Medical Decision Making current issue

12th Biennial European Meeting of the Society for Medical Decision Making Abstracts

Tue, 26 Jan 2010 10:36:33 PST

Comparison of 5 Health-Related Quality-of-Life Indexes Using Item Response Theory Analysis

Fryback, D. G., Palta, M., Cherepanov, D., Bolt, D., Kim, J.-S. — Tue, 26 Jan 2010 10:36:32 PST

Background. Five health-related quality-of-life (HRQoL) indexes—EQ-5D, HUI2, HUI3, QWB-SA, and SF-6D—are each used to assign community-based utility scores to health states, although these scores differ. Objective. The authors transform these indexes to a common scale to understand their interrelationships. Methods. Data were from the National Health Measurement Study, a telephone survey of 3844 US adults. The 5 indexes were analyzed using item response theory analysis to estimate scores on an underlying construct of summary health, . Unidimensionality was evaluated using nonlinear principal components analysis. Index scores were plotted against the estimated scores on the common underlying construct. In addition, scores on the Health and Activities Limitation Index (HALex), the Centers for Disease Control and Prevention Healthy Days questions, and self-rated health on a 5-category scale ranging from excellent to poor were plotted. Results. SF-6D and QWB-SA are nearly linear across the range of but with a shallow slope; EQ-5D, HUI2, and HUI3 are linear with a steep slope from low (poor health) into midrange of , then approximately linear with a less steep slope for higher (health just below to well above average), although the inflection points differ by index. Conclusion. Simple linear functions may serve as crosswalks among these indexes only for lower health states, albeit with low precision. Ceiling effects make crosswalks among most of the indexes ill specified above a certain level of health. Although each index measures generic health on a utility scale, these indexes are not identical but are relatively simply, if imprecisely, related.

Capacity Constraints and Cost-Effectiveness: A Discrete Event Simulation for Drug-Eluting Stents

Jahn, B., Pfeiffer, K. P., Theurl, E., Tarride, J.-E., Goeree, R. — Tue, 26 Jan 2010 10:36:32 PST

Background. Waiting times for access to care, for example, for diagnostic imaging or surgery, are a highly relevant issue in health care. Waiting or deferred treatment caused by limited resource capacities can affect treatment success, quality of life, and costs. However, when treatment alternatives are compared in economic models, often unrestricted availability of resources is assumed, and dynamic changes in waiting lines remain unconsidered. The objective of this study was to evaluate the impact of potential real-world capacity restrictions and implied waiting lines on cost-effectiveness results and additional model outcomes. Methods. A case study of drug-eluting and bare-metal stent treatment illustrates the effect of hypothetical capacity limitations of daily stenting procedures. Therefore, a decision-analytic model which allows for explicitly defined resource capacities and dynamic waiting lines was built using discrete event simulation. Cost-effectiveness, utilization, waiting time, and budgetary impact of alternative treatment scenarios are analyzed under the assumption of limited and unlimited resource capacities. Results. The compared treatment allocation scenarios in the case study demonstrate that the additional cost for waiting increases the average treatment cost per patient. The different scenarios have different impacts on waiting lines because of the number of repeated interventions. Additionally, this effect leads to changes in cost-effectiveness results for the hypothetical capacity limit. Explicitly modeled capacities allow for further analysis of capacity utilization, waiting lines, and budgetary impact. Conclusion. Our model shows that neglected limited capacities can cause wrong cost-effectiveness results. Therefore, capacities should be explicitly included in decision-analytic models if there is evidence of scarcity.

Improving Decision Making at the End of Life With Video Images

Volandes, A. E., Barry, M. J., Chang, Y., Paasche-Orlow, M. K. — Tue, 26 Jan 2010 10:36:32 PST

Background. Decision making at the end of life is frequently complex and often filled with uncertainty. We hypothesized that people with limited health literacy would have more uncertainty about end-of-life decision making than people with adequate literacy. We also hypothesized that video images would decrease uncertainty. Design. Before and after oral survey. Participants. Subjects presenting to their primary care physicians. Methods. Subjects were asked about their preferences for end-of-life care after they heard a verbal description of advanced dementia and were asked to rate the level of their uncertainty. Subjects then viewed a video of a patient with advanced dementia and were asked again about their preferences and uncertainty. Uncertainty was measured using the Decisional Conflict Scale with score ranges from 3 (high uncertainty) to 15 (no uncertainty). Health literacy was measured using the Rapid Estimate of Adult Literacy in Medicine, and subjects were divided into 3 literacy categories: low (0—45, 6th grade and below), marginal (46— 60, 7th—8th grade), and adequate (61—66, 9th grade and above). Results. A total of 146 patients completed the interview. Prior to the video, the average uncertainty scores for subjects with low, marginal, and adequate health literacy were 10.8, 12.4, and 13.5, respectively (P < 0.0001). After the video, the 3 groups had similar uncertainty about their decisions. The average uncertainty scores for subjects with low, marginal, and adequate health literacy were 13.6, 14.1, and 14.5, respectively (P = 0.046). Conclusions. Subjects with limited health literacy expressed more uncertainty about their preferences for end-of-life care than did subjects with adequate literacy. Our video decision aid improved end-of-life decision making by decreasing uncertainty regarding subjects’ preferences, especially for those with limited literacy.

The Challenge of Shared Decision Making Among Patients With Lower Literacy: A Framework for Research and Development

McCaffery, K. J., Smith, S. K., Wolf, M. — Tue, 26 Jan 2010 10:36:32 PST

There have been major advances in techniques to increase patient involvement in health decisions with the benefits of greater involvement and shared decision making now widely recognized. However, there has been little attention in the development of tools and strategies to support patient participation among adults with lower literacy, a group with poor health knowledge, limited involvement in health decisions, and poor health outcomes. The authors put forward a framework to consider the different stages of shared health decision making and the tasks and skills required to achieve each stage. They consider where current research exists in the decision making literature and where more is needed if adults with limited literacy are to be better engaged in shared decision making in health care.

Patients' Preferences for Treatment of Hepatitis C

Fraenkel, L., Chodkowski, D., Lim, J., Garcia-Tsao, G. — Tue, 26 Jan 2010 10:36:32 PST

Background. The objective of this study was to ascertain patient preferences for treatment of hepatitis C virus (HCV). Methods. The authors recruited consecutive patients eligible for treatment of HCV and used adaptive conjoint analysis (ACA), a hybrid approach of conjoint analysis that uses both self-explicated ratings and pair-wise comparisons, to elicit preferences for pegylated-interferon and ribavirin. They examined the association between patient characteristics and treatment preferences using the Mann-Whitney U test and ² statistic for continuous and categorical variables, respectively, and subsequently calculated adjusted odds ratios and 95% confidence intervals using logistic regression. Results. A total of 140 subjects completed the ACA task. The mean (±SD) age of the sample was 51±8 y; 85% were male, and 59% were white. When described as being associated with mild side effects, 67% (n = 94) of subjects preferred treatment for HCV. The percentage of subjects preferring therapy decreased to 51% (n = 72) when it was described as being associated with severe side effects. Preferences for treatment of HCV were stronger among subjects with a higher perceived risk of developing cirrhosis, more severe underlying liver disease, and worse HCV-related quality of life. Subjects having more severe disease placed greater weight on the importance of expected benefits and less on the risk of toxicity compared with those with mild or no fibrosis. Conclusions. Whether to choose treatment for HCV is a difficult decision for many patients. Treatment is usually recommended for those with moderate to severe liver disease, and these results demonstrate that most patients’ preferences are concordant with this practice. Key words: hepatitis C; decision making; pegylated-interferon; ribavirin. (Med Decis Making 2010;30:45—57)

Health Utility Bias: A Systematic Review and Meta-Analytic Evaluation

Doctor, J. N., Bleichrodt, H., Lin, H. J. — Tue, 26 Jan 2010 10:36:32 PST

Background. A common assertion is that rating scale (RS) values are lower than both standard gamble (SG) and time tradeoff (TTO) values. However, differences among these methods may be due to method specific bias. Although SG and TTOs suffer systematic bias, RS responses are known to depend on the range and frequency of other health states being evaluated. Over many diverse studies this effect is predicted to diminish. Thus, a systematic review and data synthesis of RS-TTO and RS-SG difference scores may better reveal persistent dissimilarities. Purpose. The purpose of this study was to establish through systematic review and meta-analysis the net effect of biases that endure over many studies of utilities. Methods. A total of 2206 RS and TTO and 1318 RS and SG respondents in 27 studies of utilities participated. MEDLINE was searched for data from 1976 to 2004, complemented by a hand search of full-length articles and conference abstracts for 9 journals known to publish utility studies, as well as review of results and additional recommendations by 5 outside experts in the field. Two investigators abstracted the articles. We contacted the investigators of the original if required information was not available. Results. No significant effect for RS and TTO difference scores was observed: effect size (95% confidence interval [CI]) = 0.04 (–0.02, 0.09). In contrast, RS scores were significantly lower than SG scores: effect size (95% CI ) =–0.23 (–0.28, –0.19). Correcting SG scores for 3 known biases (loss aversion, framing, and probability weighting) eliminated differences between RS and SG scores: effect size (95% CI ) = 0.01 (–0.03, 0.05). Systematic bias in the RS method may exist but be heretofore unknown. Bias correction formulas were applied to mean not individual utilities. Conclusions. The results of this study do not support the common view that RS values are lower than TTO values, may suggest that TTO biases largely cancel, and support the validity of formulas for correcting SG bias.

Sample Size in Obesity Trials: Patient Perspective Versus Current Practice

Tue, 26 Jan 2010 10:36:33 PST

Objective. To evaluate patient opinions on acceptable risks in exchange for a given degree of weight loss and their implications for sample size determination in obesity randomized clinical trials (RCTs). Design. Survey of patients entering RCTs for weight loss in a university-based clinical research setting and power calculations based on their responses. Participants. Men (n = 8) and women (n = 66) between 24 and 73 years of age with body mass indices ranging from 26.8 to 40.5 kg/m². Measurements. Survey responses to questions assessing the added risk of serious adverse events (SAEs) or death one is willing to assume for a given degree of weight loss. Results. For 5% and 10% weight loss against risk for death per se, the mean acceptable risk tended to be about 3.5%, but the median (0.00) and mode (0.00) suggested that for most individuals, only a risk of ≤ 1% would be acceptable. Figures, estimated dropout rates, and base rates of SAEs (including deaths) from recent obesity trials indicate that 1-year 2-group obesity RCTs would need tens of thousands of participants per group to have 80% power to detect risks that are meaningful to patients at the 2-tailed 0.05 level. Conclusion. Patient education is needed to explain which risks are realistically detectable in RCTs so that patients may provide truly informed consent, or RCT standards should be modified to meet patients’ implicit expectations.

The Language of Prognostication in Intensive Care Units

White, D. B., Engelberg, R. A., Wenrich, M. D., Lo, B., Curtis, J. R. — Tue, 26 Jan 2010 10:36:33 PST

Rationale. Although misunderstandings about prognosis are common in intensive care units (ICUs), little is known about how physicians actually communicate prognostic information. Objectives. The authors sought to 1) develop a framework to describe the language physicians use to disclose prognosis, 2) determine whether physicians frame prognostic statements as estimates for populations or estimates for individual patients, and 3) determine whether physicians use the recommended ‘‘ask-tell-ask’’ approach when discussing prognosis. Methods. The authors conducted a multicenter, cross-sectional study of 51 audiotaped physician-family conferences about life support decisions in ICUs. They identified each prognostic statement and used grounded theory methods to develop a framework to understand the language physicians use to communicate prognosis. Main Results. Physicians prognosticated in 50 of 51 conferences. When discussing prognosis, physicians used qualitative probability statements in 72% (36/50) of conferences, numeric statements in 20% (10/50), absolute statements in 13% (4/32), and nonprobabilistic statements in 40% (20/50). Physicians exclusively used population-based language in 10% (5/50) of conferences, single-event probability statements in 62% (31/50), and both in 28% (14/ 50). In only 2% (1/50) of conferences did physicians ask whether the family wished to hear prognostic information prior to discussing it, and in only 14% of conferences (7/50) did physicians check to verify that families understood the prognostic information. Conclusions. There is considerable variability in the language used by physicians to disclose prognosis, with only 20% of physicians using quantitative terms. Very few physicians checked whether families understood prognostic information. These findings may provide potential targets for interventions to improve communication about prognosis in ICUs.

Estimating Progression Rates for Human Papillomavirus Infection From Epidemiological Data

Jit, M., Gay, N., Soldan, K., Hong Choi, Y., Edmunds, W. J. — Tue, 26 Jan 2010 10:36:33 PST

A Markov model was constructed in order to estimate typespecific rates of cervical lesion progression and regression in women with high-risk human papillomavirus (HPV). The model was fitted to age- and type-specific data regarding the HPV DNA and cytological status of women undergoing cervical screening in a recent screening trial, as well as cervical cancer incidence. It incorporates different assumptions about the way lesions regress, the accuracy of cytological screening, the specificity of HPV DNA testing, and the age-specific prevalence of HPV infection. Combinations of assumptions generate 162 scenarios for squamous cell carcinomas and 54 scenarios for adenocarcinomas. Simulating an unscreened cohort of women infected with high-risk HPV indicates that the probability of an infection continuing to persist and to develop into invasive cancer depends on the length of time it has already persisted. The scenarios and parameter sets that produce the best fit to available epidemiological data provide a basis for modeling the natural history of HPV infection and disease.

The Association Between Individual Time Preferences and Health Maintenance Habits

Bradford, W. D. — Tue, 26 Jan 2010 10:36:33 PST

Context. Encouraging healthy behaviors, including disease screening, exercise, and tobacco avoidance, has been a significant focus of clinical attention in recent decades. Little is known about the association between individual preferences with respect to time play and preventive health care use and healthy lifestyles. Objective. To determine whether rates of these health behaviors are associated with latent time preferences. Design. Interval regression analysis was used to impute individual level discount rates. The difference in means for the rates of health behaviors were assessed for high vs. low to moderate discounting groups using one-factor probit models. Participants. The 2004 wave of the Health and Retirement Survey included in a time preferences module (1,039 respondents aged 24 to 65 years). Main Outcome Measures. Rates of recent mammograms, breast exams, Pap smears, prostate exams, cholesterol testing, flu shots, and dental visits, and non-smoking status. Results. Respondents in the upper 20th percentile of the distribution have an average imputed annual discount rate of 0.335 (33.5%). High discount rate status is found to have a negative marginal association on the probability that respondents had recent mammogram use (—15.1%; P = 0.001), Pap smear use (—8.3%; P = 0.049), prostate examination use (—20.4%; P =0.003), dental visits (—24.8%; P = 0.001), cholesterol testing (—12.4%; P = 0.001), flu shot usage (—11.1%; P = 0.005), rates of vigorous exercise (—15.1%; P = 0.001), nonsmoking status (—10.4%; P= 0.001), and undertook all measured health habits (—7%; P = 0.001). Conclusions. Differences in underlying preferences for the present over the future may be a substantial barrier for people’s propensity to adopt healthy lifestyles.

An Equivalent Relative Utility Metric for Evaluating Screening Mammography

Abbey, C. K., Eckstein, M. P., Boone, J. M. — Tue, 26 Jan 2010 10:36:33 PST

Comparative studies of performance in screening mammography are often ambiguous. A new method will frequently show a higher sensitivity or detection rate than an existing standard with a concomitant increase in false positives or recalls. The authors propose an equivalent relative utility (ERU) metric based on signal detection theory to quantify screening performance in such comparisons. The metric is defined as the relative utility, as defined in classical signal detection theory, needed to make 2 systems equivalent. ERU avoids the problem of requiring a predefined putative relative utility, which has limited application of utility theory in receiver operating characteristic analysis. The metric can be readily estimated from recall and detection rates commonly reported in comparative clinical studies. An important practical advantage of ERU is that in prevalence matched populations, the measure can be estimated without an independent estimate of disease prevalence. Thus estimating ERU does not require a study with long-term follow-up to find cases of missed disease. The approach is applicable to any comparative screening study that reports results in terms of recall and detection rates, although the authors focus exclusively on screening mammography in this work. They derive the ERU from the definition of utility given in classical treatments of signal detection theory. They also investigate reasonable values of relative utility in screening mammography for use in interpreting ERU using data from a large clinical study. As examples of application of ERU, they reanalyze 2 recently published reports using recall and detection rates in screening mammography.

Evaluation of Imputation Methods in Ovarian Tumor Diagnostic Models Using Generalized Linear Models and Support Vector Machines

Dimou, I., Van Calster, B., Van Huffel, S., Timmerman, D., Zervakis, M. — Tue, 26 Jan 2010 10:36:33 PST

Neglecting missing values in diagnostic models can result in unreliable and suboptimal performance on new data. In this study, the authors imputed missing values for the CA-125 tumor marker in a large data set of ovarian tumors that was used to develop models for predicting malignancy. Four imputation techniques were applied: regression imputation, expectation-maximization, data augmentation, and hotdeck. Models using the imputed data sets were compared with models without CA-125 to investigate the important clinical issue concerning the necessity of CA-125 information for diagnostic models and with models using only complete cases to investigate differences between imputation and complete case strategies for missing values. The models are based on Bayesian generalized linear models (GLMs) and Bayesian least squares support vector machines. Results indicate that the use of CA-125 resulted in small, clinically nonsignificant increases in the AUC of diagnostic models. Minor differences between imputation methods were observed, and imputing CA-125 resulted in minor differences in the AUC compared with complete case analysis (CCA). However, GLM parameter estimates of predictor variables often differed between CCA and models based on imputation. The authors conclude that CA-125 is not indispensable in diagnostic models for ovarian tumors and that missing value imputation is preferred over CCA.

A Maximum Likelihood Estimator of a Markov Model for Disease Activity in Crohn's Disease and Ulcerative Colitis for Annually Aggregated Partial Observations

Tue, 26 Jan 2010 10:36:33 PST

Crohn’s disease (CD) and ulcerative colitis (UC) are chronic inflammatory bowel diseases that have a remitting, relapsing nature. During relapse, they are treated with drugs and surgery. The present study was based on individual data from patients diagnosed with CD or UC at Herlev University Hospital, Copenhagen, Denmark, during 1991 to 1993. The data were aggregated over calendar years; for each year, the number of relapses and the number of surgical operations were recorded. Our aim was to estimate Markov models for disease activity in CD and UC, in terms of relapse and remission, with a cycle length of 1 month. The purpose of these models was to enable evaluation of interventions that would shorten relapses or postpone future relapses. An exact maximum likelihood estimator was developed that disaggregates the yearly observations into monthly transition probabilities between remission and relapse. These probabilities were allowed to be dependent on the time since start of relapse and on the time since start of remission, respectively. The estimator, initially slow, was successfully optimized to shorten the execution time. The estimated disease activity model for CD fits well to observed data and has good face validity. The disease activity model is less suitable for UC due to its transient nature through the presence of curative surgery.

Comment on Petrou and Kupek: Use Unadjusted HUI3 Scores, Not Adjusted Disutilities

Russell, L. B. — Tue, 26 Jan 2010 10:36:33 PST

Author Index for the 12th Biennial European Meeting of the Society for Medical Decision Making Abstracts

Tue, 26 Jan 2010 10:36:33 PST