The multivariate linear-time subset scan (MLTSS) extends previous spatial and subset scanning methods to achieve timely and accurate event detection in massive multivariate datasets, efficiently optimizing a likelihood ratio statistic over proximity-constrained subsets of locations and all subsets of the monitored data streams. However, some disease outbreaks may only affect a subpopulation of the monitored population (age group, gender, individuals engaging in a specific high-risk behavior, etc.), and MLTSS is unable to use this additional information to enhance detection ability.
Objective
We present Multidimensional Subset Scan (MD-Scan), a new method for early outbreak detection and characterization using multivariate case data from individuals in a population. MD-Scan extends previous work on multivariate event detection by identifying the characteristics of the affected subpopulation, and enables more timely and accurate detection while maintaining computational tractability