Navigating Indirect Utility Models in Cancer: Mapping the EORTC QLQ-C30 onto the EQ-5D — ASN Events
  1. Schedule
  2. Health Services
  3. Posters - Health services
  4. Navigating Indirect Utility Models in Cancer: Mapping the EORTC QLQ-C30 onto the EQ-5D

Navigating Indirect Utility Models in Cancer: Mapping the EORTC QLQ-C30 onto the EQ-5D (#214)

Brett Doble 1 , Anthony Harris 1 , Paula Lorgelly 1
  1. Centre for Health Economics, Clayton, VIC, Australia

Aims

There is a plethora of papers that have mapped the non-preference based cancer specific outcome measure EORTC QLQ-C30 onto the EQ-5D, such that utilities can be derived in studies which have not included a multi-attribute utility instrument. The external validity of a number of mapping algorithms is tested by predicting EQ-5D-3L utilities from QLQ-C30 responses.

Methods

Mean differences and root mean square errors were calculated to assess the predictive ability of the reported mapping algorithms using responses to the EQ-5D-3L and QLQ-C30 questionnaires from a prospective multi-hospital cohort study of treatment naïve cancer patients from Victoria, Australia (Cancer 2015). Ordinary least squares (OLS) regression was used to examine the influence of various patient characteristics on the predicted EQ-5D scores.

Results

EQ-5D-3L and QLQ-C30 responses collected in Cancer 2015 at three time points (baseline, 3/6 months, 1 year) were pooled to create the external validation sample for the 12 published algorithms (a maximum of 1,597 observations from 982 patients). Predicted mean EQ-5D-3L scores were larger than observed mean EQ-5D-3L scores for 8 out of the 12 mapping algorithms, with differences of means ranging from 0.0322 (95% CI 0.0238-0.0406) to 0.157 (95% CI 0.147-0.166). For the 4 algorithms generating EQ-5D-3L scores smaller than the observed values, the difference of means ranged from 0.0315 (95% CI 0.0223-0.0408) to 0.383 (95% CI 0.361-0.405). The predictive performance of the algorithms as measured by RMSE ranged from 0.160 (best) to 0.546 (worst). OLS regression results showed that observed EQ-5D-3L scores, location of treating hospital (urban or rural), ECOG status scores significantly influenced predicted EQ-5D-3L scores across all mapping algorithms (p<0.10).

Conclusions

Mapping algorithms for the QLQ-C30 seem unstable when applied to an external validation data set. The use of direct mapping from QLQ-C30 profiles to EQ-5D-3L utilities using published algorithms should be approached with caution.