, Boileau P, Favard L, Gohlke F, Loew M, Molé D. A multicentre study of the long-term results of using a flat-back polyethylene glenoid component in shoulder replacement for primary osteoarthritis. J Bone Joint Surg Br. 2011;93(2):210-6. doi:10.1302/0301-620X.93B2.25086
5. Buckingham BP, Parsons IM 4th, Campbell B, Titelman RM, Smith KL, Matsen FA 3rd. Patient functional self-assessment in late glenoid component failure at three to eleven years after total shoulder arthroplasty. J Shoulder Elbow Surg. 2005;14(4):368-74.
6. Bacle G, Nové-Josserand L, Garaud P, Walch G. Long-term outcomes of reverse total shoulder arthroplasty: a follow-up of a previous study. J Bone Joint Surg Am. 2017;99(6):454-61.
7. Hackett DJ Jr, Hsu JE, Matsen FA 3rd. Primary shoulder hemiarthroplasty: what can be learned from 359 cases that were surgically revised? Clin Orthop Relat Res. 2018;476(5):1031-40. doi:10.1007/s11999.0000000000000167
8. Rockwood CA Jr, Wirth MA. Observation on retrieved Hylamer glenoids in shoulder arthroplasty: problems associated with sterilization by gamma irradiation in air. J Shoulder Elbow Surg. 2002;11(2):191-7.
9. Wagner ER, Farley KX, Higgins I, Wilson JM, Daly CA, Gottschalk MB. The incidence of shoulder arthroplasty: rise and future projections compared with hip and knee arthroplasty. J Shoulder Elbow Surg. 2020;29(12):2601-9.
10. Duey AH, Corvi JJ, Li T, Park A, Etigunta S, Patel AV, Bienstock D, Ranson W, Kantrowitz DE, Hurley E, Shukla D, Parsons BO, Flatow EL, Cagle PJ. Long-term clinical and radiographic outcomes of pegged vs. keeled glenoid components in total shoulder arthroplasty: a matched cohort study. J Orthop. 2025;65:290-5.
11. Sharareh B, Whitson AJ, Matsen FA 3rd, Hsu JE. Minimum 10-year follow-up of anatomic total shoulder arthroplasty and ream-and-run arthroplasty for primary glenohumeral osteoarthritis. J Shoulder Elbow Surg. 2024;33:1276-84. doi:10.1016/j.jse.2023.08.028
12. Stenson JF, Collins AP, Yao JJ, Sharareh B, Whitson AJ, Matsen FA 3rd, Hsu JE. Factors associated with success of ream-and-run arthroplasty at a minimum of 5 years. J Shoulder Elbow Surg. 2023;32:S85-S91. doi:10.1016/j.jse.2023.01.024
13. Somerson JS, Neradilek MB, Service BC, Hsu JE, Russ SM, Matsen FA 3rd. Clinical and radiographic outcomes of the ream-and-run procedure for primary glenohumeral arthritis. J Bone Joint Surg Am. 2017;99:1291-304.
14. Collins AP, Sheth MM, Stenson JF, Kahsai EA, Khoo KJ, Ogunleye T, Whitson AJ, Matsen FA 3rd, Hsu JE. Characterizing glenoid wear after hemiarthroplasty with concentric glenoid reaming: a study of 113 arthroplasties at a mean of 6.7 years of follow-up. J Shoulder Elbow Surg. 2026;35:995-1002.
15. Matsen FA 3rd. Objective ignorance: a flaw in our ability to predict patient outcomes. J Bone Joint Surg Am. 2025;107(22):2508-11. doi:10.2106/JBJS.25.00638
16. Australian Orthopaedic Association National Joint Replacement Registry. https://aoanjrr.sahmri.com/
17. Menendez ME, Matsen FA 3rd. Learning from surgical failures. J Bone Joint Surg Am. 2026;108(8):547-8. doi:10.2106/JBJS.25.01110
18. Matsen FA 3rd. Shoulder arthroplasty failure research. Shoulder Arthritis Blog, August 2025. https://shoulderarthritis.blogspot.com/2025/08/shoulder-arthroplasty-failure-research.html
19. Matsen FA 3rd. How a surgeon can learn from their own adverse outcomes – an example of intrapractice analysis in reverse shoulder arthroplasty. Shoulder Arthritis Blog, January 2026. https://shoulderarthritis.blogspot.com/2026/01/how-surgeon-can-learn-from-their-own.html
