Cancer Score Translator: Measuring Well-Being for Survivors

Scientists built a clever system to convert survey responses into real quality-of-life numbers. The project focused on 663 women who beat gynecological cancer. These participants came from a major Norwegian study named LETSGO. That study tracked them over time after treatment ended. In total, there were more than 2,600 health checkpoints recorded throughout their recovery.

The team tested ten distinct mathematical approaches to link the EORTC-QLQ-C30 questionnaire to EQ-5D-5L utility values. Methods included ordinary least squares, Tobit, beta, fractional logistic regression, linear mixed models, adjusted limited dependent variable mixtures, and two-part designs. Each approach considered different combinations of survey scales plus optional factors like age, other medical conditions, or treatment type. They divided the data into five equal parts and validated each model internally before comparing performance.

The best model performed exceptionally well. Mean errors stayed between 0.054 and 0.066 across most predictions. Root mean square errors ranged from 0.081 to 0.096. Nearly half of the observations achieved high precision with errors under 0.05. The correlation between predicted and actual scores reached 0.756 to 0.803. Importantly, accuracy dropped slightly at very low health levels, but overall bias remained negligible. This framework now allows clinicians to estimate quality-of-life scores directly from survey data in similar patient groups. Even in poorer health states, the system provides useful estimates without requiring costly specialized testing.

actions