Screening for a binary characteristic, for example, an infectious disease, group testing, where specimens are tested in groups initially, has been shown can reduce substantial cost. In the past, different group testing protocols were designed in order to archive more precision prevalence estimates. In this article, we proposed a two-step adaptive grouping strategy from the perspective of estimation intending for more efficient regression estimators. Rather than random pooling, we construct the groups of the individual with similar covariates information by the use of Fisher information. Our research shows that the regression performance under proposed informative grouping beats the individual testing and group testing with random pooling in terms of generalized variance.