Multivariable propensity score models are often used to mitigate the impact of selection bias in non-randomized data; however, residual confounding remains a potential problem. In this study, we used competing risk analysis to address the problem of residual confounding in a national cohort of head and neck squamous cell carcinoma (HNSCC) patients. We analyzed a cohort of 7,117 patients with locoregionally advanced, non-metastatic HNSCC of the oropharynx (OPX), oral cavity, larynx, and hypopharynx (AJCC 7 stage III-IVB) from the national Veterans Affairs Informatics and Computing Infrastructure (VINCI) database. Patients were diagnosed between 2005-2015 and treated with definitive radiation therapy (RT), with or without chemotherapy. We stratified treatment as high-intensity (any cisplatin-containing regimen or multi-agent induction) vs. medium/low intensity (any systemic therapy not qualifying as high-intensity, or RT alone), and analyzed effects of treatment intensity on overall survival (OS), progression-free survival (PFS), cancer progression, and competing mortality (CM). CM was defined as death in the absence of a cancer progression event. We applied standard multivariable Cox regression models with and without inverse probability of treatment weighting (IPTW). Covariates in standard and IPTW Cox models were age, race, performance status (PS), primary site, T/N category, P16 status, smoking status, body mass index, Charlson Comorbidity Index (CCI), marriage status, and employment status. We used the relative event hazard and observed effect of treatment on CM to calibrate the effect of high-intensity treatment on PFS in the presence of residual confounding. Patients treated with high-intensity therapy had increased OS (adjusted hazard ratio (HR): 0.68, 95% CI 0.63-0.72; p <0.01) and PFS (HR: 0.75, 95% CI 0.72-0.79; p<0.001) compared to patients treated with medium/low intensity therapy, using the IPTW Cox model. High-intensity therapy was associated with reduced risk of both cancer progression (HR 0.77, 95% CI 0.73-0.81, p<0.001) and CM (HR 0.72, 95% CI 0.67-0.77, p<0.001) in IPTW models. Covariates associated with significantly increased risk of CM were: older age, non-OPX site, higher T/N category, current smoking, lower BMI, and increased CCI. The reduction in CM associated with high-intensity treatment, even after adjusting for covariates, was considered representative of selection bias due to residual confounding. After calibration, the estimated effect of high-intensity treatment on PFS was reduced (HR: 0.86, 95% CI 0.83-0.90). Competing risk analysis may help improve the accuracy of study conclusions by addressing residual confounding resulting from selection bias. This approach is useful for comparative effectiveness research in populations at risk for competing events.