Extracting clinical care pathways correlated with outcomes

Abstract

Systems and methods for data analysis include constructing patient traces as a set of medical events for each patient of a patient population, the patient population being segmented based on patient outcomes. Medical events in one or more of the patient traces are reduced to provide processed patient traces. The processed patient traces are clustered to identify a cluster of patient traces. A process model is mined, using a processor, representing an aggregation of treatment pathways in the patient traces from the cluster. Patterns from patient traces are identified that are discriminative of patient outcomes. At least one of the patterns is represented with respect to the process model to identify treatment pathways correlated with the patient outcomes.

What is claimed is: 
     
         1 . A computer readable storage medium comprising a computer readable program for data analysis, wherein the computer readable program when executed on a computer causes the computer to perform the steps of:
 constructing patient traces as a set of medical events for each patient of a patient population, the patient population being segmented based on patient outcomes;   reducing medical events in one or more of the patient traces to provide processed patient traces;   clustering the processed patient traces to identify a cluster of patient traces;   mining a process model, using a processor, representing an aggregation of treatment pathways in the patient traces from the cluster;   identifying patterns from patient traces that are discriminative of patient outcomes; and   representing at least one of the patterns with respect to the process model to identify treatment pathways correlated with the patient outcomes.   
     
     
         2 . A system for data analysis, comprising:
 a medical records database configured to construct patient traces stored on a computer readable storage medium as a set of medical events for each patient of a patient population, the patient population being segmented based on patient outcomes;   a trace preprocess module configured to reduce medical events in one or more of the patient traces to provide processed patient traces;   a cluster module configured to cluster the processed patient traces to identify a cluster of patient traces;   a pathway extraction module configured to mine a process model representing an aggregation of treatment pathways in the patient traces from the cluster;   a pattern extraction module configured to identify patterns from patient traces that are discriminative of patient outcomes; and   a visual interface configured to represent at least one of the patterns with respect to the process model to identify treatment pathways correlated with the patient outcomes.   
     
     
         3 . The system as recited in  claim 2 , wherein the visual interface is further configured to display the at least one of the patterns overlaid on the process model. 
     
     
         4 . The system as recited in  claim 2 , wherein the visual interface is further configured to represent the at least one of the patterns with the process model based on the patient outcomes. 
     
     
         5 . The system as recited in  claim 2 , wherein the visual interface is further configured to highlight nodes of the at least one of the patterns and edges between the nodes. 
     
     
         6 . The system as recited in  claim 2 , wherein the cluster of patient traces includes at least one of a largest cluster of patient traces and a cluster having a number of patient traces meeting or exceeding a threshold number of patient traces. 
     
     
         7 . The system as recited in  claim 2 , wherein the cluster module is further configured to represent each patient trace of the processed patient traces as a string and compute a string edit distance between two patient traces of the processed patient traces to determine similarity between the two patient traces. 
     
     
         8 . The system as recited in  claim 2 , wherein the mining module is further configured to add a start event and an end event to each of the patient traces, wherein the start event has a timestamp earlier than all other events in its patient trace and the end event has a timestamp later than all other events in its patient trace. 
     
     
         9 . The system as recited in  claim 2 , wherein the mining module is further configured to define a dependency between repeating event node pairs according to a frequency of each direction of the dependencies from each of the repeating event node pairs. 
     
     
         10 . The system as recited in  claim 2 , wherein the mining module is further configured to represent medical events in the process model according to a frequency of appearance of medical events in the cluster of patient traces compared to a threshold.