Less Is More: Cross-Validation Testing of Simplified Nonlinear Regression Model Specifications for EQ-5D-5L Health State Values.

BACKGROUND: The conventional method for modeling of the five-level EuroQol five-dimensional questionnaire (EQ-5D-5L) health state values in national valuation studies is an additive 20-parameter main-effects regression model. Statistical models with many parameters are at increased risk of overfitting-fitting to noise and measurement error, rather than the underlying relationship. OBJECTIVES: To compare the 20-parameter main-effects model to simplified, nonlinear, multiplicative regression models in terms of how accurately they predict mean values of out-of-sample health states. METHODS: We used data from the Spanish, Singaporean, and Chinese EQ-5D-5L valuation studies. Four models were compared: an 8-parameter model with single parameter per dimension, multiplied by cross-dimensional parameters for levels 2, 3, and 4; 9- and 11-parameter extensions with handling of differences in the wording of level 5; and the "standard" additive 20-parameter model. Fixed- and random-intercept variants of all models were tested using two cross-validation methods: leave-one-out at the level of valued health states, and of health state blocks used in EQ-5D-5L valuation studies. Mean absolute error, Lin concordance correlation coefficient, and Pearson R between observed health state means and out-of-sample predictions were compared. RESULTS: Predictive accuracy was generally best using random intercepts. The 8-, 9-, and 11-parameter models outperformed the 20-parameter model in predicting out-of-sample health states. CONCLUSIONS: Simplified nonlinear regression models look promising and should be investigated further using other EQ-5D-5L data sets. To reduce the risk of overfitting, cross-validation is recommended to inform model selection in future EQ-5D valuation studies.

Estimating an EQ-5D-5L Value Set for China.

OBJECTIVES: To estimate a five-level EuroQol five-dimensional questionnaire (EQ-5D-5L) value set for China using the health preferences of residents living in the urban areas of the country. METHODS: The values of a subset of the EQ-5D-5L-defined health states (n = 86) were elicited using the time trade-off (TTO) technique from a sample of urban residents (n = 1271) recruited from five Chinese cities. In computer-assisted personal interviews, participants each completed 10 TTO tasks. Two additive and two multiplicative regression models were evaluated for their performance in describing the relationship between TTO values and health state characteristics using a cross-validation approach. Final values were generated using the best-performed model and a rescaling method. RESULTS: The 8- and 9-parameter multiplicative models unanimously outperformed the 20-parameter additive model using a random or fixed intercept in predicting values for out-of-sample health states in the cross-validation analysis and their coefficients were estimated with lower standard errors. The prediction accuracies of the two multiplicative models measured by the mean absolute error and the intraclass correlation coefficient were very similar, thus favoring the more parsimonious model. CONCLUSIONS: The 8-parameter multiplicative model performed the best in the study and therefore was used to generate the EQ-5D-5L value set for China. We recommend using rescaled values whereby 1 represents the value of instrument-defined full health in economic evaluation of health technologies in China whenever the EQ-5D-5L data are available.

Comparison of health state values derived from patients and individuals from the general population.

PURPOSE: Utility values are critical for cost-utility analyses that guide healthcare decisions. We aimed to compare the utility values of the 5-level EuroQoL-5Dimension (EQ-5D-5L) health states elicited from members of the general public and patients with heart disease or cancer. METHODS: In face-to-face interviews with 157 heart disease patients, 169 cancer patients, and 169 members from the general population, participants valued 10 EQ-5D-5L health states using a composite Time Trade-Off method. RESULTS: Pooling utility values for all health states, heart disease patients and cancer patients had mean utility values lower by 0.11 points (P value = 0.014) and 0.06 points (P value = 0.148), respectively, compared to the general population. Adjusting for sociodemographic characteristics, differences in health state utility values between the patient and the general populations were rendered non-significant, except that heart disease patients gave higher utility values (mean difference = 0.08; P value = 0.007) to mild health states than the general population. Difference in utility values, defined as utility value of a better health state minus that of a poorer health state, was higher among heart disease patients compared to the general population, before and after adjusting for sociodemographic characteristics. CONCLUSIONS: Patients may differ from members of the general population in the strength of their preferences for hypothetical health states. Using utility values derived from the general population may under-estimate the comparative effectiveness of healthcare interventions for certain diseases, such as heart diseases.

Does the Introduction of the Ranking Task in Valuation Studies Improve Data Quality and Reduce Inconsistencies? The Case of the EQ-5D-5L.

BACKGROUND: Time trade-off (TTO)-based valuation studies for the three-level version of the EuroQol five-dimensional questionnaire (EQ-5D) typically started off with a ranking task (ordering the health states by preference). This was not included in the protocol for the five-level EQ-5D (EQ-5D-5L) valuation study. OBJECTIVES: To test whether reintroducing a ranking task before the composite TTO (C-TTO) could help to reduce inconsistencies in C-TTO responses and improve the data quality. METHODS: Respondents were randomly assigned to three study arms. The control arm was the present EQ-5D-5L study protocol, without ranking. The second arm (ranking without sorting) preceded the present protocol by asking respondents to rank the target health states using physical cards. The states were then valued in random order using C-TTO. In the third arm (ranking and sorting), the ranked states remained visible through the C-TTO tasks and the order of valuation was determined by the ranking. The study used only 10 EQ-5D-5L health states. We compared the C-TTO-based inconsistent pairs of health states and ties. RESULTS: The final sample size was 196 in the control arm, 205 in the ranking without sorting arm, and 199 in the ranking and sorting arm. The percentages of ties by respondents were 15.1%, 12.5%, and 12.6% for the control arm, the ranking without sorting arm, and the ranking and sorting arm, respectively. The extra cost for adding the ranking task was about 15%. CONCLUSIONS: The benefit does not justify the effort involved in the ranking task. For this reason, the addition of the ranking task to the present EQ-5D-5L valuation protocol is not an attractive option.

"When I saw walking I just kind of took it as wheeling": interpretations of mobility-related items in generic, preference-based health state instruments in the context of spinal cord injury.

BACKGROUND: In health economic analyses, health states are typically valued using instruments with few items per dimension. Due to the generic (and often reductionist) nature of such instruments, certain groups of respondents may experience challenges in describing their health state. This study is concerned with generic, preference-based health state instruments that provide information for decisions about the allocation of resources in health care. Unlike physical measurement instruments, preference-based health state instruments provide health state values that are dependent on how respondents interpret the items. This study investigates how individuals with spinal cord injury (SCI) interpret mobility-related items contained within six preference-based health state instruments. METHODS: Secondary analysis of focus group transcripts originally collected in Vancouver, Canada, explored individuals' perceptions and interpretations of mobility-related items contained within the 15D, Assessment of Quality of Life 8-dimension (AQoL-8D), EQ-5D-5L, Health Utilities Index (HUI), Quality of Well-Being Scale Self-Administered (QWB-SA), and the 36-item Short Form health survey version 2 (SF-36v2). Ritchie and Spencer's 'Framework Approach' was used to perform thematic analysis that focused on participants' comments concerning the mobility-related items only. RESULTS: Fifteen individuals participated in three focus groups (five per focus group). Four themes emerged: wording of mobility (e.g., 'getting around' vs 'walking'), reference to aids and appliances, lack of suitable response options, and reframing of items (e.g., replacing 'walking' with 'wheeling'). These themes reflected item features that respondents perceived as relevant in enabling them to describe their mobility, and response strategies that respondents could use when faced with inaccessible items. CONCLUSION: Investigating perceptions to mobility-related items within the context of SCI highlights substantial variation in item interpretation across six preference-based health state instruments. Studying respondents' interpretations of items can help to understand discrepancies in the health state descriptions and values obtained from different instruments. This line of research warrants closer attention in the health economics and quality of life literature.

Influenced from the start: anchoring bias in time trade-off valuations.

PURPOSE: The de facto standard method for valuing EQ-5D health states is the time trade-off (TTO), an iterative choice procedure. The TTO requires a starting point (SP), an initial offer of time in full health which is compared to a fixed offer of time in impaired health. From the SP, the time in full health is manipulated until preferential indifference. The SP is arbitrary, but may influence respondents, an effect known as anchoring bias. The aim of the study was to explore the potential anchoring effect and its magnitude in TTO experiments. METHODS: A total of 1249 respondents valued 8 EQ-5D health states in a Web study. We used the lead time TTO (LT-TTO) which allows eliciting negative and positive values with a uniform method. Respondents were randomized to 11 different SPs. Anchoring bias was assessed using OLS regression with SP as the independent variable. In a secondary experiment, we compared two different SPs in the UK EQ-5D valuation study TTO protocol. RESULTS: A 1-year increase in the SP, corresponding to an increase in TTO value of 0.1, resulted in 0.02 higher recorded LT-TTO value. SP had little impact on the relative distance and ordering of the eight health states. Results were similar to the secondary experiment. CONCLUSION: The anchoring effect may bias TTO values. In this Web-based valuation study, the observed anchoring effect was substantial. Further studies are needed to determine whether the effect is present in face-to-face experiments.

Learning and satisficing: an analysis of sequence effects in health valuation.

OBJECTIVE: To estimate the effect of sequence on response precision and response behavior in health valuation studies. METHODS: Time trade-off (TTO) and paired comparison responses from six health valuation studies-four US, one Spanish, and one Dutch-were examined (22,225 respondents) to test whether task sequence influences response precision (e.g., rounding), response changes, and median response times. Each study used a computer-based instrument that randomized task sequence among a national sample of adults, age 18 years or older, from the general population. RESULTS: For both TTO and paired comparisons, median response times decreased with sequence (i.e., learning), but tended to flatten after the first three tasks. Although the paired comparison evidence demonstrated that sequence had no effect on response precision, the frequency of rounded TTO responses (to either 1-year or 5-year units) increased with sequence. CONCLUSIONS: Based on these results, randomizing or reducing the number of paired comparison tasks does not appear to influence response precision; however, generalizability, practicality, and precautionary considerations remain. Overall, participants learned to respond efficiently within the first three tasks and did not resort to satisficing, but may have rounded their TTO responses.

Do health preferences contradict ordering of EQ-5D labels?

PURPOSE: The aim of this study was to test whether the ordering of item labels in EQ-5D instruments disagrees with the preferences of US adults. METHODS: A preference inversion occurs when "worse" health along a scale or score is preferred. As a sub-study of the 2013 United States Measurement and Valuation of Health Study, we tested for 33 EQ-5D preference inversions using paired comparisons with unique samples of 50 or more US adults, aged 18 or older. Specifically, we tested whether health preferences contradicted ordering of EQ-5D labels. RESULTS: The EQ-5D-3L and EQ-5D-Y item labels had no significant preference inversions. The EQ-5D-5L version had preference inversions between Levels 4 and 5. For example, 30 out of 59 respondents (51 %) preferred being "extremely" over "severely anxious or depressed," contrary to the ordering of labels for that item. CONCLUSIONS: Preference inversions between Levels 4 and 5 on the EQ-5D-5L were tested and confirmed; therefore, valuation studies may find that Levels 4 and 5 have the same value. To mitigate such inversions, labels could be revised or a 4-level version could be considered.

Examining the relationship between health-related quality of life and increasing numbers of diagnoses.

PURPOSE: Little is known about estimating utilities for comorbid (or 'joint') health states. Several joint health state prediction models have been suggested (for example, additive, multiplicative, best-of-pair, worst-of-pair, etc.), but no general consensus has been reached. The purpose of the study is to explore the relationship between health-related quality of life (HRQoL) and increasing numbers of diagnoses. METHODS: We analyzed a large dataset containing respondents' ICD-9 diagnoses and preference-based HRQoL (EQ-5D and SF-6D). Data were stratified by the number of diagnoses, and mean HRQoL values were estimated. Several adjustments, accounting for the respondents' age, sex, and the severity of the diagnoses, were carried out. Our analysis fitted additive and multiplicative models to the data and assessed model fit using multiple standard model selection methods. RESULTS: A total of 39,817 respondents were included in the analyses. Average HRQoL values were represented well by both linear and multiplicative models. Although results across all analyses were similar, adjusting for severity of diagnoses, age, and sex strengthened the linear model's performance measures relative to the multiplicative model. Adjusted R (2) values were above 0.99 for all analyses (i.e., all adjusted analyses, for both HRQoL instruments), indicating a robust result. CONCLUSIONS: Additive and multiplicative models perform equally well within our analyses. A practical implication of our findings, based on the presumption that a linear model is simpler than an additive model, is that an additive model should be preferred unless there is compelling evidence to the contrary.

Time trade-off and attitudes toward euthanasia: implications of using 'death' as an anchor in health state valuation.

PURPOSE: Health state values are by convention anchored to 'perfect health' and 'death.' Attitudes toward death may consequently influence the valuations. We used attitudes toward euthanasia (ATE) as a sub-construct for attitudes toward death. We compared the influence on values elicited with time trade-off (TTO), lead-time TTO (LT-TTO) and visual analogue scale (VAS).Since the 'death' anchor is most explicit in TTO, we hypothesized that TTO values would be most influenced by ATE. METHODS: Respondents valued eight EQ-5D health states with VAS, then TTO (n = 328) or LT-TTO (n = 484). We measured ATE on a scale from -2 (fully disagree) to 2 (fully agree) and used multiple linear regressions to predict VAS, TTO, and LT-TTO values by ATE, sex, age, and education. RESULTS: A one-point increase on the ATE scale predicted a mean TTO value change of -.113 and LT-TTO change of -.072. Demographic variables, but not ATE, predicted VAS values. CONCLUSIONS: TTO appears to measure ATE in addition to preferences for health states. Different ways of incorporating death in the valuation may impact substantially on the resulting values. 'Death' is a metaphysically unknown concept, and implications of attitudes toward death should be investigated further to evaluate the appropriateness of using 'death' as an anchor.

Time trade-off and ranking exercises are sensitive to different dimensions of EQ-5D health states.

BACKGROUND: One method suggested for creating preference-based tariffs for the new five-level EuroQol five-dimensional (EQ-5D) questionnaire is combining time trade-off (TTO) and discrete choice exercises. Rank values from previous valuation studies can be used as proxies for discrete choice exercises. This study examined rank and TTO data to determine whether the methods differ in sensitivity to the EQ-5D questionnaire dimensions. METHODS: We used rank and TTO data for 42 EQ-5D questionnaire health states from the US and UK three-level EQ-5D questionnaire valuation studies, extracting overall ranks of mean TTO and mean rank values, ranging from 1 (best) to 42 (worst). We identified pairs of health states with reversed overall ranks between TTO and rank data and regressed overall rank differences (TTO - ranking) on dummy variables representing impairments on EQ-5D questionnaire dimensions. RESULTS: Forty-three (US) and 41 (UK) health state pairs displayed reversed rank order. Both US and UK regression models on rank differences indicated that respondents rated impairments involving pain/discomfort and anxiety/depression as relatively worse in TTO than in the ranking task. DISCUSSION: Different dimension sensitivity between TTO and ranking methods suggests that combining them could lead to inconsistent tariffs. Differences could be caused by respondents focusing on the first presented dimensions when ranking states or could be related to the longest endurable time for health states involving pain/discomfort or anxiety/depression. The observed differences call into question which method best represents the preferences of the population.

Learning effects in time trade-off based valuation of EQ-5D health states.

OBJECTIVES: In EuroQol five-dimensional questionnaire valuation studies, each participant typically assesses more than 10 hypothetical health states by using the time trade-off (TTO) method. We wanted to explore potential learning effects when using the TTO method, that is, whether the valuations were affected by the number of previously rated health states (the sequence number). METHODS: We included 3773 respondents from the US EQ-5D valuation study, each of whom valued 12 health states (plus unconscious) in random order. With linear regression, we used sequence number to predict mean and standard deviations across all health states. We repeated the analysis separately for TTO responses indicating a state better than death and a state worse than death. Each TTO value requires a specific number of choice iterations. To test whether respondents used fewer iterations with experience, we used linear regression with sequence number as the independent variable and number of iterations as the dependent variable. RESULTS: Mean TTO values were fairly stable across the sequence number, but analyzing state better than death and state worse than death values separately revealed a tendency toward more extreme values: state better than death values increased by 0.02, while state worse than death values decreased by 0.21 (P < 0.0001) over the full sequence. The standard deviations increased slightly, while the number of choice iterations was the same over the sequence number. The findings were stable across the levels of health state severity, age, and sex. CONCLUSIONS: TTO values become more extreme with increasing experience. Because of the randomized valuation order, these effects do not bias specific health states; however, they reduce the overall validity and reliability of TTO values.

A critical re-evaluation of the regression model specification in the US D1 EQ-5D value function.

BACKGROUND: The EQ-5D is a generic health-related quality of life instrument (five dimensions with three levels, 243 health states), used extensively in cost-utility/cost-effectiveness analyses. EQ-5D health states are assigned values on a scale anchored in perfect health (1) and death (0).The dominant procedure for defining values for EQ-5D health states involves regression modeling. These regression models have typically included a constant term, interpreted as the utility loss associated with any movement away from perfect health. The authors of the United States EQ-5D valuation study replaced this constant with a variable, D1, which corresponds to the number of impaired dimensions beyond the first. The aim of this study was to illustrate how the use of the D1 variable in place of a constant is problematic. METHODS: We compared the original D1 regression model with a mathematically equivalent model with a constant term. Comparisons included implications for the magnitude and statistical significance of the coefficients, multicollinearity (variance inflation factors, or VIFs), number of calculation steps needed to determine tariff values, and consequences for tariff interpretation. RESULTS: Using the D1 variable in place of a constant shifted all dummy variable coefficients away from zero by the value of the constant, greatly increased the multicollinearity of the model (maximum VIF of 113.2 vs. 21.2), and increased the mean number of calculation steps required to determine health state values. DISCUSSION: Using the D1 variable in place of a constant constitutes an unnecessary complication of the model, obscures the fact that at least two of the main effect dummy variables are statistically nonsignificant, and complicates and biases interpretation of the tariff algorithm.

Comparison of hypothetical and experienced EQ-5D valuations: relative weights of the five dimensions.

PURPOSE: EQ-5D tariffs are typically based on general population valuations studies, but whether valuations of experienced health (EH) or hypothetical health (HH) are more appropriate is disputed. Previous comparisons of valuations of EH and HH have focused on absolute differences in dimension-specific regression coefficients. We examined differences in the relative importance attributed to the EQ-5D dimensions between EH and HH valuations of EQ-5D states in the United States. METHODS: We used the regression model from the US EQ-5D valuation study on EH ratings from the 2000-2003 Medical Expenditure Panel Survey and on HH ratings from the US EQ-5D valuation study conducted in 2001. We then compared patterns in the relative magnitudes of coefficients that corresponded to the five dimensions. RESULTS: In the HH model, self-care and pain/discomfort were the most important dimensions, while usual activities were the least important. In the EH model, usual activities were the most important dimension, while self-care was one of the least important. DISCUSSION: The findings reveal considerable differences between stated preferences for HH and ratings of EH, particularly for self-care and usual activities. The findings accentuate the importance of the debate about which groups' values should be used in medical priority setting.

Globalization, neo-liberalism and community psychology.

A longitudinal analysis (1984-2005) of media language in Norway is presented, demonstrating how the current globalized capitalist market ideology is now permeating this long-established Scandinavian welfare state. This ideological shift carries powerful implications for community psychology, as traditional welfare state values of equal services based on a universalistic principle are set aside, and social and material inequalities are increasingly accepted. The methodology developed in the present study may serve as a "barometer of community changes", to borrow a metaphor used by Sarason (2000).

EQ-5D-5L Valuation for the Malaysian Population.

OBJECTIVES: The aim of this study was to develop an EQ-5D-5L value set reflecting the health preferences of the Malaysian adult population. METHODS: Respondents were sampled with quotas for urbanicity, gender, age, and ethnicity to ensure representativeness of the Malaysian population. The study was conducted using a standardized protocol involving the EuroQol Valuation Technology (EQ-VT) computer-assisted interview system. Respondents were administered ten composite time trade-off (C-TTO) tasks and seven discrete choice experiment (DCE) tasks. Both linear main effects and constrained non-linear regression models of C-TTO-only data and hybrid models combining C-TTO and DCE data were explored to determine an efficient and informative model for value set prediction. RESULTS: Data from 1125 respondents representative of the Malaysian population were included in the analysis. Logical consistency was present in all models tested. Using cross-validation, eight-parameter models for C-TTO only and C-TTO + DCE hybrid data displayed greater out-of-sample predictive accuracy than their 20-parameter, main-effect counterparts. The hybrid eight-parameter model was chosen to represent the Malaysian value set, as it displayed greater out-of-sample predictive accuracy over C-TTO data than the C-TTO-only model, and produced more precise estimates. The estimated value set ranged from - 0.442 to 1. CONCLUSIONS: The constrained eight-parameter hybrid model demonstrated the best potential in representing the Malaysian value set. The presence of the Malaysian EQ-5D-5L value set will facilitate its application in research and health technology assessment activities.

EuroQol Protocols for Time Trade-Off Valuation of Health Outcomes.

The time trade-off (TTO) valuation technique is widely used to determine utility values of health outcomes to inform quality-adjusted life-year (QALY) calculations for use in economic evaluation. Protocols for implementing TTO vary in aspects such as the trade-off framework, iteration procedure and its administration model and method, training of respondents and interviewers, and quality control of data collection. The most widely studied and utilized TTO valuation protocols are the Measurement and Valuation of Health (MVH) protocol, the Paris protocol and the EuroQol Valuation Technology (EQ-VT) protocol, all developed by members of the EuroQol Group. The MVH protocol and its successor, the Paris protocol, were developed for valuation of EQ-5D-3L health states. Both protocols were designed for a trained interviewer to elicit preferences from a respondent using the conventional TTO framework with a fixed time horizon of 10 years and an iteration procedure combining bisection and titration. Developed for valuation of EQ-5D-5L health states, the EQ-VT protocol adopted a composite TTO framework and made use of computer technology to facilitate data collection. Training and monitoring of interviewers and respondents is a pivotal component of the EQ-VT protocol. Research is underway aiming to further improve the EuroQol protocols, which form an important basis for the current practice of health technology assessment in many countries.

Drop-out from addiction treatment: a systematic review of risk factors.

Completion of addiction treatment is one of the most consistent factors associated with a favorable treatment outcome. Unfortunately, it is more common for a patient to drop-out of addiction treatment than to complete the treatment. To prevent drop-out, risk factors must be identified. This box-score review focuses on studies investigating the risk factors associated with drop-out from addiction treatment published in peer-reviewed journals from 1992 to 2013. A total of 122 studies involving 199,331 participants met the inclusion criteria. Contrary to recommendations from previous reviews, 91% of the included studies focused primarily on enduring patient factors, mainly demographics. The most consistent risk factors across the different study designs, samples, and measurement methods were cognitive deficits, low treatment alliance, personality disorder, and younger age. With the exception of younger age, none of the demographic factors emerged as consistent risk factors. Further research on the relationship between simple demographic factors and drop-out risk is of limited value. However, little is known about the potential risk factors related to treatment programs and to the treatment processes. Based on the review, clinical recommendations include assessing cognitive functioning and personality disorders at baseline and continuous monitoring of treatment alliance.

Impact of transformation of negative values and regression models on differences between the UK and US EQ-5D time trade-off value sets.

BACKGROUND: National EQ-5D value sets are developed because preferences for health may vary in different populations. UK values are lower than US values for most of the 243 possible EQ-5D health states. Although similar protocols were used for data collection, analytic choices regarding how to model values from the collected data may also influence national value sets. Participants in the UK and US studies assessed the same subset of 42 EQ-5D health states using the time trade-off (TTO) method. However, different methods were used to transform negative values to a range bounded by 0 and -1, and values for all 243 health states were estimated using two different regression models. The transformation of negative values is inconsistent with expected utility theory, and the choice of which transformation method to use lacks a theoretical foundation. OBJECTIVES: Our objectives were to assess how much of the observed difference between the UK and US EQ-5D value sets may be explained by the choice of transformation method for negative values relative to the choice of regression model and the differences between elicited TTO values in the respective national studies (datasets). METHODS: We applied both transformation methods and both regression models to each of the two datasets, resulting in eight comparable value sets. We arranged these value sets in pairs in which one source of difference (transformation method, regression model or dataset) was varied. For each of these paired value sets, we calculated the mean difference between the two matching values for each of the 243 health states. Finally, we calculated the mean utility gain for all possible transitions between pairs of EQ-5D health states within each value set and used the difference in transition scores as a measure of impact from changing transformation method, regression model or dataset. RESULTS: The mean absolute difference in values was 1.5 times larger when changing the transformation method than when using different datasets. The choice of transformation method had a 3.2 times larger effect on the mean health gain (transition score) than the choice of dataset. The mean health gain in the UK value set was 0.09 higher than in the US value set. Using the UK transformation method on the US dataset reduced this absolute difference to 0.02. The choice of regression model had little overall impact on the differences between the value sets. CONCLUSIONS: Most of the observed differences between the UK and US value sets were caused by the use of different transformation methods for negative values, rather than differences between the two study populations as reflected in the datasets. Changing the regression model had little impact on the differences between the value sets.