Statistical Alchemy: Conceptual Validity and Mapping to Generate Health State Utility Values

Mapping from SIBDQ to EQ-5D-5L for patients with inflammatory bowel disease

OBJECTIVE

Clinical studies commonly use disease-specific measures to assess patients' health-related quality of life. However, economic evaluation often requires preference-based utility index scores to calculate cost per quality-adjusted life-year (QALY). When utility index scores are not directly available, mappings are useful. To our knowledge, no mapping exists for the Short Inflammatory Bowel Disease Questionnaire (SIBDQ). Our aim was to develop a mapping from SIBDQ to the EQ-5D-5L index score with German weights for inflammatory bowel disease (IBD) patients.

METHODS

We used 3856 observations of 1055 IBD patients who participated in a randomised controlled trial in Germany on the effect of introducing regular appointments with an IBD nurse specialist in addition to standard care with biologics. We considered five data availability scenarios. For each scenario, we estimated different regression and machine learning models: linear mixed-effects regression, mixed-effects Tobit regression, an adjusted limited dependent variable mixture model and a mixed-effects regression forest. We selected the final models with tenfold cross-validation based on a model subset and validated these with observations in a validation subset.

RESULTS

For the first four data availability scenarios, we selected mixed-effects Tobit regressions as final models. For the fifth scenario, mixed-effects regression forest performed best. Our findings suggest that the demographic variables age and gender do not improve the mapping, while including SIBDQ subscales, IBD disease type, BMI and smoking status leads to better predictions.

CONCLUSION

We developed an algorithm mapping SIBDQ values to EQ-5D-5L index scores for different sets of covariates in IBD patients. It is implemented in the following web application: https://www.bwl.uni-hamburg.de/hcm/forschung/mapping.html .

A Comparison of Items and Constructs of Standardized Health-Related Quality of Life and Mental Well-Being Measures

OBJECTIVES

This study aimed to explore the internal constructs of the concepts being measured by EQ-5D-5L (a health-related quality of life measure that can produce preference-based utility values) and the 12-item General Health Questionnaire (GHQ-12, a mental well-being measure) and to understand to what extent the items of EQ-5D-5L and GHQ-12 associate with each other.

METHODS

We used data from 12 701 respondents participating in a Belgian survey in 2022. Correlation coefficients between GHQ-12 and EQ-5D-5L were calculated at both the aggregate and item levels. Multidimensional scaling, exploratory factor analysis, and regression models were performed to investigate the underlying constructs that are associated with the items.

RESULTS

Despite a moderate correlation (0.39) between the EQ-5D-5L and GHQ-12 total scores, only a trivial or weak correlation (<0.3) was observed between the first 4 EQ-5D-5L items and any GHQ-12 item. Multidimensional scaling and exploratory factor analysis showed the first 4 EQ-5D-5L dimensions were clustered together with EuroQol visual analog scale and positively phrased GHQ-12 items were close to each other, whereas EQ-anxiety/depression and negatively phrased GHQ-12 items were grouped with overall life satisfaction. In the regression models, not all GHQ-12 items had a significant coefficient to predict EQ-5D-5L responses.

CONCLUSIONS

To the best of our knowledge, we present the first comparison of items and underlying constructs of GHQ-12 and EQ-5D-5L. The results showed that GHQ-12 can only partially predict the responses of EQ-5D-5L and the 2 instruments measure different constructs. Researchers should carefully consider conceptual legitimacy while applying the mapping technique and consider sensitivity analyses for the mapping estimates.

Prediction of the SF-6D utility score from Lung cancer FACT-L: a mapping study in China

OBJECTIVE

To develop a mapping algorithm for generating the Short Form Six-Dimension (SF-6D) utility score based on the Functional Assessment of Cancer Therapy-Lung (FACT-L) of lung cancer patients.

METHODS

Data were collected from 625 lung cancer patients in mainland China. The Spearman rank correlation coefficient and principal component analysis were used to evaluate the conceptual overlap between the FACT-L and SF-6D. Five model specifications and four statistical techniques were used to derive mapping algorithms, including ordinary least squares (OLS), Tobit and beta-mixture regression models, which were used to directly estimate health utility, and ordered probit regression was used to predict the response level. The prediction performance was evaluated using the correlations between the root mean square error (RMSE), mean absolute error (MAE), concordance correlation coefficient (CCC), Akaike information criterion (AIC) and Bayesian information criterion (BIC) and the observed and predicted SF-6D scores. A five-fold cross-validation method was used to test the universality of each model and select the best model.

RESULTS

The average FACT-L score was 103.024. The average SF-6D score was 0.774. A strong correlation was found between FACT-L and SF-6D scores (ρ = 0.797). The ordered probit regression model with the total score of each dimension and its square term, as well as age and sex as covariates, was most suitable for mapping FACT-L to SF-6D scores (5-fold cross-validation: RMSE = 0.0854; MAE = 0.0655; CCC = 0.8197; AEs > 0.1 (%) = 53.44; AEs > 0.05 (%) = 21.76), followed by beta-mixture regression for direct mapping. The Bland‒Altman plots showed that the ordered probit regression M5 had the lowest proportion of prediction scores outside the 95% agreement limit (-0.166, 0.163) at 4.96%.

CONCLUSIONS

The algorithm reported in this paper enables lung cancer data from the FACT-L to be mapped to the utility of the SF-6D. The algorithm allows the calculation of quality-adjusted life years for cost-utility analyses of lung cancer.

Mapping the World Health Organization Disability Assessment Schedule (WHODAS 2.0) onto SF-6D Using Swedish General Population Data

BACKGROUND AND OBJECTIVE

Mapping algorithms can be used for estimating quality-adjusted life years (QALYs) when studies apply non-preference-based instruments. In this study, we estimate a regression-based algorithm for mapping between the World Health Organization Disability Assessment Schedule (WHODAS 2.0) and the preference-based instrument SF-6D to obtain preference estimates usable in health economic evaluations. This was done separately for the working and non-working populations, as WHODAS 2.0 discriminates between these groups when estimating scores.

METHODS

Using a dataset including 2258 participants from the general Swedish population, we estimated the statistical relationship between SF-6D and WHODAS 2.0. We applied three regression methods, i.e., ordinary least squares (OLS), generalized linear models (GLM), and Tobit, in mapping onto SF-6D from WHODAS 2.0 at the overall-score and domain levels. Root mean squared error (RMSE) and mean absolute error (MAE) were used for validation of the models; R2 was used to assess model fit.

RESULTS

The best-performing models for both the working and non-working populations were GLM models with RMSE ranging from 0.084 to 0.088, MAE ranging from 0.068 to 0.071, and R2 ranging from 0.503 to 0.608. When mapping from the WHODAS 2.0 overall score, the preferred model also included sex for both the working and non-working populations. When mapping from the WHODAS 2.0 domain level, the preferred model for the working population included the domains mobility, household activities, work/study activities, and sex. For the non-working population, the domain-level model included the domains mobility, household activities, participation, and education.

CONCLUSIONS

It is possible to apply the derived mapping algorithms for health economic evaluations in studies using WHODAS 2.0. As conceptual overlap is incomplete, we recommend using the domain-based algorithms over the overall score. Different algorithms must be applied depending on whether the population is working or non-working, due to the characteristics of WHODAS 2.0.

Mapping the Peds QLTM 4.0 onto CHU-9D: a cross-sectional study in functional dyspepsia population from China

Objective

The study aims to develop a mapping algorithm from the Pediatric Quality of Life Inventory™ 4. 0 (Peds QL 4.0) onto Child Health Utility 9D (CHU-9D) based on the cross-sectional data of functional dyspepsia (FD) children and adolescents in China.

Methods

A sample of 2,152 patients with FD completed both the CHU-9D and Peds QL 4.0 instruments. A total of six regression models were used to develop the mapping algorithm, including ordinary least squares regression (OLS), the generalized linear regression model (GLM), MM-estimator model (MM), Tobit regression (Tobit) and Beta regression (Beta) for direct mapping, and multinomial logistic regression (MLOGIT) for response mapping. Peds QL 4.0 total score, Peds QL 4.0 dimension scores, Peds QL 4.0 item scores, gender, and age were used as independent variables according to the Spearman correlation coefficient. The ranking of indicators, including the mean absolute error (MAE), root mean squared error (RMSE), adjusted R², and consistent correlation coefficient (CCC), was used to assess the predictive ability of the models.

Results

The Tobit model with selected Peds QL 4.0 item scores, gender and age as the independent variable predicted the most accurate. The best-performing models for other possible combinations of variables were also shown.

Conclusion

The mapping algorithm helps to transform Peds QL 4.0 data into health utility value. It is valuable for conducting health technology evaluations within clinical studies that have only collected Peds QL 4.0 data.

Mapping the obesity problems scale to the SF-6D: results based on the Scandinavian Obesity Surgery Registry (SOReg)

BACKGROUND

Obesity Problem Scale (OP) is a widely applied instrument for obesity, however currently calculation of health utility based on OP is not feasible as it is not a preference-based measure. Using data from the Scandinavian Obesity Surgery Registry (SOReg), we sought to develop a mapping algorithm to estimate SF-6D utility from OP. Furthermore, to test whether the mapping algorithm is robust to the effect of surgery.

METHOD

The source data SOReg (n = 36 706) contains both OP and SF-36, collected at pre-surgery and at 1, 2 and 5 years post-surgery. The Ordinary Least Square (OLS), beta-regression and Tobit regression were used to predict the SF-6D utility for different time points respectively. Besides the main effect model, different combinations of patient characteristics (age, sex, Body Mass Index, obesity-related comorbidities) were tested. Both internal validation (split-sample validation) and validation with testing the mapping algorithm on a dataset from other time points were carried out. A multi-stage model selection process was used, accessing model consistency, parsimony, goodness-of-fit and predictive accuracy. Models with the best performance were selected as the final mapping algorithms.

RESULTS

The final mapping algorithms were based on OP summary score using OLS models, for pre- and post-surgery respectively. Mapping algorithms with different combinations of patients' characteristics were presented, to satisfy the user with a different need.

CONCLUSION

This study makes available algorithms enabling crosswalk from the Obesity Problem Scale to the SF-6D utility. Different mapping algorithms are recommended for the mapping of pre- and post-operative data.

Mapping health assessment questionnaire disability index onto EQ-5D-5L in China

Objective

This research aimed to develop the more accurate mapping algorithms from health assessment questionnaire disability index (HAQ-DI) onto EQ-5D-5L based on Chinese Rheumatoid Arthritis patients.

Methods

The cross-sectional data of Chinese RA patients from 8 tertiary hospitals across four provincial capitals was used for constructing the mapping algorithms. Direct mapping using Ordinary least squares regression (OLS), the general linear regression model (GLM), MM-estimator model (MM), Tobit regression model (Tobit), Beta regression model (Beta) and the adjusted limited dependent variable mixture model (ALDVMM) and response mapping using Multivariate Ordered Probit regression model (MV-Probit) were carried out. HAQ-DI score, age, gender, BMI, DAS28-ESR and PtAAP were included as the explanatory variables. The bootstrap was used for validation of mapping algorithms. The average ranking of mean absolute error (MAE), root mean square error (RMSE), adjusted R ² (adjR ²) and concordance correlation coefficient (CCC) were used to assess the predictive ability of the mapping algorithms.

Results

According to the average ranking of MAE, RMSE, adjR ², and CCC, the mapping algorithm based on Beta performed the best. The mapping algorithm would perform better as the number of variables increasing.

Conclusion

The mapping algorithms provided in this research can help researchers to obtain the health utility values more accurately. Researchers can choose the mapping algorithms under different combinations of variables based on the actual data.

Obtaining EQ-5D-3L utility index from the health status scale of traditional Chinese medicine (TCM-HSS) based on a mapping study

BACKGROUND

Almost all traditional Chinese medicine (TCM) quality of life measures are non-preference-based measures (non-PBMs), which do not provide utilities for cost-utility analysis in pharmacoeconomic evaluation. Whereas the mapping has become a new instrument to obtain utilities, which builds a bridge between non-PBMs and PBMs.

PURPOSE

To develop mapping algorithms from the health status scale of traditional Chinese medicine (TCM-HSS) onto the three-level EuroQol five-dimensional questionnaire (EQ-5D-3L).

METHODS

The cross-sectional data were collected by questionnaire survey from a tertiary hospital visit population and community residents in China, and randomly divided into training and validation set by 2:1. Based on the training set, direct and indirect mapping methods (7 regression methods and 4 model specifications) were conducted to establish alternative models, which were comprehensively evaluated based on the validation set by mean absolute error, root mean square error, and Spearman correlation coefficient between predicted and observed values. Based on the whole sample, the preferred mapping algorithm was developed.

RESULTS

A total of 639 samples were included, with an average age of 45.24 years and 61.66% of respondents were female. The mean EQ-5D-3L index was 0.9225 [SD = 0.1458], and the mean TCM-HSS index was 3.4144 [SD = 3.1154]. The final mapping algorithm was a two-part regression model including the TCM-HSS subscales, interaction terms, and demographic covariates (age and gender). The prediction performance was good. The mean error was 0.0003, the mean absolute error was 0.0566, the root mean square error was 0.1039, and 83.10% of the prediction errors were within 0.1; the Spearman correlation coefficient between predicted and observed EQ-5D-3L values was 0.6479.

CONCLUSION

It is the first study to develop a mapping algorithm between the TCM-HSS and EQ-5D-3L, which demonstrates excellent prediction accuracy and estimates utility value for economic evaluation from TCM quality of life measures.

Generating EQ-5D-5L health utility scores from BASDAI and BASFI: a mapping study in patients with axial spondyloarthritis using longitudinal UK registry data

BACKGROUND

Preference-based health-state utility values (HSUVs), such as the EuroQol five-dimensional questionnaire (EQ-5D-5L), are needed to calculate quality-adjusted life-years (QALYs) for cost-effectiveness analyses. However, these are rarely used in clinical trials of interventions in axial spondyloarthritis (axSpA). In these cases, mapping can be used to predict HSUVs.

OBJECTIVE

To develop mapping algorithms to estimate EQ-5D-5L HSUVs from the Bath Ankylosing Disease Activity Index (BASDAI) and the Bath Ankylosing Spondylitis Functional Index (BASFI).

METHODS

Data from the British Society for Rheumatology Biologics Register in Ankylosing Spondylitis (BSRBR-AS) provided 5122 observations with complete BASDAI, BASFI, and EQ-5D-5L responses covering the full range of disease severity. We compared direct mapping using adjusted limited dependent variable mixture models (ALDVMMs) and optional inclusion of the gap between full health and the next feasible value with indirect response mapping using ordered probit (OPROBIT) and generalised ordered probit (GOPROBIT) models. Explanatory variables included BASDAI, BASFI, and age. Metrics to assess model goodness-of-fit and performance/accuracy included Akaike and Bayesian information criteria (AIC/BIC), mean absolute error (MAE) and root mean square error (RMSE), plotting predictive vs. observed estimates across the range of BASDAI/BASFI and comparing simulated data with the original data set for the preferred/best model.

RESULTS

Overall, the ALDVMM models that did not formally include the gap between full health and the next feasible value outperformed those that did. The four-component mixture models (with squared terms included) performed better than the three-component models. Response mapping using GOPROBIT (no squared terms included) or OPROBIT (with squared terms included) offered the next best performing models after the three-component ALDVMM (with squared terms). Simulated data of the preferred model (ALDVMM with four-components) did not significantly underestimate uncertainty across most of the range of EQ-5D-5L values, however the proportion of data at full health was underrepresented, likely due in part to model fitting on a small number of observations at this point in the actual data (4%).

CONCLUSIONS

The mapping algorithms developed in this study enabled the generation of EQ-5D-5L utilities from BASDAI/BASFI. The indirect mapping equations reported for the EQ-5D-5L facilitate the calculation of the EQ-5D-5L utility scores using other UK and country-specific value sets.

Mapping the Haem-A-QoL to the EQ-5D-5L in patients with hemophilia

OBJECTIVE

This study's objective was to develop an algorithm that mapping the Haem-A-QoL scores to EQ-5D-5L utility scores in patients with hemophilia in China.

METHODS

A national sample of 862 patients with hemophilia completed both the EQ-5D-5L and Haem-A-QoL instruments. Eight regression models were selected to develop the mapping algorithm, they were: the ordinary least squares, general linear regression, Tobit regression, censored least absolute deviation, mixture beta regression, adjusted limited dependent variable mixture, the two-part, and robust MM-estimator model. Root mean squared error (RMSE), mean absolute error (MAE), and R-square (R²) calculated using the tenfold cross-validation and random sample validation methods were used to assess the predictive ability of the models.

RESULTS

Based on RMSE, MAE, and R², the mixture beta regression model with selected Haem-A-QoL subscale scores as the predicted variables showed the best performance.

CONCLUSIONS

Our mapping algorithm bolsters the calculation of QALYs while conducting an economic evaluation of hemophilia-related interventions when only Haem-A-QoL data are available. The external validity of the algorithm should be further assessed in the other populations.

Predicting presenteeism using measures of health status

Objectives

To identify whether it is feasible to develop a mapping algorithm to predict presenteeism using multiattribute measures of health status.

Methods

Data were collected using a bespoke online survey in a purposive sample (n = 472) of working individuals with a self-reported diagnosis of Rheumatoid arthritis (RA). Survey respondents were recruited using an online panel company (ResearchNow). This study used data captured using two multiattribute measures of health status (EQ5D-5 level; SF6D) and a measure of presenteeism (WPAI, Work Productivity Activity Index). Statistical correlation between the WPAI and the two measures of health status (EQ5D-5 level; SF6D) was assessed using Spearman’s rank correlation. Five regression models were estimated to quantify the relationship between WPAI and predict presenteeism using health status. The models were specified based in index and domain scores and included covariates (age; gender). Estimated and observed presenteeism were compared using tenfold cross-validation and evaluated using Root mean square error (RMSE).

Results

A strong and negative correlation was found between WPAI and: EQ5D-5 level and WPAI (r = − 0.64); SF6D (r =− 0.60). Two models, using ordinary least squares regression were identified as the best performing models specifying health status using: SF6D domains with age interacted with gender (RMSE = 1.7858); EQ5D-5 Level domains and age interacted with gender (RMSE = 1.7859).

Conclusions

This study provides indicative evidence that two existing measures of health status (SF6D and EQ5D-5L) have a quantifiable relationship with a measure of presenteeism (WPAI) for an exemplar application of working individuals with RA. A future study should assess the external validity of the proposed mapping algorithms.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11136-021-02936-9.

A video-feedback parenting intervention to prevent enduring behaviour problems in at-risk children aged 12-36 months: the Healthy Start, Happy Start RCT

BACKGROUND

Behaviour problems emerge early in childhood and place children at risk for later psychopathology.

OBJECTIVES

To evaluate the clinical effectiveness and cost-effectiveness of a parenting intervention to prevent enduring behaviour problems in young children.

DESIGN

A pragmatic, assessor-blinded, multisite, two-arm, parallel-group randomised controlled trial.

SETTING

Health visiting services in six NHS trusts in England.

PARTICIPANTS

A total of 300 at-risk children aged 12-36 months and their parents/caregivers.

INTERVENTIONS

Families were allocated in a 1 : 1 ratio to six sessions of Video-feedback Intervention to promote Positive Parenting and Sensitive Discipline (VIPP-SD) plus usual care or usual care alone.

MAIN OUTCOME MEASURES

The primary outcome was the Preschool Parental Account of Children's Symptoms, which is a structured interview of behaviour symptoms. Secondary outcomes included caregiver-reported total problems on the Child Behaviour Checklist and the Strengths and Difficulties Questionnaire. The intervention effect was estimated using linear regression. Health and social care service use was recorded using the Child and Adolescent Service Use Schedule and cost-effectiveness was explored using the Preschool Parental Account of Children's Symptoms.

RESULTS

In total, 300 families were randomised: 151 to VIPP-SD plus usual care and 149 to usual care alone. Follow-up data were available for 286 (VIPP-SD, n = 140; usual care, n = 146) participants and 282 (VIPP-SD, n = 140; usual care, n = 142) participants at 5 and 24 months, respectively. At the post-treatment (primary outcome) follow-up, a group difference of 2.03 on Preschool Parental Account of Children's Symptoms (95% confidence interval 0.06 to 4.01; p = 0.04) indicated a positive treatment effect on behaviour problems (Cohen's d = 0.20, 95% confidence interval 0.01 to 0.40). The effect was strongest for children's conduct [1.61, 95% confidence interval 0.44 to 2.78; p = 0.007 (d = 0.30, 95% confidence interval 0.08 to 0.51)] versus attention deficit hyperactivity disorder symptoms [0.29, 95% confidence interval -1.06 to 1.65; p = 0.67 (d = 0.05, 95% confidence interval -0.17 to 0.27)]. The Child Behaviour Checklist [3.24, 95% confidence interval -0.06 to 6.54; p = 0.05 (d = 0.15, 95% confidence interval 0.00 to 0.31)] and the Strengths and Difficulties Questionnaire [0.93, 95% confidence interval -0.03 to 1.9; p = 0.06 (d = 0.18, 95% confidence interval -0.01 to 0.36)] demonstrated similar positive treatment effects to those found for the Preschool Parental Account of Children's Symptoms. At 24 months, the group difference on the Preschool Parental Account of Children's Symptoms was 1.73 [95% confidence interval -0.24 to 3.71; p = 0.08 (d = 0.17, 95% confidence interval -0.02 to 0.37)]; the effect remained strongest for conduct [1.07, 95% confidence interval -0.06 to 2.20; p = 0.06 (d = 0.20, 95% confidence interval -0.01 to 0.42)] versus attention deficit hyperactivity disorder symptoms [0.62, 95% confidence interval -0.60 to 1.84; p = 0.32 (d = 0.10, 95% confidence interval -0.10 to 0.30)], with little evidence of an effect on the Child Behaviour Checklist and the Strengths and Difficulties Questionnaire. The primary economic analysis showed better outcomes in the VIPP-SD group at 24 months, but also higher costs than the usual-care group (adjusted mean difference £1450, 95% confidence interval £619 to £2281). No treatment- or trial-related adverse events were reported. The probability of VIPP-SD being cost-effective compared with usual care at the 24-month follow-up increased as willingness to pay for improvements on the Preschool Parental Account of Children's Symptoms increased, with VIPP-SD having the higher probability of being cost-effective at willingness-to-pay values above £800 per 1-point improvement on the Preschool Parental Account of Children's Symptoms.

LIMITATIONS

The proportion of participants with graduate-level qualifications was higher than among the general public.

CONCLUSIONS

VIPP-SD is effective in reducing behaviour problems in young children when delivered by health visiting teams. Most of the effect of VIPP-SD appears to be retained over 24 months. However, we can be less certain about its value for money.

TRIAL REGISTRATION

Current Controlled Trials ISRCTN58327365.

FUNDING

This project was funded by the National Institute for Health Research (NIHR) Health Technology Assessment programme and will be published in full in Health Technology Assessment; Vol. 25, No. 29. See the NIHR Journals Library website for further project information.

Mapping the Kansas City Cardiomyopathy Questionnaire (KCCQ) Onto EQ-5D-3L in Heart Failure Patients: Results for the Japanese and UK Value Sets

Background. Health technology assessment bodies in several countries, including Japan and the United Kingdom, recommend mapping techniques to obtain utility scores in clinical trials that do not have a preference-based measure of health. This study sought to develop mapping algorithms to predict EQ-5D-3L scores from the Kansas City Cardiomyopathy Questionnaire (KCCQ) in patients with heart failure (HF). Methods. Data from the randomized, double-blind PARADIGM-HF trial were analyzed, and EQ-5D-3L scores were calculated using the Japanese and UK value sets. Several different model specifications were explored to best fit EQ-5D data collected at baseline with KCCQ scores, including ordinary least square regression, two-part, Tobit, and three-part models. Generalized estimating equations models were also fitted to analyze longitudinal EQ-5D data. To validate model predictions, the data set was split into a derivation (n = 4,465) from which the models were developed and a separate sample (n = 1,892) for validation. Results. There were only small differences between the different model classes tested. Model performance and predictive power was better for the item-level models than for the models including KCCQ domain scores. R ² statistics for the item-level models ranged from 0.45 to 0.52. Mean absolute error in the validation sample was 0.10 for the models using the Japanese value set and 0.114 for the UK models. All models showed some underprediction of utility above 0.75 and overprediction of utility below 0.5, but performed well for population-level estimates. Conclusions. Using data from a large clinical trial in HF, we found that EQ-5D-3L scores can be estimated from responses to the KCCQ and can facilitate cost-utility analysis from existing HF trials where only the KCCQ was administered. Future validation in other HF populations is warranted.

Mapping of the EORTC QLQ-C30 to EQ-5D-5L index in patients with lymphomas

OBJECTIVE

The objective of this study was to develop algorithms to map the EORTC QLQ-C30 (QLQ-C30) onto EQ-5D-5L in a sample of patients with lymphomas.

METHODS

An online nationwide survey of patients with lymphoma was carried out in China. Ordinary least squares (OLS), beta-based mixture, adjusted limited dependent variable mixture regression, and a Tobit regression model were used to develop the mapping algorithms. The QLQ-C30 subscales/items, their squared and interaction terms, and respondents' demographic variables were used as independent variables. The root mean square error (RMSE), mean absolute error (MAE), and R-squared (R²) were estimated based on tenfold cross-validation to assess the predictive ability of the selected models.

RESULTS

Data of 2222/4068 respondents who self-completed the online survey were elicited for analyses. The mean EQ-5D-5L index score was 0.81 (SD 0.21, range - 0.81-1.0). 19.98% of respondents reported an index score at 1.0. In total, 72 models were generated based on four regression methods. According to the RMSE, MAE and R², the OLS model including QLQ-C30 subscales, squared terms, interaction terms, and demographic variables showed the best fit for overall and the Non-Hodgkin's lymphoma sample; for Hodgkin's lymphoma, the ALDVMM with 1-component model, including QLQ-C30 subscales, squared terms, interaction terms, and demographic variables, showed a better fit than the other models.

CONCLUSION

The mapping algorithms enable the EQ-5D-5L index scores to be predicted by QLQ-C30 subscale/item scores with good precision in patients living with lymphomas.

Using qualitative methods for a conceptual analysis of measures of health status and presenteeism prior to a mapping study

OBJECTIVES

The inclusion of productivity in economic evaluations is a contentious issue. Methods are currently being developed to assess how it may feasibly be included for specific interventions, such as workplace interventions (WPIs), where productivity is a key outcome. Mapping (also called cross-walking or prediction modelling) may offer a solution. Prior to producing a mapping algorithm, it is recommended that the conceptual validity between 'source' and 'target' measures be understood first. This study aimed to understand the conceptual validity of two existing measures of health status (EQ-5D; SF-6D) and presenteeism to inform the potential for a subsequent mapping algorithm.

METHODS

A purposive sample of individuals who were currently working and had either rheumatoid arthritis (RA), ankylosing spondylitis (AS) or psoriatic arthritis (PsA). Individuals were recruited through support groups. Semi-structured telephone interviews were conducted until data saturation (no new emerging themes) was reached. Deductive and inductive framework analysis methods were used to identify key aspects of the conditions (themes) that impact on presenteeism (working at reduced levels of health).

RESULTS

Twenty-two (RA = 10; AS = 9; PsA = 3) employed individuals were interviewed. Deductive analysis identified evidence which confirmed the domains included in the EQ-5D and SF-6D capture those key aspects of RA, AS and PsA that increase presenteeism. Inductive analysis identified an additional theme; mental clarity, not captured by the EQ-5D or SF-6D, was also found to have a direct impact on presenteeism.

CONCLUSIONS

The results of the study indicate conceptual validity of both health status measures to predict presenteeism. The next step is to develop a mapping algorithm for presenteeism.

Mapping clinical outcomes to generic preference-based outcome measures: development and comparison of methods

BACKGROUND

Cost-effectiveness analysis using quality-adjusted life-years as the measure of health benefit is commonly used to aid decision-makers. Clinical studies often do not include preference-based measures that allow the calculation of quality-adjusted life-years, or the data are insufficient. 'Mapping' can bridge this evidence gap; it entails estimating the relationship between outcomes measured in clinical studies and the required preference-based measures using a different data set. However, many methods for mapping yield biased results, distorting cost-effectiveness estimates.

OBJECTIVES

Develop existing and new methods for mapping; test their performance in case studies spanning different preference-based measures; and develop methods for mapping between preference-based measures.

DATA SOURCES

Fifteen data sets for mapping from non-preference-based measures to preference-based measures for patients with head injury, breast cancer, asthma, heart disease, knee surgery and varicose veins were used. Four preference-based measures were covered: the EuroQoL-5 Dimensions, three-level version (n = 11), EuroQoL-5 Dimensions, five-level version (n = 2), Short Form questionnaire-6 Dimensions (n = 1) and Health Utility Index Mark 3 (n = 1). Sample sizes ranged from 852 to 136,327. For mapping between generic preference-based measures, data from FORWARD, the National Databank for Rheumatic Diseases (which includes the EuroQoL-5 Dimensions, three-level version, and EuroQoL-5 Dimensions, five-level version, in its 2011 wave), were used.

MAIN METHODS DEVELOPED

Mixture-model-based approaches for direct mapping, in which the dependent variable is the health utility value, including adaptations of methods developed to model the EuroQoL-5 Dimensions, three-level version, and beta regression mixtures, were developed, as were indirect methods, in which responses to the descriptive systems are modelled, for consistent multidirectional mapping between preference-based measures. A highly flexible approach was designed, using copulas to specify the bivariate distribution of each pair of EuroQoL-5 Dimensions, three-level version, and EuroQoL-5 Dimensions, five-level version, responses.

RESULTS

A range of criteria for assessing model performance is proposed. Theoretically, linear regression is inappropriate for mapping. Case studies confirm this. Flexible, direct mapping methods, based on different variants of mixture models with appropriate underlying distributions, perform very well for all preference-based measures. The precise form is important. Case studies show that a minimum of three components are required. Covariates representing disease severity are required as predictors of component membership. Beta-based mixtures perform similarly to the bespoke mixture approaches but necessitate detailed consideration of the number and location of probability masses. The flexible, bi-directional indirect approach performs well for testing differences between preference-based measures.

LIMITATIONS

Case studies drew heavily on EuroQoL-5 Dimensions. Indirect methods could not be undertaken for several case studies because of a lack of coverage. These methods will often be unfeasible for preference-based measures with complex descriptive systems.

CONCLUSIONS

Mapping requires appropriate methods to yield reliable results. Evidence shows that widely used methods such as linear regression are inappropriate. More flexible methods developed specifically for mapping show that close-fitting results can be achieved. Approaches based on mixture models are appropriate for all preference-based measures. Some features are universally required (such as the minimum number of components) but others must be assessed on a case-by-case basis (such as the location and number of probability mass points).

FUTURE RESEARCH PRIORITIES

Further research is recommended on (1) the use of the monotonicity concept, (2) the mismatch of trial and mapping distributions and measurement error and (3) the development of indirect methods drawing on methods developed for mapping between preference-based measures.

FUNDING

This project was funded by the National Institute for Health Research (NIHR) Health Technology Assessment programme and will be published in full in Health Technology Assessment; Vol. 24, No. 34. See the NIHR Journals Library website for further project information. This project was also funded by a Medical Research Council grant (MR/L022575/1).

Development of algorithms to estimate EQ-5D and derive health utilities from WHOQOL-HIV Bref: a mapping study

PURPOSE

This study aimed to develop and evaluate different families of applicable models available for utility mapping between World Health Organization Quality of Life for HIV-abbreviated version (WHOQOL-HIV Bref) and EQ-5D-3L and to propose an optimised algorithm to estimate health utilities of people living with HIV.

METHODS

Estimation dataset was collected between July 2014 and September 2016 in a cross-sectional study including 1526 people living with HIV/Aids (PLWH) under care at the Instituto Nacional de Infectologia Evandro Chagas-FIOCRUZ, in Brazil. Data of WHOQOL-HIV Bref and EQ-5D-3L questionnaires were collected. Fisher's exact tests were used for testing WHOQOL-HIV Bref response frequencies among groups of responses to each of the five EQ-5D-3L domains. Multiple correspondence analyses (MCA) were used to inspect the relationships between both instrument responses. Different families of applicable models available for utility mapping between WHOQOL-HIV Bref and EQ-5D-3L were adjusted for the prediction of disutility.

RESULTS

Candidate models' performances using mean absolute error (MAE), mean squared error (MSE), and root mean squared error (RMSE) were similarly good, which was evidenced by the overlapping of its 95% confidence intervals of the mean tenfold cross-validation or estimated generalisation errors. However, the Hurdle Logistic-Log-Normal model was better on average according to generalisation errors both in the prediction of Brazilian utility values (MAE = 0.1037, MSE = 0.0178, and RMSE = 0.1332) and for those of the UK (MAE = 0.1476, MSE = 0.0443, and RMSE = 0.2099).

CONCLUSIONS

Mapping EQ-5D-3L responses or deriving health utilities directly from WHOQOL-HIV Bref responses can be a valid alternative for settings with no preference-based health utility data.

Preference-based measures of health-related quality of life in congenital mobility impairment: a systematic review of validity and responsiveness

INTRODUCTION

Mobility impairment is the leading cause of disability in the UK. Individuals with congenital mobility impairments have unique experiences of health, quality of life and adaptation. Preference-based outcomes measures are often used to help inform decisions about healthcare funding and prioritisation, however the applicability and accuracy of these measures in the context of congenital mobility impairment is unclear. Inaccurate outcome measures could potentially affect the care provided to these patient groups. The aim of this systematic review was to examine the performance of preference-based outcome measures for the measurement of utility values in various forms of congenital mobility impairment.

METHODS

Ten databases were searched, including Science Direct, CINAHL and PubMed. Screening of reference lists and hand-searching were also undertaken. Descriptive and narrative syntheses were conducted to combine and analyse the various findings. Results were grouped by condition. Outcome measure performance indicators were adapted from COSMIN guidance and were grouped into three broad categories: validity, responsiveness and reliability. Screening, data extraction and quality appraisal were carried out by two independent reviewers.

RESULTS

A total of 31 studies were considered eligible for inclusion in the systematic review. The vast majority of studies related to either cerebral palsy, spina bifida or childhood hydrocephalus. Other relevant conditions included muscular dystrophy, spinal muscular atrophy and congenital clubfoot. The most commonly used preference-based outcome measure was the HUI3. Reporting of performance properties predominantly centred around construct validity, through known group analyses and assessment of convergent validity between comparable measures and different types of respondents. A small number of studies assessed responsiveness, but assessment of reliability was not reported. Increased clinical severity appears to be associated with decreased utility outcomes in congenital mobility impairment, particularly in terms of gross motor function in cerebral palsy and lesion level in spina bifida. However, preference-based measures exhibit limited correlation with various other condition-specific and clinically relevant outcome measures.

CONCLUSION

Preference-based measures exhibit important issues and discrepancies relating to validity and responsiveness in the context of congenital mobility impairment, thus care must be taken when utilising these measures in conditions associated with congenital mobility impairments.

Mapping the kidney disease quality of life 36-item short form survey (KDQOL-36) to the EQ-5D-3L and the EQ-5D-5L in patients undergoing dialysis

OBJECTIVES

To develop algorithms mapping the Kidney Disease Quality of Life 36-Item Short Form Survey (KDQOL-36) onto the 3-level EQ-5D questionnaire (EQ-5D-3L) and the 5-level EQ-5D questionnaire (EQ-5D-5L) for patients with end-stage renal disease requiring dialysis.

METHODS

We used data from a cross-sectional study in Europe (France, n = 299; Germany, n = 413; Italy, n = 278; Spain, n = 225) to map onto EQ-5D-3L and data from a cross-sectional study in Singapore (n = 163) to map onto EQ-5D-5L. Direct mapping using linear regression, mixture beta regression and adjusted limited dependent variable mixture models (ALDVMMs) and response mapping using seemingly unrelated ordered probit models were performed. The KDQOL-36 subscale scores, i.e., physical component summary (PCS), mental component summary (MCS), three disease-specific subscales or their average, i.e., kidney disease component summary (KDCS), and age and sex were included as the explanatory variables. Predictive performance was assessed by mean absolute error (MAE) and root mean square error (RMSE) using 10-fold cross-validation.

RESULTS

Mixture models outperformed linear regression and response mapping. When mapping to EQ-5D-3L, the ALDVMM model was the best-performing one for France, Germany and Spain while beta regression was best for Italy. When mapping to EQ-5D-5L, the ALDVMM model also demonstrated the best predictive performance. Generally, models using KDQOL-36 subscale scores showed better fit than using the KDCS.

CONCLUSIONS

This study adds to the growing literature suggesting the better performance of the mixture models in modelling EQ-5D and produces algorithms to map the KDQOL-36 onto EQ-5D-3L (for France, Germany, Italy, and Spain) and EQ-5D-5L (for Singapore).

Using the Fatigue Severity Scale to inform healthcare decision-making in multiple sclerosis: mapping to three quality-adjusted life-year measures (EQ-5D-3L, SF-6D, MSIS-8D)

Background

Fatigue has a major influence on the quality of life of people with multiple sclerosis. The Fatigue Severity Scale is a frequently used patient-reported measure of fatigue impact, but does not generate the health state utility values required to inform cost-effectiveness analysis, limiting its applicability within decision-making contexts. The objective of this study was to use statistical mapping methods to convert Fatigue Severity Scale scores to health state utility values from three preference-based measures: the EQ-5D-3L, SF-6D and Multiple Sclerosis Impact Scale-8D.

Methods

The relationships between the measures were estimated through regression analysis using cohort data from 1056 people with multiple sclerosis in South West England. Estimation errors were assessed and predictive performance of the best models as tested in a separate sample (n = 352).

Results

For the EQ-5D and the Multiple Sclerosis Impact Scale-8D, the best performing models used a censored least absolute deviation specification, with Fatigue Severity Scale total score, age and gender as predictors. For the SF-6D, the best performing model used an ordinary least squares specification, with Fatigue Severity Scale total score as the only predictor.

Conclusions

Here we present algorithms to convert Fatigue Severity Scales scores to health state utility values based on three preference-based measures. These values may be used to estimate quality-adjusted life-years for use in cost-effectiveness analyses and to consider the health-related quality of life of people with multiple sclerosis, thereby informing health policy decisions.

Electronic supplementary material

The online version of this article (10.1186/s12955-019-1205-y) contains supplementary material, which is available to authorized users.

An Updated Systematic Review of Studies Mapping (or Cross-Walking) Measures of Health-Related Quality of Life to Generic Preference-Based Measures to Generate Utility Values

BACKGROUND

Mapping is an increasingly common method used to predict instrument-specific preference-based health-state utility values (HSUVs) from data obtained from another health-related quality of life (HRQoL) measure. There have been several methodological developments in this area since a previous review up to 2007.

OBJECTIVE

To provide an updated review of all mapping studies that map from HRQoL measures to target generic preference-based measures (EQ-5D measures, SF-6D, HUI measures, QWB, AQoL measures, 15D/16D/17D, CHU-9D) published from January 2007 to October 2018.

DATA SOURCES

A systematic review of English language articles using a variety of approaches: searching electronic and utilities databases, citation searching, targeted journal and website searches.

STUDY SELECTION

Full papers of studies that mapped from one health measure to a target preference-based measure using formal statistical regression techniques.

DATA EXTRACTION

Undertaken by four authors using predefined data fields including measures, data used, econometric models and assessment of predictive ability.

RESULTS

There were 180 papers with 233 mapping functions in total. Mapping functions were generated to obtain EQ-5D-3L/EQ-5D-5L-EQ-5D-Y (n = 147), SF-6D (n = 45), AQoL-4D/AQoL-8D (n = 12), HUI2/HUI3 (n = 13), 15D (n = 8) CHU-9D (n = 4) and QWB-SA (n = 4) HSUVs. A large number of different regression methods were used with ordinary least squares (OLS) still being the most common approach (used ≥ 75% times within each preference-based measure). The majority of studies assessed the predictive ability of the mapping functions using mean absolute or root mean squared errors (n = 192, 82%), but this was lower when considering errors across different categories of severity (n = 92, 39%) and plots of predictions (n = 120, 52%).

CONCLUSIONS

The last 10 years has seen a substantial increase in the number of mapping studies and some evidence of advancement in methods with consideration of models beyond OLS and greater reporting of predictive ability of mapping functions.

Mapping the Strengths and Difficulties Questionnaire onto the Child Health Utility 9D in a large study of children

PURPOSE

Non-preference-based measures cannot be used to directly obtain utilities but can be converted to preference-based measures through mapping. The only mapping algorithm for estimating Child Health Utility-9D (CHU9D) utilities from Strengths and Difficulties Questionnaire (SDQ) responses has limitations. This study aimed to develop a more accurate algorithm.

METHODS

We used a large sample of children (n = 6898), with negligible missing data, from the Longitudinal Study of Australian Children. Exploratory factor analysis (EFA) and Spearman's rank correlation coefficients were used to assess conceptual overlap between SDQ and CHU9D. Direct mapping (involving seven regression methods) and response mapping (involving one regression method) approaches were considered. The final model was selected by ranking the performance of each method by averaging the following across tenfold cross-validation iterations: mean absolute error (MAE), mean squared error (MSE), and MAE and MSE for two subsamples where predicted utility values were < 0.50 (poor health) or > 0.90 (healthy). External validation was conducted using data from the Child and Adolescent Mental Health Services study.

RESULTS

SDQ and CHU9D were moderately correlated (ρ = - 0.52, p < 0.001). EFA demonstrated that all CHU9D domains were associated with four SDQ subscales. The best-performing model was the Generalized Linear Model with SDQ items and gender as predictors (full sample MAE: 0.1149; MSE: 0.0227). The new algorithm performed well in the external validation.

CONCLUSIONS

The proposed mapping algorithm can produce robust estimates of CHU9D utilities from SDQ data for economic evaluations. Further research is warranted to assess the applicability of the algorithm among children with severe health problems.

Training teachers in classroom management to improve mental health in primary school children: the STARS cluster RCT

Background

Poor mental health in childhood is common, persistent and associated with a range of adverse outcomes that include persistent psychopathology, as well as risk-taking behaviour, criminality and educational failure, all of which may also compromise health. There is a growing policy focus on children’s mental health and the role of schools in particular in addressing this.

Objectives

To evaluate whether or not the Incredible Years®(IY) Teacher Classroom Management (TCM) training improved children’s mental health, behaviour, educational attainment and enjoyment of school, improved teachers’ mental health and relationship with work, and was cost-effective in relation to potential improvements.

Design

A two-arm, pragmatic, parallel-group, superiority, cluster randomised controlled trial.

Setting

A total of 80 UK schools (clusters) were recruited in three distinct cohorts between 2012 and 2014 and randomised to TCM (intervention) or teaching as usual [(TAU) control] with follow-ups at 9, 18 and 30 months. Schools and teachers were not masked to allocation.

Participants

Eighty schools (n = 2075 children) were randomised: 40 (n = 1037 children) to TCM and 40 (n = 1038 children) to TAU.

Interventions

TCM was delivered to teachers in six whole-day sessions, spread over 6 months. The explicit goals of TCM are to enhance classroom management skills and improve teacher–student relationships.

Main outcome measures

The primary planned outcome was the teacher-reported Strengths and Difficulties Questionnaire Total Difficulties (SDQ-TD) score. Random-effects linear regression and marginal logistic regression models using generalized estimating equations were used to analyse outcomes.

Results

The intervention reduced the SDQ-TD score at 9 months [adjusted mean difference (AMD) –1.0, 95% confidence interval (CI) –1.9 to –0.1;p = 0.03] but there was little evidence of effects at 18 months (AMD –0.1, 95% CI –1.5 to 1.2;p = 0.85) and 30 months (AMD –0.7, 95% CI –1.9 to 0.4;p = 0.23). Planned subgroup analyses suggested that TCM is more effective than TAU for children with poor mental health. Cost-effectiveness analysis using the SDQ-TD suggested that the probability of TCM being cost-effective compared with TAU was associated with some uncertainty (range of 40% to 80% depending on the willingness to pay for a unit improvement in SDQ-TD score). In terms of quality-adjusted life-years (QALYs), there was evidence to suggest that TCM was cost-effective compared with TAU at the National Institute for Health and Care Excellence thresholds of £20,000–30,000 per QALY at 9- and 18-month follow-up, but not at 30-month follow-up. There was evidence of reduced disruptive behaviour (p = 0.04) and reductions in inattention and overactivity (p = 0.02) at the 30-month follow-up. Despite no main effect on educational attainment, subgroup analysis indicated that the intervention’s effect differed between those who did and those who did not have poor mental health for both literacy (interactionp = 0.04) and numeracy (interactionp = 0.03). Independent blind observations and qualitative feedback from teachers suggested that teachers’ behaviour in the classroom changed as a result of attending TCM training.

Limitations

Teachers were not masked to allocation and attrition was marked for parent-reported data.

Conclusions

Our findings provide tentative evidence that TCM may be an effective universal child mental health intervention in the short term, particularly for primary school children who are identified as struggling, and it may be a cost-effective intervention in the short term.

Future work

Further research should explore TCM as a whole-school approach by training all school staff and should evaluate the impact of TCM on academic progress in a more thorough and systematic manner.

Trial registration

Current Controlled Trials ISRCTN84130388.

Funding

This project was funded by the National Institute for Health Research (NIHR) Public Health Research programme and will be published in full inPublic Health Research; Vol. 7, No. 6. See the NIHR Journals Library website for further project information. Funding was also provided by the NIHR Collaboration for Leadership in Applied Health Research and Care South West Peninsula (NIHR CLAHRC South West Peninsula).

Mapping functions in health-related quality of life: mapping from the Achilles Tendon Rupture Score to the EQ-5D

PURPOSE

Health state utility values are derived from preference-based measurements and are useful in calculating quality-adjusted life years (QALYs), which is a metric commonly used in cost-effectiveness studies. The purpose of this study was to convert the Achilles Tendon Rupture Score (ATRS) to the preference-based European Quality of Life-5 Dimension Questionnaire (EQ-5D) by estimating the relationship between the two scores using mapping.

METHODS

Data were collected from a randomised controlled trial, where 100 patients were treated either surgically or non-surgically for Achilles tendon rupture. Forty-three and forty-four patients in surgical group and non-surgical group completed the ATRS and the EQ-5D alongside each other during follow-up at three time points. Different models of the relationship between the ATRS and the EQ-5D were developed and analysed based on direct mapping and cross-validation. The model with the lowest mean absolute error was observed as the one with the best fit.

RESULTS

Among the competing models, mapping based on using a combination of the ATRS items four, five, and six associated with limitation due to pain, during activities of daily living and when walking on uneven ground, produced the best predictor of the EQ-5D score.

CONCLUSIONS

The present study provides a mapping algorithm to enable the derivation of utility values directly from the ATRS. This approach makes it feasible for researchers, as well as medical practitioners, to obtain preference-based values in clinical studies or settings where only the ATRS is being administered. The algorithm allows for the calculation of QALYs for use in cost-effectiveness analyses, making it valuable in the study of acute Achilles tendon ruptures.

LEVEL OF EVIDENCE

II.

Development of Conversion Functions Mapping the FACT-B Total Score to the EQ-5D-5L Utility Value by Three Linking Methods and Comparison with the Ordinary Least Square Method

INTRODUCTION

Health-related quality-of-life (HRQoL) measures are commonly mapped to a value that represents a utility for economic evaluation via regression models, which may lead to shrinkage of the variance.

OBJECTIVES

This study aimed to develop and compare conversion functions that map the Functional Assessment of Cancer Therapy-Breast (FACT-B) total score to the EuroQoL 5-Dimensions, 5-Levels (EQ-5D-5L) utility value via four methods.

METHODS

We used the HRQoL scores of 238 Singapore patients with breast cancer to develop the conversion function for the equipercentile, linear equating, mean rank and ordinary least squares (OLS) methods. We compared the distributions of the observed values and the four sets of mapped values and performed regression analyses to assess whether the association with risk factors was preserved by utility values derived from mapping.

RESULTS

At baseline, the observed EQ-5D-5L utility value had a mean ± standard deviation (SD) of 0.820 ± 0.152, and 24.8% of the respondents attained a value of 1. The OLS method (mean 0.820; SD 0.112; proportion 0%) better agreed with the observed data than the equipercentile (mean 0.831; SD 0.152; proportion 23.5%), linear equating (mean 0.814; SD 0.145; proportion 11.8%) and mean rank method (mean 0.821; SD 0.147; proportion 23.9%). The significance of association was preserved for all parameters involved in the regression analyses by the equipercentile and linear equating methods, but the mean rank and OLS methods were inconsistent with the observed data for one and two parameters, respectively.

CONCLUSION

The problem of shrinkage in the variance occurred in the OLS method, but it provided an unbiased estimate for the mean and better agreement. Among the other three linking methods, the mean rank method better described the distribution, whereas the equipercentile and linear equating methods better assessed the association with risk factors.

Direct Mapping of the QLQ-C30 to EQ-5D Preferences: A Comparison of Regression Methods

BACKGROUND

Several mapping or cross-walking algorithms for deriving utilities from the European Organisation for Research and Treatment of Cancer Quality of Life Questionnaire for Cancer (EORTC QLQ-C30) scores have been published in recent years. However, the large majority used ordinary least squares (OLS) regression, which proved to be not very accurate because of the specifics of the quality-of-life measures.

OBJECTIVE

Our objective was to compare regression methods that have been used to map EuroQol 5 Dimensions 3 Levels (EQ-5D-3L) utility values from the general EORTC QLQ-C30 using OLS as a benchmark while fixing the number of explanatory variables and to explore an alternative three-part model.

METHODS

We conducted a regression analysis of predicted EQ-5D-3L utilities generated using data from an observational study in ambulatory patients with non-small-cell lung cancer in a Toronto hospital. Six alternative regression methods were compared with a simple OLS regression as benchmark. The six alternative regression models were Tobit, censored least absolute deviation, normal mixture, beta, zero-one inflated beta and a mix of piecewise OLS and logistic regression.

RESULTS

The best predictive fit was obtained by a mix of OLS regression(s) for utilities lower than 1 with a cut-off point of 0.50 and a separate binary logistic regression for utilities equal to one. Zero-one inflated beta regression was also promising. However, OLS regression proved to be the most accurate for the mean. The prediction of utilities equal to one was poor in all regression approaches.

CONCLUSIONS

Three-part regression methods that separately target low, medium and high (<0.50, 0.51-0.99 or 1) utilities seem to have better prediction power than OLS with EQ-5D-3L data, although OLS also seems quite robust. Exploration of three-part approaches compared with single (OLS) regression should be further tested in other similar datasets or using individual pooled data from various clinical or observational studies. The use of alternative goodness-of-fit measures for mapping studies and their influence on the choice of the best performing methods should also be investigated.