Skip to main content

Competing Risks and Multi-State Models for Time-Dependent Outcomes

Dr. Guy Brock / University of Louisville

Competing time-dependent outcomes are frequently encountered when analyzing hospital epidemiology data.  Examples include hospital length of stay (LOS) and time for a patient to reach clinical stability (TCS) in the presence of in-hospital mortality.  These outcomes have become increasingly important when evaluating treatment efficacy, patient management, quality of care, and hospital costs associated with treating diseases like community acquired pneumonia (CAP).  Sub-optimal approaches to analyzing these competing events are commonly used, including restricting analysis to those patients who lived and/or assigning individuals who die the longest recorded LOS or TCS.  Drawbacks with both of these approaches will be demonstrated, and compared with the advocated approach of treating in-hospital mortality as a competing risk.  The competing risks model can also be extended to a more general multi-state model, which allows transitions between multiple competing and/or transitional states.  We present an R package, msSurv, which calculates nonparametric estimates of the transition probability matrix, marginal state occupation probabilities, the normalized and non-normalized state entry and exit time distributions, and marginal integrated transition hazards for a general multistate system allowing for left-truncation and right censoring.  Various applications of both competing risks and multi-state models with hospital epidemiology data will be discussed.

Open /*deleted href=#openmobile*/