Predicting utility ratings for joint health States from single health States in prostate cancer: empirical testing of 3 alternative theories.

BACKGROUND: Cost-effectiveness analyses measure quality of life by associating utilities with specific health states. Utilities are often defined by single health states, such as incontinence or impotence in the case of prostate cancer treatments. Health conditions often occur simultaneously, yielding joint health states (e.g., impotence with incontinence). Given the combinatorial mathematics involved, even a small number of conditions can result in large numbers of potential joint states, complicating utility elicitation for all relevant states. Analytic predictions for joint-state utilities have been based on 3 theoretical models: 1) multiplicative, 2) additive, and 3) minimum models. These models' empirical accuracy for joint-state utility prediction has been minimally examined. The authors compared these 3 models for predicting joint-state utilities from single-state utilities in men at the time of prostate biopsies. METHODS: Utilities were collected using time tradeoff in 2 university-based prostate biopsy clinics (N = 147). Single-state utilities were elicited for impotence, incontinence, watchful waiting, and post-prostatectomy. Joint-state utilities were elicited for states combining impotence with 1) incontinence, 2) postprostatectomy, or 3) watchful waiting. Testing 3 prediction models of joint-state utilities for bias and consistency, the predictions were compared against directly elicited joint-state utilities. RESULTS: All 3 models are biased. The minimum model is preferred, being the least biased and most efficient. CONCLUSIONS: No current model accurately predicts joint-state utility using the component single-state utilities. When possible, joint-state utilities should be elicited. If not possible, the minimum model is recommended. Research to identify better models is needed.     

A review and meta-analysis of prostate cancer utilities.

BACKGROUND: Health-related quality of life is a key issue in prostate cancer (PC) management. The authors summarized published utilities for common health-related quality of life outcomes of PC and determined how methodological factors affect them. METHODS: In their systematic review, the authors identified 23 articles in English, providing 173 unique utilities for PC health states, each obtained from 2 to 422 respondents. Data were pooled using linear mixed-effects modeling with utilities clustered within the study, weighted by the number of respondents divided by the variance of each utility. RESULTS: In the base model, the estimated utility of the reference case (scenario of a metastatic PC patient with severe sexual symptoms, rated by non-PC patients using time tradeoff) was 0.76. Disease stage, symptom type and severity, source of utility, and scaling method were associated with utility differences of 0.10 to 0.32 (P < 0.05). Utilities from PC patients rating their own health were 0.14 higher than those from the reference case, but utilities from PC patients rating scenarios were lowest. Time tradeoff yielded the highest utilities. Computer administration yielded lower utilities than personal interview (P = 0.02). Neither the scale's high anchor nor study purpose had significant effects on utilities. CONCLUSIONS: This study provides pooled utility estimates for common PC health states and describes how clinical and methodological factors can significantly affect these values. When possible, utility estimates for a modeling application should be derived similarly. Formal data synthesis methods might be useful to researchers integrating utility data from heterogeneous sources. Further exploration of these methods for this purpose is warranted.     

Health state utilities for childhood attention-deficit/hyperactivity disorder based on parent preferences in the United kingdom.

OBJECTIVES: The purpose of this study was to use standard gamble (SG) utility interviews to assess parent preferences for health states of childhood attention-deficit/hyperactivity disorder (ADHD). METHODS: The study was conducted in August 2003 in London, England. Parents (N=83) of children diagnosed with ADHD completed SG utility interviews, rating their child's current health and 14 hypothetical health states describing untreated ADHD and ADHD treated with a nonstimulant, immediate-release stimulant, or extended-release stimulant. Raw temporary utilities ranging from 0 (worst health) to 1 (best health) were adjusted to a chronic utility scale ranging from 0 (death) to 1 (best health) using a linear transformation. Parents rated the severity of their children's ADHD symptoms using the Attention-Deficit/Hyperactivity Disorder Rating Scale-IV (ADHD-RS) and their children's health-related quality of life using the EuroQol EQ-5D. RESULTS: Raw and adjusted SG ratings of hypothetical health states ranged from 0.63-0.90 and 0.88-0.96, respectively. Parents' raw SG scores of their child's current health state (mean=0.72) were significantly correlated with inattentive, hyperactive, and overall ADHD symptoms (r=-0.25, -0.27, -0.27; P <0.05) and the EQ-5D visual analogue scale (r=0.26; P <0.05). CONCLUSION: This UK-based study suggests that parent SG interviews are a valid method for obtaining utilities for child ADHD-related health states. The utilities obtained in this study would be appropriate for use in a cost-utility analysis evaluating the costs and benefits of childhood ADHD treatments in the United Kingdom.     

Health-state utilities in liver disease: a systematic review.

OBJECTIVES: Health-state utilities are essential for cost-utility analysis. Few estimates exist for liver disease in the literature. The authors' aim was to conduct a systematic review of health-state utilities in liver disease, to look at the variation of study designs used, and to pool utilities for some liver disease states. METHODS: A search of MED-LINE, EMBASE, and CINAHL from 1966 to September 2006 was conducted including key words related to liver disease and utility measuring tools. Articles were included if health-state utility tools or expert opinion were used. Variance-weighted mean utility estimates were pooled using metaregression adjusting for disease state and utility assessment method. RESULTS: Thirty studies measured utilities of liver diseases/disease states. Half of these estimated utilities for hepatitis viruses: hepatitis A (n = 1), hepatitis B (n = 4), and hepatitis C (n = 10). Others included liver transplant (n= 6) and chronic liver disease (n= 5) populations. Twelve utility methods were used throughout. The EQ-5D (n = 10) was most popular method, followed by visual analogue scale (n = 9), time tradeoff (n = 6), and standard gamble (n = 4). Respondents were patients (n= 16), an expert panel (n = 10), non-liver diseases adults ( n=2), patient and expert (n = 1), and patient and healthy adult (n = 1). Type of perspective included community (n=21), patient (n=4), and both (n = 5). The pooled mean estimates in hepatitis C with moderate disease, compensated cirrhosis, decompensated cirrhosis, and post-liver transplant using the EQ-5D were 0.75, 0.75, 0.67, and 0.71, respectively. The change in these utilities using different methods were -0.07 (visual analogue scale), -0.01 (health utilities index version 3), +0.04 (standard gamble), + 0.08 (health utilities index version 2), + 0.12 (time tradeoff), and + 0.15 (standard gamble-transformed visual analogue scale). CONCLUSIONS: The authors have created a valuable liver disease- based utility resource from which researchers and policy makers can easily view all available utility estimates from the literature. They have also estimated health-state utilities for major states of hepatitis C.     

Infectious disease prediction with kernel conditional density estimation

Creating statistical models that generate accurate predictions of infectious disease incidence is a challenging problem whose solution could benefit public health decision makers. We develop a new approach to this problem using kernel conditional density estimation (KCDE) and copulas. We obtain predictive distributions for incidence in individual weeks using KCDE and tie those distributions together into joint distributions using copulas. This strategy enables us to create predictions for the timing of and incidence in the peak week of the season. Our implementation of KCDE incorporates 2 novel kernel components: a periodic component that captures seasonality in disease incidence and a component that allows for a full parameterization of the bandwidth matrix with discrete variables. We demonstrate via simulation that a fully parameterized bandwidth matrix can be beneficial for estimating conditional densities. We apply the method to predicting dengue fever and influenza and compare to a seasonal autoregressive integrated moving average model and HHH4, a previously published extension to the generalized linear model framework developed for infectious disease incidence. The KCDE outperforms the baseline methods for predictions of dengue incidence in individual weeks. The KCDE also offers more consistent performance than the baseline models for predictions of incidence in the peak week and is comparable to the baseline models on the other prediction targets. Using the periodic kernel function led to better predictions of incidence. Our approach and extensions of it could yield improved predictions for public health decision makers, particularly in diseases with heterogeneous seasonal dynamics such as dengue fever.     

Utilities derived from visual analog scale scores in patients with HIV/AIDS.

BACKGROUND: Visual analog scale (VAS) scores are used as global quality-of-life indicators and, unlike true utilities (which assess the desirability of health states v. an external metric), are often collected in HIV-related clinical trials. The purpose of this study was to derive and evaluate transformations relating aggregate VAS scores to utilities for current health in patients with HIV/AIDS. METHODS: HIV-specific transformations were developed using linear and nonlinear regression to attain models that best fit mean VAS and standard gamble (SG) utility values directly derived from 299 patients with HIV/AIDS participating in a multicenter study of health values. The authors evaluated the transformations using VAS and SG utility values derived directly from patients in other HIV/AIDS studies. Derived transformations were also compared with published transformations. RESULTS: A simple linear transformation was derived (u = 0.44v + 0.49), as was the exponent for a curvilinear model (u = 1 - [1 - v]1.6), where u = the sample mean utility and v the sample mean VAS score. The curvilinear transformation predicted values within 0.10 of the actual SG utility in 5 of 8 estimates and within 0.05 in 3 of 8 estimates (absolute error ranged from -0.01 to +0.21). The linear transformation performed somewhat better, predicting within 0.10 of the actual SG value in 6 of 8 cases and within 0.05 in 5 of 8 estimates (absolute error ranged from -0.05 to +0.13). An alternative linear model (u = v + 0.018) derived from the literature performed similarly to our linear model (7 of 8 predictions within 0.10, 1 of 8 estimates within 0.05, and absolute error ranging from -0.15 to +0.10), whereas an alternative published curvilinear model (u = 1 - [1 - v]2.3) performed the least well (2 of 8 estimates within 0.10 of the actual values and no estimates within 0.05). CONCLUSIONS: Predicted utilities are a reasonable alternative for use in HIV/AIDS decision analyses and cost-effectiveness analyses. Linear transformations performed better than curvilinear transformations in this context and can be used to convert aggregate VAS scores to aggregate SG values in large HIV/AIDS studies that collect VAS data but not utilities.     

Towards a better QALY model

This paper presents a test of the predictive validity of various classes of QALY models (i.e. linear, power and exponential models). We first estimated TTO utilities for 43 EQ-5D chronic health states and next these states were embedded in nonchronic health profiles. The chronic TTO utilities were then used to predict the responses to TTO questions with nonchronic health profiles. We find that the power QALY model clearly outperforms linear and exponential QALY models. Optimal power coefficient is 0.65. Our results suggest that TTO-based QALY calculations may be biased. This bias can be corrected using a power QALY model.     

A comparison of EQ-5D index scores derived from the US and UK population-based scoring functions.

The authors recently introduced a new preference-based scoring function for the EQ-5D (D1 model) based on time tradeoff valuations from the general adult US population: In this study, they compared the EQ-5D index scores derived from the US (D1) algorithm to the more familiar UK (N3) algorithm. They compared preference-based EQ-5D index scores for all possible EQ-5D health states and differences in EQ-5D index scores between pairs of EQ-5D health states predicted by the D1 and N3 models. The responsiveness of D1- and N3-predicted EQ-5D index scores was assessed using simulated transitions between EQ-5D health states. The mean (SD) EQ-5D index scores for all 243 health states predicted by the D1 and N3 models were 0.37 (0.23) and 0.14 (0.31), respectively. The mean (SD) absolute difference in EQ-5D index scores for all 29,403 pairs of health states was 0.25 (0.19) and 0.35 (0.27), according to the D1 and N3 models, respectively. The D1 and N3 models were consistent in predicting gains/losses for 27,592 (94%) transitions between EQ-5D health state pairs; Cohen effect size, calculated using these 27,592 consistent transitions, was 1.58 and 1.59 for the D1 and N3 models, respectively. Based on these simulation results, it appears that the D1 model would lead to smaller gains in quality-adjusted life years than the N3 model; however, their responsiveness appears to be similar. Empirical studies are needed to examine whether these 2 EQ-5D scoring functions would lead to different conclusions in cost-utility analyses.     

Assessing health state utilities in elderly patients at cardiovascular risk.

BACKGROUND: Health state preferences can be a crucial component of cost-effectiveness analyses, but off-the-shelf health state utilities specifically for older people are not available. OBJECTIVES: Among participants in PROSPER, a trial of pravastatin in patients>70 years, the authors assessed utilities for the health states that were relevant for the trial's cost-utility analysis.Subjects and METHODS: The authors cross-sectionally administered the Health Utilities Index, Mark 3 (HUI) to all PROSPER participants to assess each patient's health state at the time of interview; they then used the scale's multiattribute utility function to estimate the resulting utilities. The population was then stratified into 3 health states, and the mean utility function for each was calculated: recent myocardial infarction (MI, within 3 months), previous MI (>3 months), or no prior MI. Linear and logistic regression were used to control for potential demographic and clinical characteristics. RESULTS: Of the 5804 patients enrolled in the trial, 4677 were administered the HUI instrument. The likelihood of having a complete HUI response set decreased with higher age (P<0.001) but not with the other variables studied. A complete utility score could be calculated for 3390 participants. Of these, 2755 (81.3%) had no history of MI, 546 (16.1%) had an MI>3 months previously, and 89 (2.6%) had an MI within 3 months. The mean (median) utilities were virtually identical for these states: 0.75 (0.84), 0.74 (0.84), and 0.74 (0.84), respectively. From multivariate analyses, utilities decreased with higher age and the presence of several other comorbidities (diabetes, stroke, peripheral vascular disease); women had lower utilities than men (all P<0.01). CONCLUSIONS: In this large implementation of the HUI in elderly patients, the instrument did not detect any differences in estimated utilities related to having a MI. Potential causes of nondiscrimination for MI include the possibility that competing comorbidities may reduce the impact of MI on quality of life in this age group, as well as the possibility that a standard instrument derived from and validated in younger populations may not perform as well in elderly people.     

Mapping analyses to estimate EQ-5D utilities and responses based on Oxford Knee Score

Purpose

The Oxford Knee Score (OKS) is a validated 12-item measure of knee replacement outcomes. An algorithm to estimate EQ-5D utilities from OKS would facilitate cost-utility analysis on studies analyses using OKS but not generic health state preference measures. We estimate mapping (or cross-walking) models that predict EQ-5D utilities and/or responses based on OKS. We also compare different model specifications and assess whether different datasets yield different mapping algorithms.

Methods

Models were estimated using data from the Knee Arthroplasty Trial and the UK Patient Reported Outcome Measures dataset, giving a combined estimation dataset of 134,269 questionnaires from 81,213 knee replacement patients and an internal validation dataset of 45,213 questionnaires from 27,397 patients. The best model was externally validated on registry data (10,002 observations from 4,505 patients) from the South West London Elective Orthopaedic Centre. Eight models of the relationship between OKS and EQ-5D were evaluated, including ordinary least squares, generalized linear models, two-part models, three-part models and response mapping.

Results

A multinomial response mapping model using OKS responses to predict EQ-5D response levels had best prediction accuracy, with two-part and three-part models also performing well. In the external validation sample, this model had a mean squared error of 0.033 and a mean absolute error of 0.129. Relative model performance, coefficients and predictions differed slightly but significantly between the two estimation datasets.

Conclusions

The resulting response mapping algorithm can be used to predict EQ-5D utilities and responses from OKS responses. Response mapping appears to perform particularly well in large datasets.     

Testing alternative regression models to predict utilities: mapping the QLQ-C30 onto the EQ-5D-5L and the SF-6D

Purpose

The purpose of the study was to compare alternative statistical techniques to find the best approach for converting QLQ-C30 scores onto EQ-5D-5L and SF-6D utilities, and to estimate the mapping algorithms that best predict these health state utilities.

Methods

772 cancer patients described their health along the cancer-specific instrument (QLQ-C30) and two generic preference-based instruments (EQ-5D-5L and SF-6D). Seven alternative regression models were applied: ordinary least squares, generalized linear model, extended estimating equations (EEE), fractional regression model, beta binomial (BB) regression, logistic quantile regression and censored least absolute deviation. Normalized mean absolute error (NMAE), normalized root mean square error (NRMSE), r-squared (r²) and concordance correlation coefficient (CCC) were used as model performance criteria. Cross-validation was conducted by randomly splitting internal dataset into two equally sized groups to test the generalizability of each model.

Results

In predicting EQ-5D-5L utilities, the BB regression performed best. It gave better predictive accuracy in terms of all criteria in the full sample, as well as in the validation sample. In predicting SF-6D, the EEE performed best. It outperformed in all criteria: NRMSE?=?0.1004, NMAE?=?0.0798, CCC?=?0.842 and r²?=?72.7% in the full sample, and NRMSE?=?0.1037, NMAE?=?0.0821, CCC?=?0.8345 and r²?=?71.4% in cross-validation.

Conclusions

When only QLQ-C30 data are available, mapping provides an alternative approach to obtain health state utility data for use in cost-effectiveness analyses. Among seven alternative regression models, the BB and the EEE gave the most accurate predictions for EQ-5D-5L and SF-6D, respectively.

     

The reliability and internal consistency of an Internet-capable computer program for measuring utilities

Objective: The purpose of this study was to assess the reliability and internal consistency of measurements of utilities performed with a computer program (iMPACT2) designed for Internet surveys and Internet patient decision-support systems. Methods: We implemented the Internet Multimedia Preference Assessor Construction Tool, version 2 (iMPACT2) program using the combination of a web server, HTML files, and a web-accessible database. The program randomized subjects, screened their responses for missing data and failures of internal consistency, assisted patients with resolving certain inconsistencies, and, upon a subject's completion of the protocol, provided a report of results to the research assistant administering the program. To validate the iMPACT2 program, we recruited 60 healthy community volunteers and elicited preferences in a research-lab setting using a visual analog scale (VAS) and the standard gamble (SG) for subject's current health and three hypothetical states. For purposes of comparison, we also administered a Short Form-12 (SF-12) health-assessment questionnaire. Subjects used the computer software on two occasions separated by 2–4 weeks of time. Results: Visual analog scale and standard gamble ratings for subjects' current health were reliable (intraclass correlation coefficient (ICC) of 0.82 and 0.84 (two outliers excluded − 0.60 without exclusions), respectively) were comparable with the reliability of the Physical and Mental Component scales of the SF-12 (ICCs of 0.84 and 0.75, respectively). Subjects could easily discriminate between hypothetical states (D scores 0.74 for SG and 0.90 for VAS), and 94% had a completely internally consistent ordering of preference ratings for states. Conclusions: iMPACT2 produces measurements of standard gamble utilities that are reliable and have a high degree of internal consistency. Procedures for assessment of utilities developed for desktop computer programs can be translated to software designed for the Internet, facilitating the use of utilities and endpoints in clinical trials and development of web-based decision-support applications for patients. However, further testing, including direct comparisons with traditional interviewer administered utility elicitation protocols, is needed. This revised version was published online in June 2006 with corrections to the Cover Date.     

Agreement between prostate cancer patients and their clinicians about utilities and attribute importance

PURPOSE: To examine the agreement between prostate cancer patients' utilities for selected health states and their rankings of the importance of six attributes of the health states and the clinicians' judgements of what would be in the patients' best interests. METHOD: Patients with newly diagnosed localized prostate cancer individually completed a time trade-off utility assessment shortly after being diagnosed. The health states evaluated were constructed from a multi-attribute utility model that incorporated six aspects of living with the disease and outcomes of treatment. Each patient assessed his current health state and three hypothetical states that might occur in the future, and provided rankings of the importance of the six attributes. The clinicians caring for each patient independently provided their views of what utilities and importance rankings would be in the patient's best interest. RESULTS: The across-participant correlations between patients' and clinicians' utilities were very low and not statistically significant. Across-participant correlations between patient and clinician importance rankings for the six attributes were also low. Across-health state and across-attribute correlations between utilities or importance rankings were highly variable across patient-clinician pairs. CONCLUSION: In the clinical settings studied, there is not a strong relationship between valuations of current and possible future health states by patients with newly diagnosed prostate cancer and their clinicians. Implications of these results for substituted judgement, when clinicians advise their patients or recommend a treatment strategy, are discussed.     

Assessing convergent validity of health-state utilities obtained using different scaling methods.

A number of empirical studies have attempted to assess the convergent validity of health-state utilities obtained using two or more scaling methods (standard gamble, time tradeoff, rating scale, magnitude estimation, equivalence technique, and willingness-to-pay). The data from these studies can be mapped onto an N x K matrix, where N and K are the numbers of respondents and health states, respectively, and each matrix cell consists of a pair of health-state utilities, one obtained using scaling method X and the other obtained using scaling method Y. The Pearson's rassessing convergent validity can then be computed as 1) the unraveled correlation over all N x K data pairs, 2) the mean within-respondent correlation, 3) the mean within-health-state correlation, or 4) the correlation of the across-respondents means. These four different ways of computing the correlation do not necessarily yield the same results. The appropriateness of each method of computing the correlation is considered.     

Analysing the length of care episode after hip fracture: a nonparametric and a parametric Bayesian approach

Effective utilisation of limited resources is a challenge for health care providers. Accurate and relevant information extracted from the length of stay distributions is useful for management purposes. Patient care episodes can be reconstructed from the comprehensive health registers, and in this paper we develop a Bayesian approach to analyse the length of care episode after a fractured hip. We model the large scale data with a flexible nonparametric multilayer perceptron network and with a parametric Weibull mixture model. To assess the performances of the models, we estimate expected utilities using predictive density as a utility measure. Since the model parameters cannot be directly compared, we focus on observables, and estimate the relevances of patient explanatory variables in predicting the length of stay. To demonstrate how the use of the nonparametric flexible model is advantageous for this complex health care data, we also study joint effects of variables in predictions, and visualise nonlinearities and interactions found in the data.     

Validity of utilities of patients with esophageal cancer

OBJECTIVES: The convergent validity between utility assessment methods was assessed. METHODS: Investigated were patients with esophageal cancer treated surgically with curative intent. Patients were interviewed in a period from 3 to 12 months after surgical resection. Patients evaluated their actual health and seven other states. Visual analogue scale (VAS) and standard gamble (SG) utilities were obtained for the health states in an interview. Patients also indicated whether or not they preferred death to living in a health state (worse than dead [WTD] preferences). RESULTS: Fifty patients completed the interview. Convergent validity was excellent at the aggregate and individual level. However, the relation between VAS and SG differed strongly across individuals. On a scale from 0 (dead) to 100 (perfect health), SG scores were lower for patients with WTD preferences (mean difference d = 35; p = .002); however, VAS scores did not vary by WTD preferences. CONCLUSIONS: In general, there is good agreement between VAS and SG measures, although patients disagree about how the VAS and SG are related. The standard gamble varied by WTD preferences, however, the VAS did not.     

The construction of standard gamble utilities

Health effects for cost-effectiveness analysis are best measured in life years, with quality of life in each life year expressed in terms of utilities. The standard gamble (SG) has been the gold standard for utility measurement. However, the biases of probability weighting, loss aversion, and scale compatibility have an inconclusive effect on SG utilities. We determined their effect on SG utilities using qualitative data to assess the reference point and the focus of attention. While thinking aloud, 45 healthy respondents provided SG utilities for six rheumatoid arthritis health states. Reference points, goals, and focuses of attention were coded. To assess the effect of scale compatibility, correlations were assessed between focus of attention and mean utility. The certain outcome served most frequently as reference point, the SG was perceived as a mixed gamble. Goals were mostly mentioned with respect to this outcome. Scale compatibility led to a significant upward bias in utilities; attention lay relatively more on the low outcome and this was positively correlated with mean utility. SG utilities should be corrected for loss aversion and probability weighting with the mixed correction formula proposed by prospect theory. Scale compatibility will likely still bias SG utilities, calling for research on a correction.