Medical Decision Making最新文献

BackgroundAn artificial intelligence (AI)-enabled rule-out device may autonomously remove patient images unlikely to have cancer from radiologist review. Many published studies evaluate this type of device by retrospectively applying the AI to large datasets and use sensitivity and specificity as the performance metrics. However, these metrics have fundamental shortcomings because sensitivity will always be negatively affected in retrospective studies of rule-out applications of AI.MethodWe reviewed 2 performance metrics to compare the screening performance between the radiologist-with-rule-out-device and radiologist-without-device workflows: positive/negative predictive values (PPV/NPV) and expected utility (EU). We applied both methods to a recent study that reported improved performance in the radiologist-with-device workflow using a retrospective US dataset. We then applied the EU method to a European study based on the reported recall and cancer detection rates at different AI thresholds to compare the potential utility among different thresholds.ResultsFor the US study, neither PPV/NPV nor EU can demonstrate significant improvement for any of the algorithm thresholds reported. For the study using European data, we found that EU is lower as AI rules out more patients including false-negative cases and reduces the overall screening performance.ConclusionsDue to the nature of the retrospective simulated study design, sensitivity and specificity can be ambiguous in evaluating a rule-out device. We showed that using PPV/NPV or EU can resolve the ambiguity. The EU method can be applied with only recall rates and cancer detection rates, which is convenient as ground truth is often unavailable for nonrecalled patients in screening mammography.HighlightsSensitivity and specificity can be ambiguous metrics for evaluating a rule-out device in a retrospective setting. PPV and NPV can resolve the ambiguity but require the ground truth for all patients. Based on utility theory, expected utility (EU) is a potential metric that helps demonstrate improvement in screening performance due to a rule-out device using large retrospective datasets.We applied EU to a recent study that used a large retrospective mammography screening dataset from the United States. That study reported an improvement in specificity and decrease in sensitivity when using their AI as a rule-out device retrospectively. In terms of EU, we cannot conclude a significant improvement when the AI is used as a rule-out device.We applied the method to a European study that reported only recall rates and cancer detection rates. Since there is no established EU baseline value in European mammography screening workflow, we estimated the EU baseline using data from previous literature. We cannot conclude a significant improvement when the AI is used as a rule-out device for the European study.In this work, we investigated the use of EU to evaluate rule-out devices using large retrospective d

BackgroundCervical cancer, driven predominantly by persistent high-risk human papillomavirus (HPV) infection, ranks as the fourth most common malignancy in women worldwide. China faces barriers to achieving the World Health Organization (WHO) 2030 elimination targets due to low vaccination rates and complex demographics. Strategic intervention optimization is critical for accelerating elimination.MethodsWe developed an age-stratified deterministic compartmental model integrating demographic data and HPV transmission dynamics, capturing heterogeneity in age, sex, sexual activity, and intervention efficacy. The model simulated cervical cancer natural history, including HPV infection, progression to precancerous lesions, and invasive cancer and was calibrated using epidemiological data from the Global Burden of Disease. We evaluated multiple vaccination scenarios (varying coverage rates, age groups, and durations) to project incidence trajectories, estimate elimination timelines, and calculate the reproduction number. Sensitivity analyses were conducted to assess parameter effects.ResultsWithout vaccination, HPV infection becomes endemic (R₀ = 1.38), causing 2.92 million cervical cancer cases in China during 2021 to 2070. Maintaining the 2020 vaccination rate would prevent 1.01 million cases in this period. While prioritizing females aged 15 to 26 y maximizes the per-dose impact, expanding vaccination to all females aged ≥15 y is essential for achieving elimination before 2040. Even single-year vaccination would confer >50-y protection. A higher vaccination rate accelerates elimination: annual rates of 0.09, 0.15, and 0.21 among females aged ≥15 y achieve elimination by 2037, 2035, and 2034, respectively, accelerating timelines by 15 to 20 y compared with strategies targeting only 15- to 26-y-olds.ConclusionsHPV vaccination is pivotal for reducing cervical cancer burden in China, with prioritizing women aged 15 to 26 y as the optimal strategy. Expanding vaccination to all women aged ≥15 y can accelerate the achievement of WHO elimination targets.HighlightsAn age-stratified model simulates HPV transmission patterns and assesses cervical cancer interventions.Without intervention, HPV remains endemic (R₀ = 1.38), causing 2.92 million cervical cancer cases in China (2021-2070).Prioritizing 15- to 26-y-olds maximizes the per-dose impact, but expanding to 15+ y cohorts is essential for elimination.Even a single year of vaccination offers >50 y of protection.Females ≥15 y vaccinated annually at rates of 0.09, 0.15, and 0.21 achieve elimination by 2037, 2035, and 2034, respectively.

Network meta-analysis (NMA) synthesizes data from randomized controlled trials to estimate the relative treatment effects among multiple interventions. When treatments can be grouped into classes, class effect NMA models can be used to inform recommendations at the class level and can also address challenges with sparse data and disconnected networks. Despite the potential of NMA class effects models and numerous applications in various disease areas, the literature lacks a comprehensive guide outlining the range of class effect models, their assumptions, practical considerations for estimation, model selection, checking assumptions, and presentation of results. In addition, there is no implementation available in standard software for NMA. This article aims to provide a modeling framework for class effect NMA models, propose a systematic approach to model selection, and provide practical guidance on implementing class effect NMA models using the multinma R package. We describe hierarchical NMA models that include random and fixed treatment-level effects and exchangeable and common class-level effects. We detail methods for testing assumptions of heterogeneity, consistency, and class effects, alongside assessing model fit to identify the most suitable models. A model selection strategy is proposed to guide users through these processes and assess the assumptions made by the different models. We illustrate the framework and structured approach for model selection using an NMA of 41 interventions from 17 classes for social anxiety.HighlightsProvides a practical guide and modelling framework for network meta-analysis (NMA) with class effects.Proposes a model selection strategy to guide researchers in choosing appropriate class effect models.Illustrates the strategy using a large case study of 41 interventions for social anxiety.

BackgroundAny sample of individuals has its own unique distribution of preferences for choices that they make. Discrete choice models try to capture these distributions. Mixed logits are by far the most commonly used choice model in health. Many parametric specifications for these models are available. We test a range of alternative assumptions and model averaging to test if or how model outputs are affected.DesignScoping review of current modeling practices. Seven alternative distributions and model averaging over all distributional assumptions were compared on 4 datasets: 2 were stated preference, 1 was revealed preference, and 1 was simulated. Analyses examined model fit, preference distributions, willingness to pay, and forecasting.ResultsAlmost universally, using normal distributions is the standard practice in health. Alternative distributional assumptions outperformed standard practice. Preference distributions and the mean willingness to pay varied significantly across specifications and were seldom comparable to those derived from normal distributions. Model averaging offered distributions allowing for greater flexibility and further gains in fit, reproduced underlying distributions in simulations, and mitigated against analyst bias arising from distribution selection. There was no evidence that distributional assumptions affected predictions from models.LimitationsOur focus was on mixed logit models since these models are the most common in health, although latent class models are also used.ConclusionsThe standard practice of using all normal distributions appears to be an inferior approach for capturing random preference heterogeneity. Implications. Researchers should test alternative assumptions to normal distributions in their models.HighlightsHealth modelers use normal mixing distributions for preference heterogeneity.Alternative distributions offer more flexibility and improved model fit.Model averaging offers yet more flexibility and improved model fit.Distributions and willingness to pay differ substantially across alternatives.

PurposeWhen using stochastic models for cost-effectiveness analysis (CEA), run-to-run outcome variability arising from model stochasticity can sometimes exceed the change in outcomes resulting from an intervention, especially when individual-level efficacy is small, leading to counterintuitive results. This issue is compounded for probabilistic sensitivity analyses (PSAs), in which stochastic noise can obscure the influence of parameter uncertainty. This study evaluates meta-modeling as a variance-reduction technique to mitigate stochastic noise while preserving parameter uncertainty in PSAs.MethodsWe applied meta-modeling to 2 simulation models: 1) a 4-state Sick-Sicker model and 2) an agent-based HIV transmission model among men who have sex with men (MSM). We conducted a PSA and applied 3 meta-modeling techniques-linear regression, generalized additive models, and artificial neural networks-to reduce stochastic noise. Model performance was assessed using R² and root mean squared error (RMSE) values on a validation dataset. We compared PSA results by examining scatter plots of incremental costs and quality-adjusted life-years (QALYs), cost-effectiveness acceptability curves (CEACs), and the occurrence of unintuitive results, such as interventions appearing to reduce QALYs due to stochastic noise.ResultsIn the Sick-Sicker model, stochastic noise increased variance in incremental costs and QALYs. Applying meta-modeling techniques substantially reduced this variance and nearly eliminated unintuitive results, with R² and RMSE values indicating good model fit. In the HIV agent-based model, all 3 meta-models effectively reduced outcome variability while retaining parameter uncertainty, yielding more informative CEACs with higher probabilities of being cost-effective for the optimal strategy.ConclusionsMeta-modeling effectively reduces stochastic noise in simulation models while maintaining parameter uncertainty in PSA, enhancing the reliability of CEA results without requiring an impractical number of simulations.HighlightsWhen using complex stochastic models for cost-effectiveness analysis (CEA), stochastic noise can overwhelm intervention effects and obscure the impact of parameter uncertainty on CEA outcomes in probabilistic sensitivity analysis (PSA).Meta-modeling offers a solution by effectively reducing stochastic noise in complex stochastic simulation models without increasing computational burden, thereby improving the interpretability of PSA results.

IntroductionThe potential for multicancer early detection (MCED) tests to detect cancer at earlier stages is currently being evaluated in screening clinical trials. Once trial evidence becomes available, modeling will be necessary to predict the effects on final outcomes (benefits and harms), account for heterogeneity in determining clinical and cost-effectiveness, and explore alternative screening program specifications. The natural history of disease (NHD) component will use statistical, mathematical, or calibration methods. This work aims to identify, review, and critically appraise the existing literature for alternative modeling approaches proposed for MCED that include an NHD component.MethodsModeling approaches for MCED screening that include an NHD component were identified from the literature, reviewed, and critically appraised. Purposively selected (non-MCED) cancer-screening models were also reviewed. The appraisal focused on the scope, data sources, evaluation approaches, and the structure and parameterization of the models.ResultsFive different MCED models incorporating an NHD component were identified and reviewed, alongside 4 additional (non-MCED) models. The critical appraisal highlighted several features of this literature. In the absence of trial evidence, MCED effects are based on predictions derived from test accuracy. These predictions rely on simplifying assumptions with unknown impacts, such as the stage-shift assumption used to estimate mortality impacts from predicted stage shifts. None of the MCED models fully characterized uncertainty in the NHD or examined uncertainty in the stage-shift assumption.ConclusionThere is currently no modeling approach for MCEDs that can integrate clinical study evidence. In support of policy, it is important that efforts are made to develop models that make the best use of data from the large and costly clinical studies being designed and implemented across the globe.HighlightsIn the absence of trial evidence, published estimates of the effects of multicancer early detection (MCED) tests are based on predictions derived from test accuracy.These predictions rely on simplifying assumptions, such as the stage-shift assumption used to estimate mortality effects from predicted stage shifts. The effects of such simplifying assumptions are mostly unknown.None of the existing MCED models fully characterize uncertainty in the natural history of disease; none examine uncertainty in the stage-shift assumption.Currently, there is no modeling approach that can integrate clinical study evidence.

ObjectivesTransgender (trans) people have disproportionately high HIV risk, yet adherence to preexposure prophylaxis (PrEP) remains low in this population. We aimed to determine which factors matter most in the decision of HIV-negative transgender adults to adhere to long-acting injectable PrEP (LA-PrEP), and the acceptability of providing incentives conditional on LA-PrEP program engagement.MethodsFrom March to April 2023, 385 trans adults in Washington State completed a discrete-choice experiment (DCE) eliciting preferences for a conditional economic incentive program that would provide free LA-PrEP and gender-affirming care during bimonthly visits. We used the best-best preference elicitation method across 2 hypothetical programs with an opt-out option. Program attributes included incentive format and amount, method for determining PrEP adherence, and type of hormone co-prescription. We used a rank-ordered mixed logit model for main results and estimated respondents' marginal willingness to accept each program attribute. We plotted the probability of choosing an incentivized LA-PrEP program over a range of respondent characteristics.ResultsThe optimal program design would 1) deliver incentives in cash, 2) confirm PrEP adherence via blood testing, 3) provide counseling in person, and 4) provide prescriptions for injectable gender-affirming hormones. From a maximum incentive amount of $1,200/year, respondents were willing to forgo up to $689 to receive incentives in cash (instead of voucher) and up to $547 to receive injectable (instead of oral) hormones. The probability of choosing a hypothetical program over no program waned as adults aged (>40 y) and as income increased (>$75,000/y).ConclusionsConditional economic incentives are likely acceptable and effective for improving LA-PrEP adherence, especially among younger trans adults with fewer financial resources. A randomized trial is needed to confirm the DCE's validity for predicting actual program uptake.HighlightsGender-related stigma, economic barriers, and medical concerns about hormone interactions can keep transgender (trans) adults from engaging in HIV prevention behaviors.Combining gender-affirming care with conditional economic incentives may help reduce present bias and increase trans persons' motivation to adhere to long-acting injectable preexposure prophylaxis (LA-PrEP).From a maximum yearly incentive of $1,200, trans discrete-choice experiment respondents were willing to forgo up to $689 to receive a cash (rather than voucher) incentive and up to $547 to receive co-prescriptions for injectable (rather than oral) hormones as part of a hypothetical HIV prevention program.The probability of choosing an LA-PrEP program over no program begins to wane as adults age (>40 y) and as annual income increases (>$75,000/year), such that incentivized LA-PrEP programs may be especially salient for younger trans adults with fewer financial resources.

Introduction. Calibration, a critical step in the development of simulation models, involves adjusting unobservable parameters to ensure that the outcomes of the model closely align with observed target data. This process is particularly vital in cancer simulation models with a natural history component, where direct data to inform natural history parameters are rarely available. Methods. We conducted a scoping review of studies published from 1980 to August 11, 2024, using keyword searches in PubMed and Web of Science. Eligible studies included cancer simulation models with a natural history component that used calibration methods for parameter estimation. Results. A total of 117 studies met the inclusion criteria. Nearly all studies (n = 115) specified calibration targets, while most studies (n = 91) described the parameter search algorithms used. Goodness-of-fit metrics (n = 87), acceptance criteria (n = 53), and stopping rule (n = 46) were reported less frequently. The most commonly used calibration targets were incidence, mortality, and prevalence, typically drawn from cancer registries and observational studies. Mean squared error was the most commonly used goodness-of-fit measure. Random search was the predominant method for parameter search, followed by the Bayesian approach and the Nelder-Mead method. Discussion. Despite recent advances in machine learning, such algorithms remain underutilized in the calibration of cancer simulation models. Further research is needed to compare the efficiency of different parameter search algorithms used for calibration.HighlightsThis work reviewed the literature of cancer simulation models with a natural history component and identified the calibration approaches used in these models with respect to the following attributes: cancer type, calibration target data source, calibration target type, goodness-of-fit metrics, search algorithms, acceptance criteria, stopping rule, computational time, modeling approach, and model stochasticity.Random search has been the predominant method for parameter search, followed by Bayesian approach and Nelder-Mead method.Machine learning-based algorithms, despite their fast advancement in the recent decade, have been underutilized in the cancer simulation models. Furthermore, more research is needed to compare different parameter search algorithms used for calibration.

Background. Observational studies are used to evaluate the effect of breast cancer screening programs, but their validity depends on use of different study designs. One of these is the evaluation model, which extends follow-up after screening only if women have been diagnosed with breast cancer during the screening program. However, to avoid lead-time bias, the inclusion of risk time should be based on screening invitation and not breast cancer diagnosis. The aim of this study is to investigate potential bias induced by the evaluation model. Methods. We used large-scale simulated datasets to investigate the evaluation model. Simulation model parameters for age-dependent breast cancer incidence, survival, breast cancer mortality, and all-cause mortality were obtained from Norwegian registries. Data were restricted to women aged 48 to 90 y and a period before screening implementation, 1986 to 1995. Simulation parameters were estimated for each of 2 periods (1986-1990 and 1991-1995). For the simulated datasets, 50% were randomly assigned to screening and 50% were not. Simulation scenarios depended on the magnitude of screening effect and level of overdiagnosis. For each scenario, we applied 2 study designs, the evaluation model and ordinary incidence-based mortality, to estimate breast cancer mortality rates for the screening and nonscreening groups. For each design, these rates were compared to assess potential bias. Results. In scenarios with no screening effect and no overdiagnosis, the evaluation model estimated 6% to 8% reductions in breast cancer mortality due to lead-time bias. Bias increased with overdiagnosis. Conclusions. The evaluation model was biased by lead time, especially in scenarios with overdiagnosis. Thus, the attempt to capture more of the screening effect using the evaluation model comes at the risk of introducing bias.HighlightsThe validity of observational studies of breast cancer screening programs depends on their study design being able to eliminate lead-time bias.The evaluation model has been used to evaluate breast cancer screening in recent studies but introduces a study design based on breast cancer diagnosis that may introduce lead-time bias.We used large-scale simulated datasets to compare study designs used to evaluate screening.We found that the evaluation model was biased by lead time and estimated reductions in breast cancer mortality in scenarios with no screening effect.

BackgroundIt is still a matter of debate whether a reduction in cancer-specific mortality due to cancer screening fully translates into a reduction in all-cause mortality and thus into a gain in life expectancy. Nevertheless, decision-analytic models simulating the health consequences of screening compared with no screening predict substantial gains in life expectancy.PurposeThe aim of this review was to systematically assess methodological competing mortality risk features that affect the translation of cancer-specific mortality reductions into gains in life expectancy in decision-analytic screening models for prostate, lung, breast, and colorectal cancer.Data SourcesLiterature databases were systematically searched for clinical and economic decision-analytic models evaluating the effect of screening for prostate, lung, breast, and colorectal cancer compared with no screening.Study SelectionForty-two clinical and economic decision-analytic models were included for narrative synthesis.Data ExtractionBasic information and specific methodological features of the included decision-analytic models were extracted using a standardized approach.Data SynthesisCharacteristics and methodological features of the identified studies were summarized in evidence tables.LimitationsThe review focused on models that reported undiscounted outcomes of life-years gained for standard screening strategies.ConclusionsThis review highlights key modeling features related to competing mortality risks that should be considered in decision-analytic models assessing the effects of cancer screening. All included models predicted gains in life expectancy with screening, although the magnitude of these gains varied both within and across cancer types. Models that considered competing mortality risks tended to predict smaller lifetime gains from screening interventions. Future studies should prioritize the use of advanced modeling approaches that account for competing mortality risks to improve the accuracy of benefit-harm assessments in cancer screening.HighlightsThis is the first systematic assessment of methodological competing mortality risk features of decision-analytic screening models across 4 cancer types.Models vary greatly regarding predicted gains in life expectancy, natural history assumptions (onset and progression rates), methodological model features, and screening strategies.Models that considered competing mortality risks or adjusted life expectancy for comorbidities predicted smaller lifetime gains for screening compared with no screening.