Prospective validation of the Cancer and Aging Research Group fall-risk score for older adults with cancer.

Abstract

e23195 Background: In older adults with cancer, falls are associated with poorer quality of life, chemotherapy toxicity & shorter survival, yet fall risk factors in this population are unclear. In a prior cross-sectional study, we developed & validated a model of factors associated with falls in older adults on systemic therapy. In this study, we aimed to prospectively validate the Cancer & Aging Research Group fall-risk score (CARG FRS) & explore whether additional variables or time-varying covariates improved the model’s performance. Methods: This prospective cohort study enrolled participants aged ≥70 years with cancer, receiving systemic cancer therapy & life expectancy > 1 year. Baseline assessment comprised measures of functional status, prior falls, medications, cognition, physical function, depression, neuropathy, and pain. The primary outcome was falls, prospectively ascertained over 6-months follow-up. Descriptive statistics were calculated on all study variables and risk model scores were computed. Non-parametric correlations assessed the relationship between baseline CARG FRS and falls during the study. Receiver operating characteristic (ROC) analyses determined area under the curve (AUC) values. A linear mixed model (LMM) examined if CARG scores changed over time. Extended Cox proportional hazards models examined prediction of falls based on a time-varying fall-risk score. IBM SPSS version 28 was used for all analyses; a .05 alpha level determined statistical significance. Results: 213 participants had a mean age 72.2 yrs ±SD = 5.5; 55% were male, 90% white. One-quarter (54/213, 25.4%) had ≥1 fall during follow-up. The average baseline CARG FRS was 5.7 ± 3.5, range 0-16. The baseline fall-risk score as previously published had a small, but not significant, relationship with falls (r = .131, p = .057); modifications of the previously published cutpoints for polypharmacy and depression scores based on the distribution in our cohort improved prediction of falls using baseline CARG scores (r = .163, p = .017). However, neither the original CARG FRS or the modified scores provided good discrimination among fallers and non-fallers (area under the curve .586 and .608, respectively). While individuals with prior falls were significantly more likely to fall during follow-up than those without prior falls (41% vs 18%, p < .001), stratifying by prior falls did not result in significance of the FRS for either group. The CARG FRS as a time-varying predictor did not have a statistically significant relationship with falls. Conclusions: In this external validation of the CARG FRS, the modified model was statistically significant in predicting falls over 6-months of follow-up, but with poor discrimination. Stratifying by prior falls or adding time-varying predictors did not yield an improved risk prediction model. Future research is needed to create accurate risk-prediction models for clinical use.