研究园 · 期刊文章

统计学期刊 2016 年以来的文章,共 15867 篇,按年份、期刊、主题看。想跨年份找某个词,用上面的搜索框。

有原文:本站能直接读全文的(1 项)

按年份

最近收录的 20 篇(每周周报)

JASA Other · 发表于 2026-11-06

Scalable Estimation of Crossed Random Effects Models via Multi-way Discretization

Shota Takeishi, Shonosuke Sugasawa

摘要

Cross-classified data frequently arise in scientific fields such as education, healthcare, and social sciences. A common modeling strategy is to introduce crossed random effects within a regression framework. However, this approach often encounters serious computational bottlenecks, particularly for non-Gaussian outcomes. In this paper, we propose a scalable and flexible method that approximates the distribution of each random effect by a discrete distribution, effectively partitioning the random effects into a finite number of representative values. This approximation allows us to express the model as a multi-way discrete structure, which can be efficiently estimated using a simple and fast iterative algorithm. The proposed method accommodates a wide range of outcome models and remains applicable even in settings with more than two-way cross-classification. We theoretically establish the consistency and asymptotic normality of the estimator under general settings of classification levels. Through simulation studies and real data applications, we demonstrate the practical performance of the proposed method in logistic, Poisson, and ordered probit regression models involving cross-classified structures.

Biometrics Latent Variables & Mixtures Networks & Graphs · 发表于 2026-11-06

Improving interpretation of latent class models for diagnostic tests: distinguishing target condition from measurands using directed acyclic graphs

Nandini Dendukuri, Ian Schiller, Else Margreet Bijker, Michael Libman, Paul Gustafson, Patrick M Bossuyt, Joanna Merckx

摘要

In the absence of a perfect diagnostic test for a target condition, one or more imperfect tests may be used to arrive at a clinical diagnosis. Latent class analysis can be used to model data from imperfect tests with the objective of estimating test accuracy and target condition prevalence. Such models typically assume there are 2 latent classes underlying the observed data—target condition positive and target condition negative. However, as we illustrate, this would be an oversimplification when any of the observed tests has a measurand that is different from the target condition. We show how a directed acyclic graph can be used to illustrate the relationships between the relevant variables—the observed imperfect test results, their latent measurands, the latent target condition, covariates affecting observed or latent variables, and sources of conditional dependence. We then show how to determine the underlying latent classes and their labels. Finally, we examine how consideration of the measurand of each test impacts the likelihood and identifiability of the latent class model. Using simulation studies, we show how ignoring measurands, which are distinct from the target condition, can lead to mislabeling of the latent classes. We illustrate the value of the proposed approach by re-analyzing 2 datasets used in previously published latent class analyses of tests for pediatric tuberculosis and leptospirosis.

SMMR Survival Analysis Causal Inference · 发表于 2026-11-05,收录于 2026-11-06

Self-selection bias in evaluating the effectiveness of mammography screening: A sensitivity analysis on data from the German screening program

Maria Veronica Vinattieri, Laura Buschmann, Jürgen Wellmann, André Karch, Heike Minnerup, Keith Humphreys

摘要

Self-selection bias in observational studies assessing the effectiveness of a breast cancer screening program can arise when non-participants of screening have worse general health conditions or lower compliance with treatments than screening participants. Building on recently described methodology for tumour progression modelling and causal survival analysis of screening data, our main contribution is to describe approaches to conduct sensitivity analyses for self-selection effects when studying screening effects on all-cause survival and breast cancer cumulative incidence, among screen-detected cases, in a population where all individuals are offered screening. Sensitivity analyses are inspired by an approach to correcting for self-selection bias that has been used for observational studies of mortality. We report simulation studies that demonstrate the impact that a correction factor has on all-cause survival and on the cumulative incidence of breast cancer death following diagnosis. We also apply our methods to estimate the benefit of early detection in non-overdiagnosed screen-detected cancers using data from the German mammography screening program. For sensitivity parameter values, we leverage on data on other (non-breast cancer) cause survival among participants and non-participants of mammography screening in the same study and estimate how counterfactual all-cause survival is improved, and breast cancer cumulative incidence is reduced for patients with screen-detected cancers, when accounting for self-selection bias.

Biometrics Latent Variables & Mixtures · 发表于 2026-11-04,收录于 2026-11-06

Scalable finite mixture of regression models for clustering species responses in ecology

Francis K C Hui, Patricia Menéndez, Scott Foster, Skip Woolley

摘要

When modeling species assemblages in ecology, clustering species with similar environmental responses can facilitate a more parsimonious understanding of species compositions and improve the prediction of the overall community. One statistical method that implements this idea is species archetype models (SAMs), a type of finite mixture of regression model where species are grouped according to the shape of their environmental response. However, a current shortcoming of SAMs is their computational complexity, hindering their usefulness in cases involving hundreds or thousands of sites and/or species. In this paper, we propose approximate and scalable SAMs (asSAMs) to overcome this drawback. The key to asSAMs lies in constructing a quadratic approximation to individual species-archetype contributions to the marginal likelihood. This results in an objective function resembling a finite mixture of multivariate normal densities, which can be maximized via fully closed-form iterative updates. We demonstrate how asSAMs also promote fast uncertainty quantification using bootstrapping, and variable selection through an additional regularization penalty. Simulation studies and an application to presence-absence records of over 230 species from the Great Barrier Reef Seabed biodiversity project demonstrate that asSAMs can achieve strong estimation, selection, and predictive performance, while being computationally much faster than current methods for fitting SAMs. This makes asSAMs a practical approach for clustering species' environmental responses in high-dimensional species assemblage datasets.

Biometrics Survival Analysis Testing & Uncertainty Missing Data & Measurement Error · 发表于 2026-11-04,收录于 2026-11-06

Robustness intervals for competing risks analysis with causes of failure missing not at random

Giorgos Bakoyannis, Aristofanis Rontogiannis, Ying Zhang, Wanzhu Tu, Ann Mwangi, Constantin T Yiannoutsos

摘要

Analysis of competing risks data is often complicated by incomplete or selectively missing information on the cause of failure. Standard approaches typically assume that the cause of failure is missing at random (MAR), an assumption that is generally untestable and frequently implausible in observational studies. We propose a novel sensitivity analysis framework for the proportional cause-specific hazards model that accommodates missing-not-at-random (MNAR) scenarios. A sensitivity parameter is used to quantify the association between missingness and the unobserved cause of failure. Regression coefficients are estimated as functions of this parameter, and a simultaneous confidence band is constructed via a wild bootstrap procedure. This allows identification of a range of MNAR scenarios for which effects remain statistically significant; we refer to this range as a robustness interval. The validity of the proposed approach is justified both theoretically, via empirical process theory, and empirically, through simulation studies. We apply the method to the analysis of data from an HIV cohort study in sub-Saharan Africa, where a substantial proportion of causes of failure are missing and the MAR assumption is implausible. The analysis shows that key findings regarding risk factors for care interruption and mortality are robust across a broad spectrum of MNAR scenarios, underscoring the method's utility in situations with MNAR causes of failure.

SMMR Survival Analysis Meta-Analysis & Data Integration · 发表于 2026-11-04,收录于 2026-11-06

Robust transfer learning for accelerated failure time model

Chaoqi Wang, Jie Ding, Xiaoguang Wang

摘要

Limited sample sizes can impair the statistical accuracy of traditional survival models, resulting in potentially unsatisfactory conclusions. To address this issue, we propose a robust transfer learning framework for the accelerated failure time model (RTrans-AFT) that enables efficient and flexible information borrowing from an external source. The method accommodates both individual-level and summary-level source data and models differences in regression coefficients between cohorts without requiring identical covariate distributions. To control the degree of transfer, we impose an adaptive lasso penalty to control the discrepancy between the source and target parameters. This allows the method to selectively borrow information from the source cohort in the presence of heterogeneity, while fully incorporating external information when the two cohorts are comparable. Importantly, the framework is robust to potential data irregularities, such as outliers, which are common in real-world survival datasets. We establish the consistency and asymptotic normality of the proposed estimator. Simulation studies demonstrate that RTrans-AFT substantially improves estimation accuracy and achieves notable variance reduction, leading to enhanced statistical efficiency. The practical utility of the method is further illustrated through an analysis of lung adenocarcinoma data, where the Surveillance, Epidemiology, and End Results (SEER) registry data serve as the external source.

SiM High-Dimensional Causal Inference Testing & Uncertainty · 发表于 2026-11-04,收录于 2026-11-06

Regression Adjusted Target Effects Using Auxiliary Strata Under Design-Based Inference

Zixuan Zhao, Yuehan Yang

摘要

We propose a design-based covariate adjustment that leverages information from auxiliary strata to improve the efficiency of average treatment effect estimation when target samples are sparse. The estimator borrows strength from related strata that may be correlated with the target, yielding gains for studies with limited observations or rare subpopulations. We present the estimator, develop a Neyman-style conservative variance estimator for hypothesis testing and confidence interval construction, and establish its large-sample properties. The procedure is straightforward to implement within standard randomization-based workflows and does not require a correctly specified outcome model or stochastic error assumptions. In simulations, the method delivers lower variance and narrower confidence intervals than conventional design-based or regression-adjusted estimators across a range of sparsity and correlation patterns. Applications to a schooling experiment and the ACTG 175 randomized trial illustrate practical gains in precision.

Biometrics Other · 发表于 2026-11-04,收录于 2026-11-06

Parsimonious modeling of category-specific effects in partial proportional odds models based on trees

Nikolai Spuck, Ingo Gräff, Moritz Berger

摘要

The proportional odds model is arguably the most widely applied regression approach for ordinal response variables. It assumes that the effects of the covariates are constant across the response categories allowing a parsimonious model specification. A violation of this assumption, however, may lead to biased effect estimates and decreased predictive ability. To overcome this issue, partial proportional odds models, where proportionality is only enforced for a smaller subset of covariates, can be applied. While statistical tests to check the proportional odds assumption are available, a generalized framework for data-driven selection of category-specific effects has not been established. In this paper, we introduce a novel tree-structured algorithm that forms clusters of categories sharing the same effect on the ordinal response, with a focus on model parsimony and ease of interpretation. The proposed model appeared to be superior to alternatives in a clinical application to data from the University Hospital Bonn on patients presenting in the emergency department. The predictive ability and selection properties of the proposed approach were also compared to alternative approaches, such as LASSO-penalized regression, in a simulation study.

JRSSB Nonparametric & Semiparametric Bayesian Methods · 发表于 2026-11-04,收录于 2026-11-06

Nonparametric modelling of continuous-time Markov chains with scalable exact and approximate gradients

Filippo Monti, Xiang Ji, Yucai Shao, Marc A Suchard

摘要

Inferring the infinitesimal rates of continuous-time Markov chains (CTMCs) is a central challenge in many scientific domains. This task is hindered by 3 factors: quadratic growth in the number of rates as the CTMC state space expands, strong dependencies among rates, and incomplete information for many transitions. We introduce a new Bayesian framework that flexibly models CTMC rates by incorporating covariates through Gaussian processes. This approach improves inference by integrating new information and deepens our understanding of CTMC stochastic behaviour by shedding light on potential external drivers. Unlike previous approaches limited to linear covariate effects, our method captures complex nonlinear relationships, enabling fuller use of covariate information and more accurate characterization of their influence. For efficient posterior inference, we employ Hamiltonian Monte Carlo. To avoid the prohibitive cost of repeated matrix-exponential derivatives under existing approaches, we develop 2 scalable strategies for exact and approximate gradient computation. Given N observations and K CTMC states, these strategies reduce the cost from O(NK3), with a substantial constant, to O(K3+NK2), with cheaper constants. We demonstrate the method in Bayesian phylogenetic and phylogeographic inference, showing strong performance on synthetic and real datasets and empirical quadratic scaling in K, even when N<K.

Biometrics Other · 发表于 2026-11-04,收录于 2026-11-06

Linear projection inference for Individual Variable Importance

Xinjie Zhou, Zhengbang Li

摘要

Quantifying how the predictive value of biomarkers varies across patient characteristics is central to precision medicine. While Individual Variable Importance (IVI) offers a rigorous framework for this task, existing non-parametric estimators often prove unstable and inefficient in clinical studies with limited sample sizes ($N\lt 500$). To address this, we propose a Linear Projection IVI framework that prioritizes inferential stability over functional flexibility. We formulate the estimator as a first-order approximation to high-dimensional importance surfaces using linear interaction structures. Theoretical results establish that, even under model misspecification, the estimator converges to a well-defined pseudo-true parameter corresponding to the best linear projection of the underlying IVI function. We derive closed-form expressions for local and global importance, prove $\sqrt{N}$-consistency, and introduce a principled "Gatekeeper" hypothesis test. This procedure utilizes a robust joint Wald statistic to control Type I error for heterogeneity detection prior to subgroup stratification, effectively preventing spurious discovery. Simulation studies demonstrate superior finite-sample efficiency relative to kernel-based alternatives and robustness to binary outcome misspecification. Applications to NHANES and Pima Indian diabetes datasets illustrate the framework's utility in distinguishing reproducible biological signals from noise in data-constrained settings.

SMMR Bayesian Methods High-Dimensional Genetics & Omics · 发表于 2026-11-04,收录于 2026-11-06

Improving gene-based analysis of rare variants through Bayesian variable selection

Jingxiong Xu, Nanwei Wang, Yonathan Brhane, Christopher I Amos, Laurent Briollais

摘要

A common approach for detecting rare variants (RVs) associated with complex human diseases is to perform a gene-based analysis or a region-based test of association. Typically, in a gene-based analysis, all the RVs pertaining to the gene considered are included in the statistical test, which potentially reduces its power since most RVs are likely not associated with the outcome of interest. As paradigm shift, we propose to add a variable selection step to choose the RVs that compose the gene-based test statistic as a way to enhance the power of the test. We propose a Bayes factor (BF) test statistic derived from the generalized linear model and its conjugate prior. A key component of our approach is the selection of the important RVs within a gene, which is performed through a birth-death Markov Chain Monte Carlo algorithm. Our approach can also leverage functional annotation at the RV level through the elicitation of the conjugate prior. Through simulation studies, we show that the proposed BF outperformed competing approaches both in terms of gene ranking and power to detect gene-based associations. The power of BF was improved by the use of the variable selection procedure and this gain of power was more substantial when functional annotation was included in this procedure. The application of our new approach to a large whole-exome sequencing data set comparing 1658 individuals with lung cancer to 1492 healthy controls was able to identify new genes associated with lung cancer and pointed towards interesting cancer-related pathways.

SMMR Clinical Trials Bayesian Methods · 发表于 2026-11-04,收录于 2026-11-06

Dynamic borrowing via covariate-adjusted self-adapting mixture priors in clinical trials

Yeonhee Park

摘要

Incorporating historical data into clinical trials can enhance statistical power, but it risks introducing bias if historical and current populations differ. Existing Bayesian borrowing methods often apply a single, prespecified weight to historical data, failing to account for patient-level covariate differences. To address this, we propose a covariate-adjusted self-adapting mixture (CASAM) prior, a Bayesian framework that enables covariate-dependent dynamic borrowing from external data. CASAM adaptively determines the degree of information borrowing by assessing the compatibility between historical and current data conditional on covariates, automatically down-weighting historical information when residual heterogeneity is detected. To further address covariate distribution mismatch between historical and current trials, we develop a propensity-score-integrated implementation of CASAM, which first re-weights historical controls to align their covariate distribution with that of the current control arm before applying the CASAM prior. This layered strategy allows CASAM to focus on structural similarity while the propensity-score adjustment corrects for population-level imbalance. Together, CASAM and its propensity-score-based implementation provide an empirical-Bayes framework for adaptive borrowing that combines covariate alignment with outcome-model compatibility assessment.

Biometrics Latent Variables & Mixtures Genetics & Omics · 发表于 2026-11-04,收录于 2026-11-06

Dissecting microbial community structure and heterogeneity via multivariate covariate-adjusted clustering

Zhongmao Liu, Xiaohui Yin, Yanjiao Zhou, Gen Li, Kun Chen

摘要

In microbiome studies, it is often of great interest to identify clusters or partitions of microbiome profiles within a study population and to characterize the distinctive attributes of each resulting microbial community. While raw counts or relative compositions are commonly used for such analysis, variations between clusters may be driven or distorted by subject-level covariates, reflecting underlying biological and clinical heterogeneity across individuals. Simultaneously detecting latent communities and identifying covariates that differentiate them can enhance our understanding of the microbiome and its association with health outcomes. To this end, we propose a Dirichlet-multinomial mixture regression (DMMR) model that enables joint clustering of microbiome profiles while accounting for covariates with either homogeneous or heterogeneous effects across clusters. A novel symmetric link function is introduced to facilitate covariate modeling through the compositional parameters. We develop efficient algorithms with convergence guarantees for parameter estimation and establish theoretical properties of the proposed estimators. Extensive simulation studies demonstrate the effectiveness of the method in clustering, feature selection, and heterogeneity detection. We illustrate the utility of DMMR through a comprehensive application to upper-airway microbiota data from a pediatric asthma study, uncovering distinct microbial subtypes and their associations with clinical characteristics.

JRSSB Causal Inference · 发表于 2026-11-04,收录于 2026-11-06

Causal inference with dyadic data in randomized experiments

Yilin Li, Lu Deng, Yong Wang, Wang Miao

摘要

Estimating treatment effects in networked settings is a central challenge in online controlled experiments, particularly on social media platforms. We investigate a scenario where the unit-level outcome of interest comprises a series of dyadic outcomes that record pairwise interactions between units, ranging from point-to-point messaging at the microscale to bilateral trade flows at the macroscale. Because the response is defined at the dyadic level, the treatment assigned to one unit can affect the outcomes of all dyads that involve it, inducing a form of network interference. We propose a design-based causal inference framework for randomized experiments with dyadic outcomes. Within this framework, we propose estimators of the global average treatment effect under Bernoulli, complete, and cluster randomization, derive the convergence rates, and establish a central limit theorem for Bernoulli randomization. We further construct a class of variance estimators that are asymptotically conservative and establish asymptotic confidence-interval guarantees under degree and variance-scale conditions. Numerical studies show that the proposed estimators can reduce bias and mean squared error relative to estimators based on unit-level outcomes in a range of finite-sample settings. We illustrate the methods using two large-scale online experiments, evaluating the impact of a recommendation algorithm and a calling feature.

SiM Causal Inference · 发表于 2026-11-04,收录于 2026-11-06

Beyond Pre-Trends: The Case for Routine Placebo Falsification in Difference-in-Differences Health Policy Research

S M Hasan Ul Bari, Marshia Rahman

摘要

No new empirical data were generated. The simulation code (Python) that reproduces Table 1, Figure 1, and Supporting Figure S1 is provided as Supporting Information. Figure S1: Rejection rates (α = 0.05) of the joint Wald test of the five leads, the time-shifted placebo (Periods 4–6 vs. 1–3), and the negative control outcome DiD, as the size of each violation increases. (a) Linear differential drift of 0 to 0.30 error standard deviations per period (Scenario B). (b) Shock coinciding with the policy of 0 to 1.5 error standard deviations (Scenario C). The x-axis shows the resulting bias of the main DiD estimate. Same design as Table 1 (50 units, 25 treated, 10 periods, policy from Period 7, AR(1) errors with ρ = 0.5); 2000 replications per point. The negative control outcome receives the shock in (b) but no drift in (a). In (a), bias = 5 × drift per period (drift 0 to 0.30). The dashed line marks the nominal 5% level; in (b) the two pre-period diagnostics overlap at this level (4.3% to 5.4%). Monte Carlo standard error ≤ 0.011. Supplementary S1: Data S1: Python simulation code reproducing Table 1, Figure 1, and Figure S1. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article. *(abstract via OpenAlex)*

Biometrics Bayesian Methods · 发表于 2026-11-04,收录于 2026-11-06

An efficient Bayesian data augmentation algorithm for continuous-time stochastic epidemic models fit to partially observed incidence data

Haoyu Wu, David A Stephens, Erica E M Moodie

摘要

Statistical modeling of infectious disease outbreaks with incidence data is crucial to understanding epidemic dynamics and informing public health interventions. Stochastic epidemic models (SEMs) account for the inherent stochasticity in epidemic evolution and are particularly useful in small outbreaks or low-prevalence settings. In SEMs, the epidemic is naturally represented as a continuous-time Markov jump process (MJP), while surveillance data are collected at discrete times and subject to underdetection. Fitting SEMs to partially observed epidemic trajectories requires integrating over all possible transition paths, which is often intractable. Individual-level data augmentation (DA) methods have been developed to address these challenges. However, existing approaches either rely on per-subject updates with limited scalability or impose typically unrealistic assumptions to simplify the complex dimensional change problem for state-space SEMs. We present an efficient DA Markov chain Monte Carlo (MCMC) framework for fitting continuous-time state-space SEMs to incidence data. Our method uses a block sampler to jointly update individual trajectories and proposes transition times using a discrete-time approximation to the latent MJP. Through simulation studies, we demonstrate that our method significantly improves MCMC mixing and scalability compared to the previous DA approach and remains computationally robust in scenarios where particle MCMC algorithms struggle. We apply the method to the 2014 Ebola outbreak in a rural village in Sierra Leone and the 2009 H1N1 outbreak in a remote community in Canada.

JASA Testing & Uncertainty · 发表于 2026-11-04,收录于 2026-11-06

ACS: An interactive framework for conformal selection

Yu Gui, Ying Jin, Yash Nair, Zhimei Ren

摘要

This paper presents adaptive conformal selection (ACS), an interactive framework for model-free selection with guaranteed error control. Building on conformal selection (Jin and Candès, 2023b), ACS generalizes the approach to support human-in-the-loop adaptive data analysis. Under the ACS framework, we can partially reuse the data to boost the selection power, make decisions on the fly while exploring the data, and incorporate new information or preferences as they arise. The key to ACS is a carefully designed principle that controls the information available for decision making, allowing the data analyst to explore the data adaptively while maintaining rigorous control of the false discovery rate (FDR). Based on the ACS framework, we provide concrete selection algorithms for various goals, including model update/selection, diversified selection, and incorporating newly available labeled data. The effectiveness of ACS is demonstrated through extensive numerical simulations and real-data applications in large language model (LLM) deployment and drug discovery. *(abstract via OpenAlex)*

SiM Clinical Trials · 发表于 2026-11-03,收录于 2026-11-06

Sample Size and Power Calculation for Continuous Outcomes in Open-Cohort Stepped Wedge Designs With Multiple Interventions

Shiyu Feng, Meng Zheng, Yi Zhang, Xuezhi Liang, Minglu Yuan, Wen Chen

摘要

Stepped wedge designs with multiple interventions (M-SWDs) are increasingly used to evaluate complex implementation strategies. In many such trials, participants enter and leave over time, giving rise to open-cohort sampling and complex correlation structures. Existing sample size and power calculation methods are largely restricted to single-intervention designs, closed cohorts, or simplified correlation assumptions. We develop a unified, parameter-based framework for sample size and power calculation in open-cohort M-SWDs. The proposed methodology accommodates several common multi-intervention designs, including concurrent, cross-over, supplementation, and factorial designs. Flexible open-cohort sampling schemes are incorporated to account for participant turnover, and both block-exchangeable and decay correlation structures are considered to capture within-cluster and within-individual dependence over time. Closed-form and generalized variance expressions for intervention effect estimators are derived, enabling efficient power calculation under a wide range of design settings. Extensive simulation studies demonstrate that the proposed formulas provide accurate power estimates, while further power analyses illustrate how the key correlation parameters and participant retention jointly influence power. This work provides practical and theoretically grounded guidance for planning multi-intervention stepped wedge trials under open-cohort sampling, facilitating more realistic and transparent design-stage decision-making.

Biometrika Testing & Uncertainty · 发表于 2026-11-03,收录于 2026-11-06

Optimal e-values for testing the mean of a bounded random variable against a composite alternative

S Arnold, E Clerico

摘要

We derive explicitly the e-values with optimal (relative) growth rate in the worst case for testing the mean of a bounded random variable, thereby providing the first application of the (RE)GROW quality criteria beyond the assumption of mutually absolutely continuous hypotheses for e-values originally proposed by Grünwald et al. (2024). For both criteria, we explicitly characterise the alternatives that are most difficult to test against and show that they admit a meaningful interpretation. We give two important examples in which REGROW provides a powerful quality criterion to choose optimal e-variables whereas GROW leads to trivial solutions.

SiM Genetics & Omics · 发表于 2026-11-03,收录于 2026-11-06

Factor-Augmented Neural Conditional Density Estimation with Application to Drug Response Prediction

Weineng Zhu, Xiaoyu Sheng

摘要

High-dimensional pharmacogenomic data pair thousands of molecular features with hundreds of cancer cell lines, complicating conditional density estimation. We study factor-augmented CINDES, which estimates conditional response densities from principal-component scores through binary classification. For an honest pilot/main-sample construction, assuming fresh-score PCA recovery, we derive a risk bound separating conditional-density learning, PCA score error, and representation bias. This guarantee does not cover the implemented procedure, which reuses observations for PCA and CINDES. In favorable strong-factor simulations, PCA improved a fixed CINDES pipeline, but PCA-MDN achieved lower negative log-likelihood (NLL) and integrated L 1 error in the matched comparison. Stress tests showed losses from omitted response-relevant low-variance directions and weak factor recovery, with sample-size-dependent factor-count effects. Across all 692 eligible source-specific genomics of drug sensitivity in cancer (GDSC) units and five splits, mean held-out NLL was 1.5414 for PCA-CINDES, versus 1.5661 and 1.5557 for ElasticNet and PCA/Ridge Gaussian models; PCA-MDN was better at 1.5156. A previously fixed 35-unit audit favored PCA-CINDES over Raw-X CINDES (mean paired NLL difference - 0 . 178 ) and found lower measured computational cost. The evidence supports PCA compression when retained factors capture response information and are accurately estimated; it does not establish robustness to general nonlinear or nonfactor structures, or CINDES as the preferred second-stage estimator. The application concerns preclinical prediction rather than clinical utility.