Algorithmic Reconstruction of High-Fidelity Synthetic SIRD Data

During infectious disease outbreaks, epidemiologists are faced with problems similar to those of state estimation. Often, only population-level measures of the outbreak are available, such as infection counts per week or test positivity rates. However, these aggregate statistics yield little insight into the individual-level dynamics of an epidemic. Further, empirical data does not directly measure transmission network structure and producing this information via contact tracing is extremely resource-intensive and scales poorly. In this poster, I present an algorithm that produces synthetic, individual-level disease state measurements that both obey the mechanics of an SIRD model and, in aggregate, match empirical infection count data. Inspired by the structure of Kalman Filtering, this approach iteratively simulates an SIRD model and successively corrects to account for measurements. As a result, the algorithm can be considered as sampling the space of candidate datasets generated by SIRD mechanics that can produce the observed data. This synthetic dataset opens several avenues for analysis, including transmission network inference, sensitivity analysis, and public health policy simulation.