Predicting Patient-Level 3-Level Version of EQ-5D Index Scores From a Large International Database Using Machine Learning and Regression Methods.

OBJECTIVES: This study aimed to evaluate the performance of machine learning and regression methods in the prediction of 3-level version of EQ-5D (EQ-5D-3L) index scores from a large diverse data set. METHODS: A total of 30 studies from 3 countries were combined. Predictions were performed via eXtreme Gradient Boosting classification (XGBC), eXtreme Gradient Boosting regression (XGBR) and ordinary least squares (OLS) regression using 10-fold cross-validation and 80%/20% partition for training and testing. We evaluated 6 prediction scenarios using 3 samples (general population, patients, total) and 2 predictor sets: demographic and disease-related variables with/without patient-reported outcomes. Model performance was evaluated by mean absolute error and percent of predictions within clinically irrelevant error range and within correct health severity group (EQ-5D-3L index <0.45, 0.45-0.926, >0.926). RESULTS: The data set involved 26 318 individuals (clinical settings n = 6214, general population n = 20 104) and 26 predictor variables plus diagnoses. Using all predictors and the total sample, mean absolute error values were 0.153, 0.126, and 0.131, percent of predictions within clinically irrelevant error range were 47.6%, 39.5%, and 37.4%, and within the correct health severity group were 56.3%, 64.9%, and 63.3% by XGBC, XGBR, and OLS, respectively. The performance of models depended on the applied evaluation criteria, the target population, the included predictors, and the EQ-5D-3L index score range. CONCLUSIONS: Regression models (XGBR and OLS) outperformed XGBC, yet prediction errors were outside the clinically irrelevant error range for most respondents. Our results highlight the importance of systematic patient-reported outcome (EQ-5D) data collection. Dialogs between artificial intelligence and outcomes research experts are encouraged to enhance the value of accumulating data in health systems.

Measuring the acceptability of EQ-5D-3L health states for different ages: a new adaptive survey methodology.

BACKGROUND: Acceptable health and sufficientarianism are emerging concepts in health resource allocation. We defined acceptability as the proportion of the general population who consider a health state acceptable for a given age. Previous studies surveyed the acceptability of health problems separately per EQ-5D-3L domain, while the acceptability of health states with co-occurring problems was barely explored. OBJECTIVE: To quantify the acceptability of 243 EQ-5D-3L health states for six ages from 30 to 80 years: 1458 health state-age combinations (HAcs), denoted as the acceptability set of EQ-5D-3L. METHODS: In 2019, an online representative survey was conducted in the Hungarian general population. We developed a novel adaptive survey algorithm and a matching statistical measurement model. The acceptability of problems was evaluated separately per EQ-5D-3L domain, followed by joint evaluation of up to 15 HAcs. The selection of HAcs depended on respondents' previous responses. We used an empirical Bayes measurement model to estimate the full acceptability set. RESULTS: 1375 respondents (female: 50.7%) were included with mean (SD) age of 46.7 (14.6) years. We demonstrated that single problems that were acceptable separately for a given age were less acceptable when co-occurring jointly (p < 0.001). For 30 years of age, EQ-5D-3L health states of '11112' (11.9%) and '33333' (1%), while for 80 years of age '21111' (93.3%) and '33333' (7.4%) had highest and lowest acceptability (% of population), respectively. CONCLUSION: The acceptability set of EQ-5D-3L quantifies societal preferences concerning age and disease severity. Its measurement profiles and potential role in health resource allocation needs further exploration.

A comparison of European, Polish, Slovenian and British EQ-5D-3L value sets using a Hungarian sample of 18 chronic diseases.

BACKGROUND: In the Central and Eastern European region, the British EQ-5D-3L value set is used commonly in quality of life (QoL) studies. Only Poland and Slovenia have country-specific weights. Our study aimed to investigate the impact of value set choice on the evaluation of 18 chronic conditions in Hungary. METHODS: Patients' EQ-5D-3L index scores were calculated using the VAS-based Slovenian and European and the time-trade-off-based Polish and British value sets. We performed pairwise comparisons of mean index values by dimensions, diagnoses and age groups. We evaluated disease burden by comparing index values matched by age and gender in each condition with those of the general population of the CEE region in all four value sets. RESULTS: Altogether, 2421 patients (55% female) were included in our sample with the average age of 55.87 years (SD = 17.75). The average Slovenian, European, Polish and British EQ-5D-3L scores were 0.598 (SD = 0.279), 0.661 (SD = 0.257), 0.770 (SD = 0.261) and 0.644 (SD = 0.279), respectively. We found highly significant differences in most diagnoses, with the greatest difference between the Polish and Slovenian index values in Parkinson's disease (0.265). Systematic pairwise comparison across all conditions and value sets revealed greatest differences between the time-trade-off (TTO) and VAS-based value sets as well as varying sensitivity of the disease burden evaluations of chronic disease conditions to the choice of value sets. CONCLUSIONS: Our results suggest that the choice of value set largely influences the health state utility results in chronic diseases, and might have a significant impact on health policy decisions.

Determinants of the acceptability of health problems in different ages: exploring a new application of the EQ VAS.

BACKGROUND: We aimed to determine the acceptability of non-perfect health states with age using the EQ VAS and analyse the influencing factors. METHODS: We conducted a cross-sectional survey on a convenience sample from the general population (N = 200). Respondents were asked to indicate on the EQ VAS the health states that are still acceptable for ages between 30 and 80 years in 10-year intervals (VAS acceptable health curve, AHCvas). We recorded respondents' current health, health-related lifestyle, demographic background and explored the reference person they imagined when evaluating acceptable health states. We evaluated the AHCvas by estimating linear multilevel models including a random intercept (estimated at age 30) and a random slope for age. RESULTS: AHCvas scores were available for 194 respondents (mean age = 42.8 years, range 19-93, 58% female). For ages of 30, 40, 50, 60, 70 and 80 years, mean AHCvas scores were 93, 87, 80, 73, 65 and 57, respectively. The decline of AHCvas was linear with age. Respondents' age, health status, lifestyle and health-related experiences, as well as their reference point taken (e.g. imagining themselves, others or both during the valuation task) influenced significantly the acceptability of health problems. CONCLUSIONS: When measured with the EQ VAS, health problems were increasingly acceptable with age. Capturing well the individual variability in the assessment of acceptable health states at different ages, the EQ VAS is a useful addition to EQ-5D-3L descriptive system-based measures of acceptable health.