Reviewing surveillance videos from nuclear safeguards facilities is a manually-intensive, tedious effort undertaken by nuclear safeguards inspectors who must identify any malicious activity or deviations from established protocols. While deep learning methods have demonstrated the potential for reducing the manual overhead, they require significant computational resources and specialized hardware. Inspectors, however, must often work on-site with the data and are frequently constrained to laptop computers. To meet the lower computational requirements, we extend compression-based analytics for efficient and effective spatial-temporal anomaly detection in video. Compression-based analytics are a class of machine learning algorithms that utilize data compression algorithms. At a high level, data compression algorithms aim to encode data in fewer bits than the original representation by learning and removing statistical redundancy. In the case of video anomaly detection, compression algorithms are used to learn patterns of standard operating activity. Normal operating activity will compress well and events that deviate from the standard operating behavior (i.e., anomalies) will not compress as well. On a variety of surveillance video data sets, we show that our method is competitive with state-of-the-art deep learning methods while requiring only a fraction of the computational resources. By including methods such as compression-base analytics into the Next Generation Surveillance Review tool, we seek to reduce the large burden placed on nuclear safeguards inspectors reviewing surveillance videos.