|BEC||Behavioral Economics||ESP||Applied Health Economics, Services, and Policy Research|
|DEC||Decision Psychology and Shared Decision Making||MET||Quantitative Methods and Theoretical Developments|
* Candidate for the Lee B. Lusted Student Prize Competition
Purpose: There is evidence that patient preferences are systematically higher than societal preferences for a patient’s self-reported health state, possibly due to adaption to chronic illness by patients. It is less clear whether stated preferences for hypothetical health states differ between persons with and without specific conditions. The aim of this study was to determine if presence of specific chronic conditions affected the values estimated for hypothetical EQ-5D health states.
Methods: Data were taken from the US Valuation of EQ-5D Health States. Study participants (N = 3,773) comprised a probability sample of the US adult population in 2002. Each participant valued 12 of a subset of 45 of the 243 EQ-5D health states in a TTO exercise and reported on the presence or absence of 18 chronic conditions. A novel conceptual model was developed to explain the direct and indirect effects of illness experience on values for hypothetical health states. The analyses focused on six conditions: arthritis, diabetes, depression, congestive heart failure, cancer, and allergic rhinitis. Multivariable linear regression was used to estimate differences in health state preferences among persons with a given condition alone, that condition plus one or more other conditions, one or more other conditions, or no chronic conditions while controlling for the satisfaction attributed to own health, other interpersonal differences, and the perceived severity of the valued states. All analyses accounted for the complex sampling design of the US EQ-5D valuation study.
Results: There were no statistically significant differences in mean health state preferences among the four condition-related strata for any of the six chronic conditions. No trend towards adaptation was suggested among those with specific conditions as the direction of the relationship was inconsistent. The strongest predictors of health state preferences were race/ethnicity, age, and marital status.
Conclusions: Results suggest self-reported chronic conditions have a trivial impact on preferences for hypothetical health states while race/ethnicity has a strong effect, consistent with results of a previous study. These results have important implications for researchers who seek to use patient preferences to generate preference-weighting algorithms for condition-specific health state classifiers. However, due to data limitations, including reliance on self-reported data and lack of data on severity/treatment of disease, further investigation is needed.
Purpose: Existing “gold standard” preference assessment methods may suffer from problems with validity and reliability. Conjoint analysis, a method of consumer preference measurement, may have superior comparative performance characteristics. We aimed to conduct a randomized comparison of methods in men at risk for prostate cancer.
Method: Men who had undergone treatment for localized prostate cancer were interviewed individually to identify attributes of prostate cancer treatment that were critical to patients. Using these attributes, we developed 3 preference assessment applications: rating scale (RS), time trade-off, (TTO) or conjoint analysis-based (CA). These applications were first piloted in men who had been treated for prostate cancer. We then randomized men who had undergone prostate biopsy with negative results to preference assessment with the CA and either TTO or RS applications. Validity of preference measurement was compared by assessing the ability of the utility functions derived from each application to successfully predict the patient’s preference for novel multi-attribute health states that he had not yet seen or rated. We compared the internal consistency and predictive validity of each method at the individual level as well as the perceived difficulty and effectiveness of each task. We compared the most important treatment attributes identified by each method.
Result: 17 subjects have been randomized to date. Average age was 64 years, range 55 – 71, 29% were Caucasian, 47% were African-American, 25% other. Educational attainment was 12% High School, 53% some college, 35% college graduate. The CA and RS methods had high internal consistency compared with TTO (average r2 of 85% (CA), 86% (RS) and 44% (TTO). Utility functions derived from CA and RS were superior at prediction of preference for novel multi-attribute health states compared with that of TTO. The most important three attributes to patients as determined using CA were effect on urinary function, sexual function and surgery avoidance. These differed from those identified using RS and TTO in the inclusion of surgery avoidance instead of bowel function. Patients felt CA was the most difficult method, but also the most effective at expressing their values.
Conclusion: Conjoint analysis is a feasible method of preference assessment in men at risk for prostate cancer, and is viewed as effective by such patients. Both RS and CA outperform TTO based on preliminary results.
Purpose: The Patient Oriented Prostate Utility Scale (PORPUS-U) is a 10-item disease-specific multiattribute utility instrument with utility weights from prostate cancer patients. The Prostate Cancer Index (PCI) is a descriptive quality of life instrument producing function and bother scores ranging from 0 (poor outcome) to 100 (good outcome) for urinary, sexual, and bowel problems. The study objective was to develop a function to predict utility scores from PCI scores.
Method: We used patient-level data from previous studies in which the PCI and PORPUS were administered concurrently. Study 1 included 248 prostate cancer patients from an outpatient clinic interviewed on 3 occasions within 18 months. Study 2 included 676 community-dwelling prostate cancer patients who completed the questionnaires by mail. The derivation sample (Study 2) was used to fit three linear regression models, chosen based on previous work. Study 1 data were used to validate the models. PCI scores were divided by 100 to range from 0 to 1. One model used the original PORPUS-U scores, and two used log-transformed PORPUS-U scores, one with a hierarchy constraint and one without. Also, all models were run with and without patient age. Model selection was performed with PORPUS-U score as the dependent variable and PCI score as the covariate, using stepwise selection and 5-fold cross validation. The predictive ability of the models was assessed.
Result: The best-fitting model used the log-transformed PORPUS-U with no hierarchy constraint. Inclusion of age did not improve the model. Scores were untransformed for validation, and Dunn’s smearing estimator applied to correct potential bias in the estimate. The r-squared was 0.72. The RMSE ranged from 0.041 to 0.061 for the 3 validation datasets. We compared the observed PORPUS-U scores to scores predicted from PCI responses. The mean predicted and observed scores were similar (eg., 0.966 vs 0.956). The mean predicted scores were also similar across quartiles of observed scores but slightly overestimated the lowest 5% of observed PORPUS-U scores.
Conclusion: We developed an algorithm to predict PORPUS-U utility scores from PCI scores. This facilitates the estimation of patient-derived utilities for clinical and health economic studies from the many published studies using the PCI. This is also, to our knowledge, the only attempt to map a disease-specific quality of life instrument to a disease-specific utility measure.
Purpose: Caregivers report diminished quality of life and negative physical effects of caring for ill individuals. This study measured the spillover disutility of chronic conditions on household members in the US.
Method: Medical Expenditures Panel Survey (MEPS) data from 2000-2003 were analyzed to identify the independent effect of the presence of individuals with categories of chronic conditions (by ICD 9 codes) on household members’ utility scores. Bivariate regressions of categories of conditions on adults’ utility scores were conducted to identify those which significantly affected utility for inclusion in the multivariable model. A two-stage, multivariable regression model was built to predict EuroQol-5D index weights (EQ-5D) based on the presence of mental health and non-mental health chronic conditions within the household while controlling for other known predictors of utility (including own health status).
Result: In bivariate analyses, mental disorders was the only category of household chronic conditions that significantly affected adults’ utility scores, so multivariable models included mental and non-mental disorders as categories of conditions. In the first-stage, logistic model, the presence of at least one child in the household with a chronic mental health condition decreased the odds of a co-habiting adult reporting perfect health by 28% (OR for EQ-5D score of 1.0=0.72, 95%CI=0.62,0.82); the presence of an adult in the household with a chronic mental health condition decreased the odds of other adults reporting perfect health by 34% (OR=0.66, 95%CI=0.59,0.73). In the second-stage, linear model, among adults reporting less-than-perfect health (EQ-5D score<1.0), the presence of a child or adult with a mental health condition in the household reduced their EQ-5D score by 0.02 (95%CI =-0.03,-0.01 for both). In comparison, chronic non-mental health conditions among children and adults in the household reduced co-habiting adults’ odds of reporting perfect health by 12-13% (95%CI for child= 0.81,0.94, for adult=0.82,0.92), and among those reporting less-than-perfect health, a child with non-mental health condition had no spillover effect on adults and an adult reduced others’ EQ-5D score by 0.01 (95%CI=-0.01,0).
Conclusion: In a US national sample, all health conditions produced spillover disutility on household members, but mental disorders more substantially affected parents, spouses and other adults in the household. Benefits of mental health interventions may be more accurately captured by including the spillover effects of these conditions on family members.
Purpose: To assess women’s preferences for breast cancer chemoprevention (i.e., tamoxifen or raloxifene) using conjoint analysis.
Methods: Eight attributes related to taking a pill to prevent breast cancer were identified and assigned levels (lifetime risk of breast cancer, length of time the pill must be taken, breast cancer risk reduction, risk of endometrial cancer, risk of blood clots, risk of hormone symptoms, risk reduction of bone fractures, and availability of a biomarker). The SAS conjoint analysis program was used to develop a balanced and efficient design consisting of 36 scenarios. Each scenario presented a hypothetical pill description, including each of the 8 attributes with different levels, and asked respondents to indicate how likely they would be to take that pill on a scale of 0 (not at all likely) to 9 (very likely). A randomized block design was used to equally divide the 36 scenarios. An Internet sample of women aged 40-74 was invited to complete one set of 18 scenarios plus a dominant scenario. The responses were combined and conjoint analysis was used to generate attribute importance scores and part-worth utilities of each level.
Results: The 1365 respondents had a mean age of 57 and 78% were white. The mean value for likelihood of taking the pill was 5.5 (SD 3.2) for the dominant scenario and ranged from 2.1 (SD 2.4)-5.7 (SD 2.4) for other scenarios. The order of attribute importance was lifetime risk (17.4%), time (17%), risk of blood clots (12.3%), risk of endometrial cancer (12%), breast cancer risk reduction (11.2%), biomarker availability (10.9%), reduction in bone fracture risk (9.7%) and risk of hormone symptoms (9.7%). Part-worth utility values indicated that women preferred a pill with the following features: 90% breast cancer risk reduction, had a biomarker, no additional risks for all side effects, and could be taken for 1 year.
Conclusions: There was low interest in taking a pill as a means of preventing breast cancer in this Internet sample even when the pill had high benefits and low risks. The similarity of attribute importance values suggests that all were somewhat important, with lifetime risk of breast cancer and serious, but rare, side effects being most important. Further research evaluating associations between preferences and chemoprevention adherence in high risk patients is needed.
Purpose: In analysing DCE, we typically assume that individual respondents evaluate each and every attribute offered in each alternative, and choose their most preferred. This study explores the effect of respondent attribute processing, using ‘attribute importance’, on parameter estimation, model fit and marginal rates of substitution (MRS), in colorectal cancer screening
Methods: The survey, a fractional factorial design of a two-alternative, unlabeled experiment, was mailed to a sample of 1920 subjects in NSW, Australia. Attributes included: test accuracy for cancer and for large polyps, false positive rate, cost, dietary & medication restrictions and sample collection. The importance of each attributes was assessed using a Likert scale, where 1= very important and 5 = not important at all, dichotomised for analysis (1-2 = important, 3-5= not important/neutral). Two analyses were conducted where it was assumed that 1) all attributes are attended to and influence choices (usual analysis practice); and 2) attributes were stratified by their importance on the Likert scale, using interaction terms to indicate whether attributes were important, or not. Mixed logit models were used to estimate preferences.
Results: 1152 from 1920 surveys (60%) were returned. Both choice models significantly predicted respondent test preferences. In comparing models, Model 2 was significantly better than Model 1 (chi-square equal to 485.4 (with 6 degrees of freedom, p<0.00001). There was also an improvement in McFadden’s pseudo R2 with Model 2; the reduction in AIC moving from Model 1 to Model 2 indicated that this improvement remained even after penalising for the loss of parsimonious specification. Respondents who reported the attribute was important to them had significantly higher parameter estimates compared to those who considered the attribute not important or neutral. This was consistent across all attributes, and also resulted in significant differences in MRS and WTP.
Conclusions: Rather than assuming all attributes are equally attended to by respondents, our analysis suggests that taking account of respondent-reported attribute importance (as a proxy for attribute processing) may result in models that better explain respondent’s choice behaviour and preferences. This issue and other attribute processing strategies should be further explored in different settings and data sets