MAPPING THE EDMONTON SYMPTOM ASSESSMENT SYSTEM-REVISED (ESAS-R) TO THE EQ-5D-5L IN PATIENTS WITH CANCER
Author(s)
Hilary Short, MSc1, Jiabi Wen, PhD1, Fatima Al Sayah, PhD1, Siwei Qi, MSc2, Claire Link, MA3, Linda Watson, PhD4, Jeffrey A Johnson, PhD1, Lisa Barbera, PhD3.
1School of Public Health, University of Alberta, Edmonton, AB, Canada, 2Surveillance & Reporting, Health Shared Services, Alberta Health Services, Edmonton, AB, Canada, 3Cancer Care Alberta, Alberta Health Services, Edmonton, AB, Canada, 4Cancer Care Alberta, Alberta Health Services, Calgary, AB, Canada.
1School of Public Health, University of Alberta, Edmonton, AB, Canada, 2Surveillance & Reporting, Health Shared Services, Alberta Health Services, Edmonton, AB, Canada, 3Cancer Care Alberta, Alberta Health Services, Edmonton, AB, Canada, 4Cancer Care Alberta, Alberta Health Services, Calgary, AB, Canada.
OBJECTIVES: This study aimed to develop a mapping algorithm to predict the Canadian preference-based EQ-5D-5L index scores from the Edmonton Symptom Assessment System-Revised (ESAS-r) in cancer populations.
METHODS: Cross-sectional data collected between January 2018 and December 2021 from cancer clinics across Alberta, Canada, were utilized. Spearman correlation coefficients were computed between instruments, and mapping models were developed separately for four tumor groups (breast, gastrointestinal, genitourinary, and intrathoracic). Response mapping was conducted using ordered logistic regression to predict response levels for each of the EQ-5D-5L dimensions and various direct mapping model types were fitted. A 5-fold cross validation method was used. Four model specifications were explored for each mapping type: 1) all ESAS-r items; 2) all ESAS-r items + age + sex + cancer stage; 3) select ESAS-r items; 4) select ESAS-r items + age + sex + cancer stage. Purposeful selection procedures were performed in models 3 and 4 to select significant ESAS-r variables. Model performance was assessed using mean square error (MSE) and mean absolute error (MAE).
RESULTS: A total of 1,223 patients were included across four tumor groups: breast (n=258), gastrointestinal (n=309), genitourinary (n=353), and intrathoracic (n=303). ESAS-r items were strongly correlated with their conceptually related EQ-5D-5L dimensions, most notably ESAS-r pain with the EQ-5D-5L pain/discomfort dimension (r=0.76-0.78) and ESAS-r anxiety/depression items with the EQ-5D-5L anxiety/depression dimension (r=0.68-0.76) in all groups. Among the mapping models evaluated, ordered logistic regression, ordinary least squares, and Tobit consistently produced the lowest prediction errors (MAE: 0.136-0.157; MSE: 0.031-0.043), while censored least absolute deviations and generalized linear models performed less well, particularly when demographic covariates were included, across all groups.
CONCLUSIONS: Mapping the ESAS-r to the EQ-5D-5L is feasible across multiple cancer types. These models can estimate EQ-5D-5L index scores from routine ESAS-r data, supporting economic evaluations when direct utility measurements are unavailable.
METHODS: Cross-sectional data collected between January 2018 and December 2021 from cancer clinics across Alberta, Canada, were utilized. Spearman correlation coefficients were computed between instruments, and mapping models were developed separately for four tumor groups (breast, gastrointestinal, genitourinary, and intrathoracic). Response mapping was conducted using ordered logistic regression to predict response levels for each of the EQ-5D-5L dimensions and various direct mapping model types were fitted. A 5-fold cross validation method was used. Four model specifications were explored for each mapping type: 1) all ESAS-r items; 2) all ESAS-r items + age + sex + cancer stage; 3) select ESAS-r items; 4) select ESAS-r items + age + sex + cancer stage. Purposeful selection procedures were performed in models 3 and 4 to select significant ESAS-r variables. Model performance was assessed using mean square error (MSE) and mean absolute error (MAE).
RESULTS: A total of 1,223 patients were included across four tumor groups: breast (n=258), gastrointestinal (n=309), genitourinary (n=353), and intrathoracic (n=303). ESAS-r items were strongly correlated with their conceptually related EQ-5D-5L dimensions, most notably ESAS-r pain with the EQ-5D-5L pain/discomfort dimension (r=0.76-0.78) and ESAS-r anxiety/depression items with the EQ-5D-5L anxiety/depression dimension (r=0.68-0.76) in all groups. Among the mapping models evaluated, ordered logistic regression, ordinary least squares, and Tobit consistently produced the lowest prediction errors (MAE: 0.136-0.157; MSE: 0.031-0.043), while censored least absolute deviations and generalized linear models performed less well, particularly when demographic covariates were included, across all groups.
CONCLUSIONS: Mapping the ESAS-r to the EQ-5D-5L is feasible across multiple cancer types. These models can estimate EQ-5D-5L index scores from routine ESAS-r data, supporting economic evaluations when direct utility measurements are unavailable.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
MSR254
Topic
Economic Evaluation, Methodological & Statistical Research, Patient-Centered Research
Disease
Oncology