Long-term outcomes of hemiarthroplasty and anatomic total shoulder arthroplasty for atraumatic humeral head avascular necrosis

Abstract

Background

Avascular necrosis (AVN) of the humeral head arises from diverse etiologies, yet comparative outcome data across etiologic subgroups remain limited. This study evaluated whether AVN etiology influences functional recovery, implant survival, and mortality following shoulder arthroplasty.

Methods

We conducted a retrospective review of arthroplasty cases for primary atraumatic humeral head AVN performed between 2003 and 2023. Patients who underwent total shoulder arthroplasty, hemiarthroplasty, humeral head resurfacing, or partial humeral head resurfacing were included. Revision-free survival and overall survival were evaluated with Kaplan–Meier analysis, and predictors of mortality were assessed with multivariable binary logistic regression. American Shoulder and Elbow Surgeons (ASES) score, Simple Shoulder Test, and range of motion (ROM) data were compared in shoulders with at least 1 year of follow-up using linear mixed-effects models. A post hoc Cruess grade sensitivity analysis was performed in shoulders with adequate pre-operative radiographs by 2 blinded graders.

Results

Of 51 shoulders in 42 patients, 60.8% were female, and the mean age at surgery was 55.1 ± 11.5 years. Five revisions (9.8%) occurred. Kaplan–Meier revision-free survivorship was 90.2% at 5 years and 82.0% at 10 years, with no significant differences by procedure type or AVN etiology. Overall patient survival was 79.7% at 10 years and 75.9% at 15 years. Pre-operative American Society of Anesthesiologists physical status was the only variable significantly associated with mortality after adjustment for age, sex, procedure group, and etiology group (odds ratio, 13.47 per 1-point increase; 95% CI, 1.65–110.15; P = .015; the wide confidence interval reflects the small number of deaths and complete separation, and this estimate should be regarded as imprecise). In the patient-reported outcome measure subgroup (n = 31), idiopathic AVN cases had lower modeled baseline ASES scores but faster improvement over time in ASES, Simple Shoulder Test, and multiple ROM outcomes than known-etiology cases (n = 19). A Cruess grade sensitivity analysis (n = 43) showed no difference in radiographic stage between groups by either blinded grader (P = .889 and .773).

Conclusion

In this cohort, idiopathic AVN demonstrated more rapid improvement in patient-reported outcomes and ROM over time despite worse modeled baseline function. This patient-reported outcome finding derives from an exploratory subgroup (n = 31) and is hypothesis-generating. With only 5 revision events, revision comparisons should be interpreted cautiously. In contrast, mortality was associated with overall comorbidity burden, with pre-operative American Society of Anesthesiologists status the only significant predictor of long-term survival.

The humeral head accounts for approximately 7% of all avascular necrosis (AVN) cases, trailing only the femoral head.^5,7,13,14,21 Etiologies for humeral head AVN can be divided into post-traumatic and atraumatic causes. Atraumatic AVN, also referred to as primary AVN, includes both cases with identifiable systemic etiologies and idiopathic cases. Reported causes of atraumatic AVN include chronic corticosteroid use, hematologic dyscrasias, human immunodeficiency virus (HIV), ethanol use, and others.¹⁶

Surgical management of humeral head AVN includes joint-preserving procedures such as core decompression and arthroscopic débridement (15.8% of surgical interventions); however, arthroplasty remains the predominant treatment modality (82.4%).¹² Options include hemiarthroplasty (HA), anatomic total shoulder arthroplasty (TSA), reverse total shoulder arthroplasty (rTSA), and humeral head resurfacing (HHR).

Prior arthroplasty outcomes for humeral head AVN are compared with the present findings in the Discussion.

The purpose of this study was to evaluate functional outcomes, revision-free survival (RFS), and mortality after anatomic shoulder arthroplasty (TSA, HA, HHR, and partial HHR [PHHR]) for primary atraumatic humeral head AVN, with particular emphasis on differences between idiopathic and known etiologies. We restricted the cohort to atraumatic disease and to anatomic and humeral-side procedures: post-traumatic AVN is a mechanistically distinct entity with a different natural history, and rTSA was represented by only 2 cases, too few for meaningful comparison. We hypothesized that AVN etiology would be associated with post-operative functional recovery, implant survival, and mortality, and that other patient-related factors may also contribute to these outcomes. Longitudinal functional recovery was the primary outcome; RFS and mortality were prespecified secondary, exploratory outcomes.

Materials and methods

This retrospective study reviewed all patients diagnosed with primary AVN who underwent shoulder arthroplasty by the senior author (M.A.F.) between January 1, 2003, and December 31, 2023, at a single institution. Institutional review board approval was obtained prior to the initiation of this retrospective study, and the requirement for individual patient consent was waived. The inclusion criteria were a diagnosis of primary (atraumatic) AVN of the humeral head, supported by characteristic radiographic findings (subchondral lucency or sclerosis, the crescent sign, or articular collapse consistent with Cruess staging) in the absence of antecedent fracture or dislocation. Patients who underwent rTSA were excluded. A total of 106 shoulders in 94 patients were identified; 29 were excluded due to post-traumatic AVN, 16 were excluded due to insufficient clinical data, 8 were excluded due to co-existing glenohumeral osteoarthritis, arthropathy, or septic arthritis, and 2 who underwent rTSA were excluded, resulting in a cohort of 51 shoulders from 42 patients (Fig. 1). Post-traumatic AVN was excluded because it is a mechanistically and prognostically distinct entity.¹¹ It was further excluded because the cohort was abstracted specifically for atraumatic disease; over the long study interval, an institutional electronic medical record transition limited retrieval of the clinical data and serial radiographs required for these analyses in post-traumatic patients. rTSA was excluded because it is a non-anatomic reconstruction with distinct biomechanics and failure modes, and only 2 such cases were identified.

Figure 1.

Flowchart of patient inclusion and exclusion. A total of 106 shoulders in 94 patients were identified; 29 were excluded due to post-traumatic AVN, 16 due to insufficient clinical data, 8 due to co-existing glenohumeral osteoarthritis or arthropathy, and 2 due to reverse total shoulder arthroplasty, resulting in a final cohort of 51 shoulders from 42 patients. AVN, avascular necrosis.

This study conforms to the Strengthening the Reporting of Observational Studies in Epidemiology guidelines for observational cohort studies (Supplemental Table S1).³⁷ Operative notes and post-operative imaging were reviewed to confirm correct identification of the following implant types: TSA, HA, HHR, and PHHR.

All patient records were reviewed, and the American Shoulder and Elbow Surgeons (ASES) score, Simple Shoulder Test (SST), and range of motion (ROM) (forward flexion [FF], abduction [Abd], internal rotation [IR], and external rotation [ER]) at pre-operative and all available follow-up visits were collected. American Society of Anesthesiologists (ASA) physical status classification, body mass index, smoking status, anxiety status, AVN etiology, and comorbidities were also recorded. A thorough review of patient records and death records was conducted to determine whether mortality had occurred.

Mortality and revision analysis

Patients without confirmed mortality were censored at the study end date for mortality analyses, whereas shoulders without revision were censored at the date of last follow-up for analyses of RFS. Continuous variables were summarized as means and standard deviations. Categorical variables were reported as frequencies and percentages, with comparisons performed using 2-sided Fisher's exact test to account for small cell frequencies. Etiology was grouped as idiopathic versus known (steroid, HIV, Caisson disease, chemotherapy, or sickle cell disease [SCD]); procedure type was grouped as TSA versus non-TSA (HA, HHR, PHHR), a grouping based on glenoid resurfacing that follows prior comparative work; and ASA status was grouped as low (ASA 1–2) versus high (ASA 3–4) for descriptive and survival analyses.^4,10 RFS and overall mortality were analyzed using the Kaplan–Meier method with group comparisons performed using the log-rank test. Multivariable analyses of mortality were performed using binary logistic regression including age at surgery, sex, procedure group, etiology group, and ASA status. With 10 deaths and 5 candidate predictors, this model had fewer than the commonly recommended 5–10 events per variable and was therefore overfit; it is reported as hypothesis-generating. Because complete separation occurred when ASA was entered as a grouped variable—all 10 deaths occurred in the high-ASA group and none in the low-ASA group—the final multivariable mortality model used the raw ASA score to estimate the effect of each 1-point increase in ASA on mortality. Complete separation of this kind indicates that the data cannot stably estimate the ASA effect and is the source of the wide confidence interval (CI), so this estimate is interpreted as hypothesis-generating rather than as a precise effect size. Statistical analyses were performed using IBM SPSS Statistics for Mac, version 29.0 (IBM Corp., Armonk, NY, USA).

One-year follow-up subgroup

For the patient-reported outcome measure (PROM) analysis, we excluded 20 shoulders that were missing PROM data or had less than 1 year of PROM follow-up. ASES, SST, and ROM data were compared in shoulders with greater than 1 year of PROM follow-up (n = 31) using linear mixed-effects models in Python using the statsmodels MixedLM module.³⁴ Separate models were fit for ASES total, SST, and ROM measures (FF, Abd, ER, and IR). Follow-up time was modeled continuously in months, and age at surgery was mean-centered. Fixed effects included etiology (known vs. idiopathic), procedure type (TSA vs. other), and ASA group (low vs. high), with time-by-predictor interaction terms to evaluate trajectories across groups. For each outcome, analyses used complete cases with non-missing values for the outcome being modeled, follow-up time, age, etiology, procedure type, and ASA. Models were estimated using restricted maximum likelihood; fixed effects are reported as coefficients with 95% CI and 2-sided P values. Interaction terms were additionally evaluated using likelihood ratio tests (maximum likelihood fits) comparing full models with interactions to nested models without interactions. The nominal significance threshold was set to α = 0.05.

Cruess grade sensitivity analysis

A post hoc radiographic sensitivity analysis was performed in the subset of shoulders with adequate pre-operative Grashey radiographs (n = 43 of 51). Each radiograph was independently reviewed by 2 fellowship-trained shoulder surgeons, blinded to clinical etiology and to one another's assessments, and assigned a Cruess stage (I–V).⁵ Interrater agreement was quantified using exact percent agreement and Cohen's kappa with both linear and quadratic ordinal weighting. Cruess grade was compared between idiopathic and known-etiology AVN using the Mann-Whitney U test, dichotomized as early-stage (I–II) versus advanced-stage (III–V) disease and compared with Fisher's exact test, and compared across individual etiology categories using the Kruskal-Wallis test. All comparisons were repeated for each blinded grader rather than relying on consensus grading.

Results

Demographics

A total of 51 shoulders in 42 patients were analyzed. The mean age at surgery was 55.1 ± 11.5 years, and 60.8% were female. AVN etiologies included idiopathic (n = 17) and known causes (n = 34), the latter comprising steroid-associated (n = 26), HIV-associated (n = 4), and smaller numbers of Caisson disease, chemotherapy, and SCD (Table I).

Table I.

Baseline demographic and clinical characteristics of the full cohort (N = 51 shoulders, 42 patients) by procedure type.

Variable	Total (N = 51 shoulders)	TSA (n = 24)	PHHR (n = 15)	HA (n = 7)	HHR (n = 5)	P value
Age at surgery (yr), mean ± SD	55.1 ± 11.5	57.8 ± 10.1	50.2 ± 13.0	55.6 ± 12.3	56.9 ± 9.9	.255
Female sex, n (%)	31 (60.8)	16 (66.7)	9 (60.0)	2 (28.6)	4 (80.0)	.295
Right side, n (%)	27 (52.9)	13 (54.2)	6 (40.0)	6 (85.7)	2 (40.0)	.241
ASA physical status ≥3, n (%)	38 (74.5)	19 (79.2)	9 (60.0)	6 (85.7)	4 (80.0)	.571
Follow-up (yr), median (IQR)	3.6 (1.0-9.3)	4.5 (1.4-9.7)	2.1 (0.8-8.7)	6.4 (3.5-7.5)	2.6 (0.9-9.9)	.830
AVN etiology, n (%)						.091
Idiopathic	17 (33.3)	12 (50.0)	4 (26.7)	1 (14.3)	0 (0.0)
Corticosteroid use	26 (51.0)	9 (37.5)	6 (40.0)	6 (85.7)	5 (100.0)
HIV	4 (7.8)	1 (4.2)	3 (20.0)	0 (0.0)	0 (0.0)
Caisson disease	2 (3.9)	0 (0.0)	2 (13.3)	0 (0.0)	0 (0.0)
Chemotherapy	1 (2.0)	1 (4.2)	0 (0.0)	0 (0.0)	0 (0.0)
Sickle cell disease	1 (2.0)	1 (4.2)	0 (0.0)	0 (0.0)	0 (0.0)
Clinical outcomes, n (%)
Revision event (per shoulder)	5 (9.8)	1 (4.2)	2 (13.3)	1 (14.3)	1 (20.0)
Death during follow-up (deceased shoulders)	12 (23.5)	5 (20.8)	3 (20.0)	2 (28.6)	2 (40.0)
Unique deceased patients (per patient)	10/42 (23.8)	5/21 (23.8)	2/12 (16.7)	1/5 (20.0)	2/4 (50.0)

HA, hemiarthroplasty; HHR, humeral head resurfacing; PHHR, partial humeral head resurfacing; SD, standard deviation; TSA, total shoulder arthroplasty; IQR, interquartile range; ASA, American Society of Anesthesiologists; AVN, avascular necrosis; HIV, human immunodeficiency virus.

Table II.

Characteristics of the patient-reported outcome measure (PROM) subgroup (N = 31 shoulders) stratified by AVN etiology

Variable	Idiopathic AVN (n = 12)	Known etiology AVN (n = 19)	P value
Age at surgery (yr), mean ± SD	59.0 ± 10.1	54.5 ± 9.0	.221
Right side, n (%)	6 (50.0)	12 (63.2)	.710
ASA physical status ≥3, n (%)	9 (75.0)	13 (68.4)	1.000
Procedure type, n (%)			.738
Anatomic TSA (aTSA)	8 (66.7)	9 (47.4)
Partial HHR (PHHR)	3 (25.0)	5 (26.3)
Hemiarthroplasty (HA)	1 (8.3)	3 (15.8)
Humeral head resurfacing (HHR)	0 (0.0)	2 (10.5)
Known etiology breakdown, n (%)
Corticosteroid use	—	12 (63.2)
HIV	—	3 (15.8)
Caisson disease	—	2 (10.5)
Chemotherapy	—	1 (5.3)
Sickle cell disease	—	1 (5.3)
Pre-operative patient-reported outcome measures, mean ± SD
ASES total score	18.5 ± 15.0i (n = 11)	35.8 ± 16.1i (n = 13)	.013
SST score	1.4 ± 2.0 (n = 11)	3.5 ± 3.3 (n = 13)	.070
Pre-operative range of motion (°), mean ± SD
Forward flexion	50 ± 37 (n = 8)	94 ± 60 (n = 13)	.053
Abduction	48 ± 31 (n = 8)	87 ± 51 (n = 13)	.042
External rotation	12 ± 39 (n = 7)	28 ± 55 (n = 9)	.514
Internal rotation (Rowe scale)	1.7 ± 1.0 (n = 7)	4.6 ± 2.7 (n = 11)	.006
Clinical outcomes
Revision, n (%)	3 (25.0)	2 (10.5)	.350
Death during follow-up, n (%)	1 (8.3)	3 (15.8)	1.000

ASES, American Shoulder and Elbow Surgeons; ASA, American Society of Anesthesiologists; AVN, avascular necrosis; HA, hemiarthroplasty; HHR, humeral head resurfacing; PHHR, partial humeral head resurfacing; ROM, range of motion; SD, standard deviation; SST, Simple Shoulder Test; aTSA, anatomic total shoulder arthroplasty; HIV, human immunodeficiency virus; TSA, total shoulder arthroplasty; PROM, patient-reported outcome measure.

The idiopathic group includes patients with no identifiable precipitating cause. The known etiology group comprises all patients with an identified contributing factor. Pre-operative PROM and ROM sample sizes vary due to missing data at the pre-operative visit.

ⁱSuperscript denotes that pre-operative ASES data were not available for all patients in that group.

P values compare the idiopathic and known-etiology groups (independent-samples t-test for continuous variables; Fisher exact test for categorical variables). Range of motion values are reported in whole degrees.

Patient-reported outcome measure subgroup analysis

Within the mixed-effects model subgroup (Table II), idiopathic AVN cases (12/31 shoulders) had lower modeled baseline ASES scores than cases with known etiologies (19/31), with a mean difference of 27.0 points (P < .001). Known-etiology cases improved significantly more slowly across ASES, SST, FF, Abd, and ER, whereas IR did not differ by etiology (Table III; Figure 2, Figure 3, Figure 4). The modeled ASES difference narrowed to 16.7 points at 12 months and 6.5 points at 24 months.

Table III.

Fixed-effect estimates from linear mixed-effects models for longitudinal post-operative patient-reported outcome and range-of-motion measures after arthroplasty for atraumatic humeral head avascular necrosis.

Outcome	Month 0 difference, β (95% CI)	P value	Slope difference per month, β (95% CI)	P value
A. Etiology (known vs. idiopathic)
ASES total	26.97 (11.80 to 42.15)	<.001	−0.85 (−1.16 to −0.55)	<.001
SST	3.84 (1.35 to 6.33)	.0025	−0.10 (−0.15 to −0.05)	<.001
Forward flexion, °	24.78 (−11.57 to 61.12)	.1816	−1.15 (−1.92 to −0.38)	.0036
Abduction, °	31.03 (−2.28 to 64.33)	.0678	−1.60 (−2.31 to −0.88)	<.001
External rotation, °	11.12 (−11.04 to 33.27)	.3254	−0.69 (−1.21 to −0.17)	.0098
Internal rotation	0.64 (−3.21 to 4.49)	.7455	−0.07 (−0.18 to 0.04)	.2063
B. Procedure type (TSA vs. other procedures)
ASES total	3.92 (−10.86 to 18.71)	.6030	−0.03 (−0.19 to 0.13)	.7237
SST	−0.24 (−2.67 to 2.18)	.8434	0.02 (−0.01 to 0.05)	.1918
Forward flexion, °	−1.77 (−37.07 to 33.53)	.9217	0.12 (−0.31 to 0.55)	.5736
Abduction, °	−2.41 (−34.85 to 30.03)	.8842	0.05 (−0.34 to 0.44)	.8117
External rotation, °	−3.08 (−23.17 to 17.01)	.7636	−0.12 (−0.41 to 0.18)	.4377
Internal rotation	−1.55 (−5.14 to 2.05)	.3987	−0.01 (−0.07 to 0.05)	.6552
C. ASA group (ASA 3–4 vs. ASA 1–2)
ASES total	−16.92 (−33.35 to −0.48)	.0436	0.32 (0.14 to 0.49)	<.001
SST	−1.70 (−4.40 to 1.00)	.2173	0.02 (−0.01 to 0.06)	.1330
Forward flexion, °	−8.25 (−49.40 to 32.90)	.6942	0.30 (−0.18 to 0.77)	.2201
Abduction, °	−6.69 (−44.35 to 30.96)	.7276	0.06 (−0.38 to 0.50)	.7969
External rotation, °	−3.64 (−30.22 to 22.94)	.7885	0.09 (−0.24 to 0.43)	.5914
Internal rotation	−0.26 (−4.51 to 4.00)	.9054	−0.00 (−0.07 to 0.06)	.9579

ASES, American Shoulder and Elbow Surgeons; ASA, American Society of Anesthesiologists; CI, confidence interval; SST, Simple Shoulder Test; TSA, total shoulder arthroplasty; HIV, human immunodeficiency virus.

Month 0 difference represents the adjusted between-group difference at time 0. Slope difference per month represents the adjusted between-group difference in longitudinal rate of change. Reference groups were idiopathic AVN, other procedures, and ASA 1–2. Known etiology included steroid-associated AVN, HIV-associated AVN, sickle cell disease-associated AVN, caisson disease-associated AVN, and chemotherapy-associated AVN.

Bold P values indicate statistical significance (P < .05).

Figure 2.

Linear mixed-effects model–estimated trajectories of ASES total score over time by AVN etiology group (idiopathic vs. known). Shaded bands represent 95% confidence intervals. AVN, avascular necrosis; ASES, American Shoulder and Elbow Surgeons; TSA, total shoulder arthroplasty; ASA, American Society of Anesthesiologists.

Figure 3.

Linear mixed-effects model–estimated trajectories by SST score over time by AVN etiology group (idiopathic vs. known). Shaded bands represent 95% confidence intervals. AVN, avascular necrosis; SST, Simple Shoulder Test; TSA, total shoulder arthroplasty; ASA, American Society of Anesthesiologists.

Figure 4.

Linear mixed-effects model–estimated trajectories of range-of-motion outcomes (forward flexion, abduction, and external rotation) over time by AVN etiology group (idiopathic vs. known). Shaded bands represent 95% confidence intervals. AVN, avascular necrosis; TSA, total shoulder arthroplasty; ASA, American Society of Anesthesiologists.

Unadjusted analyses

Procedure group differed significantly by etiology group (P = .036). Idiopathic AVN was more frequently treated with TSA (12/17, 70.6%), whereas shoulders with known etiology were more frequently treated with non-TSA procedures (22/34, 64.7%). Procedure group did not differ by ASA group or sex.

ASA group differed significantly by sex (P = .008), with high ASA status more common among males than females. The ASA group did not differ by etiology or procedure group.

Death was significantly associated with the ASA group, with all observed deaths occurring in the high-ASA group (P = .040). Death was not associated with etiology (P = .270), procedure group (P = 1.000), or sex (P = .259). Revision was not significantly associated with procedure group, etiology group, ASA group, or sex (Table IV).

Table IV.

Unadjusted associations among key clinical variables, mortality, and revision

Comparison	Group 1, n/N (%)	Group 2, n/N (%)	P value
Procedure group (TSA), n/N (%)
Etiology	Idiopathic: 12/17 (70.6)	Known: 12/34 (35.3)	.036
ASA group	Low ASA: 5/13 (38.5)	High ASA: 19/38 (50.0)	.534
Sex	Female: 16/31 (51.6)	Male: 8/20 (40.0)	.567
ASA group (high ASA), n/N (%)
Sex	Male: 19/20 (95.0)	Female: 19/31 (61.3)	.008
Etiology	Idiopathic: 13/17 (76.5)	Known: 25/34 (73.5)	1.000
Procedure group	Non-TSA: 19/27 (70.4)	TSA: 19/24 (79.2)	.534
Death, n/N (%)∗
ASA group	High ASA: 10/30 (33.3)	Low ASA: 0/12 (0.0)	.040
Etiology	Known: 8/26 (30.8)	Idiopathic: 2/16 (12.5)	.270
Procedure group	Non-TSA: 5/21 (23.8)	TSA: 5/21 (23.8)	1.000
Sex	Male: 5/14 (35.7)	Female: 5/28 (17.9)	.259
Revision, n/N (%)
Procedure group	Non-TSA: 4/27 (14.8)	TSA: 1/24 (4.2)	.354
Etiology	Idiopathic: 3/17 (17.6)	Known: 2/34 (5.9)	.318
ASA group	Low ASA: 2/13 (15.4)	High ASA: 3/38 (7.9)	.591
Sex	Female: 4/31 (12.9)	Male: 1/20 (5.0)	.636

ASA, American Society of Anesthesiologists; TSA, total shoulder arthroplasty.

Bold P values indicate statistical significance (P < .05).

^∗Death analyses based on 42 unique patients; all other analyses based on 51 shoulders. P values are from 2-sided Fisher exact tests.

Revision-free survival

Kaplan–Meier analysis identified 5 revisions (9.8%) among 51 shoulders. Overall revision-free survivorship was 90.2% at 5 years and 82.0% at 10 years (Table V). Among censored shoulders, the median follow-up was 4.4 years (interquartile range, 0.9-9.7 years). Log-rank analysis demonstrated no significant differences in revision-free survivorship by procedure group, etiology group, or ASA group (Fig. 5). With only 5 revision events, these subgroup comparisons are underpowered and should not be interpreted as evidence of equivalence.

Table V.

Kaplan-Meier overall revision-free survival

Shoulders, n	Revisions, n (%)	1-yr survival, %	5-yr survival, %	10-yr survival, %	Mean revision-free survival, yr (95% CI)
51	5 (9.8)	95.4	90.2	82.0	18.4 (15.9-20.9)

CI, confidence interval.

Survival estimates were derived from Kaplan-Meier analysis of 51 shoulders. Estimated mean revision-free survival was 18.4 ± 1.3 years (95% CI, 15.9–20.9). Median revision-free survival was not reached.

Figure 5.

Kaplan-Meier revision-free survival after shoulder arthroplasty for atraumatic humeral head avascular necrosis. Five revisions occurred among 51 shoulders. Overall revision-free survivorship was 95.4% at 1 year, 90.2% at 5 years, and 82.0% at 10 years. Numbers at risk are shown below the x-axis. P values for subgroup comparisons were calculated with the log-rank test. TSA, total shoulder arthroplasty; ASA, American Society of Anesthesiologists.

Mortality

Kaplan–Meier analysis identified 10 deaths (23.8%) among 42 patients. Overall patient survival was 85.6% at 5 years, 79.7% at 10 years, and 75.9% at 15 years (Table VI). Among censored patients, median follow-up was 15.5 years (interquartile range, 9.6–19.2 years). Log-rank analysis demonstrated no significant differences in overall survival by procedure group or etiology group, whereas overall survival differed significantly by ASA group, with no deaths in the low-ASA group and all 10 deaths occurring in the high-ASA group (Table VII; Fig. 6).

Table VI.

Kaplan–Meier overall patient survival estimates at 1, 5, 10, and 15 years for the full cohort

Patients, n	Deaths, n (%)	1-yr survival, %	5-yr survival, %	10-yr survival, %	15-yr survival, %	Mean overall survival, yr (95% CI)
42	10 (23.8)	95.2	85.6	79.7	75.9	17.99 (15.48-20.50)

CI, confidence interval.

Table VII.

Kaplan-Meier overall survival stratified by pre-operative ASA group

ASA group	Patients, n	Deaths, n (%)	1-yr survival, %	5-yr survival, %	10-yr survival, %	15-yr survival, %
Low (1–2)	12	0 (0.0)	100.0	100.0	100.0	100.0
High (3–4)	30	10 (33.3)	93.3	79.7	70.9	64.4

ASA, American Society of Anesthesiologists.

Survival estimates were derived from Kaplan-Meier analysis of 42 unique patients. Overall survival differed significantly by ASA group (log-rank P = .017). No deaths occurred in the low-ASA group during follow-up. In the high-ASA group, the far-right tail of the curve was unstable because the final drop occurred at 21.317 years when only 1 patient remained at risk.

Figure 6.

Kaplan-Meier overall survival after shoulder arthroplasty for atraumatic humeral head avascular necrosis. Ten deaths occurred among 42 patients. Overall survival was 95.2% at 1 year, 85.6% at 5 years, 79.7% at 10 years, and 75.9% at 15 years. Numbers at risk are shown below the x-axis. P values for subgroup comparisons were calculated with the log-rank test. The apparent late decline in the extreme right tail of the curve should be interpreted cautiously because few patients remained at risk at those time points; accordingly, the 1-, 5-, 10-, and 15-year estimates are more informative than the far-right tail of the curve. TSA, total shoulder arthroplasty; ASA, American Society of Anesthesiologists.

In the multivariable logistic regression analysis, pre-operative ASA status was the only variable significantly associated with mortality after adjustment for age at surgery, sex, procedure group, and etiology group (P = .015), with each 1-point increase in ASA score associated with a 13.5-fold increase in the odds of mortality (odds ratio, 13.47; 95% CI, 1.65–110.15). Age at surgery, sex, procedure group, and etiology group were not independently associated with mortality (Table VIII). All Kaplan-Meier analyses are summarized in Table IX.

Table VIII.

Multivariable binary logistic regression analysis for predictors of mortality (n = 42 patients)

Predictor	Adjusted OR	95% CI	P value
Age at surgery, per 1-yr increase	0.939	0.864-1.020	.137
Male sex	2.083	0.256-16.970	.493
TSA vs. non-TSA	1.105	0.156-7.814	.920
Known etiology vs. idiopathic	0.290	0.036-2.326	.244
ASA score, per 1-point increase	13.473	1.648-110.152	.015

ASA, American Society of Anesthesiologists; CI, confidence interval; OR, odds ratio; TSA, total shoulder arthroplasty.

ASA score was entered as a continuous variable because complete separation occurred when it was dichotomized.

Bold P values indicate statistical significance (P < .05).

Table IX.

Kaplan-Meier survivorship summary by outcome and subgroup

Outcome	Comparison	Group	N	Events, n (%)	5-yr survival, %	10-yr survival, %	15-yr survival, %	KM-estimated mean survival, y	Log-rank P value
Revision-free survival	Overall	Overall	51	5 (9.8)	90.2	82.0	82.0	18.4	—
	Procedure	Non-TSA	27	4 (14.8)	86.2	71.8	71.8	15.1	.204
		TSA	24	1 (4.2)	94.7	94.7	94.7	20.2	.204
	Etiology	Idiopathic	17	3 (17.6)	75.0	75.0	—	11.3	.074
		Known	34	2 (5.9)	97.1	87.4	87.4	19.6	.074
	ASA group	Low ASA (1–2)	13	2 (15.4)	88.9	74.1	74.1	15.6	.645
		High ASA (3–4)	38	3 (7.9)	90.5	90.5	90.5	19.4	.645
Overall survival	Overall	Overall	42	10 (23.8)	85.6	79.7	75.9	18.0	—
	Procedure	Non-TSA	21	5 (23.8)	81.0	74.7	74.7	16.5	.655
		TSA	21	5 (23.8)	90.5	84.8	78.3	18.7	.655
	Etiology	Idiopathic	16	2 (12.5)	87.5	87.5	87.5	20.2	.227
		Known	26	8 (30.8)	84.6	75.7	69.4	16.6	.227
	ASA group	Low ASA (1–2)	12	0 (0.0)	100.0	100.0	100.0	22.8∗	.017
		High ASA (3–4)	30	10 (33.3)	79.7	70.9	64.4	15.6	.017

ASA, American Society of Anesthesiologists; KM, Kaplan-Meier; TSA, total shoulder arthroplasty.

Values are presented as n (%) or percentage survival. KM-estimated mean survival refers to the Kaplan-Meier estimated mean survival time. Dashes (−) indicate not applicable or not estimable at the specified time point. Revision-free survival analyses are based on 51 shoulders; overall survival analyses are based on 42 unique patients. For idiopathic revision-free survival, 15-year survival was not estimable because the largest observed follow-up time was censored at 14.7 years.

ASA, American Society of Anesthesiologists; KM, Kaplan-Meier; TSA, total shoulder arthroplasty.

Bold P values indicate statistical significance (P < .05).

^∗The low-ASA overall survival mean is a restricted estimate (all cases censored); survival remained 100% throughout the entire follow-up period.

Cruess grade sensitivity analysis

Pre-operative Grashey radiographs were adequate for Cruess grading in 43 of 51 shoulders (84.3%); the 8 excluded shoulders included the single chemotherapy-associated case. Exact interrater agreement was 22 of 43 (51.2%; unweighted κ = 0.30, indicating only fair raw agreement driven largely by adjacent-stage disagreement) and improved with ordinal weighting (linear κ = 0.47; quadratic κ = 0.64), reflecting predominantly adjacent-stage disagreements. Cruess grade did not differ significantly between idiopathic (n = 14) and known-etiology (n = 29) AVN by either blinded grader (Mann-Whitney U, P = .889 and P = .773), nor when stages were dichotomized as early-stage (I–II) versus advanced-stage (III–V) disease (Fisher's exact, P = 1.000 for both graders). Comparison across individual etiology categories was likewise nonsignificant (Kruskal-Wallis, P = .254 and P = .306). Both graders therefore yielded the same inferential conclusion that baseline radiographic stage was comparable between etiology groups.

Discussion

This study is among the first to statistically examine predictors of long-term mortality in patients undergoing shoulder arthroplasty for primary atraumatic humeral head AVN, and to our knowledge is the first to compare longitudinal functional recovery trajectories between idiopathic and known-etiology AVN using linear mixed-effects modeling. We found no significant association between AVN etiology and patient survival. Instead, our findings suggest that pre-operative ASA status, rather than the underlying etiology, was the factor most strongly associated with long-term mortality in this cohort, although the small number of events limits certainty. In multivariable analysis, ASA status was the only variable that reached statistical significance after adjustment for age, sex, procedure group, and etiology (P = .015). Given the limited number of mortality events, the multivariable model should be interpreted as hypothesis generating. The absence of a significant association for etiology does not establish that etiology is unimportant, as the study is underpowered to exclude such an effect.

Notably, no deaths occurred among the 12 patients with low pre-operative ASA status (ASA 1-2) during the entire follow-up period, whereas 10 of 30 patients (33.3%) with ASA 3-4 died. While the sample size limits definitive conclusions, the complete absence of mortality in the low-ASA group suggests that pre-operative medical optimization may be an important consideration for high-ASA patients undergoing arthroplasty for AVN.

Prior studies, including those by Hattrup and Cofield3, 15 and Bülhoff et al,3, 15 reported deaths during follow-up in arthroplasty cohorts with humeral head AVN, but these observations were primarily descriptive rather than analytically linked to specific patient factors. To our knowledge, our study is the first to statistically analyze the relationship between specific patient factors and mortality in this population. The clinical relevance of this finding is further strengthened by the relationship between patient sex and ASA status. In our cohort, high ASA status was significantly more common among males (95%) than females (61.3%; P = .008). However, when adjusted for ASA score in our multivariable model, sex was not independently associated with mortality (P = .493; Table VIII). This suggests that while certain demographics may appear “higher risk” on the surface, it is the cumulative systemic comorbidity burden, as captured by the ASA score, that truly influences long-term patient outcomes.

Notably, idiopathic AVN was not associated with a lower comorbidity burden than known etiologies; patients with idiopathic disease were just as likely to have high ASA status, reinforcing that pre-operative physiologic status, rather than etiology, is the more informative guide to long-term survival.

Our overall RFS of 82.0% at 10 years is somewhat lower than some reports in the current literature, although these cohorts included varying proportions of post-traumatic and atraumatic AVN, which limits direct comparison (Table X). Notably, we found no statistically significant difference in implant survival among procedure types or AVN etiologies. However, only 5 revision events occurred in the present cohort, so these subgroup revision analyses should be interpreted as exploratory. No prior study has demonstrated significant differences in implant survivorship among specific atraumatic etiologies. Ristow et al³¹ found no significant differences in their mixed AVN cohort. Kennon et al¹⁹ directly compared SCD and steroid-induced AVN but did not report significant differences in arthroplasty outcomes between these groups. Accordingly, our finding of similar implant survival across idiopathic and known etiologies should be interpreted cautiously, but it suggests that etiology may not be a major determinant of implant durability in atraumatic AVN.

Table X.

Literature review of shoulder arthroplasty for humeral head avascular necrosis. Studies are organized chronologically by era.

Author, year	Study design/focus	N (shoulders)	Mean follow-up	Implant construct	Key findings	Revision/survivorship	Mortality
Foundational series (1976-2005)
Cruess, 19765	Retrospective, steroid-induced single-etiology series	19 (atraumatic)	Variable	Predominantly HA	All 19 shoulders steroid-induced; 6 prostheses placed; satisfactory function with intermittent symptoms; no comparative etiology analysis; historically important for establishing steroid-associated disease as a distinct surgical entity	NR	NR
Basso, 19992	Conference/chapter-level case series; atraumatic AVN arthroplasty outcomes	NR	NR	Arthroplasty cohort	Early outcomes of arthroplasty for atraumatic AVN; limited extractable data; bibliographic accessibility restricted	NR	NR
Hattrup, 200015	Large retrospective series; etiology-outcome comparison	127 (88 at f/u)	8.9 yr	HA/aTSA	Steroid-induced AVN yielded superior outcomes vs. post-traumatic disease; ASES mean 63 (median 68); 79.5% subjective improvement; post-traumatic group had inferior ROM and pain scores; little difference between HA and TSA overall	NR	36 patients died during f/u (censored)
Parsch, 200328	Small prospective series; nontraumatic ON arthroplasty	13 (atraumatic)	NR	Arthroplasty cohort	Nontraumatic AVN: significant improvement in Constant score; progressive glenoid erosion noted in 2 patients; hemiarthroplasty identified as source of glenoid wear failure mode	NR	NR
Mansat et al, 200521	Retrospective case series; etiology-outcome comparison	19 (atraumatic)	7 yr (2–12)	HA/aTSA	Etiology mix: steroid (10), idiopathic (6), radiation (2), Gaucher (1); >80% pain-free or satisfactory; Constant: 58 (78% age-adjusted); 7 excellent, 9 satisfactory, 3 unsatisfactory; post-radiation cases yielded worst outcomes; glenoid wear and loosening observed	NR	NR
Mid-era comparative and resurfacing studies (2007-2012)
Lau et al, 200720	Etiology-specific case series; sickle cell subgroup	8 (SCD)	4.3 yr (0.2–10)	HA/aTSA	SCD-only cohort; ASES improved +31.9 points; ROM and function improved; pain relief less predictable—only 2/8 (25%) improved on VAS; 2 post-operative sickle cell crises; perioperative medical risk documented	NR	NR
Orfaly et al, 200726	Prospective consecutive series; etiology-outcome comparison	21 (12 atraumatic)	4.7 yr (2–8)	HA/aTSA	Steroid/idiopathic group had worse pre-operative function (SSI 26 vs. 49) but similar post-operative outcomes (SSI 78 vs. 82) compared to post-traumatic; SSI 36 → 81; SST 3 → 10; elevation 88°→123°; etiology affects baseline severity more than final outcome	NR	NR
Feeley et al, 20087	Retrospective comparative series; implant type and etiology both examined	64	4.8 yr	HA vs. aTSA (comparative)	No significant difference in outcomes by etiology overall; ROM worse in post-traumatic ON; no major implant-type difference; ASES 67; flexion 127°; ER 49°; complication rates higher with TSA	NR	NR
Smith et al, 200835	Steroid-only long-term HA series; single-etiology cohort	31 (atraumatic)	12.0 yr	Predominantly HA	Steroid-only cohort: 45% unsatisfactory results at long-term f/u; 38% moderate/severe pain attributed to progressive glenoid wear; pain 4.6 → 2.6 (P < .0001); elevation 92°→139°; ER 36°→65°; 42% excellent, 13% satisfactory, 45% unsatisfactory	2 revisions; 38% moderate/severe pain at f/u	NR
Uribe & Botto-van Bemden, 200936	Resurfacing case series; clinical outcomes	12 (atraumatic)	2.5 yr	Partial humeral head resurfacing	No etiology subgroup analysis; VAS pain 75 → 16 (P < .001); elevation 94°→142° (P < .001); good-to-excellent WOOS, SSI, and Constant scores; favorable short-term function in small AVN cohort	NR	NR
Raiss et al, 200929	Surface replacement series; etiology-outcome comparison	17 (9 atraumatic, 8 post-traumatic)	3.0 yr	Cementless surface replacement	Nontraumatic superior to post-traumatic: Constant 31 → 62 (34%→75% adjusted); flexion 87°→139°; better outcomes in atraumatic vs. post-traumatic subgroup; bone-preserving implant demonstrated low complication burden	14/17 satisfied or very satisfied; no revisions	NR
Gadea et al, 20129	Mixed-indication HA series; etiology-outcome comparison	272 total (40 AVN; 139 at ≥8 y f/u)	11.2 yr (134 mo)	Predominantly HA	AVN demonstrated best outcomes among all etiologies: Constant 70.1 (vs. OA 60.7, RA 55.3, fracture 57.7, CTA 46.2); lowest complication rate 15%; highest adjusted Constant and Neer scores in the AVN group at ≥8 yr	Kaplan–Meier survivorship reported; 30 prostheses removed overall; AVN 10-yr survival 94.9%	NR
Modern comparative, registry, and systematic review studies (2016-2025)
Kennon et al, 201619	Retrospective cohort; core decompression and arthroplasty; SCD vs. steroid	25 (atraumatic): 13 SCD, 12 steroid	1–2 yr	Core decompression + arthroplasty (mixed)	All SCD patients (100%) progressed to collapse following core decompression vs. 88% overall; arthroplasty yielded 81% satisfactory-to-excellent results in advanced disease; significant functional improvement in all scores; arthroplasty superior to joint preservation in advanced SCD	NR	NR
Schoch et al (atraumatic), 201632	Retrospective cohort; atraumatic AVN; HA vs. aTSA	141 (138 at f/u)	9.3 yr	HA/aTSA (comparative)	Etiology mix: steroid (67), idiopathic (35), alcohol (17), SCD (9), other (13); significant pain improvement (P < .001); elevation and ER improved (P < .01); satisfaction >75%; Neer excellent/satisfactory >65%; HA favored when glenoid cartilage preserved	HA: 8 reoperations; TSA: 11 reoperations; 20-yr KM survivorship: HA 87%, TSA 79%	NR
Schoch et al (post-traumatic), 201633	Retrospective cohort; post-traumatic ON; HA vs. aTSA	93 (83 at f/u)	8.9 yr	HA/aTSA (comparative)	Post-traumatic cohort only; TSA superior to HA: less pain (2.1 vs. 3.0, P = .001), better satisfaction (70% vs. 56%), better Neer ratings (57% vs. 41%); inferior survivorship vs. atraumatic (15-yr vs. 20-yr data)	HA: 9 reoperations; TSA: 5 reoperations; 15-yr KM survivorship: HA 79.5%, TSA 83%	NR
Ristow et al, 201931	Retrospective cohort; mixed etiologies; clinical outcomes	29	3.9 yr (1–8.5)	HA/TSA (comparative)	All groups improved significantly (P < .01) regardless of implant, etiology, or disease stage; trauma group had numerically highest scores in 3/4 measures but differences were not significant (P > .05); TSA scored higher than HA; tempers claims of consistent etiology-driven outcome differences	No revisions; 2 minor complications	NR
Miyazaki et al, 202124	Long-term HA series; minimum 10-yr f/u; clinical outcomes	NR	≥10 yr (mean ∼16 yr)	HA	Minimum 10-yr hemiarthroplasty follow-up; radiographic glenoid erosion worsened over time but clinical function remained acceptable; implant survival 84.7%; long-term HA durability documented despite progressive glenoid changes. [Rev Bras Ortop (Sao Paulo). 2021;56(1):91–97]	Kaplan–Meier survivorship: 84.7% implant survival	NR
Hervé et al, 202117	Multicenter retrospective cohort; primary AVN patients <60 years old	127 (atraumatic): 108 HA, 19 TSA	8.0 yr	HA vs. aTSA (comparative)	No significant clinical differences between HA and TSA; both improved outcomes; primary atraumatic AVN only (post-traumatic excluded); young cohort (<60 yr); HA favored for implant durability in younger active patients	HA revision 2% (2/108, glenoid wear); TSA revision 26% (5/19, loosening/instability)	NR
Alkhateeb et al, 20211	Systematic review; SCD-specific; surgical treatment of humeral head AVN	NR (pooled SCD cohorts)	Variable	HA/aTSA (mixed)	SCD-specific systematic review; arthroplasty improves function but pain relief less predictable; 100% of SCD patients progress to collapse following core decompression; perioperative sickle crises documented; core decompression should be avoided in SCD. [Published in JSES Int. 2021;5(3):391–397]	NR (pooled)	NR
McLaughlin et al, 202223	Multinational database cohort; rTSA vs. aTSA; clinical outcomes	119 (rTSA 69, aTSA 50)	rTSA 3.9 yr/aTSA 4.5 yr	rTSA vs. aTSA (comparative)	Both rTSA and aTSA improved ROM and PROMs significantly; rTSA: greater abduction improvement (+51° vs. +32°); aTSA: greater IR improvement; difference in UCLA and pain scores did not exceed clinically important thresholds; extends implant question beyond HA vs. aTSA	NR	NR
Wang et al, 202238	Long-term single-surgeon cohort; HA vs. aTSA; survivorship >10 yr	16 (atraumatic)	HA 13.0 yr/TSA 13.8 yr	HA vs. aTSA (comparative)	Both HA and TSA improved significantly; TSA better IR (2.3 vs. −3.0°, P = .03); no significant overall difference in outcomes or survivorship; one of the cleanest long-term survivorship comparisons in the literature	KM survivorship: HA 88.9%, TSA 80% at 10 y (NS)	NR
Rhee et al, 202330	Comparative series; nontraumatic AVN only; HA vs. TSA	26 (atraumatic)	8.0 yr (96.4 mo)	HA (HHR) vs. TSA (comparative)	Nontraumatic cohort only; no etiology subgroup analysis; VAS 6.3 → 2.2; UCLA 11.6 → 28.9; ASES 35.0 → 82.6 (P < .001); HHR yielded greater abduction (123.1° vs. 96.9°); satisfactory outcomes and implant longevity in both groups	TSA: 2 reoperations for infection; no revisions for loosening	NR
Marigi et al, 202322	Matched cohort; SCD vs. matched OA controls; etiology-specific outcomes	17 SCD (matched)	5.9 yr	HA/aTSA (mixed)	SCD vs. matched OA: higher pre-operative VAS (9.1 vs. 3.8, P < .001) and post-operative VAS (3.8 vs. 1.3, P < .001); ASES 48.6 → 73.5; 2 sickle cell crises; post-operative pain persistently worse than OA controls despite functional gains; strongest contemporary etiology-specific prognostic study. [J Shoulder Elbow Surg. 2023;32(2):247–252]	29% complications; 18% reoperations	NR
Na et al, 202325	Systematic review; outcomes and complications after arthroplasty in ON of the humeral head	NR (pooled)	Variable	Mixed (HA, aTSA, rTSA)	Comprehensive systematic synthesis of outcomes and complications following shoulder arthroplasty for humeral head ON; confirms feasibility of pooled analysis in this population; essential for novelty assessment. [J Orthop Sci. 2023;28(4):772–777]	Pooled revision and complication rates reported	NR
Parel et al, 202427	Large retrospective matched cohort; registry-based; 10-yr survivorship HA vs. aTSA	6,794 (4,825 TSA; 1,969 HA)	10 yr (KM)	HA vs. aTSA (comparative)	No significant difference in 10-yr revision-free survivorship between TSA and HA for humeral head AVN; largest comparative survivorship analysis in the field; directly strengthens revision-free survival as a defensible endpoint	10-yr KM survivorship comparable between HA and aTSA (NS)	NR
Iyer et al, 202418	Retrospective case series; HA after solid-organ transplant; transplant/steroid subgroup	3 shoulders	NR	Hemiarthroplasty	Humeral head HA in solid-organ transplant recipients with immunosuppression-related AVN; perioperative considerations specific to transplant patients documented; very small series—relevant for rare etiology subgroup	NR	NR
Bülhoff et al, 20253	Long-term resurfacing series; monocentric longitudinal observational study	17 (atraumatic)	19 yr (17–22)	Cementless humeral head resurfacing	Age-adjusted CMS 44.4%→82.9% (P < .01); no high-grade glenoid erosion or loosening; 2 revisions (glenoid erosion, rotator cuff tear defect); mature long-term resurfacing durability data; only 7 shoulders available for final clinical/radiographic review	KM survivorship: 93.3% at 15 yr; 2 revisions	5 patients died with implant in situ (censored)
Daher et al, 20256	Meta-analysis; HA vs. aTSA; secondary synthesis	504 (pooled)	7.9 yr	HA vs. aTSA (meta-analysis)	Across 8 comparative studies: HA lower complications (11.6% vs. 29.8%, P < .001); HA lower revisions (7.7% vs. 17.6%, P = .05); HA better forward flexion (+19°, P < .001) and IR; no clear ASES difference; indication bias limits conclusions	HA: lower revisions (7.7% vs. 17.6%, P = .05)	NR
Current study
Cohen et al, 2026	Retrospective cohort; etiology–outcome comparison; mortality predictor analysis	51 shoulders (42 patients)	Median 4.4 yr (revision); 15.5 yr (mortality)	HA/aTSA/HHR/PHHR	Idiopathic AVN: lower modeled baseline ASES (−27.0 pts, P < .001) but significantly faster rate of improvement (+0.85 pts/mo interaction, P < .001); ASA score the only independent mortality predictor (OR 13.47 per 1-point increase, 95% CI 1.65–110.15, P = .015); no significant differences in RFS by etiology, procedure, or ASA class	RFS: 95.4% at 1 yr, 90.2% at 5 yr, 82.0% at 10 yr; 5 revisions (9.8%)	10 deaths (23.8%); overall survival 85.6% at 5 yr, 79.7% at 10 yr, 75.9% at 15 yr

ASES, American Shoulder and Elbow Surgeons score; ASA, American Society of Anesthesiologists; aTSA, anatomic total shoulder arthroplasty; AVN, avascular necrosis; CI, confidence interval; CMS, Constant–Murley score; CTA, cuff tear arthropathy; ER, external rotation; f/u, follow-up; HA, hemiarthroplasty; HHR, humeral head resurfacing; IR, internal rotation; KM, Kaplan–Meier; NR, not reported; NS, not statistically significant; OA, osteoarthritis; ON, osteonecrosis; OR, odds ratio; PHHR, partial humeral head resurfacing; PROM, patient-reported outcome measure; RA, rheumatoid arthritis; RFS, revision-free survival; ROM, range of motion; RTSA, reverse total shoulder arthroplasty; SCD, sickle cell disease; SSI, Shoulder Score Index; SST, Simple Shoulder Test; TSA, total shoulder arthroplasty; UCLA, University of California, Los Angeles shoulder score; VAS, visual analog scale; WOOS, Western Ontario Osteoarthritis of the Shoulder Index.

Superscript numbers correspond to references in the manuscript reference list.

In this study, known-etiology atraumatic humeral head AVN was associated with slower improvement in PROMs and ROM over time than idiopathic AVN after adjustment for age, procedure type, and ASA status. Patients with known-etiology AVN began with higher baseline function but improved more slowly, narrowing the modeled difference over the first 24 months. These findings suggest that AVN etiology may influence the trajectory of post-operative recovery rather than simply the baseline level of dysfunction. Because idiopathic cases began substantially lower and the groups converged over time, part of this steeper slope may reflect regression to the mean and a ceiling effect rather than etiology alone, and the trajectory difference should be read as consistent with, but not proof of, an etiology effect. Because procedure type also differed by etiology, with idiopathic cases more often treated with TSA, the etiology effect cannot be fully separated from procedure selection despite adjustment.

Because idiopathic AVN demonstrated lower modeled baseline ASES and SST scores in our PROM subgroup, we considered whether this baseline difference might reflect more advanced pre-operative radiographic disease in the idiopathic group rather than a true etiology effect. To address this, a post hoc Cruess grade sensitivity analysis was performed in the radiographically gradable subset (n = 43). Two blinded graders demonstrated no significant difference in Cruess grade between idiopathic and known-etiology AVN, whether modeled ordinally, dichotomized as early-stage versus advanced-stage disease, or compared across individual etiology categories. This supports the interpretation that the differing recovery trajectories observed between idiopathic and known-etiology AVN were not driven by baseline radiographic disease severity. These findings are consistent with Ristow et al³¹, who similarly reported that neither radiographic stage nor etiology significantly stratified post-operative outcomes in their mixed humeral head AVN arthroplasty cohort.

Our finding that idiopathic cases started with worse modeled baseline function but demonstrated more rapid post-operative improvement extends prior work by Orfaly et al²⁶, who reported that patients with steroid-associated and idiopathic AVN had worse pre-operative function but achieved similar post-operative outcomes compared to post-traumatic AVN.

The less favorable recovery pattern observed in our known-etiology cohort is consistent with prior etiology-specific literature. Smith et al³⁵ found that 45% of steroid-associated hemiarthroplasties had unsatisfactory results at 12-year follow-up. Lau et al²⁰ found that only 25% of patients with SCD improved on visual analog scale despite gains in ROM and function, and Kennon et al¹⁹ reported progression to collapse in all SCD patients after core decompression, suggesting a potentially more aggressive disease course in that population. Together, these prior studies support the possibility that certain known AVN etiologies may be associated with more resistant or less predictable functional recovery than idiopathic disease.

The strong association between pre-operative ASA status and mortality may have practical implications for surgical planning. In patients with greater comorbidity burden, it may be reasonable to consider procedures that offer more predictable early functional recovery, particularly when long-term survival may be limited by systemic health rather than implant-related factors. This interpretation should be viewed as hypothesis-generating, as our study was not designed to directly compare procedure selection strategies by comorbidity burden.

This study has several strengths. It directly compared idiopathic and known atraumatic AVN etiologies using linear mixed-effects modeling with adjustment for age, procedure type, and ASA status, allowing evaluation of longitudinal functional recovery while accounting for important confounders. Additional strengths include long-term follow-up and inclusion of mortality as a clinically meaningful outcome. The decision to analyze multiple known etiologies within a single comparison group also reflects a practical clinical framework, as these patients often share systemic disease burdens that may influence post-operative recovery and long-term survival.

This study also has several limitations. Foremost, the multivariable mortality model was overfit: 5 candidate predictors were evaluated for only 10 deaths, complete separation occurred when ASA was dichotomized, and the resulting CI was wide, so the mortality findings are hypothesis-generating rather than definitive. Second, procedure type and etiology were confounded, as idiopathic cases were more often treated with TSA, so the adjusted etiology effect cannot be fully separated from procedure selection. Its retrospective design introduces the potential for selection bias and unmeasured confounding. The PROM subgroup was modest in size (31 shoulders), which may have limited power to detect smaller differences, particularly in subgroup analyses. In addition, the known-etiology group was heterogeneous, and combining steroid-associated, HIV-associated, Caisson disease-associated, chemotherapy-associated, and sickle cell-associated AVN may have obscured etiology-specific differences. Because procedure type differed by etiology, the observed recovery differences may partly reflect selection bias related to glenoid status, surgeon preference, or implant era. Grouping HA, HHR, and partial resurfacing as non-TSA procedures also combines implants that may differ in failure mode, although these humeral-side procedures have shown clinically equivalent function in prior work; our subgroups were too small to analyze each separately.^8,39 Finally, only 5 revision events occurred, so revision subgroup analyses should be interpreted as exploratory and likely underpowered.

In addition, the post hoc radiographic sensitivity analysis was limited to the 43 of 51 shoulders with adequate pre-operative Grashey radiographs, and exact interobserver agreement on Cruess staging was modest, although ordinal-weighted agreement was substantial and the inferential conclusions were concordant between the 2 blinded graders. Cross-sectional imaging with magnetic resonance imaging would be expected to provide a more sensitive and reproducible assessment of disease stage and may be valuable in future etiology-specific investigations of humeral head AVN.

The wide CI around the ASA odds ratio (1.65–110.15) reflects the limited sample size and small number of mortality events (n = 10). This point estimate should be interpreted with this imprecision in mind. Future studies with larger cohorts will be needed to determine whether specific non-idiopathic etiologies differ meaningfully in functional recovery, implant survival, or mortality.

Conclusions

This study was undertaken to examine the effects of AVN etiology and procedure type on functional outcomes, implant survival, and mortality after arthroplasty for atraumatic humeral head AVN. Idiopathic AVN was associated with worse modeled baseline function but more rapid post-operative improvement over time, suggesting that etiology influences recovery trajectory. Revision-free survivorship was 90.2% at 5 years and 82.0% at 10 years; with only 5 revision events, subgroup revision comparisons should be interpreted cautiously. In this small, exploratory cohort, higher pre-operative ASA status was associated with reduced long-term survival, a hypothesis-generating observation that requires confirmation in larger studies. These findings suggest that systemic comorbidity burden, rather than etiology or procedure type alone, may warrant emphasis when counseling patients about long-term survival, although the limited number of events tempers any firm conclusion.

Disclaimers:

Funding: Funding was graciously provided by the Tampa General Hospital Department of Orthopaedic Surgery.

Conflicts of interest: Kaitlyn N. Christmas reports that she is a paid consultant for Enovis for work related to the subject of this article. The institution with which she is affiliated received funding for this work.

Mark A. Frankle reports that he receives royalties, is a paid presenter or speaker, and receives research support and consulting fees from Enovis, as well as consulting fees from Synchrony Healthcare Communications, for work related to the subject of this article.

Peter Simon reports that the institution with which he is affiliated received funding for this work. He has not personally received any financial payments or other benefits from any commercial entity related to the subject of this article.

Any additional authors, their immediate families, and any research foundations with which they are affiliated have not received any financial payments or other benefits from any commercial entity related to the subject of this article.

Declaration of generative AI and AI-assisted technologies in the writing process

ChatGPT (GPT-5.5 Thinking; OpenAI, April 2026) was used for language editing/rewording from author-provided SPSS outputs to help with clarity of results section. No data, analyses, tables, or figures were AI-generated.