Medical Decision Making最新文献

BackgroundWhile structured expert elicitation (SEE) is gaining traction in health technology assessment in situations in which data are scarce, its application in practice remains limited. Co-designing a practical and fit-for-purpose SEE with experts could enhance its acceptability and feasibility in clinical research.ObjectivesAn SEE was co-designed with clinicians to elicit expert opinions on 3 uncertain quantities of interest (QoIs) for a decision-analytic model in exercise oncology.MethodsA series of co-design meetings was convened to design 6 elicitation stages. Individual elicitation was conducted using the variable interval method (VIM), via videoconferencing. Linear pooling was adopted to generate group estimates. Semi-structured interviews were conducted after the elicitation exercise to gather the experts' first-hand experience of the elicitation process and to identify areas for improvement. Qualitative data were transcribed and content analyzed.ResultsTwelve experts participated in the co-designed SEE. Three beta distributions were derived and estimated from the experts' responses: the relative risk reduction of cardiovascular events of exercise for women who survived early-stage endometrial cancer (Mean: 0.362, SD: 0.15), the probability that a clinician would refer a patient to the exercise program (Mean: 0.457, SD: 0.218), and the probability that a cancer patient would use such a health service upon referral (Mean: 0.446, SD: 0.203). Most of the experts' first-hand experience of the co-designed SEE was positive. The qualitative feedback highlighted critical aspects of the elicitation process that should be designed and executed with caution when targeting clinicians with no prior experience of SEE.ConclusionsThis is the first expert elicitation conducted in exercise oncology. Engaging diverse stakeholders through co-design meetings and incorporating qualitative feedback proved effective and practical in introducing expert elicitation into clinical research.HighlightsRecent SEE guidelines aim to facilitate the conduct of expert elicitation in model-based economic evaluation, but its application in practice remains limited.Engaging experts in the design of SEE could enhance its acceptability and feasibility in clinical research.This is the first co-designed expert elicitation involving clinicians in the field of exercise oncology.This practical approach to conducting SEE could promote a wider adoption to inform health care policy decisions when the evidence is lacking or uncertain.

BackgroundIntensive care unit (ICU) hospitalizations of very old patients with acute respiratory infection have risen. The decision-making process for ICU admission is multifaceted, and the prediction of long-term survival outcome is an important component. We hypothesized that data-driven algorithms could build long-term prediction by examining massive real-life data. Our objective was to assess machine learning (ML) algorithms to predict the 1-y survival of very old patients with severe respiratory infections.MethodsA national 2011-2020 study of ICU patients ≥80 y with respiratory infection was carried out, using French hospital discharge databases. Data for the training cohort were collected from 2013 to 2016 to build the models, and the data of patients extracted in 2017 were used for external validation. Our proposed models were developed using random forest, logistic regression (LR), and XGBoost. The optimal model was selected based on its accuracy, sensitivity, specificity, Matthews coefficient correlation (MCC), receiver-operating characteristic curve (AUROC), and decision curve analysis (DCA). The local interpretable model-agnostic explanation (LIME) algorithm was used to analyze the contribution of individual features.ResultsA total of 24,270 very old patients were hospitalized in the ICU for respiratory infection (2013-2017) with a known vital status at 1 y. The 1-y survival rate was 41.3% (median survival: 3 mo [2.7-3.3]). Of the 3 ML models tested, LR exhibited promising performance with an accuracy, sensitivity, specificity, MCC, and AUROC (95% confidence interval) of 0.65, 0.76, 0.60, 0.27, and 0.70 (0.69-0.72), respectively. LR achieved an AUROC of 0.70 (0.68-0.71) in external validation by temporal splitting. LR demonstrated higher net benefits across a range of threshold probability values in DCA. The LIME algorithm identified the 10 most influential features at an individual scale.ConclusionsWe demonstrated that a ML model has the potential to predict long-term outcomes for very old patients with acute respiratory infections. As a proof of concept, we proposed a program that acts as an "explainer" for the ML model. This work represents a step forward in translating ML models into practical, transparent, and reliable clinical tools to support medical decision making.HighlightsThe decision to admit a very old patient to the ICU is one of the most complex challenges faced by intensivists, often relying on subjective judgment.In this study, we evaluated the efficacy of machine learning algorithms in predicting the 1-y survival rate of critically ill very old patients (≥80 y) with severe respiratory infections, using data available prior to the admission decision.Our findings demonstrate that machine learning can effectively predict long-term outcomes in very old patients. We used an innovative approach that aims to support medical decision making about admission in ICU.

BackgroundThis study explores patient and clinician perceptions of a patient decision aid, focusing on 2 features that are often absent: a health-literate approach (e.g., using plain language, encouraging question asking) and a tool that explicitly shows how treatment options align with patient values. The aim was to gather qualitative feedback from patients and clinicians to better understand how such features might be useful in guiding future decision aid development.MethodsWe present a secondary analysis of data collected during the development of a decision aid for patients considering surgery for sciatica (20 patients with sciatica or low-back pain; 20 clinicians). Patient and clinician feedback on the design was collected via semi-structured interviews with a think-aloud protocol. Transcripts were analyzed using framework analysis.ResultsTheme 1 explored designs that reinforced key messages about personal autonomy, including an interactive values-clarification tool. Theme 2 explored how participants valued encouragement and scaffolding to ask questions. Theme 3 described how patients preferred information they felt was complete, balanced, and understandable.LimitationsFurther experimental and observational research is needed to quantitatively evaluate these decision aid features including evaluation among patients with and without low health literacy.ConclusionsA health-literate approach to decision aid design and embedding an interactive values-clarification tool may be useful strategies for increasing patient capacity to engage in key aspects of shared decision making. These features may support patients in developing an understanding of personal autonomy in the choice at hand and confidence to ask questions.ImplicationsFindings presented here were specific to the clinical context but provide generalizable practical insights for decision aid developers. This study provides insight into potential future areas of research for decision aid design.HighlightsThis qualitative study explored clinician and patient perceptions of health literacy features and an interactive values-clarification task within a decision aid for patients considering surgery for sciatica.The first theme described how patients and clinicians appreciated sections of the decision aid that reinforced the importance of personal choice. Patients and clinicians thought the interactive values-clarification task would help patients reflect on their values and support shared decision-making discussions.The second theme described how patients and clinicians appreciated strategies to encourage patients to ask questions of the surgeon.The third theme described patients' preference for information that they felt was complete, balanced, and understandable.

PurposeThe 2023 American College of Physicians (ACP) guidelines for colorectal cancer (CRC) screening are at odds with the United States Preventive Task Force (USPSTF) guidelines, with the former recommending screening starting at age 50 y and the latter at age 45 y. This article "stress tests" CRC colonoscopy screening strategies to investigate their robustness to uncertainties stemming from the natural history of disease and sensitivity of colonoscopy.MethodsThis study uses the CRC-SPIN microsimulation model to project the life-years gained (LYG) under several colonoscopy CRC screening strategies. The model was extended to include birth cohort effects on adenoma risk. We estimated natural history parameters under 2 different assumptions about the youngest age of adenoma initiation. For each, we generated 500 parameter sets to reflect uncertainty in the natural history parameters. We simulated 26 colonoscopy screening strategies and examined 4 different colonoscopy sensitivity assumptions, encompassing the range of sensitivities consistent with prior tandem colonoscopy studies. Across this set of scenarios, we identify efficient screening strategies and report posterior credible intervals for benefits of screening (LYG), burden (number of colonoscopies), and incremental burden-effectiveness ratios.ResultsProjected absolute screening benefits varied widely based on assumptions, but strategies starting at age 45 y were consistently in the efficiency frontier. Strategies in which screening starts at age 50 y with 10-y intervals were never efficient, saving fewer life-years than starting screening at age 45 y and performing colonoscopies every 15 y while requiring more colonoscopies per person.ConclusionsDecennial colonoscopy screening initiation at age 45 y remained a robust recommendation. Colonoscopy screening with a 10-y interval starting at age 50 y did not result in an efficient use of colonoscopies in any of the scenarios evaluated.HighlightsColorectal cancer colonoscopy screening strategies initiated at age 45 y were projected to yield more life-years gained while requiring the least number of colonoscopies across different model assumptions about disease natural history and colonoscopy sensitivity.Colonoscopy screening starting at age 50 y with a 10-y interval consistently underperformed strategies that started at age 45 y.

Background(1) To estimate the prevalence of decision regret in chronic pain care, and (2) to identify factors associated with decision regret.DesignWe conducted a pan-Canadian cross-sectional online survey and reported the results following the Checklist for Reporting of Survey Studies guidelines. We recruited a sample of adults experiencing chronic noncancer pain. We used a stratified proportional random sampling based on the population and chronic pain prevalence of each province. We measured decision regret with the Decision Regret Scale (DRS) and decisional needs with the Ottawa Decision Support Framework. We performed descriptive analysis to estimate the prevalence and level of decision regret and multilevel multivariable regression analysis to identify factors associated with regret according to the STRengthening Analytical Thinking for Observational Studies recommendations.ResultsWe surveyed 1,649 people living with chronic pain, and 1,373 reported a most difficult decision from the 10 prespecified ones, enabling the collection of a DRS score. On a scale ranging from 0 to 100 where 1 reflects the presence of decision regret and 25 constitutes important decision regret, the mean DRS score in our sample was 28.8 (s = 19.6). Eighty-four percent of respondents experienced some decision regret and 50% at an important level. We identified 15 factors associated with decision regret, including 4 personal and 9 decision-making characteristics, and 2 consequences of the chosen option. Respondents with low education level and higher decisional conflict experienced more decision regret when the decision was deemed difficult.ConclusionsThis pan-Canadian survey highlighted a high prevalence and level of decision regret associated with difficult decisions for pain care. Decision making in pain care could be enhanced by addressing factors that contribute to decision regret.HighlightsWe conducted an online pan-Canadian survey and collected responses from a wide diversity of people living with chronic pain.More than 84% of respondents experienced decision regret and approximately 50% at an important level.We identified 15 factors associated with decision regret, including 4 personal and 9 decision-making characteristics, and 2 consequences of the chosen option.Our pan-Canadian survey reveals an urgent need of a shared decision-making approach in chronic pain care that can be potentiated by targeting multiple factors associated with decision regret.

BackgroundCancer diagnosis causes emotional distress, which can influence patients' treatment choice. This study aimed to investigate the effect of increased emotionally supportive surgeon communication in a virtual setting on treatment preference for thyroid cancer.DesignThis randomized trial (NCT05132478), conducted from November 2021 to February 2023, enrolled adults with ≤4-cm thyroid nodules not requiring surgery. Participants were randomized 1:1 to watch a virtual clinic visit depicting a patient-surgeon treatment discussion for low-risk thyroid cancer. Control and intervention videos were identical except for added emotionally supportive communication in the intervention. The primary outcome was treatment preference for total thyroidectomy or lobectomy. Secondary outcomes were perceived physician empathy, physician trust, decisional confidence, and disease-specific knowledge. An intention-to-treat analysis was performed using conditional regression to account for stratification by sex. Qualitative content analysis evaluated participants' open-ended responses about treatment choice and surgeon communication.ResultsOf 208 eligible patients, 118 (56.7%) participated. Participants were 85.6% female and 88.1% White. Overall, 89.0% (n = 105) of participants preferred lobectomy, which was similar between the intervention and control groups (90.0% v. 87.9%, respectively, P = 0.72). Compared with control, participants who viewed the consultation with enhanced communication perceived higher levels of physician empathy (34.5 ± 5.8 v. 25.9 ± 9.1, P < 0.001) and reported increased trust in the physician (12.0 ± 2.6 v. 10.4 ± 3.1, P < 0.001). The groups were similar in decisional confidence (7.6 ± 2.1 v. 7.7 ± 1.9, P = 0.74) and disease-specific knowledge. Prominent qualitative themes among participants choosing thyroid lobectomy included desire to avoid daily thyroid hormone (n = 53) and concerns about surgical complications (n = 25).ConclusionsIn this randomized controlled study, a significant proportion of participants preferred thyroid lobectomy if diagnosed with low-risk thyroid cancer. Participants perceived increased empathy when provided even in the virtual setting, which was associated with increased trust in the physician.HighlightsIn this single-site, randomized controlled trial, enhanced emotionally supportive surgeon communication had no effect on hypothetical treatment preference for low-risk thyroid cancer.Participants who experienced enhanced emotionally supportive surgeon communication perceived higher physician empathy and reported greater trust in the physician.The incorporation of empathetic communication during surgical consultation for low-risk thyroid cancer promotes patient trust and perception of empathy.

The effective sample size (ESS) measures the informational value of a probability distribution in terms of an equivalent number of study participants. The ESS plays a crucial role in estimating the expected value of sample information (EVSI) through the Gaussian approximation approach. Despite the significance of ESS, except for a limited number of scenarios, existing ESS estimation methods within the Gaussian approximation framework are either computationally expensive or potentially inaccurate. To address these limitations, we propose a novel approach that estimates the ESS using the summary statistics of generated datasets and nonparametric regression methods. The simulation experiments suggest that the proposed method provides accurate ESS estimates at a low computational cost, making it an efficient and practical way to quantify the information contained in the probability distribution of a parameter. Overall, determining the ESS can help analysts understand the uncertainty levels in complex prior distributions in the probability analyses of decision models and perform efficient EVSI calculations.HighlightsEffective sample size (ESS) quantifies the informational value of probability distributions, essential for calculating the expected value of sample information (EVSI) using the Gaussian approximation approach. However, current ESS estimation methods are limited by high computational demands and potential inaccuracies.We propose a novel ESS estimation method that uses summary statistics and nonparametric regression models to efficiently and accurately estimate ESS.The effectiveness and accuracy of our method are validated through simulations, demonstrating significant improvements in computational efficiency and estimation accuracy.

BackgroundIn cases of surgical necrotizing enterocolitis (NEC), the choice between laparotomy (LAP) or comfort care (CC) presents a complex, ethical dilemma. A behavioral artificial intelligence technology (BAIT) decision aid was trained on expert knowledge, providing an output as "x percentage of experts advise laparotomy for this patient." This retrospective study aims to compare this output to clinical practice.DesignVariables required for the decision aid were collected of preterm patients with NEC for whom the decision of LAP or CC had been made. These data were used in 2 BAIT model versions: one center specific, built on the input of experts from the same center as the patients, and a nationwide version, incorporating the input of additional experts. The Mann-Whitney U test compared the model output for the 2 groups (LAP/CC). In addition, model output was classified as advice for LAP or CC, after which the chi-square test assessed correspondence with observed decisions.ResultsForty patients were included in the study (20 LAP). Model output (x percentage of experts advising LAP) was higher in the LAP group than in the CC group (median 95.1% v. 46.1% in the center-specific version and 97.3% v. 67.5% in the nationwide version, both P < 0.001). With an accuracy of 85.0% by the center-specific and 80.0% by the nationwide version, both showed significant correspondence with observed decisions (P < 0.001).LimitationsWe are merely examining a proof of concept of the decision aid using a small number of participants from 1 center.ConclusionsThis retrospective study demonstrates that treatment choices by artificial intelligence align with clinical practice in at least 80% of cases.ImplicationsFollowing prospective validation and ongoing refinements, the decision aid may offer valuable support to practitioners in future NEC cases.HighlightsThis study assesses the output of behavioral artificial intelligence technology in deciding between laparotomy and comfort care in surgical necrotizing enterocolitis.The model output aligns with clinical practice in at least 80% of patient cases.Following prospective validation, the decision aid may offer valuable support to physicians working at the neonatal intensive care unit.

BackgroundHealth-state utility values (HSUVs) are key inputs into cost-utility analyses. There is debate over whether they are best derived from the community or patients, with concerns raised that community-derived preferences may devalue benefits to ill, elderly, or disabled individuals. This tutorial compares the effects of using patient-derived HSUVs versus community-derived HSUVs on incremental cost-effectiveness ratios (ICERs) and shows their implications for policy.DesignWe review published studies that compared HSUVs derived from patients and the community. We then present equations for the gains in quality-adjusted life-years (QALYs) that would be estimated for an intervention using patient versus community HSUVs and discuss the implications of those QALY gains. We present a numerical example as another way of showing how ICERs change when using patient versus community HSUVs.ResultsPatient HSUVs are generally higher than community HSUVs for severe health states. When an intervention reduces mortality, patient ratings yield more favorable ICERs than do community ratings. However, when the intervention reduces morbidity, patient ratings yield less favorable ICERs. For interventions that reduce both morbidity and mortality, the effect on ICERs of patient versus community HSUVs depends on the relative contribution of each to the resulting QALYs.ConclusionsThe use of patient HSUVs does not consistently favor treatments directed at those patients. Rather, the effect depends on whether the intervention reduces mortality, morbidity, or both. Since most interventions do both, using patient HSUVs has mixed implications for promoting investments for people with illness and disabilities. A nuanced discussion of these issues is necessary to ensure that policy matches the intent of the decision makers.HighlightsThe debate about whether health state utility values (HSUVs) are best derived from patients or the community rests in part on the presumption that using community values devalues interventions for disabled persons or those with chronic diseases.However, we show why the effect of using patient HSUVs depends on whether the intervention in question primarily reduces mortality or morbidity or has substantial effects on both.If the intervention reduces mortality, using patient HSUVs will make the intervention appear more cost-effective than using community HSUVs, but if it reduces morbidity, using patient HSUVs will make the intervention appear less cost-effective.If the intervention reduces both morbidity and mortality, a common situation, the effect of patient versus community HSUVs depends on the relative magnitudes of the gains in quality and length of life.

ObjectivesDiscrete choice experiments (DCEs) as a valuation method require preferences to be anchored on the quality-adjusted life-year scale, usually through tasks involving choices between immediate death and various impaired health states or between health states with varying durations of life. We sought to determine which anchoring approach aligns best with the composite time tradeoff (cTTO) method, with a view to informing a valuation protocol that uses DCEs in place of the cTTO.MethodsA total of 970 respondents from Trinidad and Tobago completed a DCE with duration survey. Tasks involved choosing between 2 lives with identical durations, followed by a third option, representing either full health for a number of years or immediate death. Data were analyzed using mixed logit models, both with and without exponential discounting for time preferences.ResultsAssuming linear time preferences, the estimated utility of immediate death was -2.1 (95% credible interval [CrI] -3.2 to -1.2) versus -0.28 (95% CrI -0.47, -0.10) when allowing for nonlinear time preferences. Under linear time preferences, the predicted health-state values anchored on duration had range (-1.03, 1) versus (0.34, 1) when anchored on immediate death. The ranges under nonlinear time preferences were (-0.54, 1) versus (-0.22, 1). The estimated discount parameter was 23% (95% CrI 22% to 25%).ConclusionsThe nonzero discount parameter indicates that time preferences were nonlinear. Nonlinear time preferences anchored on duration provided the closest match to the benchmark EQ-VT cTTO values in Trinidad and Tobago, whose range was (-0.6, 1). Thus, DCE with duration can provide similar values to cTTO provided that nonlinear time preferences are accounted for and anchoring is based on duration.HighlightsTime preferences for health states in Trinidad and Tobago were nonlinear.In discrete choice tasks, we show that immediate death has a utility less than zero.DCE utilities under nonlinear time preferences with anchoring on duration agreed well with cTTO utilities.