Systematic review of prognostic models for predicting recurrence and survival in patients with treated oropharyngeal cancer.

objectivesThis systematic review aims to evaluate externally validated models for individualised prediction of recurrence or survival in adults treated with curative intent for oropharyngeal cancer.

designSystematic review.

settingHospital care.

methodsSystematic searches were conducted up to September 2023 and records were screened independently by at least two reviewers. The Prediction model Risk Of Bias ASsessment Tool was used to assess risk of bias (RoB). Model discrimination measures (c-indices) were presented in forest plots. Clinical and methodological heterogeneity precluded meta-analysis.

resultsFifteen studies developing and/or evaluating 25 individualised risk prediction models were included. The majority (77%) of c-indices for model developments and validations were ≥0.7 indicating 'good' discriminatory ability for models predicting overall survival. For disease-specific measures, most (73%) c-indices for model development were also ≥0.7, but fewer (40%) were ≥0.7 for external validations. Comparisons across models and outcome measures were hampered by heterogeneity. Only two studies directly compared models in the same cohort. Since all models were subject to a high RoB, primarily due to concerns with the analysis, the trustworthiness of the findings remains uncertain. Concerns included a lack of accounting for potentially missing data, model overfitting or competing risks as well as small event numbers. There were fewer concerns related to the participant, predictor and outcome domains, although reporting was not always detailed enough to make an informed decision. Where human papilloma virus (HPV) status and/or a radiomics score were included as a variable, models had better discriminative ability.

conclusionsThere were no models assessed as being at low RoB. Given that HPV status or a radiomics score appeared to improve model discriminative performance, further external validation of existing models to assess generalisability should focus on models that include HPV status as a variable. Development and validation of future models should be considered in HPV+ or HPV- cohorts separately to ensure representativeness. PROSPERO REGISTRATION NUMBER: CRD42021248762.