Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


RanranDong_PhDdissertation.pdf (1.56 MB)

ETD Abstract Container

Abstract Header

Stepped Wedge Cluster Randomized Controlled Trials for Three-Level Data: Design and Evaluation

Dong, Ranran
http://rave.ohiolink.edu/etdc/view?acc_num=osu1514492341281384

Abstract Details

2018, Doctor of Philosophy, Ohio State University, Biostatistics.
Cluster randomized trials (CRTs) are designed to randomly allocate groups of participants rather than individuals to either the intervention or the control. As an increasingly popular type of CRT, the stepped wedge CRT (SW-CRT) is a one-way crossover design where the intervention is provided sequentially to clusters whose orders are randomly determined. In this dissertation, we propose novel SW-CRTs for three-level data such as patients (observation level) within wards (unit level) within hospitals (cluster level). The proposed designs differ in timing of allocating units within the same cluster to different treatments. We evaluate the efficiency of each design under a variety of underlying models generating three-level data. Impacts of misspecifying random unit effects and ignoring contamination on inference about the intervention effect are also evaluated via simulation studies. We derive the closed-form expression for variance of the intervention effect estimator under a standard three-level model incorporating constant random unit effect across time. The formula is flexible and can accommodate a wide variety of three-level CRTs. Using the variance formula, we compute and compare the efficiencies across all of our proposed three-level SW-CRTs under various scenarios. Results show that when there is no contamination, design 4 transferring units from the same cluster at all time points is most efficient, and design 1 transferring units from the same cluster at a single time point is least efficient. We then extend the standard three-level model by including varying random unit effects across time. Under the extended model, the order of efficiency among the proposed designs does not change. For our proposed designs, we study the impact of model misspecification on inference about the intervention effect. We fit the standard model to data generated from the aforementioned extended model. Results show minimal influence on bias of the treatment effect estimator, but potentially low coverage probabilities for the treatment effect under all designs. Other studies we conduct include incorrectly assuming a random unit effect when none are present and incorrectly omitting a random unit effect when it truly exists. In these two cases, there is minimal impact on inference about the intervention effect under all designs. We also address the problem of contamination for the proposed three-level SW-CRTs in which units from the same cluster may contaminate each other. Under each design, we consider practical scenarios where contamination could occur, describe the severity of contamination, and evaluate the impact of ignoring contamination when modeling in the presence of contamination. Our numerical studies show that designs 2 and 3 transferring units from the same cluster within two steps can still provide valid inference about the intervention effect in the presence of mild to moderate contamination, and design 4 should not be preferred even when contamination is mild.
Abigail Shoben (Advisor)
Rebecca Andridge (Committee Member)
Eloise Kaizar (Committee Member)
Michael Pennell (Committee Member)
151 p.
Biostatistics

Recommended Citations

Citations

  • Dong, R. (2018). Stepped Wedge Cluster Randomized Controlled Trials for Three-Level Data: Design and Evaluation [Doctoral dissertation, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu1514492341281384

    APA Style (7th edition)

  • Dong, Ranran. Stepped Wedge Cluster Randomized Controlled Trials for Three-Level Data: Design and Evaluation. 2018. Ohio State University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=osu1514492341281384.

    MLA Style (8th edition)

  • Dong, Ranran. "Stepped Wedge Cluster Randomized Controlled Trials for Three-Level Data: Design and Evaluation." Doctoral dissertation, Ohio State University, 2018. http://rave.ohiolink.edu/etdc/view?acc_num=osu1514492341281384

    Chicago Manual of Style (17th edition)

osu1514492341281384
130
© 2018, all rights reserved.
This open access ETD is published by The Ohio State University and OhioLINK.