Predicting locoregional treatment failure in anal cancer using multivariable clinical-imaging models

Concurrent chemoradiotherapy (cCRT) is an effective treatment for anal squamous cell carcinoma, but identifying patients with good treatment outcomes remains challenging. We aimed to develop a multivariable model combining baseline clinical and imaging (MRI, PET/CT) variables, HPV status and semi-quantitative 18F-FDG-PET/CT parameters to predict locoregional treatment failure, defined as incomplete response or local relapse, within two years after cCRT.

This retrospective cohort study included 152 anal cancer patients treated with cCRT between 2008 and 2023. All patients underwent pre-treatment 18F-FDG-PET/CT and ≥ 2 years of follow-up was available. Seven predictive models were constructed using combinations of (1) clinical baseline variables (age, sex, cTN-stage), (2) tumour hr-HPV status, and (3) PET/CT parameters (e.g. SUVmax, MTV42%, TLG). Model performance for predicting locoregional failure was evaluated using bias-corrected area under the receiver operating characteristic curve (AUC), using LASSO regression.

Locoregional failure occurred in 21 of 152 patients (14%) within two years after cCRT. The model combining hr-HPV and PET/CT parameters showed the best performance (bias-corrected AUC 0.69). In this model, hr-HPV positivity was associated with substantially lower odds of failure (OR 0.30, 95% CI 0.07-1.00), while higher MTV42% was associated with slightly increased odds (OR 1.01, 95% CI 1.00-1.05).

The model combining hr-HPV status and semi-quantitative PET/CT parameters (particularly MTV42%) performed moderately well to predict two-year locoregional failure in anal cancer patients. These findings support further exploration of baseline risk stratification to develop models that may guide more personalised treatment strategies.