Mapping of the Gastrointestinal Short Form Questionnaire (GSF-Q) into EQ-5D-3L and SF-6D in patients with gastroesophageal reflux disease

The short, self-administered Gastroesophageal Reflux Disease (GERD) Symptom Frequency Questionnaire (GSFQ) is a specific Quality of Life (QoL) instrument which measures the impact of GERD symptoms on QoL. This study aims to map the specific scores in GSFQ into two generic instruments: SF-6D and EQ-5D-3 L, in order to obtain utility estimates derived from the GERD condition.

A national representative sample of GERD patients was selected, stratified by gender, age (< 45, ≥45 years) and GERD severity (0-I, II-IV Savary-Miller score) for validation purposes. Age, gender, BMI, GERD diagnose, GERD severity, associated comorbidities and risk factors were recorded. GSFQ, SF-6D, EQ-5D-3 L, and the HRQoL Visual Analogue Scale (VAS) were answered by patients. Several mapping methods were estimated, regression using dummy variables, and linear, quadratic and cubic regression using optimal factor scores. The use of a GERD aggregated summary severity derived from the GSFQ was dimed the best predictor. Overall Mean Absolute Error (MAE), overall Mean Absolute Percentage Error (MAPE) were used as goodness-of-fit (GOF) indexes to compare models.

A total of 3405 patients were recruited by 490 clinicians. Mean age was 49 (±14.4) years and 49.8% were women. Reported comorbidities were clustered in 6 antecedents and 15 concomitant pathologies. Aggregation of levels for the frequency of symptoms items was found more suitable for estimation. Regression weights were found to follow a monotonous progressive pattern. Overall MAE ranged from 0.092 to 0.094 for SF-6D utility prediction and from 0.008 to 0.08 for EQ-5D-3 L, while MAPE values ranged from 27.9 to 29% for SF-6D and from 36.8 to 38.4% for EQ-5D-3 L. Cubic regression GOF demonstrated a better fit.

It is possible to translate specific GSFQ scores assessing GERD condition into generic SF-6D and EQ-5D-3 L utility values. Although regression using dummy variables is a suitable mapping procedure, other alternative mapping methods convey better fit, in particular cubic regression.