Purpose
To establish and compare generalized or “global” mapping relationships between QLQ-C30 and SF-6D, applicable across different cancer types.
Methods
Patients (N = 671) with breast, myeloma, colorectal, lymphoma, bone marrow, prostate, lung and gastroenteric cancer were randomly split into estimation (75 %) and validation (25 %) datasets. SF-6D was estimated from QLQ-C30 scores via ordinary least squares, generalized linear models and median (least-absolute deviations) regression approaches, and with Bayesian additive regression kernels. Predictive ability was assessed with root mean square error, mean absolute error and proportions of predictions with absolute errors >0.05 and >0.1, whereas explanatory power with adjusted R
2 or equivalent fit measures. Two external samples (breast and colorectal cancer) were used to further test the models.
Results
The QLQ-C30's global health item, the physical, emotional and social functioning scales, and the fatigue, pain and diarrhea symptom scales were significant predictors (p < 0.05 or better) in all models. Negligible deviations in models’ performance were observed. All models overpredicted utilities for patients in worst health and underpredicted them for those in better health (p < 0.01 or better). Regarding external validation, performance was better in the colorectal cancer than in the breast cancer sample.
Conclusions
This study has provided evidence to support the use of “global” mapping models to predict SF-6D utilities from QLQ-C30 in patients with different cancers. Testing with diverse patient samples is required to confirm the generalizability (or not) of mapping models across cancer conditions.
