Medical Decision Making最新文献

PurposeTo summarize the degree to which evidence from our recent Making Numbers Meaningful (MNM) systematic review of the effects of data presentation format on communication of health numbers supports recommendations from the 2021 International Patient Decision Aids Standards (IPDAS) Collaboration papers on presenting probabilities.MethodsThe MNM review generated 1,119 distinct findings (derived from 316 papers) related to communication of probabilities to patients or other lay audiences, classifying each finding by its relation to audience task, type of stimulus (data and data presentation format), and up to 10 distinct sets of outcomes: identification and/or recall, contrast, categorization, computation, probability perceptions and/or feelings, effectiveness perceptions and/or feelings, behavioral intentions or behavior, trust, preference, and discrimination. Here, we summarize the findings related to each of the 35 IPDAS paper recommendations.ResultsStrong evidence exists to support several IPDAS recommendations, including those related to the use of part-to-whole graphical formats (e.g., icon arrays) and avoidance of verbal probability terms, 1-in-X formats, and relative risk formats to prevent amplification of probability perceptions, effectiveness perceptions, and/or behavioral intentions as well as the use of consistent denominators to improve computation outcomes. However, the evidence base appears weaker and less complete for other IPDAS recommendations (e.g., recommendations regarding numerical estimates in context and evaluative labels). The IPDAS papers and the MNM review agree that both communication of uncertainty and use of interactive formats need further research.ConclusionsThe idea that no one visual or numerical format is optimal for every probability communication situation is both an IPDAS panel recommendation and foundational to the MNM project's design. Although no MNM evidence contradicts IPDAS recommendations, the evidence base needed to support many common probability communication recommendations remains incomplete.HighlightsThe Making Numbers Meaningful (MNM) systematic review of the literature on communicating health numbers provides mixed support for the recommendations of the 2021 International Patient Decision Aids Standards (IPDAS) evidence papers on presenting probabilities in patient decision aids.Both the IPDAS papers and the MNM project agree that no single visual or numerical format is optimal for every probability communication situation.The MNM review provides strong evidentiary support for IPDAS recommendations in favor of using part-to-whole graphical formats (e.g., icon arrays) and consistent denominators.The MNM review also supports the IPDAS cautions against verbal probability terms and 1-in-X formats as well as its concerns about the potential biasing effects of relative risk formats and framing.MNM evidence is weaker related to IPDAS recommendations about placing numerical estimates in context

Artificial intelligence models display human-like cognitive biases when generating medical recommendations. We tested whether an explicit forewarning, "Please keep in mind cognitive biases and other pitfalls of reasoning," might mitigate biases in OpenAI's generative pretrained transformer large language model. We used 10 clinically nuanced cases to test specific biases with and without a forewarning. Responses from the forewarning group were 50% longer and discussed cognitive biases more than 100 times more frequently compared with responses from the control group. Despite these differences, the forewarning decreased overall bias by only 6.9%, and no bias was extinguished completely. These findings highlight the need for clinician vigilance when interpreting generated responses that might appear seemingly thoughtful and deliberate.HighlightsArtificial intelligence models can be warned to avoid racial and gender bias.Forewarning artificial intelligence models to avoid cognitive biases does not adequately mitigate multiple pitfalls of reasoning.Critical reasoning remains an important clinical skill for practicing physicians.

Machine-learning (ML) models have the potential to transform health care by enabling more personalized and data-driven clinical decision making. However, their successful implementation in clinical practice requires careful consideration of factors beyond predictive accuracy. We provide an overview of essential considerations for developing clinically applicable ML models, including methods for assessing and improving calibration, selecting appropriate decision thresholds, enhancing model explainability, identifying and mitigating bias, as well as methods for robust validation. We also discuss strategies for improving accessibility to ML models and performing real-world testing.HighlightsThis tutorial provides clinicians with a comprehensive guide to implementing machine-learning classification models in clinical practice.Key areas covered include model calibration, threshold selection, explainability, bias mitigation, validation, and real-world testing, all of which are essential for the clinical deployment of machine-learning models.Following these guidance can help clinicians bridge the gap between machine-learning model development and real-world application and enhance patient care outcomes.

IntroductionEconomic evaluations identify the best course of action by a decision maker with respect to the level of health within the overall population. Traditionally, they identify 1 optimal treatment choice. In many jurisdictions, multiple technologies can be covered for the same heterogeneous patient population, which limits the applicability of this framework for directly determining whether a new technology should be covered. This article explores the impact of different decision frameworks within this context.MethodsThree alternate decision frameworks were considered: the traditional normative framework in which only the optimal technology will be covered (normative); a commonly adopted framework in which the new technology is recommended for reimbursement only if it is optimal, with coverage of other technologies remaining as before (current); and a framework that assesses specifically whether coverage of the new technology is optimal, incorporating previous reimbursement decisions and the market share of current technologies (positivist). The implications of the frameworks were assessed using a simulated probabilistic Markov model for a chronic progressive condition.ResultsResults illustrate how the different frameworks can lead to different reimbursement recommendations. This in turn produces differences in population health effects and the resultant price reductions required for covering the new technology.ConclusionBy covering only the optimal treatment option, decision makers can maximize the level of health across a population. If decision makers are unwilling to defund technologies, however, the second best option of adopting the positivist framework has the greatest relevance with respect to deciding whether a new technology should be covered.HighlightsTraditionally, economic evaluations focus on identifying the optimal treatment choice.This paper considers three alternative decision frameworks, within the context of multiple technologies being covered for the same heterogeneous patient population.This paper highlight that if decision makers are unwilling to defund therapies, current approaches to assessing cost effectiveness may be non-optimal.

BackgroundPrognostic models are crucial for predicting patient outcomes and aiding clinical decision making. Despite their availability in acute neurologic care, their use in clinical practice is limited, with insufficient reflection on reasons for this scarce implementation.PurposeTo summarize facilitators and barriers among clinicians affecting the use of prognostic models in acute neurologic care.Data SourcesSystematic searches were conducted in Embase, Medline ALL, Web of Science Core Collection, and Cochrane Central Register of Controlled Trials from inception until February 2024.Study SelectionEligible studies included those providing clinicians' perspectives on the use of prognostic models in acute neurologic care.Data ExtractionData were extracted concerning study characteristics, study aim, data collection and analysis, prognostic models, participant characteristics, facilitators, and barriers. Risk of bias was assessed using the Qualsyst tool.Data SynthesisFindings were structured around the Unified Theory of Acceptance and Use of Technology framework. Identified facilitators included improved communication with patients and surrogate decision makers (n = 9), reassurance of clinical judgment (n = 6) perceived improved patient outcomes (n = 4), standardization of care (n = 4), resource optimization (n = 3), and extension of clinical knowledge (n = 3). Barriers included perceived misinterpretation during risk communication (n = 3), mistrust in data (n = 3), perceived reduction of clinicians' autonomy (n = 3), and ethical considerations (n = 2). In total, 15 studies were included, with all but 1 demonstrating good methodological quality. None were excluded due to poor quality ratings.LimitationsThis review identifies limitations, including study heterogeneity, exclusion of gray literature, and the scarcity of evaluations on model implementation.ConclusionsUnderstanding facilitators and barriers may enhance prognostic model development and implementation. Bridging the gap between development and clinical use requires improved collaboration among researchers, clinicians, patients, and surrogate decision makers.HighlightsThis is the first systematic review to summarize published facilitators and barriers affecting the use of prognostic models in acute neurologic care from the clinicians' perspective.Commonly reported barriers and facilitators were consistent with several domains of the Unified Theory of Acceptance and Use of Technology model, including effort expectancy, social influence, and facilitating conditions, with the focus on the performance expectancy domain.Future implementation research including collaboration with researchers from different fields, clinicians, patients, and their surrogate decision makers may be highly valuable for future model development and implementation.

BackgroundAlthough sterilization is one of the most effective methods of birth control, some physicians may hesitate to perform postpartum sterilizations on patients after preterm birth, as preterm labor and delivery may preclude adequate counseling.MethodsThis is a cross-sectional study conducted at a single, tertiary care, academic institution of adult pregnant patients who experienced a spontaneous or iatrogenic preterm delivery between March 15, 2011, and May 10, 2014 and underwent postpartum female surgical sterilization within 12 wk of delivery. A validated Decision Regret Scale was administered 7 to 11 y later. Univariate and bivariate analyses were conducted. Unadjusted and multivariate logistic regression analyses identified factors associated with moderate to severe decision regret.ResultsMost participants (75.5%) with a preterm delivery reported no or mild regret associated with their sterilization. Circumstances surrounding the sterilization decision were positive, as 85.7% reported having enough information, 81.6% reported enough emotional support, and 75.5% reported adequate decision time. Adjusting for maternal and gestational age at delivery plus other covariates, only those reporting they had adequate time to make their sterilization decision remained significantly associated with no or mild regret (odds ratio: 0.002, 95% confidence interval: <0.001-0.61).DiscussionStudy results indicated high confidence in the sterilization decision, which was not affected by maternal age at delivery or the fact that the individual had a preterm delivery, emphasizing the importance of individualized counseling and support for patients during the decision-making process.ConclusionProviding adequate time for patients to decide on postpartum surgical sterilization was the most important factor for decreased sterilization regret.ImplicationsThe decision for sterilization should be made using a patient-centered, shared decision-making framework.HighlightsAmong patients with a preterm delivery who underwent postpartum surgical sterilization, maternal age at delivery was not associated with increased decision regret.Providing adequate time for patients to decide on postpartum surgical sterilization was the most important factor for decreased sterilization regret among patients with a preterm delivery.We must trust the patient knows they are making the right decision for themselves in that moment, even if this is at the time of a preterm delivery.

BackgroundEconomic models often require extrapolation of clinical time-to-event data for multiple events. Two modeling approaches in oncology that incorporate time dependency include partitioned survival models (PSM) and semi-Markov decision models estimated using multistate modeling (MSM). The objective of this simulation study was to assess the performance of PSM and MSM across datasets with varying sample size and degrees of censoring.MethodsWe generated disease trajectories of progression and death for multiple hypothetical populations with advanced cancers. These populations served as the sampling pool for simulated trial cohorts with multiple sample sizes and various levels of follow-up. We estimated MSM and PSM by fitting survival models to these simulated datasets with different approaches to incorporating general population mortality (GPM) and selected best-fitting models using statistical criteria. Mean survival was compared with "true" population values to assess error.ResultsWith near complete follow-up, both PSMs and MSMs accurately estimated mean population survival, while smaller samples and shorter follow-up times were associated with a larger error across approaches and clinical scenarios, especially for more distant clinical endpoints. MSMs were slightly more often not estimable when informed by studies with small sample sizes or short follow-up, due to low numbers at risk for the downstream transition. However, when estimable, the MSM models more commonly produced a smaller error in mean survival than the PSMs did.ConclusionsCaution should be taken with all modeling approaches when the underlying data are very limited, particularly PSMs, due to the large errors produced. When estimable and for selections based on statistical criteria, MSMs performed similar to or better than PSMs in estimating mean survival with limited data.HighlightsCaution should be taken with all modeling approaches when underlying data are very limited.Partitioned survival models (PSMs) can lead to significant errors, particularly with limited follow-up. Incorporating general population mortality (GPM) via internal additive hazards improved estimates of mean survival, but the effects were modest.When estimable, decision models based on multistate modeling (MSM) produced similar or smaller error in mean survival compared with PSM, but small samples or limited deaths after progression produce additional challenges for fitting MSMs; more research is needed to improve estimation of MSMs and similar state transition-based modeling methods with limited data.Future studies are needed to assess the applicability of these findings to comparative analyses estimating incremental survival benefits.

ObjectivesThe EQ-5D-5L and Patient-Reported Outcomes Measurement Information System (PROMIS®) preference score (PROPr) are preference-based measures. This study compares mapping and linking approaches to align the PROPr and the PROMIS domains included in PROPr plus Anxiety with EQ-5D-5L item responses and preference scores.MethodsA general population sample of 983 adults completed the online survey. Regression-based mapping methods and item response theory (IRT) linking methods were used to align scores. Mapping was used to predict EQ-5D-5L item responses or preference scores using PROMIS domain scores. Equating strategies were applied to address regression to the mean. The linking approach estimated item parameters of EQ-5D-5L based on the PROMIS score metric and generated bidirectional crosswalks between EQ-5D-5L item responses and relevant PROMIS domain scores.ResultsEQ-5D-5L item responses were significantly accounted for by PROMIS domains of Anxiety, Depression, Fatigue, Pain Interference, Physical Function, Social Roles, and Sleep Disturbance. EQ-5D-5L preference scores were accounted for by the same PROMIS domains, excluding Anxiety and Fatigue, and by the PROPr preference scores. IRT-linking crosswalks were generated between EQ-5D-5L item responses and PROMIS domains of Physical Function, Pain, and Depression. Small differences were found between observed and predicted scores for all 3 methods. The direct mapping approach (directly predicting EQ-5D-5L scores) with the equipercentile equating strategy proved superior to the linking method due to improved prediction accuracy and comparable score range coverage.ConclusionsThe PROPr and the PROMIS domains included in the PROMIS-29+2 predict EQ-5D-5L preference scores or item responses. Both methods can generate acceptably precise EQ-5D-5L preference scores, with the direct mapping approach using the equating strategy offering better precision. We summarized recommended score conversion tables based on available and desired scores.HighlightsThis study compares mapping (score prediction) and IRT-based linking approaches to align the PROPr and the PROMIS domains with EQ-5D-5L item responses and preference scores.Researchers, clinicians, and stakeholders can use this study's regression formulas and score crosswalks to convert scores between PROMIS and EQ-5D-5L.Mapping can generate more precise scores, while linking offers greater flexibility in score estimation when fewer PROMIS domain scores are collected.

BackgroundMany contributing factors can influence individuals' health, and these factors may not affect health outcomes equally. This study compared the importance of 38 predictors of health-related quality of life (HRQOL) and 2-y mortality for US older adults.MethodsData were from the Medicare Health Outcome Survey Cohort 23 (baseline 2020, follow-up 2022). This study included participants ≥65 y (N = 142,551). HRQOL measures included physically unhealthy days (PUD), mentally unhealthy days (MUD), and activity limitation days (ALD) from the Healthy Days questions and 3 measures from the Veterans RAND 12-Item Health Survey (VR-12). A variable's importance was measured as the average gain in R² after adding the variable in all submodels.ResultsFor physical health (PUD), pain interfered with daily activities was the most important predictor with an importance score (I) of 8.4, indicating that this variable contributed 8.4% variance of PUD. Other leading predictors included pain interfered with socializing (I = 7.3) and pain rating (I = 6.7). For mental health (MUD), depression (I = 11.6) was far more important than any of the other predictors, contributing 38% of the total importance. For perceived disability (ALD), pain interfered with socializing was the most important predictor (I = 8.3), followed by difficulty doing errands (I = 6.1) and pain interfered with activities (I = 6.0). Of note, this general pattern was consistent for VR-12 HRQOL measures. Variables' importance scores for 2-y morality were very different from that for HRQOL. Age (I = 2.8) and difficulty doing errands (I = 2.6) were the most important variables.ConclusionsThis study demonstrated a large discrepancy in the variables' importance for HRQOL and 2-y mortality. Functional limitations/disabilities and geriatric syndromes were more important for the prediction of HRQOL than were chronic conditions and other factors combined.HighlightsFor older adults, large differences were found in variable importance for explaining health-related quality of life (HRQOL) and 2-y mortality among 38 explanatory variables, including functional limitations, geriatric syndromes, chronic conditions, and other factors.Pain and pain interference, difficulty doing errands, difficulty concentrating, memory problems, problems with walking/balance, and depression were the most important predictors of HRQOL.Age, marital status, education, difficulty doing errands, congestive heart failure, chronic obstructive pulmonary disease, and any cancer were more important for 2-y mortality than HRQOL.Health care providers and policy makers should focus on the impact of multimorbidity and the interaction between often multifactorial conditions, as opposed to focusing only on individual diseases.