Learning analytics have the potential to improve teaching and learning in K-12 education, but as student data is increasingly being collected and transferred for the purpose of analysis, it is important to take measures that will protect student privacy. A common approach to achieve this goal is the de-identification of the data, meaning the removal of personal details that can reveal student identity. However, as we demonstrate, de-identification alone is not a complete solution. We show how we can discover sensitive information about students by linking de-identified datasets with publicly available school data, using unsupervised machine learning techniques. This underlines that de-identification alone is insufficient if we wish to further learning analytics in K-12 without compromising student privacy.