Modeling Causal Mediation by Dental Care and Risk Prediction of Jaw Osteonecrosis in Bisphosphonate Users - Project Summary Bisphosphonates (BPs) are widely prescribed antiresorptive agents that improve bone density and reduce skeletal complications, particularly among older patients with osteoporosis or cancer-related bone metastases. However, they are associated with osteonecrosis of the jaw (ONJ), a debilitating adverse event marked by exposed necrotic bone, persistent infection, and impaired oral function. ONJ presents a serious clinical dilemma: discontinuing BP therapy may increase fracture risk or facilitate cancer progression, while continuing therapy may exacerbate ONJ risk, especially in older adults with multiple comorbidities. To reduce ONJ risk, clinical guidelines, including those from the American Association of Oral and Maxillofacial Surgeons, recommend comprehensive dental care (e.g., regular exams, timely periodontal/restorative procedures, and avoidance of invasive interventions) during BP therapy. Yet these guidelines are largely based on observational evidence prone to confounding, creating a major clinical gap: the lack of causal evidence undermines providers' ability to make evidence-based decisions about whether, when, and how specific dental interventions should be prioritized or modified during BP therapy. This underscores the critical need to understand the causal mechanisms by which dental care modifies ONJ risk. Addressing this requires a rigorous mediation framework that accounts for the time-to-event nature of ONJ, the time-varying characteristics of BP exposure and dental care, and the competing risk of death. Current approaches are not equipped to handle these complexities simultaneously, revealing a methodological gap. In parallel, clinicians lack validated dynamic prognostic models for ONJ that incorporate time-varying risk factors—tools essential for translating risk factor associations and causal pathways into actionable, real-time decisions (e.g., BP interval adjustment). This creates a clinical gap: without such tools, risk assessments rely on static heuristics and clinical intuition, failing to reflect evolving patient profiles and misclassifying high- and low-risk individuals. To address these clinical and methodological gaps, we propose two specific aims leveraging 100% Texas Medicare and SEER-Medicare datasets. In Aim 1, we will develop and apply novel causal mediation methods—extending marginal structural models and the parametric mediational g-formula—to estimate the direct and indirect effects of time-varying BP exposure on ONJ risk, mediated by time-varying dental care and accounting for competing mortality. In Aim 2, we will construct and validate a dynamic prognostic model for ONJ by applying joint modeling of endogenous longitudinal risk factors (e.g., frailty, comorbidity, BP duration, dental procedures) and survival outcomes. Model performance will be assessed using time-dependent ROC curves and AUC to identify critical windows of ONJ risk post-BP initiation. Completion of the proposed aims will provide validatedcausal evidence and user-friendly risk prediction toolsto support individualized dental and medication decisions, helping preserve the therapeutic benefits of BPs while minimizing ONJ risk.
