Background/Purpose: Fragility fractures are a critical outcome for people ageing with HIV (PAH). Studies demonstrate higher prevalence, incidence, and earlier onset than for the general population. Current fragility fracture prediction models for the general population underestimate risk in PAH. To address this gap, we developed one- and three-year predictive models for risk of fragility fracture that better capture risk among PAH.

Methods: We used data from the Veterans Aging Cohort Study (VACS), a national electronic health records-based cohort that includes all individuals diagnosed with HIV receiving care within the Veterans Health Administration. We included all PAH (≥50 years of age) and followed them from 01/01/2009 through 12/31/2019. Observations consisted of one-year person-intervals and three-year person-intervals. Fragility fracture outcomes (at the spine, hip/femur, wrist/forearm, and shoulder/upper arm) were identified using validated ICD9 and ICD10 codes. From an initial pool of 27 candidate predictors (including HIV-specific variables) selected from the literature, clinical experience, and prior research, multivariable logistic regression chose 15 predictors that were subsequently reduced to a set of 10 based on statistical criteria and feedback from HIV providers. This feedback motivated two versions of each model that permit optional inclusion of race.

Results: The cohort included 32,842 individuals contributing a total of 223,607 person-years of data. Table 1 shows the distribution of the 15 short-listed candidate predictors in the first year of follow-up. Over the follow up period, there were 2,355 incident fractures. The ten final predictors, with odds ratios from the one year model, included: age OR = 1.02 (95%CI 1.01 – 1.02), albumin OR = 0.7 (0.64 – 0.76), alcohol use ≥3 drinks per day OR = 1.19 (1.11 – 1.27), BMI OR = 0.96 (0.95 – 0.97), chronic medication count OR = 1.05 (1.05 – 1.06), COPD OR = 1.36 (1.23 – 1.50), fall in previous year OR = 1.57 (1.35 – 1.82), female sex OR = 1.66 (1.33 – 2.08), fracture in previous year OR = 15.46 (13.45 – 17.77), and nonwhite race OR = 0.62 (0.57 – 0.68). The one-year and three-year models with race exhibited good discrimination, with areas under the curve (AUC) of 74.7% and 70.6%, respectively. For the models without race, the AUCs were 73.9% and 69.7% for the one- and three-year models. Figure 1 presents the ROC curves for all four models; all four exhibited excellent calibration.

Conclusion: Based upon rigorous statistical analysis and provider input, we present a new risk prediction model for fragility fracture that is statistically robust and uniquely optimized for the care of PAH. Future studies will externally validate this model using datasets from outside the VA.

Table 1. Distribution of the 15 short-listed candidate predictors in the first year of follow-up within the study cohort.

Figure 2. Receiver Operating Characteristic (ROC) Curves of Models for Risk of Fragility Fracture in One- (left) and Three-Year (right) Periods.

« Back to ACR Convergence 2025

ACR Meeting Abstracts - https://acrabstracts.org/abstract/development-of-a-fracture-prediction-model-for-patients-aging-with-hiv/