Glucagon-like Peptide-1 Treatments and Recovery After Hip Surgery for Hip Fractures in People with Type 2 Diabetes

A retrospective cross-sectional study of prospectively collected data was done using deidentified electronic health record data from eight community hospitals in North Texas that belong to a large healthcare system in the United States. The Institutional Review Board approved this study as exempt under category 4 (secondary analysis of existing data sets). The study findings are per the Strengthening the Reporting of Observational Studies in Epidemiology guidelines.

Inclusion criteria for this study were adults aged 18 years and older, with diagnosis of T2DM admitted from the community to acute inpatient units from January 1, 2016, to August 28, 2023, who received International Classification of Disease, 10th Revision, Clinical Modification (ICD-10) diagnoses of FNF (S72.0), and had Current Procedural Terminology (CPT) codes for HA (27125 or 27236). Because CPT 27236 includes internal fixation of FNF fractures, the presence of ICD-10-Procedure Coding System codes for "Replacement of Hip Joint, Femoral Surface" (0SRS, 0SRR) was also necessary to ensure only HA was included. Patients admitted for revision hip surgery, those treated with THA, patients who left against medical advice, patients without demographic variables reported, and patients with a history of rheumatoid arthritis or concurrent myocardial infarction, traumatic brain injury, or other fractures were excluded. Overall, 291 primary encounters of HA to treat a FNF in patients with T2DM taking a GLP-1RA at the time of surgery were reported.

The comparator group of patients with T2DM not taking a GLP-1RA with a FNF treated with HA was selected through 1:1 random sampling from the same hospital system during the study period. No specific matching algorithm or time-based matching was done. Although patients were not explicitly matched by year of surgery, subsequent multivariable regression analyses were conducted, adjusting for key covariates (age, sex, body mass index [BMI] class, smoking status, and Elixhauser Comorbidity Index [ECI]) to minimize confounding and address baseline differences between groups.

Demographic variables included age, sex, smoking status, race, BMI, and the ECI for medical comorbidities.^26,27 ECI scores are based on the dichotomous presence of 31 comorbidities using ICD diagnosis codes. ECI accurately predicts readmissions and mortality in total joint arthroplasty patients.^28–30 Admission hemoglobin A1c (HgbA1c) was recorded when available.

Where available, ICD-10-Procedure Coding System procedure codes were used to determine the HA fixation technique (cemented [0SR*0*9], press-fit [0SR*0*A], or no qualifier [0SR*0*Z]). Specific details regarding the surgical approach (eg, anterior, anterolateral, and posterior) and individual surgeon information were unavailable and therefore not included in the analysis.

The variables analyzed included medical complications (cerebrovascular accident, deep vein thrombosis, pulmonary embolism, myocardial infarction, pneumonia, acute kidney injury, and hypoglycemic event) and surgical site infection (SSI) within the first 30 days, 90 days, and 1 year. The incidence of aspiration pneumonitis during the index hospitalization was evaluated using ICD-10-Clinical Modification diagnostic codes (J69.0). Reencounters were evaluated as inpatient readmissions (IRs) or emergency department encounters (EDEs) within a 1-year period. Implant complications analyzed included periprosthetic fracture, instability, prosthetic joint infection (PJI), aseptic loosening, other implant complications (broken internal prosthesis, dislocation, periprosthetic osteolysis, wear of the articular bearing surface of the internal prosthesis, and other mechanical complications), and revision hip arthroplasty within 1 year after HA. Individual complications were counted separately, but IRs or EDE was uniquely counted, regardless of the presence of multiple concurrent diagnoses.

The demographic and clinical characteristics of the patient sample were described using descriptive statistics, using means and SD for continuous variables and frequencies and percentages for categorical variables. Comparisons across groups were conducted using the Pearson chi-square test for categorical outcomes with adequate expected counts and the Fisher exact test for those with small expected counts. The Wilcoxon rank sum test was used for the pairwise comparison of continuous variables. Bonferroni correction was applied to control for the familywise error rate for multiplicity. Effect sizes were determined using the rank-biserial correlation, r, for continuous variables, and Cramér V, for categorical variables.

Binary logistic regression models were used to examine the association between GLP-1RA use and outcomes while adjusting for diabetes status, age, sex, BMI group, smoking status, and ECI. Firth correction was applied for rare outcomes, including 1-year implant complications, revision surgery within 1 year, and 30-day mortality. Standard binary logistic regression was used for IRs or EDEs and 30-day medical complications. Adjusting for the same covariates, negative binomial regression was used to assess the association between GLP-1RA use and hospital length of stay (LOS). A P-value of < 0.05 was considered statistically significant. All analyses were conducted using SAS software, version 9.4 (SAS Institute).

Two hundred ninety-one patients with T2DM on a GLP1-RA who underwent HA for FNF were identified. Three hundred patients with T2DM who were not on a GLP-1 RA and underwent HA for FNF were randomly selected as comparators. After applying the exclusion criteria, 499 patients (GLP-1RA, N = 248; No GLP-1RA, N = 251) remained for analysis.

Of patients with T2DM who underwent HA for FNF, the GLP-1RA group was significantly younger (mean age 74.9 ± 7.6 versus 78.2 ± 9.4 years, P < 0.001) and had a higher mean BMI (28.9 ± 5.9 versus 26.1 ± 5.8 kg/m², P < 0.001) compared with those not on GLP-1RA (Table 1). No differences were noted in sex distribution (P = 0.530) or race group (P = 0.166). Both cohorts had similar comorbidity burdens (ECI 4.5 ± 1.9 versus 4.5 ± 1.9, P = 0.755), and there were no significant differences in length of hospital stay (P = 0.072) or time from injury to surgery (P = 0.945). However, patients on GLP-1RA had higher admission HbA1c levels (8.0 ± 2.1 versus 7.1 ± 1.7, P < 0.001). In addition, there was no significant difference in the rate of aspiration pneumonitis between GLP-1RA users and nonusers during the index admission (P = 0.750).

At 30 days postoperatively, rates of acute kidney injury (18% versus 20%, P = 0.614), hypoglycemic events (2.4% versus 2.8%, P = 0.796), and other measured complications did not differ significantly between the GLP-1RA and non–GLP-1RA groups (Table 2). Similar findings were observed at 90 days, with no significant differences in cerebrovascular accident (1.6% versus 1.2%, P = 0.723), pulmonary embolism (2.4% versus 3.2%, P = 0.603), or total medical complications (18% versus 19%, P = 0.776). At 1 year, IRs (36% versus 32%, P = 0.344) and EDEs (17% versus 14%, P = 0.143) were similar, and there was no difference in total medical complications (27% versus 29%, P = 0.538). Notably, the GLP-1RA group experienced a lower rate of discharge to hospice or mortality at 1 year (10% versus 16%, P = 0.039), although the effect size was negligible (V = 0.092).

Surgical complication rates did not significantly differ between the GLP-1RA and non–GLP-1RA cohorts at 30 days, 90 days, or 1 year (Table 3). Specifically, there were no significant differences in SSI, PJI, periprosthetic fracture, aseptic loosening, or all-cause revision surgery across all time points (P > 0.05).

No regressions were conducted for SSI within 30 days (N = 6, 1.2%) and fragility fracture within 1 year (N = 2, 0.4%) due to insufficient cases to support a reliable analysis. Binary logistic regression analyses revealed no statistically significant association between GLP-1RA use and the odds ratio of 30-day medical complications or mortality, nor with 1-year IRs or EDEs, implant complications, all-cause revision surgery, or mortality after adjusting for potential confounders (Table 4). Of the other control variables, ECI was associated with increased LOS, 30-day medical complications, IRs or EDEs, mortality, and increased 1-year mortality. Supplemental Digital Content Table 1, http://links.lww.com/JG9/A457↗, summarizes the results from the regression analysis for each control variable.

The findings from this retrospective cross-sectional analysis should be interpreted within the context of the study's limitations. Our findings are limited to patients undergoing HA for FNF within a single hospital system. These factors may affect the generalizability of the findings to other hip fracture subtypes, surgical modalities, or geographic regions. In addition, complications may have been underreported if patients received care at outside hospitals. Our data relied on ICD-10 and CPT coding, which are susceptible to errors or omissions, potentially underreporting some outcomes.

Temporal variables such as the year and hospital of presentation were not controlled for, which could introduce confounding. However, to avoid overfitting, we prioritized adjustment for clinically relevant covariates. In addition, the cohorts collected span from 2016 to 2023. Since GLP-1RA use has grown markedly since 2023, larger contemporary cohorts may better delineate effects.

Procedural details such as surgical approach, anesthesia type, and surgeon identifiers were unavailable because of database limitations. Although these variables potentially have minimal effect on long-term results, their exclusion hindered our capacity to evaluate procedural variability.^37,38 Postoperative glucose control measures, timing of GLP-1RA resumption, patients on dialysis, or deep vein thrombosis prophylaxis regimens were unavailable, limiting evaluation of these perioperative factors. Although HgbA1c values were available for a subset of patients, they were insufficient for inclusion in multivariable models. Nevertheless, adjustment for BMI and the ECI offers a partial proxy for underlying metabolic burden.

Although Firth logistic regression was used to mitigate small-sample bias, some residual bias may persist. Medical and implant complications were aggregated into a single outcome, limiting granularity. Owing to its low incidence, regression analysis was not done for SSI, and these rare outcomes should be interpreted cautiously.

Regarding statistical power, we did not conduct post hoc power analyses, which are generally discouraged in the statistical literature because of their redundancy with observed P-values and potential to mislead.^39,40 Power calculations are best done before study planning.⁴¹ Given that this was a retrospective observational study, our sample size was determined by the number of eligible patients available, making power planning inapplicable. We emphasized effect sizes and confidence intervals to ensure valid interpretation rather than relying solely on statistical significance.

Future research should focus on prospective or randomized trials assessing the effect of GLP-1RAs on both short-term and long-term outcomes after hip fracture surgery. Incorporating standardized measures of frailty, inflammatory biomarkers, and more granular glycemic control metrics could help clarify the mechanisms by which GLP-1RAs might influence surgical recovery. More extensive, multicenter studies with sufficient power to evaluate medical and surgical complications, functional recovery, and quality of life would further elucidate the role of GLP-1RAs in orthopaedic populations.