{"doi":"10.7302/1330","title":"Developing New Statistical Methods for Challenges in Evaluating Dynamic Treatment Regimes","abstract":"Personalized medicine is built upon the understanding that patients are uniquely heterogeneous in their existing and emergent comorbidities, as well as their tolerance of, response to, and even preference for different treatments. Dynamic treatment regimes (DTRs) lead to personalized medicine through a series of stage-specific decision rules that map a patient's up-to-date individual characteristics, including treatment history and disease state, to a tailored treatment assignment at each successive treatment stage. In this dissertation we develop new statistical methods to overcome several challenges arising in the field of dynamic treatment regimes. In Chapter II, we develop Restricted Sub-Tree Learning (ReST-L) to estimate optimal DTRs in a multi-stage multi-treatment setting using observational data while restricting estimated DTRs to include only the set of covariates considered to be meaningful tailoring variables. ReST-L uses a purity measure derived from the augmented inverse probability weighted estimator for the counterfactual mean outcome; it is able to correctly estimate the optimal underlying dynamic treatment regime for a relatively large number of covariates with comparatively small sample sizes. We demonstrate the utility of ReST-L in a study of treatment recommendations for patients presenting to the emergency department with traumatic amputation of digit(s) on the hand. Chapter III is motivated by a clustered sequential multiple assignment randomized trial (Clustered SMART) designed to improve the clinic-level uptake of evidence-based practices and health outcomes of patients with mood disorders. We develop estimation and inference procedures for Clustered Q-learning to inform the empirical construction of an optimal clustered adaptive intervention and address the well-known non-regularity challenge that can occur in a multi-stage estimation setting. We show that estimates of model parameters are unbiased and demonstrate near nominal coverage of confidence intervals across two intervention stages when the number of clusters is large and sample sizes within clusters are moderate. We apply Clustered Q-Learning to data from a Clustered SMART to evaluate whether a set of candidate tailoring variables may be used to additionally tailor cluster-level interventions to improve patient-level outcomes of patients with mood disorders. We develop Penalized Spline-Involved Tree-based (PenSIT) Learning in Chapter IV, which seeks to improve upon existing tree-based approaches to estimating an optimal multi-stage multi-treatment DTR. Instead of using the estimated propensity score to construct the inverse weighting, which may result in unstable estimates when weights are large, we predict missing counterfactual outcomes using regression models that incorporate a penalized spline of pre-transformed propensity scores, as well as other covariates predictive of the outcome. Our simulation results demonstrate good performance of PenSIT Learning across different scenarios, particularly when the level of confounding is high or moderate, or the sample size is small. We apply PenSIT Learning to a retrospectively-collected dataset to estimate a two-stage fluid resuscitation strategy to minimize a measure of organ dysfunction in patients with acute emergent sepsis.","journal":"Deep Blue (University of Michigan)","year":2021,"id":229848,"datarank":0.0,"base_score":0.0,"endowment":0.0,"self_citation_contribution":0.0,"citation_network_contribution":0.0,"self_endowment_contribution":0.0,"citer_contribution":0.0,"corpus_percentile":null,"corpus_rank":null,"citation_count":0,"citer_count":0,"citers_with_citation_signal":0,"citers_with_endowment":0,"datacite_reuse_total":0,"is_dataset":false,"is_dataset_confidence":0.9538,"is_data_producer":false,"deposit_databanks":null,"is_oa":true,"file_count":0,"downloads":0,"has_version_chain":false,"published_date":"2021-01-01","fair_score":null,"fair_percentile":null,"algorithm_id":"datarank_citation_only_1hop_v6","ranking_scope":"data_only","authors":[{"id":834583,"name":"Kelly Speth","orcid":"0000-0001-8045-8406","position":0,"is_corresponding":true}],"reference_count":0,"raw_metadata":null,"created_at":"2026-07-18T23:55:01.675482Z","pmid":null,"pmcid":null,"fwci":null,"citation_percentile":null,"influential_citations":0,"oa_status":null,"license":null,"views":0,"total_file_size_bytes":0,"version_count":0,"fair_f":null,"fair_a":null,"fair_i":null,"fair_r":null,"fair_zscore":null,"fair_rationale":null,"fair_model":null,"fair_agent_version":null,"fair_fulltext_source":null,"fair_has_llm":null,"fair_computed_at":null,"clinical_trials":[],"software_tools":[],"db_accessions":[],"linked_datasets":[],"topics":[]}