Erica E. M. Moodie

Dynamic treatment regimes (DTRs) formalize sequences of treatment decision rules that adapt to a patient’s evolving clinical status. In observational studies, estimation of the value of a DTR (i.e., the expected population-outcome of a treatment strategy) using inverse probability weighting (IPW) and augmented IPW can be unstable when strict regime compatibility yields few valid participant trajectories, leading to highly variable weights and noisy estimates. In this paper, we introduce two adaptive weighting frameworks: generalized compatibility weighting (GCW) and bootstrap compatibility windowing (BCW), that allow near-compatible trajectories to contribute information in limited samples, with diminishing information-borrowing in larger samples. We illustrate how these estimators reduce noise in estimation and stabilize empirical optimal treatment threshold identification. Using data from adults living with HIV in one site of the TREAT Asia HIV Observational Database in India observed from 2013-2024, we evaluate clinically interpretable threshold-based decision rules based on baseline CD4 count and weight. In this application, GCW and BCW produced more stable estimated regime values than standard IPW-based approaches, with reduced variability across candidate thresholds and more reliable identification of empirically optimal treatment rules. These methods provide practical tools for improving treatment rule evaluation in observational studies.

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.

Irregular observation times are common in longitudinal observational studies and can affect causal inferences. We use data from electronic health records of Kaiser Permanente Washington patients in the United States who initiated an antidepressant medication between 2008 and 2018, with a confirming diagnosis of depression. We are interested in the estimation of the average causal effect of citalopram compared with fluoxetine on the severity of depression symptoms measured via the Patient Health Questionnaire. The score on the Patient Health Questionnaire is measured irregularly during follow‐up, and its observation is associated with patient characteristics. We describe two approaches, one based on imputation and one based on inverse weighting, to address observation irregularity in these data and obtain a consistent causal treatment effect. The approaches are compared in large simulation studies and applied to the Kaiser Permanente Washington data. Advantages and pitfalls of these methods are discussed.

Background: Sequential multiple adaptive randomized treatment designs can generate registrational-quality evidence for the contribution of phase in perioperative oncology by preserving randomized comparisons and intent-to-treat estimation while improving efficiency relative to traditional three-arm trials, and yet they remain underappreciated. We present a sequential multiple adaptive randomized treatment design framework for perioperative regimens with neoadjuvant and adjuvant phases, motivated by perioperative immunotherapy in resectable non-small-cell lung cancer, where the benefit of neoadjuvant therapy is hypothesized to be greater than that of adjuvant therapy. Methods: We define analyses for perioperative-versus-control, neoadjuvant-only-versus-control, and contribution of phase (incremental adjuvant benefit). We specify intent-to-treat analysis sets and describe unbiased estimation. Results: In a hypothetical trial calibrated to a perioperative setting, a sequential multiple adaptive randomized treatment design increases the number of events for regimen-versus-control comparisons and isolates a randomized comparison for the contribution of phase at the second randomization. Compared with a traditional multi-arm design, simulations demonstrate controlled family-wise type I error (<0.05) and higher power. Conclusions: Sequential multiple adaptive randomized treatment designs provide a principled, practical approach to within-trial contribution of phase in confirmatory oncology. Sequential multiple adaptive randomized treatment designs offer an efficient approach that addresses regulatory concerns and provides valid and robust inference. Because sequential multiple adaptive randomized treatment designs formalize the iterative nature of clinical decision-making, they are uniquely poised to explicitly answer questions related to overtreatment or undertreatment.

Early-phase, personalized dose-finding trials for combination therapies seek to identify patient-specific optimal biological dose (OBD) combinations, which are defined as safe dose combinations that maximize therapeutic benefit for a specific covariate pattern. Given the small sample sizes that are typical of these trials, it is challenging for traditional parametric approaches to identify OBD combinations across multiple dosing agents and covariate patterns. To address these challenges, we propose a Bayesian optimization approach to dose-finding that incorporates efficacy and toxicity information into the sequential search strategy. Independent Gaussian processes are used to model the efficacy and toxicity surfaces, and an acquisition function is utilized to define the dose-finding strategy. Furthermore, we define an adaptive stopping rule using the posterior entropy for the location of the OBD. This work is motivated by a personalized dose-finding trial which considers a dual-agent therapy for obstructive sleep apnea (OSA), where OBD combinations are tailored to OSA severity. Via a simulation study, the approach is first investigated across varying degrees of response heterogeneity for both efficacy and toxicity, and then a collection of final designs for the OSA trial are compared. We demonstrate that the proposed approach toward personalized dose-finding yields good performance under the considered scenarios.

Risk of suicide attempt varies over time. Understanding the importance of risk factors measured at a mental health visit can help clinicians evaluate future risk and provide appropriate care during the visit. In prediction settings where data are collected over time, such as in mental health care, it is often of interest to understand both the importance of variables for predicting the response at each time point and the importance summarized over the time series. Building on recent advances in estimation and inference for variable importance measures, we define summaries of variable importance trajectories and corresponding estimators. Under common regularity conditions, the same approaches for inference can be applied to these measures regardless of the choice of the algorithm(s) used to estimate the prediction function under standard convergence conditions. We propose a nonparametric efficient estimation and inference procedure as well as a null hypothesis testing procedure that are valid even when complex machine learning tools are used for prediction. Through simulations we demonstrate that our proposed procedures have good operating characteristics. We use these approaches to analyze electronic health records data from two large health systems to investigate the longitudinal importance of risk factors (i.e., the importance of risk factors over time) for suicide attempt to inform future suicide prevention research and clinical workflow.

Data from the REPRIEVE trial have been used to study the association between treatment with abacavir and cardiovascular disease in people with HIV. Past exposure to abacavir before the trial was found to be just as risky as exposure during the trial. This calls into doubt the widely accepted hypothesis that recent exposure to abacavir increases the risk of cardiovascular disease but that risk rapidly attenuates after exposure ends. The evidence for this hypothesis is weak while evidence for an alternative hypothesis has been neglected. An alternative hypothesis is that risk from exposure to abacavir is not immediate but increases as exposure cumulates, plateauing after about three years. Confusion over the likely effect of abacavir has arisen partly because overly simplistic methods have been used to model the exposure outcome relationship in observational cohort data. Better methods are available and should be used in future to avoid misdirected research effort.