Development and internal validation of a predictive risk model for anxiety after completion of treatment for early stage breast cancer

Abstract

Abstract Objective To develop a predictive risk model (PRM) for patient-reported anxiety after treatment completion for early stage breast cancer suitable for use in practice and underpinned by advances in data science and risk prediction. Methods Secondary analysis of a prospective survey of > 800 women at the end of treatment and again 6 months later using patient reported outcome (PRO) the hospital anxiety and depression scale-anxiety (HADS-A) and > 20 candidate predictors. Multiple imputation using chained equations (for missing data) and least absolute shrinkage and selection operator (LASSO) were used to select predictors. Final multivariable linear model performance was assessed (R2) and bootstrapped for internal validation. Results Five predictors of anxiety selected by LASSO were HADS-A (Beta 0.73; 95% CI 0.681, 0.785); HAD-depression (Beta 0.095; 95% CI 0.020, 0.182) and having caring responsibilities (Beta 0.488; 95% CI 0.084, 0.866) increased risk, whereas being older (Beta − 0.010; 95% CI -0.028, 0.004) and owning a home (Beta 0.432; 95% CI -0.954, 0.078) reduced the risk. The final model explained 60% of variance and bias was low (− 0.006 to 0.002). Conclusions Different modelling approaches are needed to predict rather than explain patient reported outcomes. We developed a parsimonious and pragmatic PRM. External validation is required prior to translation to digital tool and evaluation of clinical implementation. The routine use of PROs and data driven PRM in practice provides a new opportunity to target supportive care and specialist interventions for cancer patients.