A new Statistical method to balance Social Network Effect and Covariate Effect in Response Adaptive Design

Lead Investigator: Yang Li, School of Statistics, Renmin University of China
Title of Proposal Research: A new Statistical method to balance Social Network Effect and Covariate Effect in Response Adaptive Design
Vivli Data Request: 9433
Funding Source: None
Potential Conflicts of Interest: None

Summary of the Proposed Research:

Traditional Randomized Control Trials (RCT) divide patients evenly between a trial group (e.g., receiving a new drug) and a control group (e.g., receiving standard therapy or placebo). A different method of assigning patients to treatment groups in clinical trials is Response adaptive randomization (RAR). RAR aims to assign more patients to the better treatment based on patient responses already accrued in the trial. Thus, RAR is more ethical than RCT, especially for those who would have been allocated to the group receiving the inferior treatment.

In real clinical trials other factors may affect the outcomes of the trial, causing an imbalance between different groups, resulting in an inaccurate estimate of the final drug effect.

One factor is the patient’s network effect (i.e., the patient’s social network relationship, for example if patient A and patient B are friends), where for example in a routine vaccine trials, if the vaccine works, a person with a friend in the vaccine group is less likely to develop the infectious disease than a person with a friend in the placebo group, because they are less likely to be exposed to the source of infection.

Another factor is the patient’s covariate effect (i.e., the patient’s characteristics, such as gender, age, ethnicity, etc.), where if the treatment group is all men and the placebo group is all women, it is difficult to tell whether the drug really works or the drug only works for one sex.

To solve this problem, we are introducing a new randomization method, Network and covariate adjusted response adaptive randomization(NCARAR), which adaptively assigns patients in a sequential manner (one by one, or group by group) based on the current level of imbalance (including network and covariate imbalances), the outcomes of previous subjects (e.g. cured or uncured after treatment), and the characteristics (e.g. gender, age of the patient) and social network relationships (e.g. the number of friends in different groups) of the patients to be assigned to a group.

This method has demonstrated significant advantages over traditional methods in terms of network and covariate balance, and estimation accuracy, and has been widely used in causal inference (A methodology for studying causal relationships between things) and clinical trials.
We require the requested clinical trial data to validate the effectiveness of our approach.

Requested Studies:

A Study to Evaluate the Safety and Efficacy of A/California/7/2009 (H1N1)V-like Vaccines GSK2340274A and GSK2340273A in Children Aged 6 Months to Less Than 10 Years of Age
Data Contributor: GSK
Study ID: NCT01051661
Sponsor ID: 114000