Early identification of complications after surgery
Our Data Science project on Early Identification of complications after surgery was one of the toughest challenges in the Data Science course, as it was an active ongoing research in the University of Twente. That is the reason we cannot share the dataset we have used in the project, as we have a NDA with the University and all the parties involved.
Our paper provides a study of the MoViSign (Mobile Vital Sign tracking in high risk surgical ward patients) data and gives insights on methods that might be used to detect complications at an early phase by analysing subject characteristics, abnormalities in vital signs and complication events. Our paper also binds new findings based on Machine Learning, predicting the occurrence of complications by having a set of sensor vital readings as parameters.
The purpose of this paper is to develop new methods that will better evaluate the medical state of patients in surgical care and to early identify signs of abnormalities.
Our research questions:
- How does the complications depend on subjects’ characteristics?
- When can abnormalities be noticed in cases of complications?
- What are the relations between complications and vital signs?
Our key findings:
- The key findings included the height, weight and the age groups which played a role in the development of complications across different groups of patients.
- We were also able to relate the time when abnormalities could be seen after or before a complication is detected.
- Our Machine Learning model provided a standardized classification method which can be used in place of MEWS score to decide whether a patient is at risk or not. We received new threshold values from the Decision Tree Model which we further used to classify the risk and non-risk set. The results show that our model can classify the data better than the MEWS score and it can be used to build new threshold for monitoring the patients. Our Machine Learning model also gave us insights into the most important sensor vitals that need to be monitored (Heart Rate and Respiration Rate).
Conclusion: We can conclude that we have accomplished our research goals by studying the subject characteristics with the complications, followed by the time frame of when abnormalities can be noticed in cases of complications and what are the relations between the complications and the vital signs of the patients.
Acknowledgements: We acknowledge the use of the data from the MoViSign study (Mobile Vital Sign tracking in high risk surgical ward patients). The MoViSign study was conducted in the ZGT Hospital in 2018-2019, in collaboration with the University of Twente.