Medical Decision Making

In this keynote address delivered at the 41st Annual North American Meeting of the Society for Medical Decision Making, I discuss the psychology behind valuing human lives. Research confirms what we experience in our daily lives. We are inconsistent and sometimes incoherent in our valuation of human life. We value individual lives greatly, but these lives lose their value when they become part of a larger crisis. As a result, we do too little to protect human lives in the face of catastrophic threats from violence, natural disasters, and other causes. In medicine, this may pose difficult choices when treating individual patients with expensive therapies that keep hope alive but are not cost-effective for the population, for example, with end of life. Lifesaving judgments and decisions are highly context-dependent, subject to many forms of response mode and framing effects and affective biases. This has implications for risk communication and the concept of shared decision making. Slower, more introspective decision making may reduce some of the biases associated with fast, intuitive decisions. But slow thinking can also introduce serious biases. Understanding the strengths and weaknesses of fast and slow thinking is a necessary first step toward valuing lives humanely and improving decisions.

Background. Informal caregivers are regularly faced with difficult housing decisions for older adults with cognitive impairment. They often regret the decision they made. We aimed to identify factors associated with decision regret among informal caregivers engaging in housing decisions for cognitively impaired older adults. Methods. We performed a secondary analysis of cross-sectional data collected from a cluster-randomized trial. Eligible participants were informal caregivers involved in making housing decisions for cognitively impaired older adults. Decision regret was assessed after caregivers’ enrollment in the study using the Decision Regret Scale (DRS), scored from 0 to 100. We used a conceptual framework of potential predictors of regret to identify independent variables. We performed multilevel analyses using a mixed linear model by estimating fixed effects (β) and 95% confidence intervals (CIs). Results. The mean (SD) DRS score of 296 informal caregivers (mean [SD] age, 62 [12] years) was 12.4 (18.4). Factors associated with less decision regret were having a college degree compared to primary education (β [95% CI]: –11.14 [–18.36, –3.92]), being married compared to being single (–5.60 [–10.05, –1.15]), informal caregivers’ perception that a joint process occurred (–0.14 [–0.25, –0.02]), and older adults’ not having a specific housing preference compared to preferring to stay at home (–4.13 [–7.40, –0.86]). Factors associated with more decision regret were being retired compared to being a homemaker (7.74 [1.32, 14.16]), higher burden of care (0.14 [0.05, 0.22]), and higher decisional conflict (0.51 [0.34, 0.67]). Limitations. Our analysis may not illustrate all predictors of decision regret among informal caregivers. Conclusions. Our findings will allow risk-mitigation strategies for informal caregivers at risk of experiencing regret.

Background. Inappropriate antibiotic prescribing can lead to antimicrobial resistance and drug side effects. Tools that assist general practitioners (GPs) in prescribing decisions may help to optimize prescribing. The aim of this study was to explore the use, acceptability, and feasibility of a clinical decision support (CDS) tool that incorporates evidence-based guidelines and consumer information that integrates with the electronic medical record (EMR). Methods. Eight GPs completed an interview and brief survey after participating in 2 simulated consultations. The survey consisted of demographic questions, perception of realism and representativeness of consultations, Post-Study System Usability Questionnaire, and System Usability Scale. Qualitative data were analyzed using framework analysis. Video data were reviewed, with length of consultation and time spent using the CDS tool documented. Results. Survey responses indicated that all GPs thought the consultations were “real” and representative of real-life consultations; 7 of 8 GPs were satisfied with usability of the tool. Key qualitative findings included that the tool assisted with clinical decision making and informed appropriate antibiotic prescribing. Accessibility and ease of use, including content (guideline and patient education resources), layout, and format, were key factors that determined whether GPs said that they would access the tool in everyday practice. Integration of the tool at multiple sites within the EMR facilitated access to guidelines and assisted in ensuring that the tool fit the clinical workflow. Conclusion. Our CDS tool was acceptable to GPs. Key features required for the tool were easy navigation, clear and useful guideline content, ability to fit into the clinical workflow, and incorporation into the EMR. Piloting of the tool in general practices to assess the impact and feasibility of use in real-world consultations will now be undertaken.

Background. Reviewing drugs to determine coverage or reimbursement level is a complex process that involves significant time and expertise. Review boards gather evidence from the submission provided, input from clinicians and patients, and results of clinical and economic reviews. This information consists of assessments on multiple criteria that often conflict with one another. Multiple-criteria decision analysis (MCDA) includes methods to address complex decision making problems with conflicting objectives and criteria. We propose an MCDA approach that infers a utility model based on reviews of previously submitted drugs. Methods. We use a recent extension of the UTilitiés Additives DIScriminantes approach, UTADIS^GMS. This disaggregation approach deconstructs a portfolio of elements such as a set of drugs that have been reviewed and for which a decision has been made. It derives global and marginal utility functions that are consistent with the preferences exhibited by the review boards in their recommendations. We apply the method to oncology drugs reviewed in Canada between 2011 and 2017. We also illustrate how to conduct scenario analyses and predict the coverage decisions for new drugs. Results. Applying the method yields a utility value for each submission along with a set of thresholds that partition the utility values based on the submission outcomes. Scenario analyses illustrate the predictive ability of the method. Conclusion. Preference disaggregation is an indirect way of eliciting an additive global utility value function. It requires less of a cognitive effort from the decision making bodies because it infers preferences from the data rather than relying on direct assessments of model parameters. We illustrate how it can be applied to validate existing decisions and to predict the recommendation of a new drug.

Well-established methods of economic evaluation are used in many countries to inform decisions about the funding of new medical interventions. To guide such decisions, it is important to consider what health gains would be expected from the same level of investment elsewhere in the health care system. Recent research in the United Kingdom has evaluated the evidence available and the methods required to estimate the health effects of changes in health care expenditure within the National Health Service. Because of the absence of sufficiently broad-ranging data, assumptions were required in the previously mentioned work to estimate health effects in terms of a broader measure of health (quality-adjusted life-years), which is more relevant for policy. These assumptions constitute important sources of uncertainty. This work presents an application of the structured elicitation of the judgments of key individuals about these uncertain quantities. This article describes the design and conduct of the exercise, including the quantities elicited, the individual (rather than consensus) approach used, how uncertainty in knowledge was elicited (mode and bounds of an 80% credible interval), and methods to generate group estimates. It also reports on a successful application involving 28 clinical experts and 25 individuals with policy responsibilities. Although, as expected, most experts found replying to the questions challenging, they were able to express their beliefs quantitatively. Consistent across the uncertainties elicited, experts’ judgments suggest that the quality-adjusted life-year (QALY) impacts of changes in expenditure from earlier work using assumptions are likely to have been underestimated and the “central” estimate of health opportunity cost from that work (£12,936 per QALY) to have been overestimated.

Introduction. Patient-level simulation models facilitate extrapolation of clinical trial data while allowing for heterogeneity, prior history, and nonlinearity. However, combining different types of uncertainty around within-trial and extrapolated results remains challenging. Methods. We tested 4 methods to combine parameter uncertainty (around the regression coefficients used to predict future events) with sampling uncertainty (uncertainty around mean risk factors within the finite sample whose outcomes are being predicted and the effect of treatment on these risk factors). We compared these 4 methods using a simulation study based on an economic evaluation extrapolating the AFORRD randomized controlled trial using the UK Prospective Diabetes Study Outcomes Model version 2. This established type 2 diabetes model predicts patient-level health outcomes and costs. Results. The 95% confidence intervals around life years gained gave 25% coverage when sampling uncertainty was excluded (i.e., 25% of 95% confidence intervals contained the “true” value). Allowing for sampling uncertainty as well as parameter uncertainty widened confidence intervals by 6.3-fold and gave 96.3% coverage. Methods adjusting for baseline risk factors that combine sampling and parameter uncertainty overcame the bias that can result from between-group baseline imbalance and gave confidence intervals around 50% wider than those just considering parameter uncertainty, with 99.8% coverage. Conclusions. Analyses extrapolating data for individual trial participants should include both sampling uncertainty and parameter uncertainty and should adjust for any imbalance in baseline covariates.

Background. To interpret cervical cancer screening model results, we need to understand the influence of model structure and assumptions on cancer incidence and mortality predictions. Cervical cancer cases and deaths following screening can be attributed to 1) (precancerous or cancerous) disease that occurred after screening, 2) disease that was present but not screen detected, or 3) disease that was screen detected but not successfully treated. We examined the relative contributions of each of these using 4 Cancer Intervention and Surveillance Modeling Network (CISNET) models. Methods. The maximum clinical incidence reduction (MCLIR) method compares changes in the number of clinically detected cervical cancers and mortality among 4 scenarios: 1) no screening, 2) one-time perfect screening at age 45 that detects all existing disease and delivers perfect (i.e., 100% effective) treatment of all screen-detected disease, 3) one-time realistic-sensitivity cytological screening and perfect treatment of all screen-detected disease, and 4) one-time realistic-sensitivity cytological screening and realistic-effectiveness treatment of all screen-detected disease. Results. Predicted incidence reductions ranged from 55% to 74%, and mortality reduction ranged from 56% to 62% within 15 years of follow-up for scenario 4 across models. The proportion of deaths due to disease not detected by screening differed across the models (21%–35%), as did the failure of treatment (8%–16%) and disease occurring after screening (from 1%–6%). Conclusions. The MCLIR approach aids in the interpretation of variability across model results. We showed that the reasons why screening failed to prevent cancers and deaths differed between the models. This likely reflects uncertainty about unobservable model inputs and structures; the impact of this uncertainty on policy conclusions should be examined via comparing findings from different well-calibrated and validated model platforms.

Introduction. The EQ-5D-5L valuation protocol contains both time tradeoff (TTO) tasks and discrete choice experiments (DCE), raising the question of how to best use these in creating a value set. The hybrid model, which combines TTO and DCE data, has emerged as a commonly used approach. However, this model assumes independence among responses from the same individual, a linear relationship between TTO and DCE utilities, and, in many implementations, homoscedastic residuals. The aims of this study are to examine alternatives to these assumptions and determine the impact of misspecification on value sets. Methods. We performed a simulation study, parameterized using the US EQ-5D-5L valuation study, to assess the impact of model misspecification. We simulated TTO and DCE data with nonlinear relationships between TTO and DCE utilities, heteroscedastic errors, and correlated responses. Simulated data were analyzed using hybrid models with and without heteroscedasticity, Tobit models with and without heteroscedasticity, a latent class model, and a mixed model. Results. Mean absolute errors (MAEs) for correctly specified models were <0.05, whereas models that incorrectly assumed a linear relationship between TTO and DCE utilities or homoscedasticity of TTO responses featured states with an MAE >0.1. When a linear relationship between TTO and DCE utilities held, using both TTO and DCE data under correct specification yielded smaller MAEs compared with using TTO data alone but yielded larger MAEs when a linear relationship did not hold. Mistakenly assuming homoscedasticity led to increased MAEs, whereas ignoring dependence did not. Conclusions. Because heteroscedasticity in TTO utilities and nonlinear associations between DCE and TTO utilities have been noted, we recommend careful assessment of scedasticity and linearity to ascertain the suitability of a hybrid model.

Background. The ICECAP-O and the ICECAP-A are validated capability well-being instruments. To be used in economic evaluations, multidimensional instruments require weighting of the distinguished well-being states. These weights are usually obtained through ex ante preference elicitation (i.e., decision utility) but could also be based on experienced utility. Objective. This article describes the development of value sets for ICECAP-O and ICECAP-A based on experienced utility and compares them with current decision utility weights. Methods. Data from 2 cross-sectional samples corresponding to the target groups of ICECAP-O and ICECAP-A were used in 2 separate analyses. The utility impacts of ICECAP-O and ICECAP-A levels were assessed through regression models using a composite measure of subjective well-being as a proxy for experienced utility. The observed utility impacts were rescaled to match the 0 to 1 range of the existing value set. Results. The calculated experienced utility values were similar to the decision utility weights for some of the ICECAP dimensions but deviated for others. The largest differences were found for weights of the ICECAP-O dimension enjoyment and the ICECAP-A dimensions attachment and autonomy. Conclusions. The results suggest a different weighting of ICECAP-O and ICECAP-A levels if experienced utility is used instead of decision utility.

Background. The English National Health Service publishes hospital performance indicators based on average postoperative EQ-5D index scores after hip replacement surgery to inform prospective patients’ choices of hospital. Unidimensional index scores are derived from multidimensional health-related quality-of-life data using preference weights estimated from a sample of the UK general population. This raises normative concerns if general population preferences differ from those of the patients who are to be informed. This study explores how the source of valuation affects hospital performance estimates. Methods. Four different value sets reflecting source of valuation (general population v. patients), valuation technique (visual analog scale [VAS] v. time tradeoff [TTO]), and experience with health states (currently experienced vs. experimentally estimated) were used to derive and compare performance estimates for 243 hospitals. Two value sets were newly estimated from EQ-5D-3L data on 122,921 hip replacement patients and 3381 members of the UK general public. Changes in hospital ranking (nationally) and performance outlier status (nationally; among patients’ 5 closest hospitals) were compared across valuations. Results. National rankings were stable under different valuations (rank correlations >0.92). Twenty-three (9.5%) hospitals changed outlier status when using patient VAS valuations instead of general population TTO valuations, the current approach. Outlier status also changed substantially at the local level. This was explained mostly by the valuation technique, not the source of valuations or experience with the health states. Limitations. No patient TTO valuations were available. The effect of value set characteristics could be established only through indirect comparisons. Conclusion. Different value sets may lead to prospective patients choosing different hospitals. Normative concerns about the use of general population valuations are not supported by empirical evidence based on VAS valuations.

Background. The Ottawa Decision Support Framework (ODSF) has guided the development of patient decision aids (PtDAs) for 20 years and needs updating across a range of decisions and hypothesized outcomes. Purpose. To determine the effectiveness of ODSF-developed PtDAs on hypothesized outcomes and to recommend framework changes. Data Source. A subanalysis of randomized controlled trials included in the 2017 Cochrane review of PtDAs comparing PtDAs to usual care in adults considering health treatment or screening decisions (searched to 2015). Study Selection. Trials in the original review that evaluated ODSF-developed PtDAs. Data Synthesis. Meta-analyses of ODSF outcomes with similar measurements and descriptions of other reported outcomes. Results. Of 105 trials, 24 evaluated ODSF-developed PtDAs. Compared with usual care, ODSF PtDAs improved knowledge (mean difference [MD] 13.85; 95% confidence interval [CI] 10.32−17.37; 14 trials), increased accurate risk perceptions (risk ratio [RR] 2.41; 95% CI 1.66−3.48; 7 trials), and increased congruence between informed values and chosen options (RR 1.32; 95% CI 1.09−1.59; 4 trials). They reduced perceived decisional needs as measured using the Decisional Conflict Scale (MD −5.92; 95% CI −8.58 to −3.26; 15 trials) and the proportion of undecided patients (RR 0.65; 95% CI 0.50−0.83; 13 trials). Non-ODSF PtDAs, designed with or without a specific framework, also outperformed usual care. Few ODSF trials measured secondary outcomes. Limitations. The included trials had heterogeneity. Conclusion. ODSF PtDAs address decisional needs and improve decision quality; the best indicator of addressing perceived uncertainty is “proportion undecided.” Secondary ODSF outcomes should be reduced to adherence to one’s chosen option and use/costs of health services, which warrant further research.

Background. Having dependent children may affect cancer treatment decisions. We sought to describe women’s surgery and chemotherapy decisions in nonmetastatic breast cancer by parental status. Methods. We conducted a secondary analysis of the 2015 cross-sectional Share Thoughts on Breast Cancer Study, conducted in 7 Midwestern states in the United States, restricted to women of prime parenting age (aged 20–50 years) who consented to the use of their medical records (N = 225). We examined treatment decisions using data visualization and logistic regression (adjusted for age, stage, family history of breast cancer, income, education, race, health insurance, and partner status). Results. Women with dependent children received bilateral mastectomy more often than women without dependent children (adjusted odds ratio 3.09, 95% confidence interval 1.44–6.62).We found no differences in the receipt of chemotherapy by parental status. Women reported more active roles in surgery than in chemotherapy decision making. Conclusions. As a likely factor in cancer treatment decisions, parental status should be addressed in clinical practice and research. Future research should assess patients’ sense of ownership in treatment decision making by treatment type.

The 10-item validated Medical Maximizer-Minimizer Scale (MMS-10) assesses patients’ preferences for aggressive v. more passive approaches to health care. However, because many research or clinical situations do not allow for use of a 10-item scale, we developed a single-item maximizer-minimizer elicitation question (the MM1) based on our experiences describing the construct to patient groups, clinical researchers, and the public. In 2 large samples of US adults (N = 368 and N = 814), the correlation between MMS-10 scores and the MM1 was .52 and .60, respectively. Both measures were robust predictors of medical preferences in a set of 12 hypothetical scenarios, and both had strong (and roughly equivalent) associations with 7 self-report measures of health care utilization. Our results demonstrate that the MM1 is a valid, brief elicitation of maximizing-minimizing preferences that can be used in clinical or research contexts where the 10-item scale is infeasible.