Defining an abnormal geriatric assessment for older adults with cancer: Which deficits matter most?

Abstract

12039 Background: At present, there is no universal, objective, and evidence-based definition of what constitutes an abnormal geriatric assessment (GA) in geriatric oncology (GO). In the literature, a threshold number of abnormal GA domains (ranging from 1-4) is often used to define an abnormal GA. However, it is not well-established whether having a specific number of abnormal domains more frequently leads to treatment plan modification (TPM), a key goal of GA, or if particular domains have a greater impact on TPM. The primary objectives of this study are: (1) to determine how well the current definitions of an abnormal GA predict TPM following GA, and (2) to identify particular GA domains associated with TPM. Methods: A retrospective review of the GO clinic database at Princess Margaret Cancer Centre was conducted. All new patients seen in clinic from May 22, 2015 to June 10, 2022 who met the following criteria were included: (1) referred for treatment decision making, (2) received a complete GA, and (3) had a proposed oncologic treatment plan. Demographic, oncologic, and GA-domain variables were extracted. Univariate and multivariate logistic regression modelling was conducted using SPSS to determine each variable’s association with TPM; age, sex, frailty (VES-13) score, and treatment intent were included in all multivariate models. Area under the curve (AUC) was calculated for each model. Results: The study cohort (n = 736) had a mean age of 80.7 years (61-100), 46.1% was female, and 78.3% had a VES-13 score indicating vulnerability. In univariate analysis, age, VES-13 score, disease stage, treatment intent, all GA domains (except Medication Optimization and Social Supports), and all threshold numbers of abnormal domains (except 1 and 7) were significantly associated (p-value < 0.050) with TPM. The best-performing threshold number of abnormal domains in univariate analysis was 4 (AUC 0.628). Overall, the best-performing multivariate model based on AUC was the model containing all 6 significant GA domains (AUC 0.710). In this model, age, treatment intent, Comorbidities, Falls Risk, and Cognition were independently associated with TPM (p-value < 0.05). The multivariate model with a threshold of 4 abnormal domains alone had an AUC of 0.689 and age, VES-13 score, treatment intent, and the threshold were independently associated with TPM. Of the models which included a single GA domain plus the threshold, the models with Comorbidities and Cognition performed best, having AUCs of 0.699 and 0.700, respectively. Conclusions: Overall, our results suggest that an abnormal GA (leading to TPM) may be best defined as one with abnormalities in the domains of Comorbidities, Falls Risk, and Cognition. In terms of a strictly numerical threshold, a GA may be best defined as abnormal if at least 4 GA domains are abnormal. When at least 4 GA domains are abnormal, abnormalities in Comorbidities and Cognition appear to best predict TPM.