Mapping the Problem Areas in Diabetes (PAID) and Diabetes Distress Scale (DDS) to the EQ-5D-5L in diabetes care

To develop and validate mapping algorithms that estimate EQ-5D health state utility values (HSUVs) from two diabetes-specific patient-reported outcome measures, the Problem Areas in Diabetes (PAID) and Diabetes Distress Scale (DDS), to support economic evaluations when EQ-5D data are unavailable. Data from 662 adults with diabetes in Germany included PAID-20, DDS-17, and EQ-5D-5L responses. Direct mapping models predicted EQ-5D utilities using Tobit and censored least absolute deviation (CLAD) regressions, while indirect mapping models used proportional and partial proportional odds regressions to predict EQ-5D dimensions. Six model specifications were tested, incorporating PAID and DDS items with covariates (age, sex, diabetes type), with selection of predictors guided by item correlations and backward stepwise procedure to retain statistically relevant variables. Model performance was assessed using a selection of performance metrics including root mean square error (RMSE) and mean absolute error (MAE). Internal validation of the models employed 10-fold cross-validation. The strongest-performing direct mapping models consisted of a model with a selection of PAID-20 items and covariates (Tobit regression; RMSE: 0.159; MAE: 0.117) and a model that incorporated PAID-20, DDS-17, and covariates (CLAD regression; RMSE: 0.146; MAE: 0.09), and a model based on DDS-17 items and covariates (Tobit regression; RMSE: 0.164; MAE: 0.121). Indirect mapping models demonstrated higher prediction errors overall (RMSE: 0.187–0.21; MAE: 0.124–0.141) than their counterparts. We present novel algorithms that enable estimation of EQ-5D utilities from PAID and DDS scores, facilitating economic evaluation in diabetes. Preferred models demonstrated predictive accuracy comparable to published mapping studies. Many people with diabetes experience emotional strain, worries, and frustrations while managing their condition. Two widely used tools that serve to measure how the condition affects their daily life are the Problem Areas in Diabetes (PAID) scale and the Diabetes Distress Scale (DDS). To assess the value of new treatments, researchers and health agencies commonly rely on a measure called EQ‑5D, which summarizes overall health and is used to calculate quality‑of‑life scores important for economic evaluations and treatment choices. However, many diabetes studies collect PAID or DDS but do not collect EQ‑5D, making it difficult to judge the wider impact of treatments. This study set out to solve that problem by creating new algorithms that can estimate EQ‑5D scores from PAID and DDS responses when EQ‑5D was not included in a study. Using data from 662 adults with diabetes in Germany, several versions of these algorithms were tested, and the best ones closely matched people’s actual EQ‑5D scores. Among the different statistical approaches we tested, the models that directly linked both PAID and DDS responses to EQ-5D produced the most accurate predictions. These models performed well across a wide range of health states and closely matched the EQ-5D scores that participants reported. In practice, this enables researchers to generate the health-related quality-of-life information required for economic evaluations, even in studies that collect PAID or DDS but not EQ-5D.