Parametric Distribution Research Articles

Survey-based inference of continental African elephant decline

Long-term quantification of temporal species trends is fundamental to the assignment of conservation status, which in turn is critical for planning and targeting management interventions. However, monitoring effort and methodologies can change over the assessment period, resulting in heterogeneous data that are difficult to interpret. Here, we develop a hierarchical, random effects Bayesian model to estimate site-level trends in density of African elephants from geographically disparate survey data. The approach treats the density trend per site as a random effect and estimates a parametric distribution of these trends for each partitioning of the data. Data were available from 475 sites, in 37 countries, between 1964 and 2016 (a total of 1,325 surveys). We implemented the model separately and in combination for the African forest ( Loxodonta cyclotis ) and savannah ( Loxodonta africana ) elephant species, as well as by region. Inference from these distributions indicates a mean site-level decline for each species over the study period, with the average forest elephant decline estimated to be more than 90% compared to 70% for the savannah elephant. In combination, there has been a mean 77% decline across all sites; but in all models, substantial heterogeneity in trends was found, with stable to increasing trends more common in southern Africa. This work provides the most comprehensive assessment undertaken on the two African elephant species, illustrating the variability in their status across populations.

Enhanced adaptive permutation test with negative binomial distribution in genome-wide omics datasets.

The permutation test has been widely used to provide the p-values of statistical tests when the standard test statistics do not follow parametric null distributions. However, the permutation test may require huge numbers of iterations, especially when the detection of very small p-values is required for multiple testing adjustments in the analysis of datasets with a large number of features. To overcome this computational burden, we suggest a novel enhanced adaptive permutation test that estimates p-values using the negative binomial (NB) distribution. By the method, the number of permutations are differently determined for individual features according to their potential significance. In detail, the permutation procedure stops, when test statistics from the permuted dataset exceed the observed statistics from the original dataset by a predefined number of times. We showed that this procedure reduced the number of permutations especially when there were many insignificant features. For significant features, we enhanced the reduction with Stouffer's method after splitting datasets. From the simulation study, we found that the enhanced adaptive permutation test dramatically reduced the number of permutations while keeping the precision of the permutation p-value within a small range, when compared to the ordinary permutation test. In real data analysis, we applied the enhanced adaptive permutation test to a genome-wide single nucleotide polymorphism (SNP) dataset of 327,872 features. We found the analysis with the enhanced adaptive permutation took a feasible time for genome-wide omics datasets, and successfully identified features of highly significant p-values with reasonable confidence intervals.

Two sample test for covariance matrices in ultra-high dimension

In this paper, we propose a new test for testing the equality of two population covariance matrices in the ultra-high dimensional setting that the dimension is much larger than the sizes of both of the two samples. Our proposed methodology relies on a data splitting procedure and a comparison of a set of well selected eigenvalues of the sample covariance matrices on the split data sets. Compared to the existing methods, our methodology is adaptive in the sense that (i). it does not require specific assumption (e.g., comparable or balancing, etc.) on the sizes of two samples; (ii). it does not need quantitative or structural assumptions of the population covariance matrices; (iii). it does not need the parametric distributions or the detailed knowledge of the moments of the two populations. Theoretically, we establish the asymptotic distributions of the statistics used in our method and conduct the power analysis. We justify that our method is powerful under weak alternatives. We conduct extensive numerical simulations and show that our method significantly outperforms the existing ones both in terms of size and power. Analysis of two real data sets is also carried out to demonstrate the usefulness and superior performance of our proposed methodology. An R package UHDtst is developed for easy implementation of our proposed methodology.

A non-stationary bias adjustment method for improving the inter-annual variability and persistence of projected precipitation

Hydrological studies depend heavily on environmental variables, such as precipitation and resulting runoff, which exhibit highly seasonal and intermittent behaviour in semiarid basins. In these basins, the use of traditional methods to adjust biases in time series projections can lead to inaccurate results regarding the impacts of climate change. This study introduces a non-stationary bias adjustment methodology (NS) specifically designed for environmental variables characterized by sporadic events and substantial intensity variability, such as precipitation. The methodology involves establishing a probability threshold to adapt the occurrence of precipitation events and utilizes a quantile mapping method based on a non-stationary theoretical and parametric distribution to adjust biases associated with precipitation intensity. The NS method is applied to daily precipitation projections from seven regional climatic models under the RCP 8.5 scenario spanning 2006–2100, alongside historical concurrent data from 1970 to 2005. The present method is compared to the widely used quantile delta mapping approach (QDM), revealing significant differences in performance related to the distribution of precipitation events throughout the year and the behaviour of mean and extreme intensity values. Both approaches show reliable performance, with root mean square errors between the empirical distribution functions of the corrected hindcast time series and the observations being lower than or close to 1 mm for the percentiles smaller than 50th. The error increases with the percentiles, particularly at stations located at higher altitudes. In these locations, QDM doubles or triples the error obtained with the non-stationary approach for percentiles higher than 75th, reaching up to 34 mm. The proposed methodology demonstrates promising potential in reducing uncertainties associated with systematic errors in inter-annual precipitation variability. It is a first step to jointly apply a similar approach to all involved driving variables. Such methods are key to assessing hydrological responses and associated impacts in semi-arid mountainous basins all around the Globe.

Comparison of Prior Distributions for the Heterogeneity Parameter in a Rare Events Meta-Analysis of a Few Studies.

Bayesian meta-analysis is a promising approach for rare events meta-analysis. However, the inference of the overall effect in rare events meta-analysis is sensitive to the choice of prior distribution for the heterogeneity parameter. Therefore, it is crucial to assign a convincing prior specification and ensure that it is both plausible and transparent. Various priors for the heterogeneity parameter have been proposed; however, the comparative performance of alternative prior specifications in rare events meta-analysis is poorly understood. Based on a binomial-normal hierarchical model, we conducted a comprehensive simulation study to compare seven heterogeneity prior specifications for binary outcomes, using the odds ratio as the metric. We compared their performance in terms of coverage, median percentage bias, width of the 95% credible interval, and root mean square error (RMSE). We illustrate the results with two recently published rare events meta-analyses of a few studies. The results show that the half-normal prior (with a scale of 0.5), the prior proposed by Turner etal. for the general healthcare setting (without restriction to a specific type of outcome) and for the adverse event setting perform well when the degree of heterogeneity is not relatively high, yielding smaller bias and shorter interval widths with similar coverage and RMSE in most cases compared to other prior specifications. None of the priors performed better when the heterogeneity between-studies were significantly extreme.

Maximal point-polyserial correlation for non-normal random distributions.

We consider the problem of determining the maximum value of the point-polyserial correlation between a random variable with an assigned continuous distribution and an ordinal random variable with categories, which are assigned the first natural values , and arbitrary probabilities . For different parametric distributions, we derive a closed-form formula for the maximal point-polyserial correlation as a function of the and of the distribution's parameters; we devise an algorithm for obtaining its maximum value numerically for any given . These maximum values and the features of the corresponding -point discrete random variables are discussed with respect to the underlying continuous distribution. Furthermore, we prove that if we do not assign the values of the ordinal random variable a priori but instead include them in the optimization problem, this latter approach is equivalent to the optimal quantization problem. In some circumstances, it leads to a significant increase in the maximum value of the point-polyserial correlation. An application to real data exemplifies the main findings. A comparison between the discretization leading to the maximum point-polyserial correlation and those obtained from optimal quantization and moment matching is sketched.

A Fast Nonparametric Sampling (NPS) Method for Time-to-Event in Individual-Level Simulation Models.

Individual-level simulation models often require sampling times to events, however efficient parametric distributions for many processes may often not exist. For example, time to death from life tables cannot be accurately sampled from existing parametric distributions. We propose an efficient nonparametric method to sample times to events that does not require any parametric assumption on the hazards. We developed a nonparametric sampling (NPS) approach that simultaneously draws multiple time-to-event samples from a categorical distribution. This approach can be applied to univariate and multivariate processes. We discretize the entire period into equal-length time intervals and then derived the interval-specific probabilities. The times to events can then be used directly in individual-level simulation models. We compared the accuracy of our approach in sampling time-to-events from common parametric distributions, including exponential, gamma, and Gompertz. In addition, we evaluated the method's performance in sampling age to death from US life tables and sampling times to events from parametric baseline hazards with time-dependent covariates. The NPS method estimated similar expected times to events from 1 million draws for the three parametric distributions, 100,000 draws for the homogenous cohort, 200,000 draws from the heterogeneous cohort, and 1 million draws for the parametric distributions with time-varying covariates, all in less than a second. Our method produces accurate and computationally efficient samples for time-to-events from hazards without requiring parametric assumptions.

Leveraging the variational Bayes autoencoder for survival analysis

Survival analysis in medical research has witnessed a growing interest in applying deep learning techniques to model complex, high-dimensional, heterogeneous, incomplete, and censored data. Current methods make assumptions about the relations between data that may not be valid in practice. Therefore, we introduce SAVAE (Survival Analysis Variational Autoencoder). SAVAE, based on Variational Autoencoders, contributes significantly to the field by introducing a tailored Evidence Lower BOund formulation, supporting various parametric distributions for covariates and survival time (if the log-likelihood is differentiable). It offers a general method that demonstrates robustness and stability through different experiments. Our proposal effectively estimates time-to-event, accounting for censoring, covariate interactions, and time-varying risk associations. We validate our model in diverse datasets, including genomic, clinical, and demographic tabular data, with varying levels of censoring. This approach demonstrates competitive performance compared to state-of-the-art techniques, as assessed by the Concordance Index and the Integrated Brier Score. SAVAE also offers an interpretable model that parametrically models covariates and time. Moreover, its generative architecture facilitates further applications such as clustering, data imputation, and synthetic patient data generation through latent space inference from survival data. This approach fosters data sharing and collaboration, improving medical research and personalized patient care.

Testing Similarity of Parametric Competing Risks Models for Identifying Potentially Similar Pathways in Healthcare.

The identification of similar patient pathways is a crucial task in healthcare analytics. A flexible tool to address this issue are parametric competing risks models, where transition intensities may be specified by a variety of parametric distributions, thus in particular being possibly time-dependent. We assess the similarity between two such models by examining the transitions between different health states. This research introduces a method to measure the maximum differences in transition intensities over time, leading to the development of a test procedure for assessing similarity. We propose a parametric bootstrap approach for this purpose and provide a proof to confirm the validity of this procedure. The performance of our proposed method is evaluated through a simulation study, considering a range of sample sizes, differing amounts of censoring, and various thresholds for similarity. Finally, we demonstrate the practical application of our approach with a case study from urological clinical routine practice, which inspired this research.

Modeling longitudinal skewed functional data.

This paper introduces a model for longitudinal functional data analysis that accounts for pointwise skewness. The proposed procedure decouples the marginal pointwise variation from the complex longitudinal and functional dependence using copula methodology. Pointwise variation is described through parametric distribution functions that capture varying skewness and change smoothly both in time and over the functional argument. Joint dependence is quantified through a Gaussian copula with a low-rank approximation-based covariance. The introduced class of models provides a unifying platform for both pointwise quantile estimation and prediction of complete trajectories at new times. We investigate the methods numerically in simulations and discuss their application to a diffusion tensor imaging study of multiple sclerosis patients. This approach is implemented in the R package sLFDA that is publicly available on GitHub.

Calculating the Expected Net Benefit of Sampling for Survival Data: A Tutorial and Case Study.

The net value of reducing decision uncertainty by collecting additional data is quantified by the expected net benefit of sampling (ENBS). This tutorial presents a general-purpose algorithm for computing the ENBS for collecting survival data along with a step-by-step implementation in R.The algorithm is based on recently published methods for simulating survival data and computing expected value of sample information that do not rely on the survival data to follow any particular parametric distribution and that can take into account any arbitrary censoring process.We demonstrate in a case study based on a previous cancer technology appraisal that ENBS calculations are useful not only for designing new studies but also for optimizing reimbursement decisions for new health technologies based on immature evidence from ongoing trials.

What Do We Know Without the Catalog? Eliciting Prior Beliefs from Experts for Aftershock Models

Abstract Fitting parametric seismological models to earthquake catalogs often comes with numerical challenges, especially when catalogs are small. An alternative way to quantify parameter values for a seismic region is by eliciting expert opinions on the seismological characteristics that each parameter corresponds to. For instance, expert beliefs on aftershock patterns can be formulated into prior distributions for aftershock parameters, for example, for the epidemic-type aftershock sequence (ETAS) model. We illustrate such a method by not only eliciting priors for ETAS parameters for the Pacific Northwest (PNW), a subduction zone with a complex tectonic environment, but also a relatively small catalog. We compare these priors with those suggested by the ETAS literature for global subduction zones, discussing implications for aftershock forecasting for the PNW.

Bayesian merged utilization of GRAPPA and SENSE (BMUGS) for in-plane accelerated reconstruction increases fMRI detection power

In fMRI, capturing brain activity during a task is dependent on how quickly the k-space arrays for each volume image are obtained. Acquiring the full k-space arrays can take a considerable amount of time. Under-sampling k-space reduces the acquisition time, but results in aliased, or “folded,” images after applying the inverse Fourier transform (IFT). GeneRalized Autocalibrating Partial Parallel Acquisition (GRAPPA) and SENSitivity Encoding (SENSE) are parallel imaging techniques that yield reconstructed images from subsampled arrays of k-space. With GRAPPA operating in the spatial frequency domain and SENSE in image space, these techniques have been separate but can be merged to reconstruct the subsampled k-space arrays more accurately. Here, we propose a Bayesian approach to this merged model where prior distributions for the unknown parameters are assessed from a priori k-space arrays. The prior information is utilized to estimate the missing spatial frequency values, unalias the voxel values from the posterior distribution, and reconstruct into full field-of-view images. Our Bayesian technique successfully reconstructed simulated and experimental fMRI time series with no aliasing artifacts while decreasing temporal variation and increasing task detection power.

Meta-Reinforcement Learning in Nonstationary and Nonparametric Environments.

Recent state-of-the-art artificial agents lack the ability to adapt rapidly to new tasks, as they are trained exclusively for specific objectives and require massive amounts of interaction to learn new skills. Meta-reinforcement learning (meta-RL) addresses this challenge by leveraging knowledge learned from training tasks to perform well in previously unseen tasks. However, current meta-RL approaches limit themselves to narrow parametric and stationary task distributions, ignoring qualitative differences and nonstationary changes between tasks that occur in the real world. In this article, we introduce a Task-Inference-based meta-RL algorithm using explicitly parameterized Gaussian variational autoencoders (VAEs) and gated Recurrent units (TIGR), designed for nonparametric and nonstationary environments. We employ a generative model involving a VAE to capture the multimodality of the tasks. We decouple the policy training from the task-inference learning and efficiently train the inference mechanism on the basis of an unsupervised reconstruction objective. We establish a zero-shot adaptation procedure to enable the agent to adapt to nonstationary task changes. We provide a benchmark with qualitatively distinct tasks based on the half-cheetah environment and demonstrate the superior performance of TIGR compared with state-of-the-art meta-RL approaches in terms of sample efficiency (three to ten times faster), asymptotic performance, and applicability in nonparametric and nonstationary environments with zero-shot adaptation. Videos can be viewed at https://videoviewsite.wixsite.com/tigr.

Confidence-based design optimization using multivariate kernel density estimation under insufficient input data

The uncertainty quantification of the input statistical model in reliability-based design optimization (RBDO) has been widely investigated for accurate reliability analysis, and it could be estimated through its characteristics, cumulative experiences, and available data. However, uncertainty quantification of random variables in existing RBDO studies has exploited parametric distributions quantifying the uncertainty through the Bayes' theorem. In addition, a correlation between random variables is often underestimated due to a lack of knowledge and difficulty to describe the high-dimensional correlation. Hence, it has been a challenge to properly quantify input statistical model and its uncertainty. Therefore, a multivariate kernel density estimation (KDE) is employed to perform data-driven confidence-based design optimization (CBDO) for effective quantification of input model uncertainty. Any assumption on input distribution is not necessary since it is established only with the given input data. Moreover, the input model uncertainty due to insufficient data is quantified using bootstrapping and optimal adaptive bandwidth matrices through the Bayes’ theorem using cross-validation error. Consequently, the proposed CBDO with given input data is capable of finding a conservative optimum of RBDO accounting for both aleatory uncertainty of random variables and epistemic uncertainty induced by a limited number of input data through the multivariate KDE.

Forecasting the design maxima water discharges of floods on the Latorica river according to the data of the Mukachevo gauging station using plotting position formulas

This article presents the results of forecasting design maxima discharges on the Latorica River within Mukachevo town based on hydrological observation data at the “Mukachevo” gauging station using plotting position formulas. While solving the task, a novel non-parametric method of forecasting using observation data is applied. The method includes extrapolating the discrepancy (divergence, disagreement) between the estimates of the statistical annual probabilities of exceedance obtained by different plotting position formulas. The task is considered in the frame of the stationarity hypothesis of the maximum river flow employing a time series of maximal discharges of the Latorica River observed at the “Mukachevo” gauging station from 1947 to 1999.We involved the thirteen plotting position formulas. There was no specific criterion for choosing them to solve the task. All applied formulas were considered admissible options, and results obtained after using them – expert judgments reflecting decision-makers’ predisposition to more cautious or less expensive decision options in flood management strategies.The epistemic uncertainty of the different plotting positions was reduced by employing the Fishburn rule. According to this rule, the significance of various plotting positions was given by arranging their estimates in descending order of importance of their values under decision-making. Depending on the selected significance option assignment of the different plotting position formulas, such rank-weighted estimates of the design peak discharges (each of them for annual exceedance probability 1%, 0.5%, and 0.2%) were computed: (1) the rank-weighted upper bound estimate (sup-estimate) corresponding to the predisposition to more cautious decision options; (2) the rank-weighted lower bound estimate (inf-estimate) corresponding to the predisposition to less expensive decision options. As possible control theoretical alternatives for forecasting design maximal discharges considered were five parametric probability distributions: 1) the Kritskyi-Menkel three-parameter gamma distribution; 2) Pearson’s type III distribution; 3) the Extreme value type I distribution (Gumbell’s type I distribution); 4) the Logarithmic Pearson type III distribution; and 5) the Two-parameters logarithmic-normal distribution. The population statistical parameters for these parametric probability distributions were estimated from the sample statistics by the method of moments.

EXPRESS: Stratified Distributional Analysis - a Novel Perspective on RT Distributions.

Response times and their distributions serve as a powerful lens into cognitive processes. We present a novel statistical methodology called Stratified Distributional Analysis (SDA) to quantitatively assess how key determinants of response times (word frequency and length) shape their distributions. Taking advantage of the availability of millions of lexical decision response times in the English Lexicon Project and the British Lexicon Project, we made important advances into the theoretical issue of linking response times and word frequency by analysing RT distributions as a function of word frequency and word length. We tested these distributions against the lognormal, Wald, and Gamma distributions and three measures of word occurrence (word form frequencies obtained from subtitles and contextual diversity as operationalized as discourse contextual diversity and user contextual diversity). We found that the RT distributions were best described by a lnorm distribution across both megastudies when word occurrence was quantified by a contextual diversity measure. The link between the lnorm distribution and its generative process highlights the power of SDA in elucidating mechanisms that govern the generation of RTs through the fitting of probability distributions. Using a hierarchical Bayesian framework, SDA yielded posterior distributions for the distributional parameters at the single-participant level, enabling probabilistic predictions of response times as a function of word frequency and word length, which has the potential to serve as a diagnostic tool to uncover idiosyncratic features of word processing. Crucially, while we applied our parsimonious methodology to lexical decision response times, it is applicable to a variety of tasks such as word-naming and eye-tracking data.

The Impact of the Approach to Accounting for Age and Sex in Economic Models on Predicted Quality-Adjusted Life-Years.

The method used to model general population mortality estimates in cohort models can make a meaningful difference in appraisals; particularly in scenarios involving potentially curative treatments where a prior National Institute for Health and Care Excellence (NICE) appraisal demonstrated that this assumption alone could make a difference of ~£10,000 to the incremental cost-effectiveness ratio. Our objective was to evaluate the impact of different methods for calculating general population mortality estimates on the predicted total quality-adjusted life expectancy (QALE) as well as absolute and proportional quality-adjusted life year (QALY) shortfall calculations. We employed three distinct methods for deriving general population mortality estimates: firstly, utilizing the population mean age at baseline; secondly, modelling the distribution of mean age at baseline by fitting a parametric distribution to patient-level data sourced from the Health Survey for England (HSE); and thirdly, modelling the empirical age distribution. Subsequently, we simulated patient age distributions to explore the effects of mean starting age and variance levels on the predicted QALE and applicable severity modifiers. Provided sample code in R and Visual Basic for Applications (VBA) facilitates the utilization of individual patient age and sex data to generate weighted average survival and health-related quality of life (utility) outputs. We observed differences of up to 10.4% (equivalent to a difference of 1.01 QALYs in quality-adjusted life-expectancy) between methods using the HSE dataset. In our simulation study, increasing variance in baseline age diminished the accuracy of predictions relying solely on mean age estimation. Differences of -0.30 to 2.24 QALYs were found at a standard deviation of 20%; commonly observed in trials. For potentially curative treatments this would represent a difference in economically justifiable price of -£4,500-+£33,600 at a cost-effectiveness threshold of £30,000 per QALY for a treatment with a 50% cure rate. For lower baseline ages, the population mean method tended to overestimate QALE, whereas for higher baseline ages, it tended to underestimate QALE compared with individual patient age-based approaches. The severity modifier assigned did not vary, however, apart from simulations with means at the extremes of the age distribution or with very high variance. Our analysis underscores the necessity of accounting for the distribution of mean age at baseline, as failure to do so can lead to inaccurate QALE estimates, thereby affecting calculations of incremental costs and QALYs in models, which base survival and quality of life predictions on general population expectations. We would recommend that patient age and sex distribution should be accounted for when incorporating general population mortality in economic models. Provided sufficient sample size, utilizing the observed empirical distribution for the expected population in clinical practice is likely to yield the most accurate results. However, in the absence of patient-level data, selecting a suitable parametric distribution is recommended.

Optimum noise mechanism for differentially private queries in discrete finite sets

The differential privacy (DP) literature often centers on meeting privacy constraints by introducing noise to the query, typically using a pre-specified parametric distribution model with one or two degrees of freedom. However, this emphasis tends to neglect the crucial considerations of response accuracy and utility, especially in the context of categorical or discrete numerical database queries, where the parameters defining the noise distribution are finite and could be chosen optimally. This paper addresses this gap by introducing a novel framework for designing an optimal noise probability mass function (PMF) tailored to discrete and finite query sets. Our approach considers the modulo summation of random noise as the DP mechanism, aiming to present a tractable solution that not only satisfies privacy constraints but also minimizes query distortion. Unlike existing approaches focused solely on meeting privacy constraints, our framework seeks to optimize the noise distribution under an arbitrary (ϵ,δ)\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$(\\epsilon , \\delta )$$\\end{document} constraint, thereby enhancing the accuracy and utility of the response. We demonstrate that the optimal PMF can be obtained through solving a mixed-integer linear program. Additionally, closed-form solutions for the optimal PMF are provided, minimizing the probability of error for two specific cases. Numerical experiments highlight the superior performance of our proposed optimal mechanisms compared to state-of-the-art methods. This paper contributes to the DP literature by presenting a clear and systematic approach to designing noise mechanisms that not only satisfy privacy requirements but also optimize query distortion. The framework introduced here opens avenues for improved privacy-preserving database queries, offering significant enhancements in response accuracy and utility.

A Cost-Effectiveness Analysis of Axicabtagene Ciloleucel versus Tisagenlecleucel in the Treatment of Diffuse Large B-cell Lymphoma Based on a Real-World French Registry.

Axicabtagene ciloleucel (axi-cel) and tisagenlecleucel (tisa-cel) are chimeric antigen receptor T-cell therapies that were evaluated in third and later line (3L+) relapsed or refractory (r/r) diffuse large B-cell lymphoma (DLBCL) in the ZUMA-1 and JULIET trials, respectively. As of October 2021, the DESCAR-T registry included 729 French patients with 3L+r/r DLBCL who received axi-cel or tisa-cel. Using these data, propensity score matching was used to conduct an adjusted comparison between axi-cel and tisa-cel. Axi-cel was associated with statistically significant improvements in overall survival (OS) and progression-free survival (PFS), and significantly more frequent Grade≥3 immune effector cell-associated neurotoxicity syndrome (ICANS), compared with tisa-cel. There was no significant difference in Grade≥3 cytokine release syndrome (CRS). The current analysis assessed the cost-effectiveness of axi-cel versus tisa-cel in the treatment of 3L+r/r DLBCL using propensity score-matched data from the DESCAR-T registry. A partitioned survival model was used to extrapolate costs and quality-adjusted life years (QALYs) over a lifetime. Survival curves for PFS and OS were based on independent mixture cure models fitted to digitized Kaplan-Meier data for the propensity score-matched DESCAR-T populations. Average duration of intensive care unit stays for each of axi-cel and tisa-cel in DESCAR-T were used to inform adverse event costs. Selected parametric survival distributions were based on clinical expert validation. Utility values were derived from ZUMA-1, and costs were obtained from French registries and published sources. List prices were used for both axi-cel and tisa-cel. Costs and outcomes were discounted at an annual rate of 2.5%. Axi-cel is associated with an incremental cost-effectiveness ratio of €15,520 per QALY compared with tisa-cel. Based on explicit willingness-to-pay thresholds applied in Europe, axi-cel is expected to be a cost-effective use of healthcare resources in real-world clinical settings compared with tisa-cel in 3L+r/r DLBCL.