Analysis of patients’ incident reports within healthcare using learning algorithms

Purpose and goal

** Denna text är maskinöversatt ** The project has carried out an initial study of how a decision support system based on machine learning can be used to make it easier for staff at the patient board to categorize the type of complaints that are received. The result indicates that learning algorithms are suitable for predicting problem types. With the help of the model, and the method, one can also extract information from the application texts to find new insights.

Expected results and effects

** Denna text är maskinöversatt ** In addition to the initial evaluation of the clustering and the prediction of problem areas, it was also investigated how the text in the notifications was designed for each problem area. Among other things, the words that affect the classification towards each problem area were analyzed, see Appendix 1 for an example of a correct classification. Finally, the clusters were transformed into two dimensions for visualization and to enable the discovery of new connections between the complaints, see Appendix 2.

Planned approach and implementation

The project results show the difficulty of clustering patients’ complaints with limited amount of data. However, accuracy is likely to increase with more access to data. We therefore chose to investigate the use of classification algorithms instead. Further, as the data-driven analyzes resulted in increased insight into the complaints, so a qualitative follow-up study is planned. In conclusion, the collaboration between the organizations turned out well and continued collaboration is planned.