We performed a retrospective cohort study using the TriNetX Research Network, a federated database of de-identified electronic health records from over 100 U.S. healthcare organizations. TriNetX provides access to de-identified patient demographics, diagnoses, procedures, medications, and laboratory results using International Classification of Diseases, 10th Revision (ICD-10), Current Procedural Terminology (CPT), and LOINC coding systems. All data are de-identified and are thus exempt from institutional review board oversight.

We identified patients who underwent primary RCR between May 1, 2005, and May 1, 2024. To avoid confounding, we excluded patients with any history of shoulder fracture, previous RCR, or shoulder arthroplasty at any point prior to surgery. Patients with documented hypoglycemia were also excluded. A full list of inclusion and exclusion criteria is provided in Supplementary Table 1.

Patients were divided into two cohorts based on preoperative glucose measurements recorded within the 24-hour period preceding surgery. Only patients with at least one recorded glucose value in this window were included in the analysis. The hyperglycemic cohort consisted of patients with at least one glucose measurement exceeding 180 mg/dL. The normoglycemic cohort consisted of patients with all recorded glucose values between 70 and 179 mg/dL. Glucose values were obtained using a harmonized TriNetX variable that aggregates multiple LOINC codes across serum, plasma, blood, and capillary samples, ensuring consistency across institutions.

To reduce baseline differences between groups and minimize the effects of confounding, we performed 1:1 propensity score matching using logistic regression within the TriNetX platform. A greedy nearest-neighbor algorithm with a caliper of 0.1 pooled standard deviations was applied, ensuring that matched pairs differed in propensity scores by no more than this threshold. Matching was performed using variables available up to and including the day of surgery.

Matching variables were demographic characteristics (age, sex, race, ethnicity, and body mass index), social history (smoking status and alcohol use), comorbidities (hypertension, coronary artery disease, congestive heart failure, chronic kidney disease, and nutritional deficiencies), presence and type of diabetes, hemoglobin A1c (HbA1c), and use of glucose-modifying agents (insulin, glucocorticoids, GLP-1 receptor agonists, sulfonylureas, sodium-glucose cotransporter 2 inhibitors, dipeptidyl peptidase 4 inhibitors, and metformin). Patients were also matched on procedural factors including type of rotator cuff tear (e.g., partial vs. full-thickness, single vs. multi-tendon) and surgical approach (open vs. arthroscopic repair). A full list of variables and their standardized mean differences before and after matching is presented in Table 1, and a schematic overview of cohort selection is shown in Fig. 1.

Postoperative complications were assessed across two predefined follow-up windows: a short-term period (1 to 90 days following surgery) and a long-term period (1 to 365 days following surgery). These intervals were selected to capture both early surgical complications and delayed medical or musculoskeletal sequelae. Short-term complications included sepsis, myocardial infarction, acute kidney injury (AKI), postoperative infection (including superficial, deep, and device-related infections), urinary tract infection, postoperative pain, pulmonary embolism, and deep vein thrombosis. Long-term complications included stress fracture of the upper arm or shoulder, frozen shoulder (adhesive capsulitis), revision RCR, shoulder arthroplasty, and removal of hardware. The full list of diagnosis and procedure codes used to define each outcome is provided in Supplementary Table 2. For each outcome, we report the cumulative incidence, along with the risk ratio (RR), 95% CI, and false discovery rate (FDR)-corrected P-value.

To evaluate whether findings were robust to the definition of the normoglycemic reference group, we conducted a subgroup sensitivity analysis in which the normoglycemic cohort was restricted to patients with all recorded preoperative glucose values between 70 and 139 mg/dL. This more stringent threshold was selected to exclude patients with mild hyperglycemia (140–179 mg/dL) from the reference group, thereby providing a cleaner comparison against confirmed normoglycemic patients. The hyperglycemic cohort (glucose ≥180 mg/dL) remained unchanged. The same 1:1 propensity score matching procedure and matching variables were applied. The same prespecified outcomes were assessed across identical follow-up windows.

For each outcome, cumulative incidence was calculated for both cohorts, and the RR with 95% CI is reported. P-values were derived from the z-test for risk difference comparing event rates between matched cohorts. To account for the risk of type I error associated with multiple comparisons, p-values were adjusted using the Benjamini-Hochberg FDR correction method. Adjusted p-values less than 0.05 were considered statistically significant.

Time-to-event analysis was performed using the Kaplan-Meier method, with between-group differences assessed using the log-rank test. The proportional hazards assumption was verified using the Schoenfeld residuals test. Hazard ratios (HRs) with 95% CIs are reported for time-to-event outcomes. All statistical analyses were performed using the TriNetX analytics platform, with the exception of FDR correction, which was performed post-hoc using Python 3.10.16 and statsmodels 0.14.5.

Before matching, the cohort comprised 6,174 patients with preoperative hyperglycemia and 23,023 patients with normoglycemia. Following 1:1 propensity score matching, 5,839 matched pairs remained for analysis. As shown in Table 1, all covariates were successfully balanced with standardized mean differences less than 0.1, indicating appropriate balancing between groups.

At 90 days postoperatively, patients in the hyperglycemic cohort experienced significantly higher rates of multiple adverse events than normoglycemic controls (Table 2). The risk of sepsis was more than doubled in the hyperglycemic group (0.70% vs. 0.26%; RR, 2.73; 95% CI, 1.51–4.93; P-FDR=0.002). AKI was also significantly elevated (1.27% vs. 0.62%; RR, 2.06; 95% CI, 1.38–3.06; P-FDR=0.002), as was myocardial infarction (0.60% vs. 0.33%; RR, 1.84; 95% CI, 1.06–3.22; P-FDR =0.047), postoperative infection (0.75% vs. 0.41%; RR, 1.83; 95% CI, 1.12–3.01; P-FDR=0.030), and postoperative pain (4.14% vs. 3.15%; RR, 1.32; 95% CI, 1.09–1.59; P-FDR=0.011). Other complications, including pulmonary embolism, urinary tract infection, and deep vein thrombosis, showed no statistically significant differences between groups after FDR correction (all P-FDR >0.05).

By one year postoperatively, patients with preoperative hyperglycemia continued to experience higher rates of certain musculoskeletal complications (Table 2). The incidence of stress fracture of the upper arm or shoulder was significantly elevated in the hyperglycemic group (1.80% vs. 1.18%; RR, 1.52; 95% CI, 1.13–2.06; P-FDR=0.015), corresponding to a number needed to harm of 162. Frozen shoulder was also significantly more common among hyperglycemic patients (7.02% vs. 5.75%; RR, 1.22; 95% CI, 1.06–1.40; P-FDR=0.015), corresponding to a number needed to harm of 79. Time-to-event analysis further confirmed a significantly higher cumulative incidence of stress fracture in the hyperglycemic cohort over the 365-day follow-up period (HR, 1.54; 95% CI, 1.13–2.08; log-rank P=0.005) (Fig. 2). There were no significant differences between cohorts in rates of shoulder replacement (0.62% vs. 0.87%, P-FDR=0.177), surgical revision (0.22% vs. 0.22%, P-FDR=1.000), or removal of hardware (0.26% vs. 0.41%, P-FDR=0.186).

To assess the robustness of our primary findings, we repeated the analysis restricting the normoglycemic reference group to patients with all preoperative glucose values between 70 and 139 mg/dL, yielding 4,671 matched pairs. Results were largely consistent with the primary analysis (Table 3). Among short-term complications, hyperglycemic patients demonstrated significantly higher rates of sepsis (0.71% vs. 0.24%; RR, 3.00; 95% CI, 1.52–5.93; P-FDR=0.004), AKI (1.18% vs. 0.36%; RR, 3.24; 95% CI, 1.88–5.57; P-FDR <0.001), postoperative infection (0.73% vs. 0.39%; RR, 1.89; 95% CI, 1.07–3.34; P-FDR=0.046), myocardial infarction (0.56% vs. 0.28%; RR, 2.00; 95% CI, 1.03–3.89; P-FDR=0.046), pulmonary embolism (0.49% vs. 0.24%; RR, 2.09; 95% CI, 1.02–4.28; P-FDR=0.046), urinary tract infection (1.28% vs. 0.81%; RR, 1.58; 95% CI, 1.05–2.37; P-FDR=0.046), and postoperative pain (4.05% vs. 3.00%; RR, 1.35; 95% CI, 1.09–1.67; P-FDR=0.021). Among long-term complications, stress fracture incidence remained significantly elevated in the hyperglycemic cohort (2.01% vs. 1.09%; RR, 1.84; 95% CI, 1.31–2.59; P-FDR=0.002). Frozen shoulder showed a consistent direction of effect but did not reach statistical significance after FDR correction (6.47% vs. 5.72%; RR, 1.13; 95% CI, 0.96–1.33; P-FDR=0.217).

Patients with preoperative hyperglycemia experienced significantly higher rates of sepsis (0.70% vs. 0.26%) and postoperative infection (0.75% vs. 0.41%) within 90 days of RCR. Urinary tract infection did not reach significance in the primary analysis but became significant in the sensitivity analysis, suggesting a similar but weaker association. These findings are consistent with prior work by Wang et al. [11], who identified preoperative hyperglycemia as a significant predictor of postoperative infection in non-orthopedic surgical populations, even among patients without diagnosed diabetes. This association is both biologically plausible and well-supported by the literature. Elevated blood sugar has been shown to impair multiple aspects of innate immunity, including neutrophil chemotaxis and phagocytosis, and can even directly foster bacterial proliferation and biofilm formation in healing tissue [20,21].

AKI was significantly more common among patients with preoperative hyperglycemia than normoglycemic controls (1.27% vs. 0.62%). While the association between diabetes and renal dysfunction is well established, the link between preoperative hyperglycemia and AKI after orthopedic surgery is less defined. This association is biologically plausible as well, however, as even short periods of hyperglycemia may impair renal function through multiple mechanisms. For instance, elevated glucose levels can directly cause renal injury through increased oxidative stress and formation of advanced glycation end-products within the kidney [22,23]. Elevated blood sugar can also lead to volume depletion via osmotic diuresis, leading to reduced renal perfusion and increasing the risk of ischemic injury. Together, these pathophysiologic changes may be especially harmful in the postoperative setting, where kidney function is already more susceptible to stress and instability.

Preoperative hyperglycemia was also associated with a significantly higher rate of myocardial infarction within 90 days of surgery (0.60% vs. 0.33%; RR, 1.84; 95% CI, 1.06–3.22; P-FDR=0.047). Pulmonary embolism did not reach significance in the primary analysis but became significant in the sensitivity analysis (RR, 2.09; 95% CI, 1.02–4.28; P-FDR=0.046). These findings are consistent with a growing body of literature linking perioperative hyperglycemia to adverse cardiovascular outcomes across surgical specialties [12,13,24]. In a large prospective cohort of nearly 12,000 patients undergoing non-cardiac surgery, Punthakee et al. [13] demonstrated that preoperative glucose concentration independently predicted myocardial injury after surgery, particularly in patients without known diabetes. Park et al. [12] further showed that preoperative hyperglycemia was associated with myocardial injury after noncardiac surgery. Crucially, HbA1c was not, which further supports the notion that acute glucose elevation may capture perioperative risk that chronic glycemic markers alone cannot. Hyperglycemia is well known to promote endothelial dysfunction, platelet aggregation, oxidative stress, and inflammation, all of which may increase the risk of acute coronary events in the perioperative period [16,25]. Given that cohorts were matched on coronary artery disease and congestive heart failure, this association is unlikely to be fully explained by baseline cardiovascular comorbidity. Still, residual confounding from unmeasured factors cannot be excluded.

Stress fracture of the upper arm or shoulder was significantly more common among patients with preoperative hyperglycemia at 1-year follow-up (1.80% vs. 1.18%). This was further supported by time-to-event analysis (HR, 1.54; log-rank P=0.005). Our findings align with a large body of literature demonstrating that individuals with diabetes, particularly those with poor glycemic control or longer disease duration, are at elevated risk for fractures [26]. However, the effects of short-term or preoperative hyperglycemia on bone health remain poorly understood, especially in the context of RCR. Several biological pathways may help explain this association. As mentioned previously, even brief episodes of hyperglycemia can lead to the formation of advanced glycation end-products, which accumulate in collagen-rich tissues like bone and disrupt normal collagen cross-linking [27]. Hyperglycemia can also alter bone remodeling by inhibiting osteoblast activity and promoting osteoclast-mediated resorption, further contributing to reduced bone strength and impaired tendon-to-bone healing [28]. In the context of RCR, this combination of weakened bone and impaired healing may leave the shoulder particularly vulnerable to fracture as patients return to functional activities. However, it is important to note that fracture risk is multifactorial, and a single preoperative glucose value is unlikely to be a direct driver of this outcome. As mentioned previously, preoperative hyperglycemia may instead serve as a marker of underlying frailty that predisposes patients to skeletal complications independent of the surgical procedure itself.

Frozen shoulder was also significantly more common among patients with preoperative hyperglycemia at the 1-year follow-up (7.02% vs. 5.75%). This finding aligns with prior studies linking diabetes to adhesive capsulitis. For example, Borton et al. found that diabetic patients were over four times more likely to develop frozen shoulder following RCR than non-diabetic controls [29], and a meta-analysis by Dyer et al. [30] reported a pooled odds ratio of 3.69. While these prior studies focused on long-standing diabetes, our findings suggest a similar association in patients with preoperative hyperglycemia. The pathogenesis for this association is likely multifactorial. One proposed mechanism involves the accumulation of advanced glycation end-products, which can cross-link collagen fibers and contribute to capsular fibrosis [31]. Hyperglycemia has also been shown to stimulate proinflammatory cytokine release, which has been found at elevated levels in the capsule and synovium of patients with frozen shoulder [32,33]. However, this association did not reach significance in the sensitivity analysis and the effect size was modest, and should therefore be interpreted with caution.

Notably, rates of revision surgery, shoulder replacement, and hardware removal did not differ between groups. This pattern suggests that preoperative hyperglycemia may have a greater effect on systemic and metabolic complications than surgical outcomes. This is also consistent with the notion that preoperative hyperglycemia serves as a marker of underlying metabolic vulnerability rather than a direct driver of surgical complications. Nonetheless, these complications carry significant clinical consequences, and preoperative glucose measurement may still serve as a useful tool for identifying patients at elevated perioperative risk.

This study has several important limitations. First, the retrospective nature of our analysis using electronic health record data from TriNetX precludes any conclusions about causality. Although we performed 1:1 propensity score matching to balance covariates such as age, body mass index, diabetes status, and medication use, unmeasured factors such as surgical complexity, perioperative glucose management, and nutritional status may have influenced outcomes. The observational nature of this study prevented us from fully distinguishing whether hyperglycemia functions as an independent risk factor or simply reflects underlying metabolic dysfunction. Second, the definition of hyperglycemia relied on a single preoperative glucose value, which introduces several sources of potential misclassification. The dataset also did not specify whether glucose values were obtained in a fasting or postprandial state, meaning that some elevations may reflect recent food intake or perioperative stress rather than true metabolic dysregulation. Furthermore, the TriNetX platform does not support continuous exposure modeling, precluding a dose-response analysis across finer glucose categories. Patients without a recorded glucose value were excluded from both cohorts, meaning our findings apply specifically to patients in whom glucose was measured preoperatively. The decision to test may have been driven by clinical suspicion of dysglycemia, thereby introducing selection bias. Third, our outcomes were defined using ICD-10 and CPT codes, which are subject to misclassification and incomplete capture. Importantly, the ICD-10 codes used to define fracture and frozen shoulder outcomes do not specify laterality, which made it impossible to confirm whether these complications occurred on the operative shoulder or the contralateral side. Fourth, the study period spans nearly two decades, during which surgical techniques, perioperative anesthesia protocols, and rehabilitation practices evolved considerably. Finally, because our analysis was limited to patients receiving care at large health systems contributing to the TriNetX network, findings may not be fully generalizable to populations treated in smaller community hospitals or international healthcare settings.

Future studies should investigate whether optimizing glucose levels prior to RCR can reduce the risk of complications such as infection, delayed healing, and postoperative stiffness. Additional research is also needed to clarify whether brief episodes of hyperglycemia are themselves modifiable contributors to these complications, or instead serve as early markers of underlying metabolic dysfunction.