Development and validation of a risk prediction model for poor performance status and severe symptoms among cancer patients.
Conferences
Overview
Research
Identity
Additional Document Info
View All
Overview
abstract
12097 Background: Existing cancer predictive tools focus on survival, but few incorporate patient-reported outcomes to predict quality-of-life domains, such as symptoms and performance status. The objective was to develop and validate a predictive cancer model (called PROVIEW) for poor performance status and severe symptoms over time. Methods: We used a retrospective, population-based, cohort study of patients, with a cancer diagnosis, in Ontario, Canada between 2008-2015. We randomly selected 60% of patients for model derivation and 40% for validation. Using the derivation cohort, we developed multivariable logistic regression models with baseline characteristics, using a backward stepwise variable selection process. The primary outcome was odds of having poor performance status six months from index date, as measured by a score < = 30 out of 100 on the Palliative Performance Scale. The index date for each model was diagnosis (Year 0), which was then re-calculated at each of 4 annual survivor marks after diagnosis (up to Year 4). Secondary outcomes included having severe pain, dyspnea, well-being, or depression, as measured by a score of > = 7 out of 10 on the Edmonton Symptom Assessment System. Covariates included demographics, clinical information, current symptoms and performance status, and healthcare utilization. Model performance was assessed by AUC statistics and calibration plots. Results: Our population-based cohort identified 125,479 cancer patients for the performance status model in Year 0. The median diagnosis age was 64 years, 57% were female, and the most common cancers were breast (24%), lung (13%), and prostate (9%). 32% had Stage 3 or 4 disease. In Year 0 after backwards selection, the odds of having a poor performance status in 6 months was increased by more than 10% when the patient had: COPD, dementia, diabetes; radiation treatment; a hospital admission in the prior 3 months; high pain or depression; a current performance status < = 30; any issues with appetite; or received end-of-life homecare. Generally, these variables were also associated with a > 10% increased odds in other years and for the secondary outcomes. The average AUC across all 25 models is 0.7676 which indicates high model discrimination. Conclusions: The PROVIEW model accurately predicts risk of having a poor performance status or severe symptoms over time among cancer patients. It has the potential to be a useful online tool for patients to integrate earlier supportive and palliative care.
