Discovering specific cascades in critical care transfer networks

Most Americans will need the services of Intensive Care Units (ICUs) at some point during their lives. There are wide variations between hospitals in the outcome of critical care and, as a result, thousands of patients who die each year in ICUs may have survived if they were at the appropriate hospital. A policy agenda---including an IOM report---calls for effectively transferring patients to more capable hospitals to improve outcomes. But there appear to be substantial inefficiencies in the existing system. In particular, patients recurrently transfer to secondary hospitals rather than to a most-preferred option. Analyzing critical care transfer data across nearly 5,000 hospitals over 10 year in Medicare, we present evidence that these transfers to secondary hospitals repeatedly cascade across multiple transfers, and that specific "hotspot" hospitals appear to be triggers of such cascades. We present data mining schemes to discover inefficient cascades of transfers in this dataset. We also present methods to determine the statistical significance of these discovered cascades. We examine the exemplar case of Michigan, suggesting a possible application to create alerts when multiple, significant cascades occur.