1. Methods for Cluster Randomized Trials
Cluster randomized trials are an attractive research design for evaluating the effect of interventions that target improved health care through modification to the way that providers, clinics, or entire organizations work to treat disease and prevent illness. This design has been increasingly adopted in embedded pragmatic clinical trials (ePCTs) and enables a rigorous evaluation of the impact of interventions that apply simultaneously to aggregate groups of subjects. I have co-authored a pair of review articles that summarized methodological development in this field (Turner et al., 2017a, 2017b AJPH). Example topics of my research include:
- Constrained randomization: Covariate-constrained randomization effectively eliminates “bad-luck” allocation of clusters to treatment conditions according to prespecified balance metrics. I have addressed the implications of constrained randomization for designing and analyzing small cluster randomized trials (Li et al., 2015; 2016 Stat Med) and with multiple treatment conditions (Zhou et al. 2022 Stat Med). I have also participated in the development of R and Stata software to assist their practical implementation (Yu et al. 2019 R Journal; Gallis et al. 2019 Stata J).
- Stepped wedge designs: Stepped wedge designs roll out interventions to clusters in a staggered fashion, and have received increasing attention in health services and implementation science research. I have contributed a body of literature for stepped wedge designs including using population-averaged models for design and analyses (Li et al. 2018 Biometrics; Li 2020 Stat Med; Li et al. 2021 Biostatistics), optimal allocation (Li et al. 2018 Stat & Prob Letters), complex clustering (Davis-Plourde et al. 2021 Biometrics) and unequal cluster sizes (Tian et al. 2021 Biom J). I have also provided technical and non-technical introductions to the recent methodological advancements on stepped wedge trials (Li et al. 2021 SMMR; Li and Wang 2022, World Neurosurgery).
- Treatment effect heterogeneity: In ePCTs, the flexible inclusion of a range of clusters and patients to mimic real-world practice necessarily induces heterogeneity in response to group-level interventions. I have contributed methods for planning cluster randomized trials to power the detection of confirmatory treatment effect heterogeneity (Yang et al. 2021 Stat Med; Tong et al. 2022 Stat Med). I am also Principal Investigator of a PCORI Methodology Award that supports the expansion of this statistical methodology to multilevel and longitudinal cluster randomized trials.
- Complex endpoints: I am interested in methods for designing cluster randomized trials to accommodate multivariate endpoints (Yang et al. 2022+ arXiv) and right-truncated count outcomes (Li and Tong, 2021 Biom J). My recent interest also includes the development and dissemination of methods for designing and analyzing cluster randomized trials with censored survival (Wang et al. 2022+ arXiv) and competing risks outcomes (Li et al. 2022 SMMR; Chen and Li, 2022 Stat Med; Chen et al., 2022+ arXiv).
2. Methods for Causal Inference
Many biomedical research questions involve leveraging randomized trials or observational studies for comparative evaluation of treatments, and are causal in nature. The increasing utility of real-world observational data sources for unbiased treatment comparisons necessitate rigorous statistical methods to address confounding, selection bias and measurement bias problems. Examples topics of my research include:
- Propensity score methods: I have been developing and disseminating novel propensity score weighting estimators for improved treatment comparisons, including the utility of overlap weights to address weak positivity (Li et al. 2019 AJE; Li 2020 Stat Sci), generalizations to multiple treatments (Li and Li 2019 AOAS), censored survival data (Cheng et al. 2022 AJE; Zeng et al. 2023 Stat Sinica), and covariate adjustment in randomized trials (Zeng et al. 2020 Stat Med). An R package PSWeight has been developed to assist the implementation of novel propensity score weighting methods (Zhou et al. 2022 R Journal).
- Generalizability and transportability: Comparative results from randomized trials may not be directly generalizable to a target population of substantive interest when, as in most cases, trial participants are not randomly sampled from the target population. I have developed methods to generalize or transport trial results to external target populations under a nested design (Li et al. 2022 Commun Stat) and a non-nested design (Li et al. 2022 JRSSC).
- Machine learning: Advancements in machine learning have provided important tools to flexibly approximate the data generating processes to robustly estimate the average causal effects, and possibly capture heterogeneous individualized causal effects. I have acquired interests in comparative evaluation of causal machine learning (Hu et al. 2021 Stat Med) and their extensions to clinical settings with complex endpoints (Chen et al. 2022+ arXiv).
3. Causal Inference in Cluster Randomized Trials
The increasing popularity of cluster randomized designs in pragmatic trial settings present unprecedented opportunity for methodological research. I have a keen interest in bridging the gap between causal inference methods and methods for embedded pragmatic clinical trials. I have studied issues on causal estimands (Wang et al. 2022 Contemp Clin. Trials), selection bias due to enrollment (Li et al. 2021 Clinical Trials) and model-robust causal inference in cluster randomized trials (Wang et al. 2022+ arXiv).
I am interested in various trial designs, including but not limited to 4-level trials (Wang et al 2021 Biom J), hierarchical 2-by-2 factorial trials (Tian et al. 2021 Stat Med), cluster randomized crossover trials (Li et al 2018 Stat Med), individually randomized group treatment trials (Tong et al 2021 Clinical Trials) and patient preference trials (Wang et al 2022 SMMR). I have also studied missing data (Turner et al 2019 SMMR) and measurement error correction methods (Chen et al 2021 SMMR). I’ve recently become interested in causal mediation (Cheng et al. 2021 BMC Med Res Methodol; Cheng et al. 2022 AJE).
Beyond methodological research, I have a broad interest in collaborative research, including cardiology, geriatrics, nephrology, internal medicine, implementation science and pragmatic trials research in general.