Nested Gaussian process modeling for high-dimensional data imputation in healthcare systems

Surgical operation involves multifarious elements with a great level of complexity. To cope with this issue and enhance information visibility, modern healthcare systems are increasingly investing in advanced sensing and information technology which gives rise to data-rich environments in hospitals. However, it is not uncommon to encounter data uncertainty and incompleteness. This adversely affects the process of healthcare decisions (e.g., multiple attributes with missing values among describing a patient's health conditions). There is a dire need to go beyond current clinical practice and develop data-driven methods that will enable the extraction of pertinent knowledge about clinical status from heterogeneous healthcare recordings. This paper presents a novel nested Gaussian process for high-dimensional tensor data imputation. We evaluate and validate the proposed methodology with real tensor data with high-level of missing information from intensive care units. Experimental results show that the proposed method effectively handles the imputation of incomplete tensor data under uncertainty, further improves the effectiveness of biomarker extraction, patient monitoring, and decision making. The nested Gaussian process method shows great potentials for general application in many of other disciplines that need high-dimension data imputation and analytics.