Modeling and clustering disease progression for correlation with genetic and demographic factors

Robert Kingan, PhD. , Senior Software Engineer
 

We consider the problem of modeling disease progression from historical clinical databases, including the issue of what mathematical model types are most appropriate for describing disease progression. The ultimate objective of this modeling is to provide a basis for stratifying patients into groups with clearly distinguishable prognoses or suitability for different treatment strategies, on the basis of clinical variables measured over time. Correlation with genetic and demographic factors are then possible, to strengthen associations and predict disease susceptibilities.  To account for the underlying physiology, models must describe the temporal behavior of several biomarkers and the relationships among them, in a way that has a clear clinical interpretation. Practically important aspects of this problem include the complicated, mixed structure of clinical databases, strong censoring along the time axis, and the prevalence of missing data and other anomalies. We will discuss these issues and describe early results from two different large-scale databases tracking transplant patients over time. In both cases, models of long-term survival were developed using quantitative clinical variables measured during a fixed-length post-transplant time period, and subsets of patients were identified who were at a particularly high risk for loss of the transplanted organ or death.