Mistry, Pratik. Goal realization framework to optimize medical workflow model and generate plan libraries. Retrieved from https://doi.org/doi:10.7282/t3-rtcy-rg97
DescriptionThe number of patient deaths due to medical errors has increased in the past decade due to either human errors or errors in computer-aided decision support systems. These systems are often modeled using an expert medical algorithm and the plans of action based on the experts' theoretical knowledge and medical experience. But these approaches are prone to errors as the algorithm and the plans might have errors of omission and commission. Hence, there is a need to validate the expert algorithm and determine the correct plans which motivated this study. I have developed the goal realization framework for a medical goal named Establish IV Access (IV Goal) that optimized the IV realization algorithm based on the ground truth, i.e., medical process logs. I have built another framework that could detect the ground truth from the dataset where the IV goal was truly met to determine the medical algorithm’s accuracy, i.e., realization framework's accuracy. After multiple experiments, analysis, conclusions, and revisions of the algorithm, the expert model (algorithm) was optimized, and thus the goal realization framework achieved an accuracy of 100%. Other statistical analyses like age distribution of patients getting single v/s multiple IV attempts and traces following different branches (paths of the workflow) helped experts understand the standard medical procedure followed in the real-world for performing the IV goal. Using the optimized algorithm, we had determined eight concise and granular plan libraries that could be sent out as part of recommendations in the trauma resuscitation process while performing the IV goal.
The optimized algorithm (expert workflow model) is often challenging to interpret. Hence, workflow discovery algorithms or process mining tools have been developed that generate interpretable workflows or process maps. The data used for workflows is often synthetic or formed manually by the experts and motivated to build a third framework that could generate the dataset for a medical goal using the ground truth, i.e., medical process logs. The workflow developed using ideas from previous work and workflow discovery algorithms like PIMA is compared with the optimized expert algorithm to determine its fitness based on paths followed by traces and the statistical results generated from the goal realization framework.