CorIL

Project leader:

• Lars Asker, Associate Professor, DSV, SU

Researchers

• Isak Samsten, Postdoctoral Researcher, DSV, SU
• Panagiotis Papapetrou, Professor, DSV, SU

Partners

• Stockholm County Council

Project period: 2017-01-01 to 2019-12-31

Funding: Stockholm County Council and Stockholm University

Budget: Approx 9M SEK

Description

Proper medical treatment and medication can substantially improve the conditions for many heart failure patients and reduce risk of premature death. However, few county councils in Sweden reach the target levels of basic medication. A better understanding of the underlying reasons for this will enable a more efficient and effective healthcare. The main purpose of this project is to analyze healthcare data in the GVR/VAL data warehouse to i) identify differences in treatment among groups of heart failure patients, ii) find factors that affect the effectiveness of treatment, and iii) provide support for improving the treatment of heart failure patients. The analysis will be undertaken by applying and developing state-of-the-art machine learning techniques for finding patterns in sequential data, clustering and predictive modeling, including survival analysis. The success of the project will be measured by the extent to which the findings allow for a better understanding of differences in treatments as well as factors for the effect of treatment, and improved practices and guidelines.

The main Work Packages (WPs) of the project are:

• WP1: Application of pattern mining, clustering and predictive modeling:
  • Sequential pattern mining on clinical units, diagnosis codes and prescribed drugs
  • Development of distance measures for clustering
  • Predicting readmission and death rate using random (survival) forests
• WP2: Combination and evaluation of derived patterns, clusters and models:
  • Employing pattern mining on combined sequential data
  • Generating clusters using discovered patterns
  • Using patterns and clusters as features for predictive modeling
• WP3: Exploitation and dissemination of results:
  • Forming recommendations based on discovered patterns
  • Using prototypical patients as a basis for refining guidelines
  • Identifying and communicating risk factors for readmission and death

People

Lars Asker, Associate Professor

representation learning, healthcare applications

Isak Samsten, Senior Lecturer

explainability, temporal data mining, fintech

Panagiotis Papapetrou, Professor

sequential and temporal data mining, explainability, healthcare applications