Estimating preference-based EQ-5D health state utilities or item responses from neuropathic pain scores

Preference-based health state utilities are required for many health economic evaluations. When the direct evidence of such is lacking and only condition-specific scores are available, establishing a 'mapping' relationship between instruments can be useful.

Our objective was to map the 11-point Pain Intensity Numerical Rating Scale (PI-NRS-11), a pain-specific instrument ranging from 0 ('no pain') to 10 ('pain as bad as you can imagine'), to the EQ-5D, a preference-based generic instrument.

We used web survey data collected from adult US respondents who (i) had ≥ 3 months of neuropathic pain (NP), either painful diabetic peripheral neuropathy (pDPN) or post-herpetic neuralgia (PHN); (ii) were receiving medications treating NP; and (iii) had completed the EQ-5D and PI-NRS-11. We explored indirect and direct mapping approaches. The indirect method took a probabilistic approach using ordered logistic models (OLMs) predicting response levels for each EQ-5D item via repeated Monte Carlo simulations before computing utilities. The direct approach simply predicted EQ-5D utilities directly using ordinary least squares (OLS). Categorical scores of PI-NRS-11 were used as the predictors. Patient age, gender, and pain duration were additionally controlled in the full model specification. Seventy percent of the data were used for estimation and 30% for prediction. Mean square errors (MSEs) and 95% confidence intervals (CIs) of prediction errors were reported.

A total of 2719 respondents were included. Mean (SD) age was 55.48 (10.65) years and 56.23% were female. Average NP duration was 61 months and 58% gave scores ≥ 6 on the PI-NRS-11. The clinical pain scores were significantly associated with all EQ-5D items, especially with the 'pain/discomfort' item (p  <  0.001). The observed mean (SD) EQ-5D index was 0.594 (0.22). Predicted utilities and responses showed good representation of the observed ones. The reduced model showed comparable results with the full model while imposing minimum data collection burden. From the reduced model, the predicted mean (SD) EQ-5D index was 0.594 (0.11) from direct estimation and 0.588 (0.19) from indirect estimation. All estimated utilities discriminated health gains/losses along the PI-NRS-11. Lower MSEs and prediction errors were found for EQ-5D >0.2.

Findings suggest that EQ-5D utilities or item responses could be estimated on the basis of NP scores. Independent testing of the external validity of the mapping algorithms developed herein is encouraged.