<p>The ability to diagnose an individual is crucial in promoting treatment and improved health. This paper presents a computationally efficient method to construct diagnostic tests, by identifying optimal cut-points of multiple diagnostic variables in large and complex health surveys. The optimal cut-points are obtained by minimizing information criteria of logistic regression models for survey data, whose diagnostic variables are dichotomized at candidate cut-points. To alleviate the computational burden associated with large survey data, a kriging-based optimization algorithm is used to search for the optimal cut-points. A simulation study based on three sampling designs shows that the proposed method identifies accurately and efficiently the optimal cut-points. The binary variables resulted by dichotomizing the diagnostic variables at the optimal cut-points are used in the construction of diagnostic tests, and further to classification problems in biomedical applications. A cardiometabolic risk application using National Health and Nutrition Examination Survey data is presented. The optimal cut-points of three diagnostic variables are determined and used in the construction of a diagnostic test to identify young people at risk of cardiometabolic disease.</p>