Aims: To evaluate whether a validated outcome prediction model influences clinician confidence and treatment recommendations for older patients undergoing radiotherapy for head and neck cancer (HNC).
Methods: Radiation oncologists treating HNC in Australia & New Zealand participated in a survey comprising 5 clinical vignettes of HNC presentations in patients ≥70 years. These represented stage II oropharynx squamous cell carcinoma (SCC), stage III larynx SCC, stage IVA oropharynx SCC, stage I tonsillar SCC and stage I larynx SCC. Each vignette included relevant patient and tumour details. Clinicians selected recommendations for treatment intent, radiotherapy and systemic therapy from predefined options and rated their confidence in decision-making (0-10 scale). Model-predicted risks of recurrence (locoregional and distant) and non-cancer mortality were presented, and clinicians updated their recommendations and confidence. Changes in confidence were analysed using paired t-tests.
Results: Of 44 clinician participants, 65.9% had ≥10 years of HNSCC practice experience. Mean baseline confidence scores (before model predictions) ranged from 6.74 to 8.63 for the five scenarios. After presentation of model-based risks, mean confidence increased significantly in three vignettes with increases of 0.79 (95% CI:0.42-1.16, p <0.01, stage III larynx), 0.53 (95% CI: 0.28-1.00, p =0.03, stage 1 oropharynx) and 0.64 (95% CI:0.06-1.00, p <0.01, stage IVA oropharynx) points; but decreased by a mean -0.45 points in one vignette (95% CI: -0.89- -0.01, p =0.04, stage I larynx).Treatment recommendations changed for 7 of 220 decision points (3.2%) across four vignettes (no changes for stage I larynx).
Conclusion: The prediction model significantly increased clinician confidence in three vignettes with minimal impact on treatment recommendations. Increased clinician confidence was observed in the context of relatively high baseline confidence scores. These findings suggest that the prognostic information provided additional reassurance for clinical decision-support, but highlight the need for further evaluation of clinical impact.