Value in Health最新文献

Background: Heterogeneous treatment effects (HTEs) refer to differences in how individual patients or subgroups respond to the same treatment. Estimating HTEs helps target care to those most likely to benefit, improving outcomes and avoiding unnecessary interventions. Machine learning (ML) enables the use of real-world data (RWD) to estimate HTEs when randomized controlled trials are not feasible. However, practical guidance for applying these methods in health economics is lacking.

Purpose: To support method selection, we identified and categorized ML approaches to estimating HTEs in RWD and assessed the methodological quality of studies applying them.

Methods: We conducted a scoping review following PRISMA-ScR guidelines. PubMed, Scopus, Web of Science, EBSCO, and MEDLINE were searched for studies published between 2014 and 2025 that applied ML to estimate HTEs from RWD. Methodological quality was assessed using a standardized checklist.

Findings: Of 1743 records screened, 74 met the inclusion criteria. We grouped the included studies into three categories: those using prediction-only approaches unsuited to HTE estimation (n=8), those applying outcome modelling (n=9), and those using customized conditional average treatment effect (CATE) estimation (n=58). Most innovations originated in the ML and statistics communities, with minimal uptake in health economics. Methodological quality was inconsistent and requires improvement.

Conclusion: ML methods for HTE estimation are increasingly applied to RWD. Tree-based models are most common, and interest in customized CATE approaches is growing. Better evaluation standards and more transparent reporting are needed for these methods to become reliable tools for health economics research.

Objective: Introduce the rescaled value set regression for estimating EQ-5D-5L health state values as an alternative way to report the nonparametric crosswalk.

Method: The rescaled value set regression and the nonparametric crosswalk methods were applied to estimate EQ-5D-5L state values from EQ-5D-3L value sets for three example countries (United Kingdom, the Netherlands, Spain). The rescaled value set regression converted the original three-level value set regression parameters comprising dichotomous independent variables into regression parameters for the five-level version. The health state values for twenty-eight common EQ-5D-5L response profiles were then estimated by the rescaled value set regression and nonparametric crosswalk to assess whether they produced the same results using value sets from the three different countries.

Results: When applied to EQ-5D-3L value sets, the rescaled value set regression demonstrated that a level-two response and level-three response using the EQ-5D-3L, respectively, corresponded with a level-three response and level-five response using the EQ-5D-5L. The analysis of twenty-eight common EQ-5D-5L response profiles produced identical health state values for the United Kingdom, the Netherlands, and Spain's value sets under both the rescaled value set regression and nonparametric crosswalk.

Conclusion: The rescaled value set regression provides improved transparency than the nonparametric crosswalk when estimating EQ-5D-5L health state values anchored to EQ-5D-3L value sets. Both methods may be used in combination for jurisdictions where new EQ-5D-5L valuation studies are not planned but a relevant EQ-5D-3L value set is available.

Objectives: Respiratory syncytial virus is known to cause severe bronchiolitis and pneumonia among infants. There are two main ways to protect infants from RSV-related diseases, namely, vaccination of pregnant women with recombinant subunit RSV pre-fusion F3 (RSVpreF) and prophylaxis of neonates/infants with nirsevimab. In 2024, both products were approved in Japan. We evaluated the cost-effectiveness of multiple immunisation strategies for protection of Japanese infants from RSV diseases by using nirsevimab and/or RSVpreF.

Methods: A decision tree with Markov model was adopted. Incremental cost-effectiveness ratio (ICER) from payers' perspective was calculated. Variables used in the model were either calculated or extracted from literature. Costs per RSVpreF and nirsevimab vaccination was assumed to be at JPY23,948/US$160 and JPY45,000/US$300, respectively.

Results: Vaccination of pregnant women with RSVpreF strategies (seasonally or year-round), prophylaxis of infants with nirsevimab strategy, and combination of seasonally RSVpreF and Nirsevimab strategies, all reduced disease treatment costs; however, the reduction could not offset the vaccination/prophylaxis cost. RSVpreF_year-round strategy and Nirsevimab strategy were either extended or absolutely dominated by the other two strategies, and were excluded from being considered as an option. The ICER of seasonal RSVpreF strategy was JPY3,227,850/US$21,519/QALY, while Combination strategy's ICER was JPY23,236,084/US$154,907/QALY per QALY gained. One-way sensitivity analyses revealed that probability of hospitalisation, vaccination costs and effectiveness of both products influence the ICER the most. Cost-effectiveness acceptance curve revealed that the curve reached 98.7% at a willingness-to-pay (WTP) of JPY5,000,000/US$33,333 per QALY.

Conclusion: Only seasonal RSVpreF strategy is cost-effective under the JPY5,000,000/US$33,333 per QALY WTP threshold.

The argument-based approach to validation, adopted in the US Food and Drug Administration's most recent Patient-Focused Drug Development draft guidance on clinical outcome assessments (COAs), emphasizes the importance of constructing explicit rationales to support proposed interpretations of COA scores. To assist researchers and sponsors in building such rationales, we describe 2 complementary strategies: (1) reviewing steps in the assessment process to identify essential assumptions and (2) evaluating potential threats to validity from construct underrepresentation and construct irrelevance. Using these strategies, we offer initial generic rationales tailored to 4 types of COAs: patient-reported outcomes (PROs), observer-reported outcomes (ObsROs), clinician-reported outcomes (ClinROs), and performance outcome (PerfO) measures. The generic rationales serve as starting points, with the expectation that they will be adapted to specific contexts of use. Greater discussion within the field is needed to advance consensus on the construction and evaluation of evidence-based rationales, with attention to the pragmatic and iterative nature of validation work.

Objective: Children of parents with a mental illness (COPMI) face a higher risk of developing mental disorders, leading to significant long-term societal and health-related costs. While preventive interventions exist, few studies assess their cost-effectiveness, and none model long-term outcomes. This study aims to develop a Markov model to evaluate the cost-effectiveness of preventive interventions for COPMI in the Netherlands.

Methods: A decision-analytic model was constructed using data from the Avon Longitudinal Study of Parents and Children. The model simulates COPMI progression over time, with health states including: healthy, depression/anxiety, comorbidity, remission, and death. The time horizon spans 28 years, from ages 7 to 35, and outcomes are evaluated from both healthcare and societal perspectives. Results are expressed as total costs, quality-adjusted life years (QALYs), and incremental cost-effectiveness ratios (ICERs). A group-based cognitive behavioral therapy (CBT) intervention was evaluated against a reference scenario.

Results: The preventive CBT intervention yielded an additional 0.02 QALYs at an additional cost of €188 per patient, resulting in an ICER of €9,495 per QALY. The intervention had a 74% probability of being cost-effective at a willingness-to-pay threshold of €20,000 per QALY.

Conclusions: The Markov model provides a flexible tool for evaluating the cost-utility of user-defined COPMI interventions to support informed decision-making in mental health care. It is freely available for academic purposes upon request by the authors. Results suggest group-based CBT may be a cost-effective strategy for preventing mental disorders in COPMI.

Introduction: We build on Cutler and colleagues' research on the value of health-care spending using a period life-expectancy framework. We use the framework to track health-adjusted life-expectancy (HALE) and lifetime spending for all ages, show the value of improvements in health care, and demonstrate the contribution of expanding research to the full age range.

Methods: We use population-level results on mortality and years lived with disability from the 2019 Global Burden of Disease, Injuries, and Risk Factor Study, and spending from the 2016 Disease Expenditure study. We use cause-replacement methods to simulate effects of changes in health care. For 132 causes, we replace cause-specific outcomes (or spending) per case from 1996 with those measures for 2016; effect is the difference between base year and simulated calculations. Spending is reported in 2016 US dollars ($).

Results: For all-cause aggregate calculated at birth, lifetime spending effect is $234,111 (95% uncertainty interval (UI): 221,395, 242,456) and HALE effect is 1.285 (95% UI: 1.161, 1.422) years per person. Value of improvements is the ratio of these two effects, $182,201 (95% UI: 181,494, 182,912) per HALE gained. Seventy-nine (60%) causes have increase in mean HALE and lifetime spending. Value is $9,315 (95% UI: 9,204, 9,427) for HIV/AIDS and $63,184 (95% UI: 62,352, 64,030) for ischemic heart disease. For drug use disorders, HALE effect is -0.331 (95%UI: -0.370, -0.296), which offset other gains. Increases in HALE often occur at older ages than lifetime spending.

Conclusion: Comprehensive measures for all ages show value of health care by cause.

Objectives: This study evaluated how uncertainty in adaptive Health Technology Assessment (aHTA) impacts reimbursement decisions and identified factors contributing to this uncertainty.

Methods: A simulation-based approach was employed to generate a distribution of aHTA Incremental Cost-Effectiveness Ratios (ICERs). ICERs sampled from systematic reviews of seven technologies were adjusted using: (i) USD adjustment for currency and inflation, and (ii) technology-price adjustment. Uncertainty in aHTA was quantified by estimating the probability of wrong reimbursement decisions using a willingness-to-pay (WTP) threshold-based decision rule. This involved comparing decisions based on the simulated aHTA ICER against a decision based on a known ICER from Thailand, serving as the "true" reference ICER. Financial risk of wrong reimbursement decisions from the aHTA approach was also quantified.

Results: The probability of wrong decisions from aHTA decreased when aHTA ICERs were clearly above or below the WTP threshold. Low variability among published ICERs, particularly when studies shared a similar methodological framework, improved confidence in aHTA. Simple adjustments to ICERs, such as technology-price adjustments, showed potential in reducing variability across studies. Technologies with modest disease burden and lower cost were associated with smaller financial risks, even under uncertain evidence.

Conclusion: The aHTA approach is likely suitable under three conditions: (a) when the aHTA ICER is clearly positioned far from a country's WTP threshold; (b) when published ICERs exhibit low variability; and (c) when disease burden and financial risks are modest.

Objectives: To generate mapping algorithms from the Patient-Reported Outcomes Measurement Information System Pediatric Profile 25 (PROMIS-25) to both EQ-5D-Y-3L responses (indirect mapping) and EQ-5D-Y-3L utilities (direct mapping).

Methods: A subset of data from the Australian P-MIC study dataset were used, including participants aged 5-18 years who completed both the EQ-5D-Y-3L and PROMIS-25 (n=1,830). Both direct and indirect mapping approaches were used, exploring a range of regression models and predictor variables for each approach. For the direct mapping approach, the EQ-5D-Y-3L Australian value set was used, and sensitivity analyses were conducted using the EQ-5D-Y-3L Dutch value set. Five-fold internal cross-validation was used to select the optimal mapping models based on goodness-of-fit indicators, including the Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Concordance Correlation Coefficient (CCC). The final mapping algorithms reported are based on full sample.

Results: The generalised logit (GLOGIT) model using the PROMIS-25 raw domain scores as predictors was selected for predicting EQ-5D-Y-3L responses in the indirect mapping (RMSE=0.1098, MAE=0.0724). The Tobit model, also using the PROMIS-25 raw item scores as predictors, was the optimal direct mapping model for predicting Australian EQ-5D-Y-3L utilities (RMSE=0.0994, MAE=0.0712). The same models performed similarly well in sensitivity analyses using Dutch utilities.

Conclusions: The mapping algorithms provide a pathway for PROMIS-25 data to be converted either directly to an Australian or Dutch EQ-5D-Y-3L utility or EQ-5D-Y-3L responses where local value sets can be applied. This broadens the usability of PROMIS-25, enabling calculation of utilities for use in economic evaluation.