Journal of the Royal Statistical Society Series C-Applied Statistics最新文献

When treating depression, clinicians are interested in determining the optimal treatment for a given patient, which is challenging given the amount of treatments available. To advance individualized treatment allocation, integrating data across multiple randomized controlled trials (RCTs) can enhance our understanding of treatment effect heterogeneity by increasing available information. However, extending these inferences to individuals outside of the original RCTs remains crucial for clinical decision-making. We introduce a two-stage meta-analytic method that predicts conditional average treatment effects (CATEs) in target patient populations by leveraging the distribution of CATEs across RCTs. Our approach generates 95% prediction intervals for CATEs in target settings using first-stage models that can incorporate parametric regression or non-parametric methods such as causal forests or Bayesian additive regression trees (BART). We validate our method through simulation studies and operationalize it to integrate multiple RCTs comparing depression treatments, duloxetine and vortioxetine, to generate prediction intervals for target patient profiles. Our analysis reveals no strong evidence of effect heterogeneity across trials, with the exception of potential age-related variability. Importantly, we show that CATE prediction intervals capture broader uncertainty than study-specific confidence intervals when warranted, reflecting both within-study and between-study variability.

This paper introduces a novel statistical methodology integrating machine learning (ML) and Bayesian modelling to facilitate personalized prognostic predictions with application to oncology. Utilizing power priors, we construct 'patient-similarity embeddings' that identify localized patterns of prognosis. The methodology is applied to study the prognostic value of markers of anticancer immunity within the tumour microenvironment of nonsmall cell lung cancer while adjusting for established clinical characteristics. The method outperforms traditional regression and ML models, while accurately identifying subgroup patterns, thereby enhancing statistical inference and hypothesis testing.

We built models to predict absolute breast cancer (BC) risk in women treated with radiotherapy for Hodgkin lymphoma (HL). We first estimated relative risks (RRs) for risk factors, including radiation dose to 10 breast segments to accommodate heterogeneity of treatment effects, using a case-control sample nested in an HL survivor cohort. To estimate RRs of case-control matching factors we developed novel weighting approaches. We then combined RRs with age-specific BC incidence and competing mortality rates from the HL survivor cohort and a population-based registry, accommodating differences between them. We compared the performance of models using segment-specific doses with using mean dose only.

Respondent-driven sampling (RDS) is a link-tracing sampling design that was developed to sample from hidden populations. Although associations between variables are of great interest in epidemiological research, there has been little statistical work on inference on relationships between variables collected through RDS. The link-tracing design, combined with homophily, the tendency for people to connect to others with whom they share characteristics, induces similarity between linked individuals. This dependence inflates the Type 1 error of conventional statistical methods (e.g. t-tests, regression, etc.). A semiparametric randomization test for bivariate association was developed to test for association between two categorical variables. We directly extend this work and propose a semiparametric randomization test for relationships between two variables, when one or both are continuous. We apply our method to variables that are important for understanding tuberculosis epidemiology among people who smoke illicit drugs in Worcester, South Africa.

Genome-wide association studies (GWAS) have identified thousands of single-nucleotide polymorphisms (SNPs) associated with complex traits, but determining the underlying causal variants remains challenging. Fine mapping aims to pinpoint the potentially causal variants from a large number of correlated SNPs possibly with group structure in GWAS-enriched genomic regions using variable selection approaches. In multi-trait fine mapping, we are interested in identifying the causal variants for multiple related traits. Existing multivariate variable selection methods for fine mapping select variables for all responses without considering the possible heterogeneity across different responses. Here, we develop a novel multivariate Bayesian variable selection method for multi-trait fine mapping to select causal variants from a large number of grouped SNPs that target at multiple correlated and possibly heterogeneous traits. Our new method is featured by its selection at multiple levels, incorporation of prior biological knowledge to guide selection and identification of best subset of traits the variants target at. We showed the advantage of our method over existing methods via comprehensive simulations that mimic typical fine-mapping settings and a real-world fine-mapping example in UK Biobank, where we identified critical causal variants potentially targeting at different subsets of addictive behaviours and risk factors.

The aim of dynamic prediction is to provide individualized risk predictions over time, which are updated as new data become available. In pursuit of constructing a dynamic prediction model for a progressive eye disorder, age-related macular degeneration (AMD), we propose a time-dependent Cox survival neural network (tdCoxSNN) to predict its progression using longitudinal fundus images. tdCoxSNN builds upon the time-dependent Cox model by utilizing a neural network to capture the nonlinear effect of time-dependent covariates on the survival outcome. Moreover, by concurrently integrating a convolutional neural network with the survival network, tdCoxSNN can directly take longitudinal images as input. We evaluate and compare our proposed method with joint modelling and landmarking approaches through extensive simulations. We applied the proposed approach to two real datasets. One is a large AMD study, the Age-Related Eye Disease Study, in which more than 50,000 fundus images were captured over a period of 12 years for more than 4,000 participants. Another is a public dataset of the primary biliary cirrhosis disease, where multiple laboratory tests were longitudinally collected to predict the time-to-liver transplant. Our approach demonstrates commendable predictive performance in both simulation studies and the analysis of the two real datasets.

Survival is poor for patients with metastatic cancer, and it is vital to examine new biomarkers that can improve patient prognostication and identify those who would benefit from more aggressive therapy. In metastatic prostate cancer, 2 new assays have become available: one that quantifies the number of cancer cells circulating in the peripheral blood, and the other a marker of the aggressiveness of the disease. It is critical to determine the magnitude of the effect of these biomarkers on the discrimination of a model-based risk score. To do so, most analysts frequently consider the discrimination of 2 separate survival models: one that includes both the new and standard factors and a second that includes the standard factors alone. However, this analysis is ultimately incorrect for many of the scale-transformation models ubiquitous in survival, as the reduced model is misspecified if the full model is specified correctly. To circumvent this issue, we developed a projection-based approach to estimate the impact of the 2 prostate cancer biomarkers. The results indicate that the new biomarkers can influence model discrimination and justify their inclusion in the risk model; however, the hunt remains for an applicable model to risk-stratify patients with metastatic prostate cancer.

Network meta-analysis is a powerful tool to synthesize evidence from independent studies and compare multiple treatments simultaneously. A critical task of performing a network meta-analysis is to offer ranks of all available treatment options for a specific disease outcome. Frequently, the estimated treatment rankings are accompanied by a large amount of uncertainty, suffer from multiplicity issues, and rarely permit possible ties of treatments with similar performance. These issues make interpreting rankings problematic as they are often treated as absolute metrics. To address these shortcomings, we formulate a ranking strategy that adapts to scenarios with high-order uncertainty by producing more conservative results. This improves the interpretability while simultaneously accounting for multiple comparisons. To admit ties between treatment effects in cases where differences between treatment effects are negligible, we also develop a Bayesian non-parametric approach for network meta-analysis. The approach capitalizes on the induced clustering mechanism of Bayesian non-parametric methods, producing a positive probability that two treatment effects are equal. We demonstrate the utility of the procedure through numerical experiments and a network meta-analysis designed to study antidepressant treatments.

The motivation for this paper is to determine factors associated with time-to-fertility treatment (TTFT) among women currently attempting pregnancy in a cross-sectional sample. Challenges arise due to dependence between time-to-pregnancy (TTP) and TTFT. We propose appending a marginal accelerated failure time model to identify risk factors of TTFT with a model for TTP where fertility treatment is included as a time-varying treatment to account for their dependence. The latter requires extending backwards recurrence survival methods to incorporate time-varying covariates with time-varying coefficients. Since backwards recurrence survival methods are a function of mean survival, computational difficulties arise in formulating mean survival when fertility treatment is unobserved, i.e. when TTFT is censored. We address these challenges by developing computationally friendly forms for the double expectation of TTP and TTFT. The performance is validated via comprehensive simulation studies. We apply our approach to the National Survey of Family Growth and explore factors related to prolonged TTFT in the U.S.

We consider the problem of predicting an individual's identity from accelerometry data collected during walking. In a previous paper, we transformed the accelerometry time series into an image by constructing the joint distribution of the acceleration and lagged acceleration for a vector of lags. Predictors derived by partitioning this image into grid cells were used in logistic regression to predict individuals. Here, we (a) implement machine learning methods for prediction using the grid cell-derived predictors; (b) derive inferential methods to screen for the most predictive grid cells while adjusting for correlation and multiple comparisons; and (c) develop a novel multivariate functional regression model that avoids partitioning the predictor space. Prediction methods are compared on two open source acceleometry data sets collected from: (a) 32 individuals walking on a $1.06$ km path; and (b) six repetitions of walking on a 20 m path on two occasions at least 1 week apart for 153 study participants. In the 32-individual study, all methods achieve at least 95% rank-1 accuracy, while in the 153-individual study, accuracy varies from 41% to 98%, depending on the method and prediction task. Methods provide insights into why some individuals are easier to predict than others.