In clinical research, candidate preventive interventions are usually assessed by randomized clinical trials. However, these trials tend to be small, and may be compromised by insufficient power due to a number of factors. One way to possibly circumvent the power loss is by harnessing external "big" data that contain valuable information on the related trials. Although not of the same scientific rigor as that of randomized clinical trials, they still may contain valuable information about disease outcomes caused by the pathogen of interest, patient characteristics, and/or candidate preventive interventions. Taking advantage of this information in the external big data, we develop both regularization and data-driven weighting methods to improve estimation of efficacy and risk prediction in the randomized clinical trials. Among regularization methods, we use L1 and L2 penalized shrinkage methods as well as their combination. We also propose a novel weighted shrinkage estimator, based on first-order approximation and its higher order versions, and compare it with the James-Stein type shrinkage estimator.